gait analysis research paper

IEEE Account

Change Username/Password
Update Address

Purchase Details

Payment Options
Order History
View Purchased Documents

Profile Information

Communications Preferences
Profession and Education
Technical Interests
US & Canada: +1 800 678 4333
Worldwide: +1 732 981 0060
Contact & Support
About IEEE Xplore
Accessibility
Terms of Use
Nondiscrimination Policy
Privacy & Opting Out of Cookies

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. © Copyright 2024 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

Information

Author Services

Initiatives

You are accessing a machine-readable page. In order to be human-readable, please install an RSS reader.

All articles published by MDPI are made immediately available worldwide under an open access license. No special permission is required to reuse all or part of the article published by MDPI, including figures and tables. For articles published under an open access Creative Common CC BY license, any part of the article may be reused without permission provided that the original article is clearly cited. For more information, please refer to https://www.mdpi.com/openaccess .

Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature Paper should be a substantial original Article that involves several techniques or approaches, provides an outlook for future research directions and describes possible research applications.

Feature papers are submitted upon individual invitation or recommendation by the scientific editors and must receive positive feedback from the reviewers.

Editor’s Choice articles are based on recommendations by the scientific editors of MDPI journals from around the world. Editors select a small number of articles recently published in the journal that they believe will be particularly interesting to readers, or important in the respective research area. The aim is to provide a snapshot of some of the most exciting work published in the various research areas of the journal.

Original Submission Date Received: .

Active Journals
Find a Journal
Proceedings Series
For Authors
For Reviewers
For Editors
For Librarians
For Publishers
For Societies
For Conference Organizers
Open Access Policy
Institutional Open Access Program
Special Issues Guidelines
Editorial Process
Research and Publication Ethics
Article Processing Charges
Testimonials
Preprints.org
SciProfiles
Encyclopedia

Article Menu

Subscribe SciFeed
Recommended Articles
PubMed/Medline
Google Scholar
on Google Scholar
Table of Contents

Find support for a specific problem in the support section of our website.

Please let us know what you think of our products and services.

Visit our dedicated information section to learn more about MDPI.

JSmol Viewer

Gait analysis in neurorehabilitation: from research to clinical practice.

Graphical Abstract

1. Introduction

2. search strategy, 3. neurodegenerative disorders, 3.1. parkinson’s disease, 3.2. multiple sclerosis, 3.3. cerebellar ataxia, 4. acquired brain injury, 4.1. stroke, 4.2. traumatic brain injury, 5. discussion, 5.1. clinical considerations about gait and postural dysfunctions, 5.2. clinical implications of nws, 5.3. clinical implications of ws, 5.4. future directions of gait analysis, 6. conclusions, author contributions, institutional review board statement, informed consent statement, data availability statement, conflicts of interest.


“Parkinson’s disease” OR “multiple sclerosis” OR “cerebellar ataxia” OR “ataxia” OR “neurodegenerative disorders” OR “acquired brain injury” OR “stroke” OR “traumatic brain injury”	“wearable sensors” OR “gait platforms” OR “non-wearable sensors” OR “instrumented gait analysis” OR “objective gait evaluation” OR “inertial measurement units” OR “motion capture systems” OR “mobile application” OR “artificial intelligence”

Salchow-Hömmen, C.; Skrobot, M.; Jochner, M.C.E.; Schauer, T.; Kühn, A.A.; Wenger, N. Review—Emerging Portable Technologies for Gait Analysis in Neurological Disorders. Front. Hum. Neurosci. 2022 , 16 , 768575. [ Google Scholar ] [ CrossRef ]
Li, S.; Francisco, G.E.; Zhou, P. Post-stroke Hemiplegic Gait: New Perspective and Insights. Front. Physiol. 2018 , 9 , 1021. [ Google Scholar ] [ CrossRef ] [ Green Version ]
Loyd, B.J.; Dibble, L.E.; Weightman, M.M.; Pelo, R.; Hoppes, C.W.; Lester, M.; King, L.A.; Fino, P.C. Volitional Head Movement Deficits and Alterations in Gait Speed Following Mild Traumatic Brain Injury. J. Head Trauma Rehabil. 2022 , 38 , E223–E232. [ Google Scholar ] [ CrossRef ]
di Biase, L.; Di Santo, A.; Caminiti, M.L.; De Liso, A.; Shah, S.A.; Ricci, L.; Di Lazzaro, V. Gait Analysis in Parkinson’s Disease: An Overview of the Most Accurate Markers for Diagnosis and Symptoms Monitoring. Sensors 2020 , 20 , 3529. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Coca-Tapia, M.; Cuesta-Gómez, A.; Molina-Rueda, F.; Carratalá-Tejada, M. Gait Pattern in People with Multiple Sclerosis: A Systematic Review. Diagnostics 2021 , 11 , 584. [ Google Scholar ] [ CrossRef ]
Benedetti, M.G.; Piperno, R.; Simoncini, L.; Bonato, P.; Tonini, A.; Giannini, S. Gait abnormalities in minimally impaired multiple sclerosis patients. Mult. Scler. J. 1999 , 5 , 363–368. [ Google Scholar ] [ CrossRef ]
Tunca, C.; Pehlivan, N.; Ak, N.; Arnrich, B.; Salur, G.; Ersoy, C. Inertial Sensor-Based Robust Gait Analysis in Non-Hospital Settings for Neurological Disorders. Sensors 2017 , 17 , 825. [ Google Scholar ] [ CrossRef ] [ Green Version ]
Das, R.; Paul, S.; Mourya, G.K.; Kumar, N.; Hussain, M. Recent Trends and Practices toward Assessment and Rehabilitation of Neurodegenerative Disorders: Insights from Human Gait. Front. Neurosci. 2022 , 16 , 859298. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Hulleck, A.A.; Mohan, D.M.; Abdallah, N.; El Rich, M.; Khalaf, K. Present and future of gait assessment in clinical practice: Towards the application of novel trends and technologies. Front. Med. Technol. 2022 , 4 , 901331. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Bonanno, M.; De Luca, R.; Torregrossa, W.; Tonin, P.; Calabrò, R.S. Moving toward Appropriate Motor Assessment Tools in People Affected by Severe Acquired Brain Injury: A Scoping Review with Clinical Advices. Healthcare 2022 , 10 , 1115. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Bhidayasiri, R.; Martinez-Martin, P. Clinical Assessments in Parkinson’s Disease: Scales and Monitoring. Int. Rev. Neurobiol. 2017 , 132 , 129–182. [ Google Scholar ] [ CrossRef ]
Wang, C.; Ruiz, A.; Mao-Draayer, Y. Assessment and Treatment Strategies for a Multiple Sclerosis Relapse. J. Immunol. Clin. Res. 2016 , 5 , 1032. [ Google Scholar ] [ PubMed ]
Power, L.; Pathirana, P.; Horne, M.; Milne, S.; Marriott, A.; Szmulewicz, D.J. Instrumented Objective Clinical Examination of Cerebellar Ataxia: The Upper and Lower Limb—A Review. Cerebellum 2022 , 21 , 145–158. [ Google Scholar ] [ CrossRef ]
Muro-de-la-Herran, A.; Garcia-Zapirain, B.; Mendez-Zorrilla, A. Gait analysis methods: An overview of wearable and non-wearable systems, highlighting clinical applications. Sensors 2014 , 14 , 3362–3394. [ Google Scholar ] [ CrossRef ] [ Green Version ]
Schniepp, R.; Huppert, A.; Decker, J.; Schenkel, F.; Schlick, C.; Rasoul, A.; Dieterich, M.; Brandt, T.; Jahn, K.; Wuehr, M. Fall prediction in neurological gait disorders: Differential contributions from clinical assessment, gait analysis, and daily-life mobility monitoring. J. Neurol. 2021 , 268 , 3421–3434. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Pieruccini-Faria, F.; Black, S.E.; Masellis, M.; Smith, E.E.; Almeida, Q.J.; Li, K.Z.H.; Bherer, L.; Camicioli, R.; Montero-Odasso, M. Gait variability across neurodegenerative and cognitive disorders: Results from the Canadian Consortium of Neurodegeneration in Aging (CCNA) and the Gait and Brain Study. Alzheimer’s Dement. 2021 , 17 , 1317–1328. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Cicirelli, G.; Impedovo, D.; Dentamaro, V.; Marani, R.; Pirlo, G.; D’Orazio, T.R. Human Gait Analysis in Neurodegenerative Diseases: A Review. IEEE J. Biomed. Health Inform. 2022 , 26 , 229–242. [ Google Scholar ] [ CrossRef ]
Marković, V.; Stanković, I.; Radovanović, S.; Petrović, I.; Lukić, M.J.; Mišković, N.D.; Svetel, M.; Kostić, V. Gait alterations in Parkinson’s disease at the stage of hemiparkinsonism—A longitudinal study. PLoS ONE 2022 , 17 , e0269886. [ Google Scholar ] [ CrossRef ]
Pistacchi, M.; Gioulis, M.; Sanson, F.; De Giovannini, E.; Filippi, G.; Rossetto, F.; Zambito Marsala, S. Gait analysis and clinical correlations in early Parkinson’s disease. Funct. Neurol. 2017 , 32 , 28–34. [ Google Scholar ] [ CrossRef ]
Del Din, S.; Godfrey, A.; Rochester, L. Validation of an Accelerometer to Quantify a Comprehensive Battery of Gait Characteristics in Healthy Older Adults and Parkinson’s Disease: Toward Clinical and at Home Use. IEEE J. Biomed. Health Inform. 2015 , 20 , 838–847. [ Google Scholar ] [ CrossRef ] [ Green Version ]
Potter, M.V.; Cain, S.M.; Ojeda, L.V.; Gurchiek, R.D.; McGinnis, R.S.; Perkins, N.C. Evaluation of Error-State Kalman Filter Method for Estimating Human Lower-Limb Kinematics during Various Walking Gaits. Sensors 2022 , 22 , 8398. [ Google Scholar ] [ CrossRef ]
Cameron, M.H.; Nilsagard, Y. Balance, gait, and falls in multiple sclerosis. Handb. Clin. Neurol. 2018 , 159 , 237–250. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Kempen, J.C.; Doorenbosch, C.A.; Knol, D.L.; de Groot, V.; Beckerman, H. Newly Identified Gait Patterns in Patients with Multiple Sclerosis May Be Related to Push-off Quality. Phys. Ther. 2016 , 96 , 1744–1752. [ Google Scholar ] [ CrossRef ] [ PubMed ] [ Green Version ]
Severini, G.; Manca, M.; Ferraresi, G.; Caniatti, L.M.; Cosma, M.; Baldasso, F.; Straudi, S.; Morelli, M.; Basaglia, N. Evaluation of Clinical Gait Analysis parameters in patients affected by Multiple Sclerosis: Analysis of kinematics. Clin. Biomech. 2017 , 45 , 1–8. [ Google Scholar ] [ CrossRef ]
Sehle, A.; Mündermann, A.; Starrost, K.; Sailer, S.; Becher, I.; Dettmers, C.; Vieten, M. Objective assessment of motor fatigue in multiple sclerosis using kinematic gait analysis: A pilot study. J. Neuroeng. Rehabil. 2011 , 8 , 59. [ Google Scholar ] [ CrossRef ] [ Green Version ]
Hu, W.; Combden, O.; Jiang, X.; Buragadda, S.; Newell, C.J.; Williams, M.C.; Critch, A.L.; Ploughman, M. Machine learning corroborates subjective ratings of walking and balance difficulty in multiple sclerosis. Front. Artif. Intell. 2022 , 5 , 952312. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Kim, S.M.; Kim, D.H.; Yang, Y.; Ha, S.W.; Han, J.H. Gait Patterns in Parkinson’s Disease with or without Cognitive Impairment. Dement. Neurocogn. Disord. 2018 , 17 , 57–65. [ Google Scholar ] [ CrossRef ] [ PubMed ]
di Biase, L.; Raiano, L.; Caminiti, M.L.; Pecoraro, P.M.; Di Lazzaro, V. Parkinson’s Disease Wearable Gait Analysis: Kinematic and Dynamic Markers for Diagnosis. Sensors 2022 , 22 , 8773. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Rehman, R.Z.U.; Buckley, C.; Mico-Amigo, M.E.; Kirk, C.; Dunne-Willows, M.; Mazza, C.; Shi, J.Q.; Alcock, L.; Rochester, L.; Del Din, S. Accelerometry-Based Digital Gait Characteristics for Classification of Parkinson’s Disease: What Counts? IEEE Open J. Eng. Med. Biol. 2020 , 1 , 65–73. [ Google Scholar ] [ CrossRef ]
Jakob, V.; Küderle, A.; Kluge, F.; Klucken, J.; Eskofier, B.M.; Winkler, J.; Winterholler, M.; Gassner, H. Validation of a Sensor-Based Gait Analysis System with a Gold-Standard Motion Capture System in Patients with Parkinson’s Disease. Sensors 2021 , 21 , 7680. [ Google Scholar ] [ CrossRef ]
Alberto, S.; Cabral, S.; Proença, J.; Pona-Ferreira, F.; Leitão, M.; Bouça-Machado, R.; Kauppila, L.A.; Veloso, A.P.; Costa, R.M.; Ferreira, J.J.; et al. Validation of quantitative gait analysis systems for Parkinson’s disease for use in supervised and unsupervised environments. BMC Neurol. 2021 , 21 , 331. [ Google Scholar ] [ CrossRef ]
Liu, R.; Wang, Z.; Qiu, S.; Zhao, H.Y.; Wang, C.; Shi, X.; Lin, F. A Wearable Gait Analysis and Recognition Method for Parkinson’s Disease Based on Error State Kalman Filter. IEEE J. Biomed. Health Inform. 2022 , 26 , 4165–4175. [ Google Scholar ] [ CrossRef ]
Lizama, L.E.C.; Khan, F.; Lee, P.V.S.; Galea, M.P. The use of laboratory gait analysis for understanding gait deterioration in people with multiple sclerosis. Mult. Scler. J. 2016 , 22 , 1768–1776. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Buckley, E.; Mazzà, C.; McNeill, A. A systematic review of the gait characteristics associated with Cerebellar Ataxia. Gait Posture 2018 , 60 , 154–163. [ Google Scholar ] [ CrossRef ]
Serrao, M.; Pierelli, F.; Ranavolo, A.; Draicchio, F.; Conte, C.; Don, R.; Di Fabio, R.; LeRose, M.; Padua, L.; Sandrini, G.; et al. Gait Pattern in Inherited Cerebellar Ataxias. Cerebellum 2011 , 11 , 194–211. [ Google Scholar ] [ CrossRef ]
Matsushima, A.; Yoshida, K.; Genno, H.; Murata, A.; Matsuzawa, S.; Nakamura, K.; Nakamura, A.; Ikeda, S.-I. Clinical assessment of standing and gait in ataxic patients using a triaxial accelerometer. Cerebellum Ataxias 2015 , 2 , 9. [ Google Scholar ] [ CrossRef ] [ PubMed ] [ Green Version ]
Zhou, H.; Nguyen, H.; Enriquez, A.; Morsy, L.; Curtis, M.; Piser, T.; Kenney, C.; Stephen, C.D.; Gupta, A.S.; Schmahmann, J.D.; et al. Assessment of gait and balance impairment in people with spinocerebellar ataxia using wearable sensors. Neurol. Sci. 2021 , 43 , 2589–2599. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Ilg, W.; Seemann, J.; Giese, M.; Traschütz, A.; Schöls, L.; Timmann, D.; Synofzik, M. Real-life gait assessment in degenerative cerebellar ataxia: Toward ecologically valid biomarkers. Neurology 2020 , 95 , e1199–e1210. [ Google Scholar ] [ CrossRef ]
Ferrarello, F.; Bianchi, V.A.M.; Baccini, M.; Rubbieri, G.; Mossello, E.; Cavallini, M.C.; Marchionni, N.; Di Bari, M. Tools for Observational Gait Analysis in Patients With Stroke: A Systematic Review. Phys. Ther. 2013 , 93 , 1673–1685. [ Google Scholar ] [ CrossRef ] [ Green Version ]
Gupta, A.; Taly, A.B. Post-stroke Gait Analysis in Rehabilitation Set-up: Observational or Instrumental! Neurol. India 2019 , 67 , 1041–1042. [ Google Scholar ] [ CrossRef ]
Mohan, D.M.; Khandoker, A.H.; Wasti, S.A.; Alali, S.I.I.I.; Jelinek, H.F.; Khalaf, K. Assessment Methods of Post-stroke Gait: A Scoping Review of Technology-Driven Approaches to Gait Characterization and Analysis. Front. Neurol. 2021 , 12 , 650024. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Latorre, J.; Colomer, C.; Alcañiz, M.; Llorens, R. Gait analysis with the Kinect v2: Normative study with healthy individuals and comprehensive study of its sensitivity, validity, and reliability in individuals with stroke. J. Neuroeng. Rehabil. 2019 , 16 , 97. [ Google Scholar ] [ CrossRef ] [ PubMed ] [ Green Version ]
Wang, Y.; Tang, R.; Wang, H.; Yu, X.; Li, Y.; Wang, C.; Wang, L.; Qie, S. The Validity and Reliability of a New Intelligent Three-Dimensional Gait Analysis System in Healthy Subjects and Patients with Post-Stroke. Sensors 2022 , 22 , 9425. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Balaban, B.; Tok, F. Gait Disturbances in Patients with Stroke. PM&R 2014 , 6 , 635–642. [ Google Scholar ] [ CrossRef ]
Patterson, K.K.; Gage, W.H.; Brooks, D.; Black, S.E.; McIlroy, W.E. Evaluation of gait symmetry after stroke: A comparison of current methods and recommendations for standardization. Gait Posture 2010 , 31 , 241–246. [ Google Scholar ] [ CrossRef ]
Laudanski, A.; Brouwer, B.; Li, Q. Measurement of Lower Limb Joint Kinematics using Inertial Sensors During Stair Ascent and Descent in Healthy Older Adults and Stroke Survivors. J. Health Eng. 2013 , 4 , 555–576. [ Google Scholar ] [ CrossRef ] [ Green Version ]
Dever, A.; Powell, D.; Graham, L.; Mason, R.; Das, J.; Marshall, S.J.; Vitorio, R.; Godfrey, A.; Stuart, S. Gait Impairment in Traumatic Brain Injury: A Systematic Review. Sensors 2022 , 22 , 1480. [ Google Scholar ] [ CrossRef ]
Reidy, J.; Mobbs, R.; Kim, J.; Brown, E.; Mobbs, R. Clinical gait characteristics in the early post-concussion phase: A systematic review. J. Clin. Neurosci. 2023 , 107 , 184–191. [ Google Scholar ] [ CrossRef ]
Martini, D.N.; Parrington, L.; Stuart, S.; Fino, P.C.; King, L.A. Gait Performance in People with Symptomatic, Chronic Mild Traumatic Brain Injury. J. Neurotrauma 2021 , 38 , 218–224. [ Google Scholar ] [ CrossRef ]
Belluscio, V.; Bergamini, E.; Tramontano, M.; Bustos, A.O.; Allevi, G.; Formisano, R.; Vannozzi, G.; Buzzi, M.G. Gait Quality Assessment in Survivors from Severe Traumatic Brain Injury: An Instrumented Approach Based on Inertial Sensors. Sensors 2019 , 19 , 5315. [ Google Scholar ] [ CrossRef ] [ Green Version ]
Pitt, W.; Chen, S.-H.; Chou, L.-S. Using IMU-based kinematic markers to monitor dual-task gait balance control recovery in acutely concussed individuals. Clin. Biomech. 2020 , 80 , 105145. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Celik, Y.; Stuart, S.; Woo, W.; Godfrey, A. Gait analysis in neurological populations: Progression in the use of wearables. Med. Eng. Phys. 2020 , 87 , 9–29. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Stolze, H.; Klebe, S.; Baecker, C.; Zechlin, C.; Friege, L.; Pohle, S.; Deuschl, G. Prevalence of gait disorders in hospitalized neurological patients. Mov. Disord. 2004 , 20 , 89–94. [ Google Scholar ] [ CrossRef ]
Ge, H.-L.; Chen, X.-Y.; Lin, Y.-X.; Ge, T.-J.; Yu, L.-H.; Lin, Z.-Y.; Wu, X.-Y.; Kang, D.-Z.; Ding, C.-Y. The prevalence of freezing of gait in Parkinson’s disease and in patients with different disease durations and severities. Chin. Neurosurg. J. 2020 , 6 , 17. [ Google Scholar ] [ CrossRef ]
Tasseel-Ponche, S.; Delafontaine, A.; Godefroy, O.; Yelnik, A.P.; Doutrellot, P.-L.; Duchossoy, C.; Hyra, M.; Sader, T.; Diouf, M. Walking speed at the acute and subacute stroke stage: A descriptive meta-analysis. Front. Neurol. 2022 , 13 , 989622. [ Google Scholar ] [ CrossRef ]
Moon, Y.; Sung, J.; An, R.; Hernandez, M.E.; Sosnoff, J.J. Gait variability in people with neurological disorders: A systematic review and meta-analysis. Hum. Mov. Sci. 2016 , 47 , 197–208. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Patel, P.; Enzastiga, D.; Casamento-Moran, A.; Christou, E.A.; Lodha, N. Increased temporal stride variability contributes to impaired gait coordination after stroke. Sci. Rep. 2022 , 12 , 12679. [ Google Scholar ] [ CrossRef ]
Niechwiej-Szwedo, E.; Inness, E.; Howe, J.; Jaglal, S.; McIlroy, W.; Verrier, M. Changes in gait variability during different challenges to mobility in patients with traumatic brain injury. Gait Posture 2007 , 25 , 70–77. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Farrell, J.W., 3rd; Merkas, J.; Pilutti, L.A. The Effect of Exercise Training on Gait, Balance, and Physical Fitness Asymmetries in Persons with Chronic Neurological Conditions: A Systematic Review of Randomized Controlled Trials. Front. Physiol. 2020 , 11 , 585765. [ Google Scholar ] [ CrossRef ]
Bishnoi, A.; Shankar, M.; Lee, R.; Hu, Y.; Hernandez, M.E. Effects of Therapeutic Intervention on Spatiotemporal Gait Parameters in Adults With Neurologic Disorder: Systematic Review and Meta-analysis. Arch. Phys. Med. Rehabil. 2023 , 104 , 451–474. [ Google Scholar ] [ CrossRef ]
Keshner, E.A.; Lamontagne, A. The Untapped Potential of Virtual Reality in Rehabilitation of Balance and Gait in Neurological Disorders. Front. Virtual Real. 2021 , 2 , 641650. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Peppe, A.; Chiavalon, C.; Pasqualetti, P.; Crovato, D.; Caltagirone, C. Does gait analysis quantify motor rehabilitation efficacy in Parkinson’s disease patients? Gait Posture 2007 , 26 , 452–462. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Guner, S.; Inanici, F. Yoga therapy and ambulatory multiple sclerosis Assessment of gait analysis parameters, fatigue and balance. J. Bodyw. Mov. Ther. 2015 , 19 , 72–81. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Collins, J.-D.; Markham, A.; Service, K.; Reini, S.; Wolf, E.; Sessoms, P. A systematic literature review of the use and effectiveness of the Computer Assisted Rehabilitation Environment for research and rehabilitation as it relates to the wounded warrior. Work 2015 , 50 , 121–129. [ Google Scholar ] [ CrossRef ] [ PubMed ] [ Green Version ]
Smith, A.J.; Lemaire, E.D. Temporal-spatial gait parameter models of very slow walking. Gait Posture 2018 , 61 , 125–129. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Imoto, D.; Hirano, S.; Mukaino, M.; Saitoh, E.; Otaka, Y. A novel gait analysis system for detecting abnormal hemiparetic gait patterns during robot-assisted gait training: A criterion validity study among healthy adults. Front. Neurorobot. 2022 , 16 , 1047376. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Khera, P.; Kumar, N. Role of machine learning in gait analysis: A review. J. Med. Eng. Technol. 2020 , 44 , 441–467. [ Google Scholar ] [ CrossRef ]
Alfayeed, S.M.; Saini, B.S. Human Gait Analysis Using Machine Learning: A Review. In Proceedings of the 2021 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE), Dubai, United Arab Emirates, 17–18 March 2021; pp. 550–554. [ Google Scholar ]
Slemenšek, J.; Fister, I.; Geršak, J.; Bratina, B.; van Midden, V.M.; Pirtošek, Z.; Šafarič, R. Human Gait Activity Recognition Machine Learning Methods. Sensors 2023 , 23 , 745. [ Google Scholar ] [ CrossRef ]
Fricke, C.; Alizadeh, J.; Zakhary, N.; Woost, T.B.; Bogdan, M.; Classen, J. Evaluation of Three Machine Learning Algorithms for the Automatic Classification of EMG Patterns in Gait Disorders. Front. Neurol. 2021 , 12 , 666458. [ Google Scholar ] [ CrossRef ]
Liuzzi, P.; Carpinella, I.; Anastasi, D.; Gervasoni, E.; Lencioni, T.; Bertoni, R.; Carrozza, M.C.; Cattaneo, D.; Ferrarin, M.; Mannini, A. Machine learning based estimation of dynamic balance and gait adaptability in persons with neurological diseases using inertial sensors. Sci. Rep. 2023 , 13 , 8640. [ Google Scholar ] [ CrossRef ]
Rastegari, E.; Azizian, S.; Ali, H. Machine learning and similarity network approaches to support automatic classification of Parkinson’s diseases using accelerometer-based gait analysis. In Proceedings of the 52nd Annual Hawaii International Conference on System Sciences, Maui, HI, USA, 8–11 January 2019; pp. 4231–4242. [ Google Scholar ]
Harris, E.J.; Khoo, I.-H.; Demircan, E. A Survey of Human Gait-Based Artificial Intelligence Applications. Front. Robot. AI 2022 , 8 , 749274. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Luo, S.; Androwis, G.; Adamovich, S.; Su, H.; Nunez, E.; Zhou, X. Reinforcement Learning and Control of a Lower Extremity Exoskeleton for Squat Assistance. Front. Robot. AI 2021 , 8 , 702845. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Romijnders, R.; Warmerdam, E.; Hansen, C.; Schmidt, G.; Maetzler, W. A Deep Learning Approach for Gait Event Detection from a Single Shank-Worn IMU: Validation in Healthy and Neurological Cohorts. Sensors 2022 , 22 , 3859. [ Google Scholar ] [ CrossRef ] [ PubMed ]
Kaur, R.; Motl, R.W.; Sowers, R.; Hernandez, M.E. A Vision-Based Framework for Predicting Multiple Sclerosis and Parkinson’s Disease Gait Dysfunctions—A Deep Learning Approach. IEEE J. Biomed. Health Inform. 2023 , 27 , 190–201. [ Google Scholar ] [ CrossRef ]

Click here to enlarge figure

Reference n°	Gait Analysis System		Technology Description	Neurological Disorder	Clinical Implication
Reference n°	Non- Wearable Sensors	Wearable Sensors	Technology Description	Neurological Disorder	Clinical Implication
[ ]	X	X	Three-dimensional gait analysis in laboratory, including optometric system, a dynamometric platform, and ad hoc software.	PD with 1.5–2 H&Y stage	Reduced gait speed and step length, showing bilateral extra rotation of knee, ankle, and foot.
[ ]		X	Triaxial accelerometer-based device placed on the fifth lumbar vertebrae and a double-sided tape.	PD with 1–3 H&Y stage	NA
[ ]		X	Instrumented force-sensitive insole placed in patients’ shoes, with eight pressure-sensitive sensors.	PD with 2–3 H&Y stage	Stride-to-stride variability due to bradykinesia, loss of muscle synergies in the lower limb, and lack of rhythmicity.
[ ]	X	X	Motion-capture based gait analysis compared to mobile sensor (inertial sensors) gait analysis, which were integrated in the mid-sole of the athletic shoes.	PD with 1–4 H&Y stage	Reduced gait speed, stride time, and length; increased duration stance phase time accompanied by a synchronic decreasing duration of swing phase time.
[ ]	X	X	Gait assessment through an optoelectronic (48 retroflected markers), inertial, and a smartphone-based capture system.	PD with <3 H&Y stage	NA
[ ]		X	Wearable device compared to Opti Track system, using an error state Kalman filter algorithm.	PD	NA
[ ]	X		Stereophotogrammetric system (Vicon Motion Systems Ltd., Oxford, UK) and reflective markers to estimate joints’ angles.	MS with a score of ≤5–6	MS patients showed reduced gait speed, which correlated with a decrease in cadence, step length, and swing time, and an increase in stance time. Additionally, authors found an increased pelvic tilt, which negatively correlates with the 6MWT.
[ ]	X	X	Wireless AS200 system, comprising three line-scanning camera system and 11 active infrared markers attached on body’s patient, with a 2-mm accuracy.	MS with a mean score of 3.6 in EDSS	MS patients manifested changes in variability of movement gait patterns due to fatigue, altered motor coordination linked to additional activity of the antagonists, or insufficient strength produced by the agonists.
[ ]	X		Walkway sensor and machine learning (XGB) process to distinguish MS patients’ degree of severity based on their gait features.	MS with a mean score of 2.11 in EDSS	Step time and step width were considered as the most important variables to distinguish level of severity of MS subjects.
[ ]	X	X	SMART-E stereophotogrammetric system (BTS, Milan, Italy) with eight infrared cameras (for acquiring kinematic data). Sensorized pathway with 2 piezoelectric force platforms (for acquiring kinetic data), 22 retro-reflective spherical markers for lower-body segments, and 15 markers for the upper body, placed on specific anatomic sites.	Spino-CA autosomal dominant (type 1 and 2) and Friedreich’s ataxia as recessive ataxia	Loss of lower limbs control during gait and of ability to stabilize a walking strategy over time. CA patients definitively lack a stable gait control behavior since the cerebellum functions of motor behavior and developing new motor patterns are altered.
[ ]		X	Triaxial accelerometer.	Spino-CA with a mean score of 3.9 for stance and gait in SARA	Gait velocity, cadence, step length, step regularity, and step repeatability are strongly correlated with disease duration.
[ ]		X	Seven inertial sensors while performing two independent trials of gait and balance assessments.	CA	NA
[ ]		X	Three Opal inertial sensors were attached on both feet and the posterior trunk at the level of L5 with elastic Velcro bands.	Spino-CA with a mean score of 3.6 for stance and gait in SARA	Minimal changes in gait spatial–temporal parameters can be considered as accurate markers for CA progression.

Reference n°	Gait Analysis System		Technology Description	Neurological Disorder	Clinical Implication
Reference n°	NWS	WS	Technology Description	Neurological Disorder	Clinical Implication
[ ]	X		A 10 m walkway with a pressure sensitive mat. Spatial–temporal parameters were registered using GaitRite mat, which contains a total of 13,824 sensors.	Post-stroke patients (both ischemic and hemorrhagic)	Most useful gait parameters are step length, swing time, and stance time. In addition, authors stated that asymmetry time values are not reliable parameters to assess gait in post-stroke patients.
[ ]		X	Inertial Measurment Unit (IMU) system (Xsens Technology B.V., Enschede, The Netherlands, Hengelo) composed of seven inertial sensors.	Post-stroke patients	NA
[ ]		X	Kinect v2, which included an 8-core Intel in addition to an ad hoc application designed to register the 3D position and orientation of the 25 human joints provided by the Kinect v2.	Post-stroke patients (both ischemic and hemorrhagic)	Results indicated that patients with a higher fall risk manifested lower gait velocity and cadence, a shorter stride and step length, and higher double support time. Additionally, the risk of falling was related to increased trunk and pelvic obliquity and tilt, and to decreased hip flexion–extension and ankle height variation.
[ ]	X		Odonate 3D motion capture system in a mobile terminal and a workstation. This innovative a binocular depth camera combined with an artificial intelligence system to capture, analyze, and calculate gait parameters automatically.	Post-stroke patients	Alterations were found in spatial–temporal and kinematic parameters; thus, this new system can perform an objective gait assessment in five minutes, also in a home-based setting.
[ ]		X	Five synchronized IMUs.	Severe TBI patients	Severe TBI patients present serious difficulties in maintaining balance during gait, especially movements of the head, which are the most impaired, probably related to vestibular dysfunctions due to traumatic events. Additionally, authors suggested to assess gait through dynamic balance skills during curved trajectories as in Figure-of-8 Walk Test.
[ ]		X	Three IMUs were attached with elastic straps over both lateral ankles to detect gait phases and over the fifth lumbar vertebrae.	TBI	TBI patients manifest great imbalances in dynamic balance, especially in antero-medial weight shifting, when compared with healthy control subjects.

The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

Bonanno, M.; De Nunzio, A.M.; Quartarone, A.; Militi, A.; Petralito, F.; Calabrò, R.S. Gait Analysis in Neurorehabilitation: From Research to Clinical Practice. Bioengineering 2023 , 10 , 785. https://doi.org/10.3390/bioengineering10070785

Bonanno M, De Nunzio AM, Quartarone A, Militi A, Petralito F, Calabrò RS. Gait Analysis in Neurorehabilitation: From Research to Clinical Practice. Bioengineering . 2023; 10(7):785. https://doi.org/10.3390/bioengineering10070785

Bonanno, Mirjam, Alessandro Marco De Nunzio, Angelo Quartarone, Annalisa Militi, Francesco Petralito, and Rocco Salvatore Calabrò. 2023. "Gait Analysis in Neurorehabilitation: From Research to Clinical Practice" Bioengineering 10, no. 7: 785. https://doi.org/10.3390/bioengineering10070785

Article Metrics

Article access statistics, further information, mdpi initiatives, follow mdpi.

Subscribe to receive issue release notifications and newsletters from MDPI journals

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Publications
Account settings
My Bibliography
Collections
Citation manager

Save citation to file

Email citation, add to collections.

Create a new collection
Add to an existing collection

Add to My Bibliography

Your saved search, create a file for external citation management software, your rss feed.

Search in PubMed
Search in NLM Catalog
Add to Search

A comprehensive survey on gait analysis: History, parameters, approaches, pose estimation, and future work

Affiliations.

1 Information Technology, Indira Gandhi Delhi Technical University for Women, New Delhi 110006, New Delhi, India. Electronic address: [email protected].
2 Information Technology, Indira Gandhi Delhi Technical University for Women, New Delhi 110006, New Delhi, India.
3 Computer Science and Engineering, National Institute of Technology, New Delhi 110040, New Delhi, India.
PMID: 35659390
DOI: 10.1016/j.artmed.2022.102314

Human gait is a periodic motion of body segments-the analysis of motion and related studies is termed gait analysis. Gait Analysis has gained much popularity because of its applications in clinical diagnosis, rehabilitation methods, gait biometrics, robotics, sports, and biomechanics. Traditionally, subjective assessment of the gait was conducted by health experts; however, with the advancement in technology, gait analysis can now be performed objectively and empirically for better and more reliable assessment. State-of-the-art semi-subjective and objective techniques for gait analysis have limitations that can be mitigated using advanced machine learning-based approaches. This paper aims to provide a narrative and a comprehensive analysis of cutting-edge gait analysis techniques and insight into clinical gait analysis. The literature of the previous surveys during the last decade is discussed. This paper presents an elaborated schema, including gait analysis history, parameters, machine learning approaches for marker-based and marker-less analysis, applications, and performance measures. This paper also explores the pose estimation techniques for clinical gait analysis that open future research directions in this area.

Keywords: Clinical gait analysis; Human gait analysis; Objective analysis review; Pose estimation; Semi-subjective analysis.

PubMed Disclaimer

Publication types

Search in MeSH

Related information

Linkout - more resources, full text sources.

Elsevier Science
Citation Manager

NCBI Literature Resources

MeSH PMC Bookshelf Disclaimer

The PubMed wordmark and PubMed logo are registered trademarks of the U.S. Department of Health and Human Services (HHS). Unauthorized use of these marks is strictly prohibited.

REVIEW article

Present and future of gait assessment in clinical practice: towards the application of novel trends and technologies.

$\r\nAbdul Aziz Hulleck$

1 Mechanical Engineering Department, Khalifa University, Abu Dhabi, United Arab Emirates
2 School of Mechanical and Aerospace Engineering, Monash University, Clayton Campus, Melbourne, Australia
3 Weill Cornell Medicine, New York City, NY, United States
4 Biomedical Engineering Department, Khalifa University, Abu Dhabi, United Arab Emirates
5 Health Engineering Innovation Center, Khalifa University, Abu Dhabi, United Arab Emirates

Background: Despite being available for more than three decades, quantitative gait analysis remains largely associated with research institutions and not well leveraged in clinical settings. This is mostly due to the high cost/cumbersome equipment and complex protocols and data management/analysis associated with traditional gait labs, as well as the diverse training/experience and preference of clinical teams. Observational gait and qualitative scales continue to be predominantly used in clinics despite evidence of less efficacy of quantifying gait.

Research objective: This study provides a scoping review of the status of clinical gait assessment, including shedding light on common gait pathologies, clinical parameters, indices, and scales. We also highlight novel state-of-the-art gait characterization and analysis approaches and the integration of commercially available wearable tools and technology and AI-driven computational platforms.

Methods: A comprehensive literature search was conducted within PubMed, Web of Science, Medline, and ScienceDirect for all articles published until December 2021 using a set of keywords, including normal and pathological gait, gait parameters, gait assessment, gait analysis, wearable systems, inertial measurement units, accelerometer, gyroscope, magnetometer, insole sensors, electromyography sensors. Original articles that met the selection criteria were included.

Results and significance: Clinical gait analysis remains highly observational and is hence subjective and largely influenced by the observer's background and experience. Quantitative Instrumented gait analysis (IGA) has the capability of providing clinicians with accurate and reliable gait data for diagnosis and monitoring but is limited in clinical applicability mainly due to logistics. Rapidly emerging smart wearable technology, multi-modality, and sensor fusion approaches, as well as AI-driven computational platforms are increasingly commanding greater attention in gait assessment. These tools promise a paradigm shift in the quantification of gait in the clinic and beyond. On the other hand, standardization of clinical protocols and ensuring their feasibility to map the complex features of human gait and represent them meaningfully remain critical challenges.

1. Introduction

Changes in the signature of gait, or the unique sequential walking pattern in humans, reveal key information about the status and progression of numerous underlying health challenges, from neurological and musculoskeletal conditions to cardiovascular and metabolic disease, and to ageing-associated ambulatory dysfunction and trauma. Accurate reliable identification of gait patterns and characteristics in clinical settings, as well as monitoring and evaluating them over time, enable effective tailored treatment, inform predictive outcome assessment, and an allow for an overall better practice of precision medicine.

In clinical gait assessment, both a person's “ability” to walk and “how” the individual walks are highly relevant. The walking ability of a person is typically based on two main aspects: how far can an individual walk and what is his/her tolerance level ( 1 ). For example, for post stroke gait assessment, 3-, 6-, or 10 min walk tests are used, in addition to Functional Ambulation Category (FAC), Short Physical Performance Battery (SPPB), and/or Motor Assessment Scale (MAS). Other clinical subjective assessment scales include the Unified Parkinson Disease Rating Scale (UPDRS) the Scale for the Rating and Assessment of Ataxia (SARA)], the Alzheimer's Disease Assessment Scale (ADAS), the Expanded Disability Status Scale (EDSS) the High-level MobilitARTIy Assessment Tool (HiMAT), and the Dynamic Gait Index ( 2 ). The quality of gait or “how” the person walks, on the other hand, highly depends on the quantification of gait patterns and accurate identification of specific gait characteristics. Despite evidence of the advantages of quantitative instrumented gait analysis (IGA) in clinical practice and recommendations by the National Institute for Health and Clinical Excellence (NICE) ( 3 ) identifying IGA is the preferrable choice for “gait-improving surgery”, it remains not well leveraged in clinical settings due to the high cost/cumbersome equipment and complex protocols/data analysis associated with traditional gait labs, as well as diverse training, experience and preference of clinical teams ( 3 – 5 ). Moreover, the use of IGA by allied health professionals, such as physiotherapists, occupational therapists and orthotists, and training also remain non standardized and limited ( 5 – 7 ).

Observational gait analysis continues to be popular among clinicians due to its inherent simplicity, availability, and low cost ( 8 ). On the other hand, the validity, reliability, specificity, and responsiveness ( 9 , 10 ) of these qualitative methods are controversial and increasingly being questioned ( 6 ). Furthermore, there is evidence to suggest that subjective clinical assessment scales may not be sensitive to disease severity and specific characteristics and may limit understanding of underlying disease mechanisms, hence adversely impacting optimal treatment ( 11 ). Examples of such scales include Multiple Sclerosis (MS), where subjective measures, such as the Expanded Disability Status Scale (EDSS), the Multiple Sclerosis Severity Scale (MSSS), Multiple Sclerosis Walking Scale (MSWS), and Multiple Sclerosis Functional Composite (MSFC), continue to be widely used in clinical practice. These scales have been criticized for lack of sensitivity ( 12 ), high interrater variability ( 13 ), as well as being prone to practice effects and variability ( 14 , 15 ). Similarly, clinical assessment of Parkinson's disease (PD) using the Unified Parkinson's Disease Rating Scale (UPDRS) is subjective and largely dependent on the expertise and experience of the clinicians, as well as the severity of the disease ( 16 ). In Stroke patients, assessment tests such as Functional Ambulation Category (FAC), Short Physical Performance Battery (SPPB), and/or Motor Assessment Scale (MAS) are typically employed, along with qualitative observational/visual gait analysis (using naked eye or video images). Nevertheless, the validity, reliability, specificity, and responsiveness of these qualitative methods are questioned ( 9 ), and although they may be useful for the rudimentary evaluation of some gait parameters, they are not adequate for analyzing the multifaceted aspects of gait variability and complexity ( 17 ).

Instrumented gait analysis (IGA), which can provide accurate and precise quantitative measurement of gait patterns and characteristics, has long been the gold standard for gait assessment in research practice ( 18 ). IGA generally refers to the use of instrumentation to capture and analyze a variety of human gait parameters (spatiotemporal, kinematic, and kinetic measures). Traditional IGA systems include motion capture systems, and force plates, instrumented walkways, and treadmills, while more recent systems comprise of miniaturized wearable sensing system, computational platforms and modalities ( 18 ). Literature on the clinical applicability and efficacy of IGA indicates that IGA-based quantitative assessment can improve the diagnosis, outcome prediction, and rehabilitation of various gait impairments as compared to conventional observational scales and techniques for gait dysfunction in a wide spectrum of diseases including MS, PD, Stroke, and Cerebral Palsy ( 9 – 13 ). A recent review on the clinical efficacy of IGA confirms that there is strong evidence that 3-D gait analysis, or 3DGA; has the potential to alter and reinforce treatment decisions; increases confidence in treatment planning and agreement among clinicians; can better identify diagnostic groups and expected treatment outcomes; and overall can improve patient outcomes if recommendations are followed ( 19 ).

Emerging at an unprecedented rate, wearable sensing systems and associated computational modalities are rapidly transforming the quality and accessibility of healthcare, spanning multiple applications from neurology and orthopedics to cardiovascular, metabolic, and mental health. Magnetic (e.g., magnetometers), inertial measurement (e.g., accelerometers and gyroscopes), and force sensors (e.g., insole foot pressure) nowadays offer unprecedented data capture opportunities that can overcome limitations of non-wearable devices due to their low-cost, less setup-time and complexity, lightweight, and portability, making them ideal for out-of-lab and continuous monitoring in the clinic and beyond ( 20 ). Magneto-inertial measurement units (MIMUs), in conjunction with force pressure sensors, have the capability of capturing spatiotemporal, kinematic, and kinetic gait data ( 2 ) rendering the concept of a mobile gait lab a reality. Such labs can inherently overcome the limitations of IGA traditional labs, providing less costly and cumbersome tools with potential for gait assessment in natural environments (clinics, homes, sports arenas, etc.), user friendly interfaces, and the opportunity to provide continuous real-time feedback to clinicians and patients, as well as tele rehabilitation capabilities. In addition, wearable systems allow for easy synchronization with other physiological measurement systems, including EMG, ECG, and EEG, towards the acquisition of invaluable multimodal continuous physiological data in various settings.

This scoping review aims to provide a summary of the current state of clinical gait assessment, including shedding light on gait pathologies and clinical indices and scales, as well as a roadmap for the development of future gait mobile labs- highlighting the clinical validity and reliability of the latest devices and data interpretation algorithms. The word novel in the title of this review reflects recent emergence/implementation of the technologies reviewed and/or recent commercialization. This includes wearable technologies, as well as AI-driven computational platforms. The remainder of the review is structured as follows: Section 2 describes the adopted methodology, including the approach, search strategy and selection criteria. Section 3 details clinical gait pathologies, relevant parameters, as well as current clinical gait assessment tools, scales, and indices, while Section 4 presents gait assessment technologies applicable to clinical settings, including state-of-the-art imaging techniques and wearable technologies, algorithms, and novel AI-driven computational platforms. Section 5 deliberates on the concept of a mobile gait lab for clinical applications. Section 6 highlights the limitations, while Section 7 presents the conclusive remarks and future work.

This review is aimed at summarizing various clinical gait pathologies and associated parameters, applicable gait analysis techniques and gait indices, and the latest trends in wearables systems and algorithms. To address this broader research objective, the authors adopted a scoping review approach rather than a systematic review approach. As reported in ( 21 ), scoping reviews are ideal for addressing a broader scope with a more expansive inclusion criterion.

2.1. Search criteria

A keyword search was performed in PubMed, Web of Science, Medline, and ScienceDirect databases, using a combination of search terms from the following groups: 1. (normal gait OR pathological gait OR gait parameters OR gait indices), 2. (gait assessment OR gait analysis), 3. (wearable systems OR wearable algorithms), 4. (inertial measurement units OR accelerometer OR gyroscope OR magnetometer OR insole sensors OR electromyography sensors). No limit for the year of publication was set, however, the search was last updated in December 2021. Only articles written in the English language were considered in this review. In addition, the reference list of the included articles was checked to identify additional relevant publications meeting the inclusion criteria. The literature search and data extraction were carried out independently by two authors (AAH, DMM) and any inconsistencies and disagreements discrepancies were resolved through following discussions with the other authors (NA, MER, KK).

This scoping review included original published works and review articles which met the following inclusion/exclusion criteria: (i) studies addressing various gait disorders and associated gait parameters, (ii) studies focusing on instrumented gait analysis techniques and gait indices, (iii) studies evaluating the use, validity, and reliability of wearable-based gait measurement devices/systems for measuring gait events, and evaluating and assessing gait dysfunction, (iv) studies concerning the applicability of sensor fusion techniques and algorithms applicable for wearable-based systems with application to gait analysis. The title and/or abstract of the studies were initially screened for suitability. The full-text articles meeting the inclusion criteria were obtained for data extraction and synthesis. A flowchart explaining the same is shown in Figure 1 .

Figure 1 . Publication selection process.

3. Clinical gait pathologies and parameters

3.1. normal gait cycle and parameters.

Normal gait can be defined as a series of rhythmic, systematic, and coordinated movements of the limbs and trunk that results in the forward advancement of the body's center of mass ( 22 ). A result of intricate dynamic interactions between the central nervous system and feedback mechanisms ( 23 ), walking is characterized by individual gait cycles and functional phases ( Figure 2 ). A gait cycle consists of two main phases, stance, and swing, which are further divided into five and three functional phases, respectively. The stance phase corresponds to the duration between heel strike and toe-off of the same foot, constituting approximately 60% of the gait cycle. The swing phase begins with toe-off and ends with heel contact of the same foot and occupies 40% of the cycle. As each functional phase contributes to successfully accomplishing the goal of walking, healthy gait involves cyclic and complementary movements of the limbs under control. It is characterized by stance stability; toe clearance during the swing; pre-positioning at swing; sufficient step length; as well as mechanical and metabolic efficiency ( 24 ). Table 1 provides gait parameter ranges based on studies on healthy adults. Determining an appropriate normal range for many of the features is highly challenging as individuals exhibit a wide range of gait patterns across different age groups and gender ( 17 ).

Figure 2 . Normal gait cycle (adapted from 1 ).

Table 1 . Gait parameters for healthy individuals ( 1 ).

3.2. Gait parameters associated with pathology

Gait disorders are typically associated with deficits in the brain, spinal cord, peripheral nerves, muscles, joints, or bones. Some medical conditions leading to pathological gait include but not limited to muscular dystrophy, myelodysplasia, cerebral palsy, arthritis, osteoarthritis, head injury, lower limb amputation, multiple sclerosis, rheumatoid, spinal cord injury, parkinsonism, and stroke ( 25 ).

In neuromuscular conditions, the loss of central control affects the motion. In general, patients walk slower than healthy individuals and with compromised spatiotemporal, kinematic, and kinetic parameters. In older adults, a walking speed decline of 0.7% per year is observed, along with significant changes in cadence and step length. The aging population also exhibits lower knee extension at heel-strike and knee flexion during the swing phase ( 23 , 26 ). The following subsections describe some of the most common gait disorders and associated pathological parameters. The associated impacted parameters are summarized in Table 2 .

Table 2 . Effect of pathology on gait disorders.

3.2.1. Neurological gait disorders in elderly people

Gait ailments associated with aging lead to reduction in the quality of life and increased morbidity and mortality. Elderly patients exhibit complex gait disorders, and their dual task ability deteriorates due to a decline in their central resources ( 23 , 26 ).

Specific gait dysfunction noted in the elderly population are summarized as follows:

3.2.1.1. Hypokinetic-rigid gait disorders

Shuffling with a reduced step height and stride length characterizes hypokinetic gait disorder ( 27 ). Reduced arm swing with slow turning movements is also present in isolation. Festination, when patients use rapid small steps to maintain the feet beneath the forward moving trunk, is also observed. Ataxic elements include broad stance width and an increased variability in timing and amplitude of steps ( 27 ). Gait associated with underlying diseases, such as Parkinson's disease, cerebrovascular disease, and ventricular widening, is classified within hypokinetic-rigid gait disorders ( 27 , 28 ).

3.2.1.2. Cautious and careless gait

Defined as gait during which people move slowly with a wider base, and shorter stride, with minimal trunk movement, while the knees and elbows are bent. Whereas careless gait is when patients appear overly confident and walk insensitively fast. Careless gait is due to confusion and delirium associated with old age ( 27 ).

3.2.1.3. Dyskinetic gait or involuntary movements

Patients with post-anoxic encephalopathy exhibit bouncing gait and stance. This is also observed in patients with Parkinson's disease-causing excessive trunk movements contributing to falls. Several dystonic patients are reported to walk on their toes ( 27 , 28 ).

3.21.1.4. Psychogenic gait disorders

Gait dysfunction is common in elderly people due to adverse effects of drugs leading to extrapyramidal side-effects, sedation, orthostatic hypotension, behavioral abnormalities, or ataxia ( 27 , 28 ).

3.2.1.5. Fluctuating or episodic gait disorders

Elderly people often exhibit fluctuating or episodic gait disorder after exercise due to fatigue, and it might be an indication of underlying vascular or neurogenic limping. Freezing gait is part of hypokinetic-rigid syndrome ( 27 , 28 ).

3.2.2. Gait disorders in Parkinson's disease

PD is a neurological disorder which leads to cognition, where gait impairment deteriorates with disease progression, increasing reliance on cognition to control gait. Due to cognitive impairment with PD, the ability to compensate for gait disorders diminishes, leading to further gait impairment. PD is characterized by deficit in amplitude and gait speed, along with increased gait variability ( 29 ).

3.2.3. Gait in diabetic peripheral neuropathy

Neuropathy of motor, sensory, and autonomic components of the nervous system are one of the many complications of Type II Diabetes (T2D). An intact central and peripheral nervous system are essential to initiate and control healthy gait, along with sufficient muscle strength, bone, and joint movements in complete range for normal locomotion. Patients diagnosed with T2D take extra steps when walking in straight paths and during turns, along with an overall reduction in walking speed, step length, cadence, and fewer acceleration patterns as compared to age-matched healthy controls. Joint range of motion is also altered in T2D, where patients with diabetic peripheral neuropathy exhibit a reduced range of motion at the ankle joint in dorsi and plantar flexion and a reduced flexion and extension range of motion at the knee joint in both, as compared to non-diabetic people ( 24 ).

3.2.4. Post stroke gait

Hemiplegia after stroke contributes to significant reduction in gait performance. In stroke survivors, function of the cerebral cortex is usually impaired, whilst that of spinal cord is preserved. Dysfunction is typically demonstrated by a marked asymmetrical deficit. Decreased walking speed and cadence, in addition to longer gait cycle and double limb support as compared to healthy individuals. For hemiplegic stroke survivors, a reduced peak extension of the hip joint in late stance, varying peak lateral pelvis displacement, knee flexion and decreased plantarflexion of ankle at toe off are reported. The GRF (Ground Reaction Force) pattern is characterized as asymmetric, along with decreased amplitude of joint moments, at the lower limb joints on the paretic side ( 30 ).

3.2.5. Total hip arthroplasty (THA)

Large deficits in gait speed ( 31 ), stride length ( 32 , 33 ), sagittal hip range of motion ( 32 , 33 ), hip abduction moment-coronal plane ( 31 ), and negligible changes in transverse plane hip range of motion ( 31 ), deficiency in single limb support time ( 31 ), are reported in patients post THA as compared to healthy controls. Peak hip extension is typically reduced, whereas peak hip flexion remains similar as compared to controls. In addition, peak hip abduction moment is reduced along with peak hip external rotation moment ( 34 ).

3.3. Clinical gait assessment measures and indices

The use of observational gait analysis and subjective rating sales continues to be widespread in clinical settings, both as a diagnostic tool and as a prognostic measure, as previously mentioned. Although these techniques can be useful for the initial rudimentary evaluation of some gait parameters, the validity, reliability, specificity, and responsiveness of these qualitative methods are highly questionable. Researchers have therefore proposed various pathology-specific gait indices and summary measures ( 35 ) based on commercially available technologies with accepted levels of accuracy Table 3 . summarizes the current clinical gait summary measures, discrete and continuous gait indices, and non-linear approaches reported in literature, along with advantages and disadvantages.

Table 3 . Clinical gait measures and indices ( 123 , 124 ).

4. Gait assessment technologies applicable to clinical settings

In the past couple of decades, remarkable technological advancement has been witnessed in the field of gait assessment and analysis, particularly in gait assessment technology. Instrumented walkways, both portable and non-portable, became a good alternative to complicated, bulky and non-portable traditional gait labs. These systems (for example the Walkway and StrideWay from Tekscan Inc., Boston, United States) are now widely used in research and to a limited extent in clinical practice. They typically include low-profile floor walkway systems equipped with grids of embedded sensors below the surface, which record foot-strike patterns as a function of time and space as an individual walks across the platform, and dedicated software which computes the various spatiotemporal gait measures Although these instrumented mats involve less setup time and are generally simple to operate as compared to traditional IGA labs, they are expensive, restrictive to specific operational environment to over-ground trials ( 36 , 37 ).

Marker-based optical motion capture (Mocap) is another rapidly emerging technology effective for obtaining 3D kinematic movement data. Passive Mocap systems [e.g., Vicon (Vicon Motion Systems Ltd, Oxford, United Kingdom) and ELITE optoelectronic system (BTS S. p .A., Milano, Italy)], include retro-reflective markers (that reflect the light emitted by high-resolution infrared cameras) attached to specific anatomic landmarks. The location of the marker is identified by decoding the camera images. Here, the markers must be calibrated for identification before the recording session commences. Active Mocap systems (e.g., Optotrak motion capture system; Northern Digital Inc., Waterloo, Canada), on the other hand, use light-emitting diode (LED) markers (reflect their own light powered by a battery), which are automatically identified ( 38 , 39 ). In the context of clinical relevance, although such systems yield extremely accurate reliable data, operational factors including infrastructure, non-portability, high cost, additional time required for initial set-up and calibration, operational complexity, and restrictions to indoor setup impose hurdles to their functional deployment in clinics and rehabilitation centers ( 84 ). Recently, more portable cost-effective alternatives, such as Microsoft Kinect (based on a depth sensor-based markerless motion capture solution) became the application of choice ( 40 ).

Optoelectronic systems (e.g., Optogait®, Microgate, Italy) have also been used to capture spatiotemporal gait parameters. These mainly consist of a transmitting and a receiving bar containing an infrared light. Interruptions of the communication between the emitter and receiver are detected by the system to calculate the various gait parameters ( 41 ).

An evolution in the measurement of gait kinetic parameters can also be witnessed in the last two decades. These parameters include ground reaction forces, and intersegmental joint reaction forces, moments, and powers. Instrumented walkways offer dynamic plantar pressure mapping but are expensive and do not provide joint kinetic data. Force plates are also used in various gait analysis studies ( 38 , 39 , 42 ). These are able to provide intersegmental joint reaction forces by using the ground reaction forces measured along with inverse dynamics models (Winters book) Chen et al. ( 93 ) developed a novel remote sensing technology called “Electrostatic Field Sensing (EFS)” for measuring human gait including stepping, walking, and running, and further extended the work to post-stroke gait. This technology is credited with several advantages, such as being non-contact, affordable, and allows long-time monitoring ( 43 ). Shoe insole systems represent another category of gait quantification tools and techniques. These systems are designed to allow for the recording of both dynamic plantar pressure and spatiotemporal data. F-scan (Tekscan Inc., Boston, United States) is an ultra-thin in-shoe pressure measurement system utilizing Force-Sensitive Resistive films (FSR) technology ( 44 ).

The characteristics of different measurement systems applicable to clinical settings are summarized in Table 4 , and the pros and cons of these systems are listed in Table 5 .

Table 4 . Portable wearable gait assessment tools.

Table 5 . Pros and cons of different IGA systems.

Computational pipeline using computer vision techniques has been proposed as an ecological and precise method to quantify gait in children with neurodevelopmental disorders, along with the pose estimation software to obtain whole-body gait synchrony and balance ( 45 ). Speed, arm swing, postural control, and smoothness (or roughness) of movement features of gait for Parkinson's patients were extracted using videos processed by ordinal random forest classification model. Significant correlation between clinician labels and model estimates was reported, which provides gait impairment severity assessment in Parkinson's disease using single patient video, thereby reducing the need for sophisticated gait equipment ( 46 ). Computer vision-based gait assessment tools promise frequent gait monitoring using minimal resources ( 46 ). Deep learning to detect human subject in 2D images and then combining 3D sensing data to measure gait features has proven to be more robust than depth cameras in gait parameter acquisition ( 47 ).

4.1. Imaging techniques for gait assessment

As previously mentioned, marker-based optoelectronic systems are currently the most widely used systems in IGA among both research and clinical communities. On the other hand, one of the main sources of error inherent to these systems is the degree of movement of the skin, muscle, and other soft tissues, or the so- called soft tissue artifacts (STA), under the markers in relation to bony landmarks, hence violating the rigid body assumption underlying these methods ( 48 , 49 ). Moreover, STA varies by marker location in a unique and unpredictable manner, particularly during dynamic activities, which can make it unreliable for clinical applications ( 50 ).

Although not yet widespread in biomechanics, computer vision based markerless gait assessment methods offer a promising tool for gait assessment in research, as well clinical and sports biomechanics applications. By leveraging modern technologies, such as improved solvers, advanced image features and modern machine learning, markerless vision-based systems can reduce the required number of cameras, incorporating moving cameras, increasing the number of tracked individuals, and offering robust detection and fitting in diverse environments. On the other hand, issues such as accuracy and field-based feasibility remain to be addressed ( 51 ).

Three-dimensional imaging techniques have been successfully used to directly determine bone movements during walking as a gold reference standard to validate/improve current motion capture techniques ( 54 ). For example, researchers have resorted to quantifying STA by comparing with reference 3D kinematics of bone reconstructed from fluoroscopy-based tracking ( 53 ). Fluoroscopy has also emerged as a means for tracking position and orientation of underlying skeletal anatomy of the foot/ankle ( 54 ). Although single plane fluoroscopy yielded large errors when used to evaluate the accuracy of multi-segment foot models ( 49 ), dual fluoroscopy (DF) was found reliable and is considered as the current reference standard to compare joint angles ( 55 ). Combined with 2D/3D registration, video-fluoroscopy allows for accurate quantification of 3D joint motion free of STA ( 56 ). High-speed dual fluoroscopy (DF) has been reported to measure in-vivo bone motion of the foot and ankle with sub-millimeter and sub-degree errors ( 57 ). DF has also been used to evaluate multi-segment foot models and reported good agreement between DF and skin-marker data for the first metatarsal and sagittal plane measurements of the longitudinal arch ( 48 ).

Various researchers investigated the use of DF for clinical applications . In-vivo dual fluoroscopy was used to quantify the hip joint kinematics of patients with Femoroactabular impingement syndrome (FAIS) relative to asymptomatic, morphologically normal control participants during standing, level walking, incline walking and an unweighted functional activity. The kinematic position of the hip joint was obtained by registering projections of 3D computed Tomography models with DF images ( 58 ). Knee kinematic profiles were also obtained using 3D video-fluoroscopy and compared to actual and nominal flexion-extension, internal-external rotations, and antero-posterior translations profiles with optical mocap during stair climbing ( 59 ). Joint function for total talonavicular replacement after a complex articular fracture was evaluated using a full body gait analysis and 3D joint kinematics based on single-plane fluoroscopy ( 60 ). The 3D video fluoroscopic analysis was performed to assess joint motion of the replaced ankle ( 60 ). DF and CT imaging techniques were both employed to calculate in-vivo hip kinematics, along with model-based tracking, to compare the effect of different coordinate systems ( 61 ). Since marker-based systems are unable to accurately analyze talocrural or subtalar motion because the talus lacks palpable landmarks to place external markers ( 54 ), digitized video fluoroscopy was reportedly used to determine the sagittal plane motion of the medial longitudinal arch during dynamic gait ( 62 ). Characteristics of knee joint motion were also analyzed in 6DOF during treadmill walking using a dual fluoroscopy imaging system at different speeds ( 63 ).

DF uses anatomical landmarks visible on 3D CT reconstructions which substantially reduces errors due to STA ( 58 ). Computed tomography (CT) scans of participants are usually needed in DF to determine bone position from the DF images. Single plane fluoroscopy is restricted to 2D motion capture, while using a second FS allows for a full 3D analysis although a single gantry system has lower radiation than the biplane system with reported ionizing radiation levels of 10 µSv per trial ( 54 ). Stationary image intensifiers and static systems have a restricted field of view limiting their application to highly restricted movements ( 56 ). Moving fluoroscopes, consisting of a fluoroscopic unit mounted on a moving trolley which moves with the subject and is controlled by wire sensors to ensure that it remains in the field of view of the image intensifier ( 56 ), provide an enhanced field of view ideal for dynamic scenarios and moving joints.

Fluoroscopic systems designed for precise capture of bone movement and joint kinematics, unlike optical or inertial systems, are not yet commercially available, generally requiring in-house instrumentation and further performance evaluation. The evaluation would typically include determining the resolution of the hardware imaging chain, assessing how the hardware and software reduce or eliminate various distortions, and measuring static and dynamic accuracies and precisions based on precisely known motions and positions ( 64 ). Image quality is a major determinant of error in fluoroscopic applications ( 62 ). Pulse imaging of fluoroscopes, such as pulse width, limits image quality at a given frame rate. Increasing the pulse rate, which is function of pulse width, may add to radiation exposure, leading to an important tradeoff consideration between image quality and radiation exposure ( 63 ). Moving video-fluoroscopes reported lower gait velocity, step length, and cadence as compared to control conditions, indicating altered time distance parameters towards those of slow walking ( 56 ). So far, dynamic MRI used to define in- vivo talocrural and subtalar kinematics ( 65 ) does not allow data collection during normal gait.

Continued multidisciplinary collaborative efforts among biomechanists, imaging and computer vision experts, and clinicians are essential for fully leveraging these highly promising techniques in clinical applications.

4.2. Portable wearable systems for gait assessment

Wearable technology – the use of body-worn sensors to measure the characteristics of human locomotion, has recently emerged as an efficient, convenient, and most importantly, inexpensive option to quantitative gait analysis for both clinical and research-based applications ( Figure 3 ). In general, it uses individual sensor elements, such as accelerometers, gyroscopes, magneto resistive sensors, force/pressure sensors, goniometers, inclinometers, and electromyographic (EMG) sensors, or combined as an inertial measurement unit (IMU) ( 66 ). In comparison to conventional counterparts (e.g., walkway and camera based Mocap), wearable sensing enables continuous gait monitoring (> 2 h) outside the lab or clinic, allowing for replication of natural patterns of walking. Moreover, gait patterns over an ample distance could be measured as opposed to limited walking distance in a lab-based setting.

Figure 3 . Wearable gait lab

Accelerometers are often used in gait analysis for assessing stability and risk of fall. In a study which used a single tri-axial accelerometer mounted on the sacrum to analyze the risk of fall among 80 participants, accelerometry-based techniques were found to be able to detect subjects with increased risk of fall by employing appropriate machine learning techniques ( 66 ). In ( 67 ), a 3D accelerometer attached to the lower back was used for stability assessment of older adults. The applicability of a single accelerometer, worn on the back was further examined in ( 68 ), which highlights promising results for implementation in routine clinical practices. Considerable work has also been carried out to assess the consistency of gait characteristics obtained from accelerometers, where discrepancies in sensors positioning yield to critical errors ( 69 ). Furthermore, in ( 70 ), the authors have provided a comprehensive review on the use of accelerometry-based gait analysis techniques and their application to clinical settings.

Gyroscopes are also increasingly employed for gait studies. These devices measure angular velocity and are often combined with accelerometers and other micro-electromechanical systems (MEMS) devices to enhance performance through sensor fusion techniques. They have found applications in step detection, gait event detection, segmental kinematics, and more. For instance, a single gyroscope placed in the instep of the foot was successfully used to detect gait events, including heel strike, foot flat, heel off, and toe-off ( 71 ). Another study involved two gyroscopes, mounted on the lower left and right side of the waist to calculate walking steps and step length ( 72 ).

Magnetometers measure the magnetic field direction and intensity at a specific point. In combination with other inertial sensors (accelerometers and gyroscopes), they form a so-called inertial measurement unit (IMU), which can produce a drift-free estimation of gait parameters ( 73 ). Sophisticated commercialized IMUs (Physiolog 5 IMU, Gait Up, Switzerland, MTw Awinda, Xsens Technologies B.V., Netherlands), as well as in-house developed systems, were equally used for gait studies ( 74 ). In the context of human motion analysis, IMUs are employed for several possible goals, for example, to estimate the joint angles ( 74 ), to detect the risk of fall in an elderly population, long term monitoring of activities and symptoms ( 75 ), measurement of gait events, spatiotemporal parameters ( 76 , 77 , 78 ), ground reaction forces and moments ( 79 ), and estimation of gait symmetry ( 80 ). Mariani et al. (2010) used IMUs to measure foot kinematics in a study involving both young and elderly and reported the suitability of the system to clinical practice ( 81 ). Parisi et al. developed a low-cost system with a single IMU attached to the lower trunk to examine the gait characteristics of both hemiparetic and normal control subjects through measurement of spatiotemporal parameters, which showed excellent correlation with the parameters obtained from a standard reference system ( 78 ).

Insole systems for gait measurement and analysis represent a major category, which is cost-effective, portable, and applicable for both indoor and outdoor settings. Over the years, various technologies were developed ( 82 ), tested, and commercialized. These include capacitive sensors (Pedar system, Novel GmbH, Germany) ( 83 ), force-sensing resistors (FSR) (F-Scan, Tekscan Inc., United States) ( 84 ), and piezoresistive sensors (FlexiForce system, Tekscan, United States and ParoTec system, Paromed, Germany) ( 82 ). Researchers have adopted different approaches about the design, fabrication, and applications of insole systems. Both prefabricated and in-house fabricated insole systems have been tested for healthy and pathological gait ( 85 , 86 ). Some studies have also integrated inertial measurement units (IMU) with shoe insoles to enhance their capabilities. Despite the fact that these shoe-based systems have successfully been used for various gait analysis applications, they suffer from some drawbacks, such as (i) distortion of the flexible contact surface due to repeated loading, which leads to changes in the sensor response, (ii) drift in the output due to prolonged load application that causes heat inside the shoe, and (iii) need for subject-specific calibration that may alter accuracy ( 87 ). Mancinelli et al. (2012) presented ActiveGait – a novel sensorized shoe system for real-time monitoring of gait deviations associated with Cerebral Palsy in children. They reported that the severity of gait deviations can be estimated with an accuracy greater than 80% using the features derived from the center of pressure trajectories gathered from the shoe system ( 88 ). In ( 87 ), the authors designed a novel flexible foot insole system using an optoelectronic sensing technology for monitoring plantar pressure deviations in real-time. The system consists of an array of 64 sensing elements and onboard electronics for signal processing and transmission. Experimental validation was conducted on healthy subjects while walking at self-selected slow and normal speed. A commercial force plate (AMTI, Watertown, United States) was used as a reference system for benchmarking. Jagos et al. (2017), on the other hand, developed the eSHOE, which consists of four FSR sensors, a three-axis accelerometer, and a three-axis gyroscope, and reported good agreement with the gait parameters obtained from the GAITRite mat ( 89 ). Various other studies have also examined the applicability of shoe-based systems for gait analysis ( 85 , 90 – 92 ).

Another class of sensors that found major applications in gait studies is electromyography (EMG) sensors. Surface EMG is a non-invasive technique used to measure muscle activity. In ( 93 ), Lee et al. proposed a method using EMG signals to obtain biometrics from gait for personal identification methods. Another study adopted EMG techniques to understand the co-contraction patterns of thigh muscle during free walking using surface EMG ( 94 ). These research efforts emphasize the importance of wearable sensors in the study of human gait. The wearable systems discussed in this section are summarized in Table 6 .

Table 6 . Instrumented gait analysis (IGA) systems and their features.

Although emerging new wearable technologies promise to enhance gait assessment and rehabilitation, there is limited research on the use of wearable technology to assess gait and mobility and its efficacy in clinical settings. According to a recently published review by Peters et. al. on the use of wearable technology to assess gait and mobility in stroke patients ( 95 ), most of the available studies are intervention studies conducted in laboratory settings that have used sensors to investigate change in cadence, step time variability, and gait speed. As wearable technologies continue to progress in affordability and accessibility, it is expected that such technologies would enable the gathering of movement-related data in “real-world” and various clinical settings. Importantly, these researchers indicated that so far only a limited number of studies examined reliability and validity of existing wearable devices, highlighting the need for more studies to examine psychometric and other properties when collecting gait and mobility information to determine which wearable technologies are most effective. Another recent review on the evaluation of the use of wearables in PD also indicates that novel technologies and wearables have the potential to enable early or differential diagnosis of PD, monitoring of motion state, prevention, or reduction of off-stage status, and assessing of movement complications. On the other hand, more research is required for the validation and the identification of more accurate markers of PD progression ( 96 ). Importantly, these authors warn that wearable devices may not be appropriate in cases of severe motor impairment, off-stage state, cognitive impairment, and for elderly patients and that further research is required for clinical validation.

4.3. Wearable-based gait computational algorithms

Besides sensor technology, sensor fusion algorithms play a critical role in predicting the accuracy/precision of these wearable-based systems. Most of the research has focused mainly on gait feature detection, daily physical activity monitoring, and gait data classification targeting disease diagnosis and user recognition. These algorithms are based on different data mining and AI technology, including machine learning, fuzzy computing, wavelet transforms, genetic algorithms, and data fusions. Alaqtash et al. ( 97 ) developed an intelligent fuzzy computational algorithm for characterizing gait in healthy, as well as impaired subjects. McCamley et al. established a method to calculate initial and final contact of gait using continuous wavelet transforms, employing waist-mounted inertial sensors ( 98 ). Another study cited the use of a single accelerometer mounted at the lower trunk and a corresponding algorithm to identify gait spatiotemporal parameters ( 68 ). A real-time gait event detection algorithm was proposed in ( 99 ) making use of adaptive decision rules. Further in ( 100 ), an original signal processing algorithm is developed to extract heel strike, toe strike, heel-off, and toe-off from an accelerometer positioned on the feet.

A novel gyroscope only (GO) algorithm was proposed in ( 101 ) to calculate knee angle through the integration of gyroscope-derived knee angular velocity. A zero-angle update algorithm was implemented to eliminate drift in the integral value. In addition, published work on noise-zero crossing (NZC) gait phase algorithm was also adapted. This method is applicable for continuous monitoring of gait data. Nukala et al. used support vector machines (SVM), KNN, binary decision trees (BDT), and backpropagation artificial neural network (BP-ANN) to classify the gait of patients from normal subjects, where features extracted from raw signals from gyroscopes and accelerometers were used as inputs. This study reported the highest overall classification accuracy of 100% with BP-ANN, 98% with SVM, 96% with KNN, and 94% with BDT ( 102 ).

Li et al. proposed DTW algorithm, sample entropy method, and empirical mode decomposition to calculate 3 main gait features of post-stroke subjects: symmetry, complexity character, and stepping stability. A k-nearest neighbor (KNN) classifier trained on the acquired features showed a promising result (area under the curve (AUC) of 0.94), which suggests the feasibility of such techniques to automatic gait analysis systems ( 43 ). Rastegari et al. employed a feature selection technique called maximum information gain minimum correlation (MIGMC) to extract gait data of subjects with Parkinson's Disease ( 103 ). The performance of several machine learning classifiers, including Support Vector Machines, Random Forest, AdaBoost, Bagging, and Naïve Bayes were also assessed to test the power of the feature set obtained.

The use of novel computational platforms, including Machine Learning, Support Vector Machine, and Neural Network approaches, are increasingly commanding greater attention in gait and rehabilitation research. Although their use in clinical settings are not yet well leveraged, these tools promise a paradigm shift in stroke gait quantification and rehabilitation, as they provide means for acquiring, storing and analyzing multifactorial complex gait data, while capturing its non-linear dynamic variability and offering the invaluable benefits of predictive analytics ( 1 ). A recent review article discussed the potential value of ML in gait analysis towards quantification and rehabilitation ( 104 ). The authors concluded that further evidence is required although preliminary data demonstrates that the control strategies for gait rehabilitation benefit from reinforcement learning and (deep) neural-networks due to their ability to capture participants' variability. This review paper demonstrated the success of ML techniques in detecting gait disorders, predicting rehabilitation length, and control of rehabilitation devices. Further work is needed for verification in clinical settings.

4.4. Data-driven gait rehabilitation in clinical settings

Quantitative gait assessment is invaluable towards disease-specific and patient specific rehabilitation/therapeutic interventions. Spatiotemporal, kinematic, and kinetic parameters obtained during instrumented gait assessment can help clinicians benchmark, devise strategies, and evaluate the effect of various rehabilitation interventions. Gait disorders not only affect these parameters, and patterns and time spent in the various gait phases, but can also highly impact gait symmetry, and regularity, depending on the disease and severity ( 105 ). Increasing evidence supports a data-driven physical rehabilitation approach to the treatment of functional gait disturbance ( 106 ). There are multiple examples in literature on the effective use of quantitative gait measures towards more effective data-driven rehabilitation. A recent review by Biase et al. ( 107 ) studied the most relevant technologies used to evaluate gait features and the associated algorithms that have shown promise to aid diagnosis and symptom monitoring towards rehabilitation in Parkinson's disease (PD) patients. They reported physical kinematic features of pitch, roll and yaw rotations of the foot during walking, based on which feature extraction and classification techniques, such as principal component analysis (PCA) and support vector machines (SVM) method were used to classify the PD patients. They also used gait features, including step duration, rise and fall gradients of the swing phase, as well as standard deviation of the minima as quantitative measures, for benchmarking and monitoring PD motor status during rehabilitation. Interestingly, this review sheds light on need to change the evaluated gait features as a function of disease progression. Another study was Pistacchi et al. ( 108 ) suggested spatiotemporal gait parameters, such as speed and step length, where reduced step length seems to be a specific feature of Parkinson's disease gait particularly in early disease stages. On the other hand, asymmetry, step shuffling, double-limb support and increased cadence are more common in mild to moderate stages, while advanced stages are more frequent freezing of gait (FOG) and motor blocks, reduced balance and postural control, motor fluctuations and dyskinesia ( 109 ). Researchers have also investigated the evaluation of ambulatory systems for gait analysis post hip replacement ( 110 ). They found gait characteristics such as stride length and velocity, as well as thigh and shank rotations different from healthy individuals and recommended their use to monitor post-surgical rehabilitation efficacy. Spatiotemporal gait parameters, such as step length, width and cadence have been used ( 111 ) to assess the effect of swing resistance and assistance rehabilitation on gait symmetry in hemiplegic patients. Investigators have also studied whether specific variables measured routinely at a rehabilitation center were predictors of gait performance of hemiparetic stroke patients ( 112 ). They found that motor control and balance were the best predictors of gait performance. A recent review article on assessment methods of post stroke gait suggests that multiple spatiotemporal, kinematic, and kinetic parameters can be useful in diagnosing post-stroke gait dysfunction and as quantitative measures to evaluate rehabilitation outcomes ( 1 ). Spatiotemporal characteristics of post-stroke gait include reduced step or stride length, increased step length on the hemiparetic side, wider base of support, greater toe-out angle, reduced walking speed and cadence. Stride time, stance period on both lower limb, and double support time are also increased, in addition to less time in stance and more time in swing phase for the paretic side, as well as asymmetries in spatial and temporal factors. Kinematic parameters associated with hemiplegic gait (reduced mean peak extension of the hip joint in late stance, alterations in the lateral displacement of the pelvis and flexion of the knee, and decreased plantarflexion of the ankle at toe-off, in addition to a significant decrease in peak hip and knee flexion during the swing phase, reduced knee extension prior to initial contact, as well as decreased ankle dorsiflexion during swing), and kinetic parameters (asymmetric patterns, as well as decreased amplitudes of the joint moments and joint powers at the hip, knee, and ankle joints on the paretic side) can be used as quantitative means to design and evaluate effective rehabilitation ( 113 – 115 ). IGA has also been successfully used to quantify and improve gait dysfunction associated with ageing and assess the risk of falling ( 116 ). Spatiotemporal gait parameters such as velocity, swing time, stride length, stride time- and double support time variability, as well as heel strike and toe off angles, and foot clearance, have been suggested as plausible indicative quantitative measures ( 116 ) to assess the risk of falling in elderly subjects. Inertial sensor-equipped shoes additionally provided heel strike and toe off angles, and foot clearance ( 116 ). The study ( 117 ) summarizes that multi-component exercise therapy which consisted of strength, ROM exercise, balance, flexibility and stretching exercises, circuit exercise training, and gait training was found to enhance gait function for individuals suffering with diabetic peripheral neuropathy compared to control groups using spatiotemporal gait parameters like velocity, cadence, step length, step time, double support time, stride length, stride time, ankle ROM. Gait assessment has potential to develop patient training paradigms for overcoming gait disorders ( 111 ).

5. Mobile gait lab for clinical applications and beyond

In recent decades, the healthcare field has witnessed a tremendous interest in the use of wearable sensing modalities and AI-driven data management/analysis techniques for patient diagnosis, monitoring, and rehabilitation. The portability, lightweight, ease of use, and high-power efficiency are some of the factors that promote applicability to a clinical platform.

There are few examples in literature demonstrating the potential success of using wearable-based systems for gait assessment in clinical settings. Prajapati et al. assessed the walking activity of inpatients with subacute stroke using commercial accelerometers attached above the ankle. They found that the walking bouts were shorter in duration and gait was more asymmetric ( 118 ). Studies have established test-retest reliability and accuracy of different sensor technologies; however, further validation trials are recommended prior to any clinical use. Hsu et al. assessed the test-retest reliability of an accelerometer-based system with infrared assist for measuring spatiotemporal parameters, including walking speed, step length, and cadence, as well as trunk control parameters, including gait symmetry, gait regularity, acceleration root mean square, and acceleration root mean square ratio of healthy subjects in hospital ( 119 ). This study showed excellent test-retest reliability of the parameters considered, and thus highlighting the reliability of an infrared assisted, trunk accelerometer-based device for clinical gait analysis. Another study investigated the concurrent validity and test-retest reliability of gait parameters (cadence, gait velocity, step time, step length, step time variability, and step time asymmetry) acquired from elderly subjects, using a tri-axial accelerometer attached to the center of body mass ( 120 ). In comparison to a reference GAITRite system, the acquired parameters showed good validity and reliability. Poitras et al. performed a systematic review of 42 studies assessing the reliability and validity of wearable sensors, specifically, IMUs, for quantifying the joint motion ( 121 ). Evidence suggests that IMU could be an alternative solution to an expensive motion capture system, as it shows good validity for lower-limb analysis involving fewer complex tasks. However, more work is needed to draw a better conclusion with regards to its reliability, as well as to standardize the protocol to get more accurate data in a clinical setting. Importantly, additional research efforts are needed to examine the responsiveness of wearables in free-living conditions in hospital settings.

6. Limitations

This review aimed to summarize available published work on the present and future of gait analysis in clinical settings. The focus was to highlight current systems, scales, and indices, as well as recent technology-driven gait characterization and analysis approaches and their applicability to clinical settings. Within this context, pathological gait associated with different disease, as well as ageing was briefly discussed. As such, this article may have not covered the complete spectrum of gait pathologies and associated parameters. A scoping (non-systematic) search methodology was selected to broaden the scope and integration of the three main aspects of focus (gait pathology, clinical assessment, recent tools, and technologies). In addition, we do not recommend any specific protocol over the other, as most of the papers incorporate different inclusion/exclusion criteria for subject selection, as well as different sampling sizes, which may render comparisons unrealistic.

7. Conclusive remarks and future work

This scoping review aimed to shed light on the status of gait assessment in clinical settings, as well as the state-of-the-art emerging tools and technologies and their potential clinical applicability. Clinical gait analysis continues to rely mainly on observational gait and quantitative scales and is hence subjective and suffers from variability and the lack of sensitivity influenced by the observer's background and experience. Based on the reviewed literature, quantitative IGA-based gait analysis, commonly used in research labs, has the capability of providing clinicians with accurate and reliable gait data for informed diagnosis and continuous monitoring. On the other hand, several factors, including high cost and infrastructure challenges; data variability, complexity, and multidimensionality; lack of sufficient knowledge and standardized training in clinical environments; and time constraints, continue to limit its wide-spread deployment. Rapidly emerging smart wearable technology and AI, including Machine Learning, Support Vector Machine, and Neural Network approaches, are increasingly playing a bigger role in gait assessment. Although their use in clinical settings is not yet well leveraged, these tools promise an unprecedented paradigm shift in the quantification of gait in the clinic and beyond, as they provide means for acquiring, storing, and analyzing multifactorial complex gait data, while capturing its non-linear dynamic variability and offering the invaluable benefits of predictive analytics.

Researchers are also paying increased attention to multisource and multi-modality sensor fusion approaches, which can further add value by integrating the output of multiple sensors to capture the complexity and variability of gait. Multimodality sensor fusion also allows for simultaneous monitoring of various physiological signals during locomotion, such as EMG, ECG, and EEG, where fusing these with various gait measures (spatiotemporal, kinematic, and kinetic) can shed light on underlying health conditions and disease etiology towards better informed outcome prediction and clinical decisions. As the volume of data from the variety of sensors, including electroencephalography, electro-oculography, electro-cardiography, and electromyography, motion capture and force sensors data, substantially increases, more AI-driven sophisticated data management and modeling are needed to quantify and interpret complex network AI/NN models. Models which include static and dynamic features, combined with sophisticated data reduction and individualized feature selection of the most relevant gait characteristics are needed to close the loop for this paradigm shift. Future work is warranted on a multidisciplinary level: to validate the clinical applicability and integration of the various sensing modalities, to ensure proper synchronization of the various systems for accurate continuous real-time monitoring, to develop and validate fast and reliable computational platforms, and to implement modular user-friendly interfaces easy to use in any environment.

In summary, instrumented gait analysis is a well-established tool for the quantitative assessment of gait dysfunction which could be effectively used for functional diagnosis, treatment/surgery/rehabilitation/planning, and progression monitoring for a wide spectrum of disease. The literature indicates that recent advancement in wearable technology and computationally advanced data analytics, including AI, can overcome the challenges of traditional gait labs, allowing for less costly, portable, and relatively simple gait testing protocols in clinical settings, as well as user-friendly data management, analysis, and interpretation computational platforms. On the other hand, the development of clinically driven standardized methodology and procedures is of paramount significance and remains largely unaddressed. These standardized practices should not only focus on quantitative gait diagnosis but should also incorporate sophisticated objective measures and 3-D dynamic gait profiles and markers for monitoring progress and outcome prediction and evaluation. Proper gait protocols should be devised and leveraged towards identifying gait characteristics that could be effectively used as early disease diagnostic markers. Importantly, training clinical teams at various levels, from doctors and surgeons to physiotherapists and other allied health professionals, on properly using these novel assessment and computational tools is equally important and warrants an equally rapid paradigm shift in training and practice in clinical settings towards patient-specific precise medicine.

Author contributions

AAH, MR and KK conceived the idea. AAH, DMM, KK and MR formulated the objective for this review. AAH designed the search strategy, conducted abstract screening and full text review, extracted the data, and drafted the manuscript. KK, NA, DMM, and AAH contributed to writing the manuscript. DMM and NA performed a part of the literature survey, including abstract screening, full text review, and data extraction. MR, and KK provided significant guidance on the content of the manuscript, overall supervision, and critical feedback. All authors contributed to the article and approved the submitted version.

This publication is based upon work supported by the HEIC at Khalifa University of Science and Technology.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

1. Mohan DM, Khandoker AH, Wasti SA, Ismail Ibrahim Ismail Alali S, Jelinek HF, Khalaf K. Assessment methods of post-stroke gait: a scoping review of technology-driven approaches to gait characterization and analysis. Front Neurol . (2021) 12:650024. doi: 10.3389/fneur.2021.650024

PubMed Abstract | CrossRef Full Text | Google Scholar

2. Celik Y, Stuart S, Woo WL, Godfrey A. Gait analysis in neurological populations: progression in the use of wearables. Med Eng Phys . (2021) 87:9–29. doi: 10.1016/j.medengphy.2020.11.005

3. Rathinam C, Bateman A, Peirson J, Skinner J. Observational gait assessment tools in paediatrics–a systematic review. Gait Posture . (2014) 40:279–85. S0966-6362(14)00478-0 24798609

PubMed Abstract | Google Scholar

4. Eastlack ME, Arvidson J, Snyder-Mackler L, Danoff JV, McGarvey CL. Interrater reliability of videotaped observational gait-analysis assessments. Phys Ther . (1991) 71:465–72. doi: 10.1093/ptj/71.6.465

5. Hebda-Boon A, Zhang B, Amankwah A, Shortland AP, Morrissey D. Clinicians’ experiences of instrumented gait analysis in management of patients with cerebral palsy: a qualitative study. Phys Occup Ther Pediatr . (2022) 42(4):403–15. doi: 10.1080/01942638.2022.2037808

6. Toro B, Nester C, Farren P. A review of observational gait assessment in clinical practice. null . (2003) 19:137–49. doi: 10.1080/09593980307964

CrossRef Full Text | Google Scholar

7. Franki I, Desloovere K, De Cat J, Feys H, Molenaers G, Calders P, et al. The evidence-base for basic physical therapy techniques targeting lower limb function in children with cerebral palsy: a systematic review using the international classification of functioning, disability and health as a conceptual framework. J Rehabil Med . (2012) 44:385–95. doi: 10.2340/16501977-0983

8. Wallmann HW. Introduction to observational gait analysis. Home Health Care Manag Pract . (2009) 22:66–8. doi: 10.1177/1084822309343277

9. Ferrarello F, Bianchi VA, Baccini M, Rubbieri G, Mossello E, Cavallini MC, et al. Tools for observational gait analysis in patients with stroke: a systematic review. Phys Ther . (2013) 93:1673–85. doi: 10.2522/ptj.20120344

10. Turani N, Kemiksizoğlu A, Karataş M, Ozker R. Assessment of hemiplegic gait using the Wisconsin gait scale. Scand J Caring Sci . (2004) 18:103–8. doi: 10.1111/j.1471-6712.2004.00262.x

11. Dasgupta P, VanSwearingen J, Godfrey A, Redfern M, Montero-Odasso M, Sejdic E. Acceleration gait measures as proxies for motor skill of walking: a narrative review. IEEE Trans Neural Syst Rehabil Eng . (2021) 29:249–61. doi: 10.1109/TNSRE.2020.3044260

12. Noseworthy JH. Clinical scoring methods for multiple sclerosis. Ann Neurol . (1994) 36:S80–5. doi: 10.1002/ana.410360718

13. Noseworthy JH, Vandervoort MK, Wong CJ, Ebers GC. Interrater variability with the expanded disability Status scale (EDSS) and functional systems (FS) in a multiple sclerosis clinical trial. The Canadian cooperation MS study group. Neurol . (1990) 40:971–5. doi: 10.1212/wnl.40.6.971

14. Roxburgh RH, Seaman SR, Masterman T, Hensiek AE, Sawcer SJ, Vukusic S, et al. Multiple sclerosis severity score: using disability and disease duration to rate disease severity. Neurol . (2005) 64:1144–51. doi: 10.1212/01.WNL.0000156155.19270.F8

15. Vienne-Jumeau A, Quijoux F, Vidal PP, Ricard D. Value of gait analysis for measuring disease severity using inertial sensors in patients with multiple sclerosis: protocol for a systematic review and meta-analysis. Syst Rev . (2019) 8:15. doi: 10.1186/s13643-018-0918-z

16. Balaji E, Brindha D, Elumalai VK, Umesh K. Data-driven gait analysis for diagnosis and severity rating of Parkinson's Disease. Med Eng Phys . (2021) 91:54–64. S1350-4533(21)00028-X 34074466

17. Mohan DM, Khandoker AH, Wasti SA, Ismail Ibrahim Ismail Alali S, Jelinek HF, Khalaf K. Assessment methods of post-stroke gait: a scoping review of technology-driven approaches to gait characterization and Analysis. Front Neurol . (2021) 12:650024. doi: 10.3389/fneur.2021.650024

18. Cappozzo A. Gait analysis methodology. Hum Mov Sci . (1984) 3:27–50. doi: 10.1016/0167-9457(84)90004-6

19. Wren TAL, Tucker CA, Rethlefsen SA, Gorton GE 3rd, Õunpuu S. Clinical efficacy of instrumented gait analysis: systematic review 2020 update. Gait Posture . (2020) 80:274–9. S0966-6362(20)30181-8 32563727

20. Saboor A, Kask T, Kuusik A, Alam MM, Le Moullec Y, Niazi IK, et al. Latest research trends in gait analysis using wearable sensors and machine learning: a systematic review. IEEE Access . (2020) 8:167830–64. doi: 10.1109/ACCESS.2020.3022818

21. Munn Z, Peters MDJ, Stern C, Tufanaru C, McArthur A, Aromataris E. Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC Med Res Methodol . (2018) 18:143. doi: 10.1186/s12874-018-0611-x

22. Winter DA. The biomechanics and motor control of humna gait: Normal, elderly and pathological . Waterloo: Ont.: Waterloo Biomechanics (1991).

23. Ali A, Sundaraj K, Ahmad B, Ahamed N, Islam A. Gait disorder rehabilitation using vision and non-vision based sensors: a systematic review. Bosnian Journal of Basic Medical Sciences . (2012) 12:193–202. doi: 10.17305/bjbms.2012.2484

24. Alam U, Riley DR, Jugdey RS, Azmi SS, Rajbhandari SS, D’août K, et al. Diabetic neuropathy and gait: a review. Diabetes Ther . (2017) 8:1253. doi: 10.1007/s13300-017-0295-y

25. Whittle MW. Clinical gait analysis: a review. Hum Mov Sci . (1996) 15:369–87. doi: 10.1016/0167-9457(96)00006-1

26. Fukuchi CA, Fukuchi RK, Duarte M. Effects of walking speed on gait biomechanics in healthy participants: a systematic review and meta-analysis. Syst Rev . (2019) 8:153. doi: 10.1186/s13643-019-1063-z

27. Snijders AH, van de Warrenburg BP, Giladi N, Bloem BR. Neurological gait disorders in elderly people: clinical approach and classification. Lancet Neurol . (2007) 6:63–74. S1474-4422(06)70678-0 [pii].17166803

28. Snijders AH, van de Warrenburg BP, Giladi N, Bloem BR. Neurological gait disorders in elderly people: clinical approach and classifi cation. Lancet Neurol . (2007) 6(1):63–74. doi: 10.1016/S1474-4422(06)70678-0

29. Intzandt B, Beck EN, Silveira CRA. The effects of exercise on cognition and gait in Parkinson's Disease: a scoping review. Neurosci Biobehav Rev . (2018) 95:136–69. doi: 10.1016/j.neubiorev.2018.09.018

30. Belda-Lois J, Mena-del Horno S, Bermejo-Bosch I, Moreno JC, Pons JL, Farina D, et al. Rehabilitation of gait after stroke: a review towards a top-down approach. J Neuroeng Rehabil . (2011) 8:66. doi: 10.1186/1743-0003-8-66

31. Bahl JS, Nelson MJ, Taylor M, Solomon LB, Arnold JB, Thewlis D. Biomechanical changes and recovery of gait function after total hip arthroplasty for osteoarthritis: a systematic review and meta-analysis. Osteoarthritis Cartilage . (2018) 26:847–63. S1063-4584(18)31013-6 [pii].29474993

32. Ewen AM, Stewart S, St Clair Gibson A, Kashyap SN, Caplan N. Post-operative gait analysis in total hip replacement patients—a review of current literature and meta-analysis. Gait Posture . (2012) 36:1–6. doi: 10.1016/j.gaitpost.2011.12.024

33. Kolk S, Minten MJ, van Bon GE, Rijnen WH, Geurts AC, Verdonschot N, et al. Gait and gait-related activities of daily living after total hip arthroplasty: a systematic review. Clin Biomech (Bristol, Avon) . (2014) 29:705–18. S0268-0033(14)00127-2 [pii].24951319

34. Kolk S, Minten MJM, van Bon GEA, Rijnen WH, Geurts ACH, Verdonschot N, et al. Gait and gait-related activities of daily living after total hip arthroplasty: a systematic review. Clin Biomech . (2014) 29:705–18. doi: 10.1016/j.clinbiomech.2014.05.008

35. Ladha C, Din S, Nazarpour K, Hickey A, Morris R, Catt M, et al. Toward a low-cost gait analysis system for clinical and free-living assessment. Annu Int Conf IEEE Eng Med Biol Soc . (2016) 2016:1874–7. doi: 10.1109/EMBC.2016.7591086. PMID: 28268692

36. Patterson KK, Gage WH, Brooks D, Black SE, McIlroy WE. Evaluation of gait symmetry after stroke: a comparison of current methods and recommendations for standardization. Gait Posture . (2010) 31:241–6. doi: 10.1016/j.gaitpost.2009.10.014

37. Patterson KK, Gage WH, Brooks D, Black SE, McIlroy WE. Changes in gait symmetry and velocity after stroke: a cross-sectional study from weeks to years after stroke. Neurorehabil Neural Repair . (2010) 24:783–90. doi: 10.1177/1545968310372091

38. Laudanski A, Brouwer B, Li Q. Measurement of lower limb joint kinematics using inertial sensors during stair ascent and descent in healthy older adults and stroke survivors. J Healthc Eng . (2013) 4:555–76. doi: 10.1260/2040-2295.4.4.555

39. Rastegarpanah A, Scone T, Saadat M, Rastegarpanah M, Taylor SJ, Sadeghein N. Targeting effect on gait parameters in healthy individuals and post-stroke hemiparetic individuals. J Rehabil Assist Technol Eng . (2018) 5:2055668318766710. doi: 10.1177/2055668318766710

40. Latorre J, Colomer C, Alcañiz M, Llorens R. Gait analysis with the kinect v2: normative study with healthy individuals and comprehensive study of its sensitivity, validity, and reliability in individuals with stroke. J Neuroeng Rehabil . (2019) 16:97. doi: 10.1186/s12984-019-0568-y

41. Iosa M, Bini F, Marinozzi F, Fusco A, Morone G, Koch G, et al. Stability and harmony of gait in patients with subacute stroke. J Med Biol Eng . (2016) 36:635–43. doi: 10.1007/s40846-016-0178-0 [pii].27853414

42. Sousa ASP, Silva A, Santos R, Sousa F, Tavares JMRS. Interlimb coordination during the stance phase of gait in subjects with stroke. Arch Phys Med Rehabil . (2013) 94:2515–22. S0003-9993(13)00537-6 [pii].23871877

43. Li M, Tian S, Sun L, Chen X. Gait analysis for post-stroke hemiparetic patient by multi-features fusion method. Sens (Basel) . (2019) 19(7):1737. doi: 10.3390/s19071737

44. Krishnan V, Khoo I, Marayong P, DeMars K, Cormack J. Gait training in chronic stroke using walk-even feedback device: a pilot study. Neurosci J . (2016) 2016:1–8. doi: 10.1155/2016/6808319

45. Ardalan A, Yamane N, Rao AK, Montes J, Goldman S. Analysis of gait synchrony and balance in neurodevelopmental disorders using computer vision techniques. Health Informatics J . (2021) 27:14604582211055650. doi: 10.1177/14604582211055650

46. Rupprechter S, Morinan G, Peng Y, Foltynie T, Sibley K, Weil RS, et al. A clinically interpretable computer-vision based method for quantifying gait in Parkinson's Disease. Sens (Basel) . (2021) 21:5437. doi: 10.3390/s21165437

47. Li Y, Zhang P, Zhang Y, Miyazaki K. 2019 41st annual international conference of the IEEE engineering in medicine and biology society (EMBC). Gait analysis using stereo camera in daily environment (2019). p. 1471–5

48. Kessler SE, Rainbow MJ, Lichtwark GA, Cresswell AG, D'Andrea SE, Konow N, et al. A direct comparison of biplanar videoradiography and optical motion capture for foot and ankle kinematics. Front Bioeng Biotechnol . (2019) 7:199. doi: 10.3389/fbioe.2019.00199

49. Tranberg R, Karlsson D. The relative skin movement of the foot: a 2-D roentgen photogrammetry study. Clin Biomech (Bristol, Avon) . (1998) 13:71–6. S0268003397000521 [pii].11415773

50. Nester C, Jones RK, Liu A, Howard D, Lundberg A, Arndt A, et al. Foot kinematics during walking measured using bone and surface mounted markers. J Biomech . (2007) 40:3412–23. doi: 10.1016/j.jbiomech.2007.05.019

51. Colyer SL, Evans M, Cosker DP, Salo AIT. A review of the evolution of vision-based motion analysis and the integration of advanced computer vision methods towards developing a markerless system. Sports Med Open . (2018) 4:24. doi: 10.1186/s40798-018-0139-y

52. Baker R. Gait analysis methods in rehabilitation. J Neuroeng Rehabil . (2006) 3:4. doi: 10.1186/1743-0003-3-4

53. Stagni R, Fantozzi S, Cappello A, Leardini A. Quantification of soft tissue artefact in motion analysis by combining 3D fluoroscopy and stereophotogrammetry: a study on two subjects. Clin Biomech (Bristol, Avon) . (2005) 20:320–9. doi: 10.1016/j.clinbiomech.2004.11.012

54. McHenry BD, Exten E, Long JT, Harris GF. Sagittal fluoroscopy for the assessment of hindfoot kinematics. J Biomech Eng . (2016) 138:4032445. doi: 10.1115/1.4032445

55. Roach KE, Foreman KB, MacWilliams BA, Karpos K, Nichols J, Anderson AE. The modified shriners hospitals for children Greenville (mSHCG) multi-segment foot model provides clinically acceptable measurements of ankle and midfoot angles: a dual fluoroscopy study. Gait Posture . (2021) 85:258–65. doi: 10.1016/j.gaitpost.2021.02.004

56. Hitz M, Schütz P, Angst M, Taylor WR, List R. Influence of the moving fluoroscope on gait patterns. PLoS One . (2018) 13:e0200608. doi: 10.1371/journal.pone.0200608

57. Roach KE, Wang B, Kapron AL, Fiorentino NM, Saltzman CL, Bo Foreman K, et al. In vivo kinematics of the tibiotalar and subtalar joints in asymptomatic subjects: a high-speed dual fluoroscopy study. J Biomech Eng . (2016) 138:910061. doi: 10.1115/1.4034263

58. Atkins PR, Fiorentino NM, Hartle JA, Aoki SK, Peters CL, Foreman KB, et al. In vivo pelvic and hip joint kinematics in patients with cam femoroacetabular impingement syndrome: a dual fluoroscopy study. J Orthop Res . (2020) 38:823–33. doi: 10.1002/jor.24509

59. Battaglia S, Belvedere C, Jaber SA, Affatato S, D'Angeli V, Leardini A. A new protocol from real joint motion data for wear simulation in total knee arthroplasty: stair climbing. Med Eng Phys . (2014) 36:1605–10. doi: 10.1016/j.medengphy.2014.08.010

60. Belvedere C, Cadossi M, Mazzotti A, Giannini S, Leardini A. Fluoroscopic and gait analyses for the functional performance of a custom-made total talonavicular replacement. J Foot Ankle Surg . (2017) 56:836–44. doi: 10.1053/j.jfas.2017.02.004

61. Uemura K, Atkins PR, Anderson AE. The effect of using different coordinate systems on in-vivo hip angles can be estimated from computed tomography images. J Biomech . (2019) 95:109318. doi: 10.1016/j.jbiomech.2019.109318

62. Wearing SC, Urry S, Perlman P, Smeathers J, Dubois P. Sagittal plane motion of the human arch during gait: a videofluoroscopic analysis. Foot Ankle Int . (1998) 19:738–42. doi: 10.1177/107110079801901105

63. Li G, Kozanek M, Hosseini A, Liu F, Van de Velde SK, Rubash HE. New fluoroscopic imaging technique for investigation of 6DOF knee kinematics during treadmill gait. J Orthop Surg Res . (2009) 4:6. doi: 10.1186/1749-799X-4-6

64. Iaquinto JM, Tsai R, Haynor DR, Fassbind MJ, Sangeorzan BJ, Ledoux WR. Marker-based validation of a biplane fluoroscopy system for quantifying foot kinematics. Med Eng Phys . (2014) 36:391–6. doi: 10.1016/j.medengphy.2013.08.013

65. Sheehan FT, Seisler AR, Siegel KL. In vivo talocrural and subtalar kinematics: a non-invasive 3D dynamic MRI study. Foot Ankle Int . (2007) 28:323–35. doi: 10.3113/FAI.2007.0323

66. Tao W, Liu T, Zheng R, Feng H. Gait analysis using wearable sensors. Sens (Basel, Switzerland) . (2012) 12:2255–83. doi: 10.3390/s120202255

67. Ihlen EAF, Weiss A, Beck Y, Helbostad JL, Hausdorff JM. A comparison study of local dynamic stability measures of daily life walking in older adult community-dwelling fallers and non-fallers. J Biomech . (2016) 49:1498–503. doi: 10.1016/j.jbiomech.2016.03.019

68. Bugané F, Benedetti MG, Casadio G, Attala S, Biagi F, Manca M, et al. Estimation of spatial-temporal gait parameters in level walking based on a single accelerometer: validation on Normal subjects by standard gait analysis. Comput Methods Programs Biomed . (2012) 108:129–37. doi: 10.1016/j.cmpb.2012.02.003

69. Rispens SM, Pijnappels M, van Schooten KS, Beek PJ, Daffertshofer A, van Dieën JH. Consistency of gait characteristics as determined from acceleration data collected at different trunk locations. Gait Posture . (2014) 40:187–92. doi: 10.1016/j.gaitpost.2014.03.182

70. Jarchi D, Pope J, Lee TKM, Tamjidi L, Mirzaei A, Sanei S. A review on accelerometry-based gait analysis and emerging clinical applications. IEEE Rev Biomed Eng . (2018) 11:177–94. doi: 10.1109/RBME.2018.2807182

71. Felix P, Figueiredo J, Santos C, Moreno J. ADAPTIVE REAL-TIME TOOL FOR HUMAN GAIT EVENT DETECTION USING A WEARABLE GYROSCOPE. (2017). 653 p.

72. Nasr A, Nadeem T. A Novel Technique for Gait Analysis Using Two Waist Mounted Gyroscopes. In: Anonymous - 2019 IEEE Global Communications Conference (GLOBECOM); (2019). p. 1-6.

73. Kun L, Inoue Y, Shibata K, Enguo C. Ambulatory estimation of knee-joint kinematics in anatomical coordinate system using accelerometers and magnetometers. IEEE Trans on Biomed Eng . (2011) 58:435–42. doi: 10.1109/TBME.2010.2089454

74. Rodríguez-Martín D, Pérez-López C, Samà A, Cabestany J, Català A. A wearable inertial measurement unit for long-term monitoring in the dependency care area. Sens . (2013) 13(10):14079–104. doi: 10.3390/s131014079

75. Seel T, Raisch J, Schauer T. IMU-Based Joint angle measurement for gait analysis. Sens . (2014) 14(4):6891–909. doi: 10.3390/s140406891

76. Teufl W, Lorenz M, Miezal M, Taetz B, Fröhlich M, Bleser G. Towards inertial sensor based Mobile gait analysis: event-detection and spatio-temporal parameters. Sens . (2019) 19(1):38. doi: 10.3390/s19010038

77. Hwang T-H, Reh J, Effenberg AO, Blume H. Real-Time gait analysis using a single head-worn inertial measurement unit. IEEE Trans on Cons Elec . (2018) 64:240–8. doi: 10.1109/TCE.2018.2843289

78. Parisi F, Ferrari G, Baricich A, D'Innocenzo M, Cisari C, Mauro A. Accurate gait analysis in post-stroke patients using a single inertial measurement unit. In: Anonymous - 2016 IEEE 13th International Conference on Wearable and Implantable Body Sensor Networks (BSN); (2016). p. 335-340).

79. Karatsidis A, Bellusci G, Schepers HM, De Zee M, Andersen MS, Veltink PH. Estimation of ground reaction forces and moments during gait using only inertial motion capture. Senss . (2017) 17(1):75. doi: 10.3390/s17010075

80. Sant’ Anna A, Wickström N, Eklund H, Zügner R, Tranberg R. Assessment of gait symmetry and gait normality using inertial sensors: in-lab and in-situ evaluation . Berlin, Heidelberg: Springer (2013).

81. Mariani B, Hoskovec C, Rochat S, Büla C, Penders J, Aminian K. 3D Gait assessment in young and elderly subjects using foot-worn inertial sensors. J Biomech . (2010) 43:2999–3006. doi: 10.1016/j.jbiomech.2010.07.003

82. Razak AHA, Zayegh A, Begg RK, Wahab Y. Foot plantar pressure measurement system: a review. Sens (Basel, Switzerland) . (2012) 12:9884–912. doi: 10.3390/s120709884

83. Novel® Pedar® System [Internet]. []. Available at: https://www.novel.de/products/pedar/

84. Tekscan® F-Scan® System.

85. Leitch KM, Birmingham TB, Jones IC, Giffin JR, Jenkyn TR. In-shoe plantar pressure measurements for patients with knee osteoarthritis: reliability and effects of lateral heel wedges. Gait Posture . (2011) 34:391–6. doi: 10.1016/j.gaitpost.2011.06.008

86. Malvade PS, Joshi AK, Madhe SP. In-sole Shoe Foot Pressure Monitoring for Gait Analysis. In: Anonymous - 2017 International Conference on Computing, Communication, Control and Automation (ICCUBEA); (2017). p. 1-4.

87. Crea S, Donati M, De Rossi S, Marco Maria Oddo CM, Vitiello N. A wireless flexible sensorized insole for gait analysis. Sens (Basel, Switzerland) . (2014) 14:1073–93. doi: 10.3390/s140101073

88. Mancinelli C, Patel S, Deming LC, Nimec D, Chu JJ, Beckwith J, et al. A novel sensorized shoe system to classify gait severity in children with cerebral palsy. Annu Int Conf IEEE Eng Med Biol Soc . (2012) 2012:5010–3. doi: 10.1109/EMBC.2012.6347118

89. Jagos H, Pils K, Haller M, Wassermann C, Chhatwal C, Rafolt D, et al. Mobile gait analysis via eSHOEs instrumented shoe insoles: a pilot study for validation against the gold standard GAITRite(®). J Med Eng Technol . (2017) 41:375–86. doi: 10.1080/03091902.2017.1320434

90. Arafsha F, Hanna C, Aboualmagd A, Fraser S, El Saddik A. Instrumented wireless SmartInsole system for Mobile gait analysis: a validation pilot study with tekscan strideway. J Sens and Actuator Netw . (2018) 7(3):36. doi: 10.3390/jsan7030036

91. Nagano H, Begg RK. Shoe-Insole technology for injury prevention in walking. Sens (Basel) . (2018) 18:1468. doi: 10.3390/s18051468

92. Lin F, Wang A, Zhuang Y, Tomita MR, Xu W. Smart insole: a wearable sensor device for unobtrusive gait monitoring in daily life. IEEE Trans on Indust Informat . (2016) 12:2281–91. doi: 10.1109/TII.2016.2585643

93. Lee M, Ryu J, Youn I. Biometric personal identification based on gait analysis using surface EMG signals. In: Anonymous - 2017 2nd IEEE International Conference on Computational Intelligence and Applications (ICCIA); (2017). p. 318-321).

94. Strazza A, Mengarelli A, Fioretti S, Burattini L, Agostini V, Knaflitz M, et al. Surface-EMG analysis for the quantification of thigh muscle dynamic co-contractions during Normal gait. Gait Posture . (2017) 51:228–33. doi: 10.1016/j.gaitpost.2016.11.003

95. Peters D, O’Brien E, Kamrud K, Roberts S, Rooney T, Thibodeau K, et al. Utilization of wearable technology to assess gait and mobility post-stroke: a systematic review. J Neuroeng Rehabil . (2021) 18(1):67. doi: 10.1186/s12984-021-00863-x

96. Lu R, Xu Y, Li X, Fan Y, Zeng W, Tan Y, et al. Evaluation of wearable sensor devices in Parkinson's Disease: a review of current Status and future prospects. Parkinson's Disease . (2020) 2020:4693019. doi: 10.1155/2020/4693019

97. Alaqtash M, Yu H, Brower R, Abdelgawad A, Sarkodie-Gyan T. Application of wearable sensors for human gait analysis using fuzzy computational algorithm. Eng Appl Artif Intell . (2011) 24:1018–25. doi: 10.1016/j.engappai.2011.04.010

98. McCamley J, Donati M, Grimpampi E, Mazzà C. An enhanced estimate of initial contact and final contact instants of time using lower trunk inertial sensor data. Gait Posture . (2012) 36:316–8. doi: 10.1016/j.gaitpost.2012.02.019

99. Felix P, Figueiredo J, Santos C, Moreno J. ADAPTIVE REAL-TIME TOOL FOR HUMAN GAIT EVENT DETECTION USING A WEARABLE GYROSCOPE. (2017). 653 p).

100. Boutaayamou M, Schwartz C, Stamatakis J, Denoël V, Maquet D, Forthomme B, et al. Development and validation of an accelerometer-based method for quantifying gait events. Med Eng Phys . (2015) 37:226–32. doi: 10.1016/j.medengphy.2015.01.001

101. Allseits E, Kim KJ, Bennett C, Gailey R, Gaunaurd I, Agrawal V. A novel method for estimating knee angle using two leg-mounted gyroscopes for continuous monitoring with Mobile health devices. Sens (Basel) . (2018) 18:2759. doi: 10.3390/s18092759

102. Nukala BT, Nakano T, Rodriguez A, Tsay J, Lopez J, Nguyen TQ, et al. Real-Time classification of patients with balance disorders vs. Normal subjects using a low-cost small wireless wearable gait sensor. Biosens . (2016) 6(4):58. doi: 10.3390/bios6040058

103. Rastegari E, Azizian S, Ali H. Machine Learning and Similarity Network Approaches to Support Automatic Classification of Parkinson's Diseases Using Accelerometer-based Gait Analysis. (2019).).

104. Khera P, Kumar N. Role of machine learning in gait analysis: a review. J Med Eng Technol . (2020) 44:441–67. doi: 10.1080/03091902.2020.1822940

105. Zhao H, Wang Z, Qiu S, Shen Y, Wang J. IMU-based gait analysis for rehabilitation assessment of patients with gait disorders . In: Anonymous (2017). 622–6.

106. Butz C, Iske C, Truba N, Trott K. Treatment of functional gait abnormality in a rehabilitation setting: emphasizing the physical interventions for treating the whole child. Innov Clin Neurosci . (2019) 16:18–21.31832259

107. di Biase L, Di Santo A, Caminiti ML, De Liso A, Shah SA, Ricci L, et al. Gait analysis in Parkinson's Disease: an overview of the most accurate markers for diagnosis and symptoms monitoring. Sens (Basel) . (2020) 20:3529. doi: 10.3390/s20123529

108. Pistacchi M, Gioulis M, Sanson F, De Giovannini E, Filippi G, Rossetto F, et al. Gait analysis and clinical correlations in early Parkinson's Disease. Funct Neurol . (2017) 32:28–34. doi: 10.11138/FNeur/2017.32.1.028

109. Mirelman A, Bonato P, Camicioli R, Ellis TD, Giladi N, Hamilton JL, et al. Gait impairments in Parkinson's Disease. Lancet Neurol . (2019) 18:697–708. doi: 10.1016/S1474-4422(19)30044-4

110. Aminian K, Trevisan C, Najafi B, Dejnabadi H, Frigo C, Pavan E, et al. Evaluation of an ambulatory system for gait analysis in hip osteoarthritis and after total hip replacement. Gait Posture . (2004) 20:102–7. doi: 10.1016/S0966-6362(03)00093-6

111. Yen SC, Schmit BD, Wu M. Using swing resistance and assistance to improve gait symmetry in individuals post-stroke. Hum Mov Sci . (2015) 42:212–24. doi: 10.1016/j.humov.2015.05.010

112. Bohannon RW. Gait performance of hemiparetic stroke patients: selected variables. Arch Phys Med Rehabil . (1987) 68:777–81.3675175

113. Nadeau S, Betschart M, Bethoux F. Gait analysis for poststroke rehabilitation: the relevance of biomechanical analysis and the impact of gait speed. Phys Med Rehabil Clin N Am . (2013) 24:265–76. doi: 10.1016/j.pmr.2012.11.007

114. Moseley A, Wales A, Herbert R, Schurr K, Moore S. Observation and analysis of hemiplegic gait: stance phase. Aust J Physiother . (1993) 39:259–67. doi: 10.1016/S0004-9514(14)60486-4

115. von Schroeder HP, Coutts RD, Lyden PD, Billings E Jr, Nickel VL. Gait parameters following stroke: a practical assessment. J Rehabil Res Dev . (1995) 32:25–31.7760264

116. Schülein S, Barth J, Rampp A, Rupprecht R, Eskofier BM, Winkler J, et al. Instrumented gait analysis: a measure of gait improvement by a wheeled walker in hospitalized geriatric patients. J Neuroeng Rehabil . (2017) 14:18. doi: 10.1186/s12984-017-0228-z

117. Melese H, Alamer A, Hailu Temesgen M, Kahsay G. Effectiveness of exercise therapy on gait function in diabetic peripheral neuropathy patients: a systematic review of randomized controlled trials. Diabetes Metab Syndr Obes . (2020) 13:2753–64. doi: 10.2147/DMSO.S261175

118. Prajapati SK, Gage WH, Brooks D, Black SE, McIlroy WE. A novel approach to ambulatory monitoring: investigation into the quantity and control of everyday walking in patients with subacute stroke. Neurorehabil Neural Repair . (2011) 25:6–14. doi: 10.1177/1545968310374189

119. Hsu C, Tsai Y, Yau C, Shie H, Wu C. Test-Retest reliability of an automated infrared-assisted trunk accelerometer-based gait analysis system. Sens . (2016) 16(8):1156. doi: 10.3390/s16081156

120. Byun S, Han JW, Kim TH, Kim KW. Test-Retest reliability and concurrent validity of a single tri-axial accelerometer-based gait analysis in older adults with Normal cognition. PLoS One . (2016) 11:e0158956. doi: 10.1371/journal.pone.0158956

121. Poitras I, Dupuis F, Bielmann M, Campeau-Lecours A, Mercier C, Bouyer LJ, et al. Validity and reliability of wearable sensors for joint angle estimation: a systematic review. Sens . (2019) 19(7):1555. doi: 10.3390/s19071555

122. Ro DH, Lee J, Lee J, Park JY, Han HS, Lee MC. Effects of knee osteoarthritis on hip and ankle gait mechanics. Adv Orthop . (2019) 2019:9757369. doi: 10.1155/2019/9757369

123. Viteckova S, Kutilek P, Svoboda Z, Krupicka R, Kauler J, Szabo Z. Gait symmetry measures: a review of current and prospective methods. Biomed Signal Process Control . (2018) 42:89–100. doi: 10.1016/j.bspc.2018.01.013

124. Cimolin V, Galli M. Summary measures for clinical gait analysis: a literature review. Gait Posture . (2014) 39:1005–10. doi: 10.1016/j.gaitpost.2014.02.001

Keywords: clinical gait assessment, gait technologies, gait measures, mobile gait lab, gait pathologies

Citation: Hulleck AA, Menoth Mohan D, Abdallah N, El Rich M and Khalaf K (2022) Present and future of gait assessment in clinical practice: Towards the application of novel trends and technologies. Front. Med. Technol. 4:901331. doi: 10.3389/fmedt.2022.901331

Received: 21 March 2022; Accepted: 17 November 2022; Published: 16 December 2022.

Reviewed by:

© 2022 Hulleck, Menoth Mohan, Abdallah, El Rich and Khalaf. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

* Correspondence: Kinda Khalaf [email protected]

Specialty Section: This article was submitted to Diagnostic and Therapeutic Devices, a section of the journal Frontiers in Medical Technology

Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

View all journals
Explore content
About the journal
Publish with us
Sign up for alerts
Open access
Published: 25 August 2022

Biometric recognition through gait analysis

Claudia Álvarez-Aparicio 1 ,
Ángel Manuel Guerrero-Higueras 1 ,
Miguel Ángel González-Santamarta 1 ,
Adrián Campazas-Vega 1 ,
Vicente Matellán 1 &
Camino Fernández-Llamas 1

Scientific Reports volume 12 , Article number: 14530 ( 2022 ) Cite this article

5153 Accesses

10 Citations

24 Altmetric

Metrics details

Engineering
Mathematics and computing

The use of people recognition techniques has become critical in some areas. For instance, social or assistive robots carry out collaborative tasks in the robotics field. A robot must know who to work with to deal with such tasks. Using biometric patterns may replace identification cards or codes on access control to critical infrastructures. The usage of Red Green Blue Depth (RGBD) cameras is ubiquitous to solve people recognition. However, this sensor has some constraints, such as they demand high computational capabilities, require the users to face the sensor, or do not regard users’ privacy. Furthermore, in the COVID-19 pandemic, masks hide a significant portion of the face. In this work, we present BRITTANY, a biometric recognition tool through gait analysis using Laser Imaging Detection and Ranging (LIDAR) data and a Convolutional Neural Network (CNN). A Proof of Concept (PoC) has been carried out in an indoor environment with five users to evaluate BRITTANY. A new CNN architecture is presented, allowing the classification of aggregated occupancy maps that represent the people’s gait. This new architecture has been compared with LeNet-5 and AlexNet through the same datasets. The final system reports an accuracy of 88%.

OPERAnet , a multimodal activity recognition dataset acquired from radio frequency and vision-based sensors

Enhancing gait recognition by multimodal fusion of mobilenetv1 and xception features via PCA for OaA-SVM classification

A semi-automatic toolbox for markerless effective semantic feature extraction

Introduction.

User identification has become increasingly important in different research areas. For example, in the field of cybersecurity, to prevent access to critical facilities or in social and assistive robotics to improve Human-Robot Interaction (HRI).

In the cybersecurity field, it is necessary to take into account that the number of cyber-attacks in different environments has increased exponentially in recent years 1 , 2 . On access control to critical infrastructures, cybercriminals have many techniques to gain access to facilities either remotely, by obtaining credentials, exploiting vulnerabilities in some of the systems, or physically, by breaching access to the facilities 3 , 4 . If the access is done remotely, they can be detected in the system when they start to perform malicious actions such as privilege escalation or lateral movements. These techniques are detected by analysing network traffic, system logs, or the system’s overall behaviour. Nevertheless, if the access is physical, an attack named tailgating 5 , a cybercriminal only could be detected by building employees or security personnel if it exists. Thus, more and more companies nowadays implement biometric systems in their infrastructure. The standard Radio Frequency Identification (RFID) cards or passcodes have become outdated.

In robotics, the tasks that a mobile robot has to accomplish are becoming more complex, specifically if we focus on social or assistive robotics. These complex tasks are usually divided into little skills that the robot can solve. First, a robot needs to know its location in the environment where it operates 6 —this is known as localization. Then, it needs to calculate the best path avoiding obstacles or damaging people or objects in their trajectory 7 —aka navigation. Finally, the robot interacts with people and sometimes works with them on specific tasks 8 —aka Human-Robot Interaction (HRI). There are robust solutions for the first two skills because these areas have been extensively studied in the literature. The third one is probably the most complex skill, so many researchers are currently working on it. Human-Robot Interaction (HRI) has to be as similar as possible to human-human interaction 9 . Interaction refers to collision avoidance, but it also involves approaching skills or communication. Such elements are associated with two essential skills: people tracking and recognition. It is necessary to know where the people are every time, but it is also crucial to know who they are.

Tracking people helps to improve navigation skills in mobile robots and promote socially acceptable robots 10 . Many solutions in the literature attempt to solve this problem with Red Green Blue Depth (RGBD) cameras to detect people in the environments as shown in 11 . Other researchers combine data from both Laser Imaging Detection and Ranging (LIDAR) and Red Green Blue Depth (RGBD) cameras 12 , 13 , 14 . However, these approaches have a high computing demand, and it may be a drawback if the tracking runs onboard a robot. Therefore, some solutions have been proposed in the literature to solve the tracking problem using 2D Laser Imaging Detection and Ranging (LIDAR) sensors. For example, the work 15 reviewed different methods for robot navigation in crowded indoor environments. Furthermore, the clustering and centre point estimation combined with the walking centre line estimation is used to detect people with or without a walker 16 .

The above approaches are not robust enough when dealing with occlusions or changes in gait speed. To address such issues, the authors proposed a Convolutional Neural Network (CNN)-based tool that allows for locating people within the robot surroundings using the data provided by a single Laser Imaging Detection and Ranging (LIDAR) sensor 17 . This tool, called People Tracking (PeTra), creates an occupancy map from the Laser Imaging Detection and Ranging (LIDAR) sensor’s readings. Maps are processed by a Convolutional Neural Network (CNN) which returns segmented data belonging to the people in the robot’s surroundings. A centre-of-mass calculation provides the people’s location estimates from the segmented data. Several versions of the People Tracking (PeTra) have been released. Then, it was included a correlation method of location estimates using Kalman filters, as well as an optimization for the Convolutional Neural Network (CNN) 18 . Finally, a bootstrapping method was proposed to improve the accuracy of the tool in specific locations 19 . People Tracking (PeTra) is the tool selected to detect people in this work.

People recognition is a hot topic. It is not only required in mobile robotics, but also prevalent in our daily life. The use of biometry technologies is very extended. Biometry uses information about a specific part of the human body or behaviour, allowing us to distinguish people by analyzing such data 20 . Biometric technology splits into two main groups. The physiological biometric technologies analyze a specific feature of the body 21 . It is a well-known method since a significant part of society uses it daily. Within this group of techniques, we can find fingerprint 22 , facial 23 , or iris 24 identification. On the other hand, behavioural biometric technologies analyze actions carried out by people 25 . In this second group, a time component is required since any action has a beginning, developing, and ending 26 . Within this group, we can find voice 27 , hand-writing 28 , or gait identification 29 . The last mentioned work presents a review of the methods used to capture gait information based on vision, sound, pressure, and accelerometry. Gait features can be extracted from a sequence of visual images or video, an underfoot pressure image sequence obtained using a pressure mat sensor, an acceleration trace recorded by an accelerometer in a wearable device, or an audio recording.

As mentioned above, a robot must recognize the person interacting with to promote socially acceptable robots. In robotics, most studies use physiological biometric technologies—specifically face recognition using a camera 30 . However, this sensor has a critical drawback—data processing has a higher computational cost than other sensors. Furthermore, on face recognition, people should face the camera constantly. In addition, the worldwide COVID-19 situation and the consequent masks make this method even more complex. To meet such issues, some authors propose a multi-modal biometric identification combining face and voice identification 31 , 32 . Finally, it is essential to point out that the use of cameras bounds the users’ privacy 33 . Regarding Laser Imaging Detection and Ranging (LIDAR) sensors, several proposals use these sensors to perform human detection and tracking. This sensor solves the user privacy problem as well as the usability of the system, since the user does not have to constantly look at the robot. Recent studies analyze people’s gait using 2D and 3D Laser Imaging Detection and Ranging (LIDAR) sensors to improve the tracking of individuals.

The work 34 proposes an Long Short-Term Memory (LSTM)-based method for gait recognition using a multi-line Laser Imaging Detection and Ranging (LIDAR) sensor. The study proposes to create a silhouette from the point cloud extracted from the multi-line LIDAR and process it through an LTSM-based CNN network. The experimental results revealed that the proposed approach achieved a 60% of performance classification for 30 people.

Benedek et al. 35 presents two approaches, one for gait analysis and the other one for activity analysis, both based on data streams of a Rotating Multi-Beam (RMB) Laser Imaging Detection and Ranging (LIDAR) sensor. The gait analysis is used to do the person re-identification during tracking and recognition of specific activity patterns. They use a silhouette-based approach to projecting the 3D point cloud of a person obtained through the Rotating Multi-Beam (RMB) Laser Imaging Detection and Ranging (LIDAR) sensor to an image 2D plane. This image is preprocessed and evaluated through a Convolutional Neural Network (CNN). The experimental results revealed that the proposed approach achieved an average of 87% of performance classification for 28 people. Finally, a proposal to preserve the user’s privacy, the work 36 , uses a 2D Laser Imaging Detection and Ranging (LIDAR) sensor located at ankle level for people tracking. The study also performed a gait analysis to get person height estimation using Laser Imaging Detection and Ranging (LIDAR) data.

In this work, we present Biometric RecognITion Through gAit aNalYsis (BRITTANY), a tool that allows for identifying people through gait analysis using a 2D Laser Imaging Detection and Ranging (LIDAR) sensor and a Convolutional Neural Network (CNN) to process its data. A 2D Laser Imaging Detection and Ranging (LIDAR) sensor was selected because of its low computational requirements and its benefits for users’ privacy. The gait analysis has been chosen because it is a behavioural biometry method. Physiological biometry-based systems only process data gathered in a specific time instant. On the other hand, behavioural biometry-based strategies collect input data during a time interval. It allows for analysing more than one sample of user data, so they get more robust evidence without using any other external identification source.

The remainder of the paper organises as follows: “ Materials and methods ” section describes the materials and evaluation methods used to carry out our research; the results are presented and discussed in “ Results and discussion ” section; finally, conclusions and future works are proposed in “ Conclusion ” section.

( a ) Leon@Home Testbed. ( b ) Orbi-One Robot. ( c ) Leon@Home Testbed schema. The robot icons on the map point out the robot’s location and orientation during the experiments; the arrows show the trajectory of people.

Materials and methods

A set of experiments were carried out to evaluate Biometric RecognITion Through gAit aNalYsis (BRITTANY). In this section, the main elements of the research are in-depth depicted. Besides, we describe the methodology used to assess the accuracy of the proposed system.

Leon@Testbed

The experiments have been done in the mock-up apartment Leon@Home Testbed 37 , a certified testbed 38 of the European Robotics League (ERL) located in the Robotics Group’s laboratory at the University of Leon—see Fig. 1 a. It is used to test mobile service robots in a realistic environment. The apartment simulates a single-room home built in an 8 m $\times$ 7 m space. 60 cm-high walls—for allowing seeing—split the space into a living room, kitchen, bedroom, and bathroom.

Orbi-One robot

Orbi-One, the mobile service robot shown in Fig. 1 b has been used to gather data. It is manufactured by Robotnik 39 . It accommodates several sensors, such as a Red Green Blue Depth (RGBD) camera, a Laser Imaging Detection and Ranging (LIDAR) sensor, and an inertial unit. It also operates a six-degrees-freedom manipulator arm attached to its torso and a wheeled base for moving around the floor. Inside, an Intel Core i7 CPU with 8 GB of RAM allows it to run the software to control the robot hardware. The software runs the Robot Operating System (ROS) framework 40 . Specifically, we used its onboard Laser Imaging Detection and Ranging (LIDAR) sensor to collect data for our research.

People Tracking (PeTra) 17 is developed by the Robotics Group at the University of León 41 in recent years. People Tracking (PeTra) allows for locating people in the environment. It uses data provided by a Laser Imaging Detection and Ranging (LIDAR) sensor accommodated 20cm above the floor. It was evaluated by using an open-access dataset 42 . Starting from Laser Imaging Detection and Ranging (LIDAR) data People Tracking (PeTra) builds a 2D occupancy map that depicts the main features of the robot environment. Such occupancy map is then processed by a Convolutional Neural Network (CNN) that returns a new occupancy map containing only the points that belong to people. People Tracking (PeTra)’s Convolutional Neural Network (CNN) is based on the U-net architecture 43 , commonly used to perform biomedical image segmentation 44 . A centre-mass algorithm is computed over the new occupancy map to estimate the people’s location. New versions of the tool were released in later years. People Tracking (PeTra) can correlate location estimates by using Euclidean distances 45 or using Kalman filters 46 —a more robust correlation method. An optimized design for the Convolutional Neural Network (CNN) that allows People Tracking (PeTra) for working in real-time 18 . Finally, a bootstrapping-based method that improves the accuracy at specific locations, such as empty rooms or corridors 19 .

Biometric RecognITion Through gAit aNalYsis (BRITTANY) allows for recognizing people by their gait. The system is based on a Convolutional Neural Network (CNN) which uses an aggregation of occupancy maps provided by People Tracking (PeTra) as input. We pose that such aggregated occupancy maps are unique for each person and may be used to identify them.

Aggregated occupancy maps are processed to get probability values for each legitimate user. For instance, for five people, Biometric RecognITion Through gAit aNalYsis (BRITTANY) might get [0.01, 0.96, 0.2, 0.24, 0.09] probability values, meaning that input data belongs to the first person with a 0.01 probability, to the second one with a 0.96 probability, to the third one with a 0.2 probability, to the fourth one with a 0.24 probability, and the fifth one with a 0.09 probability. Thus, we might assert that the input data belongs to the second person.

We consider several predictions during a time interval to prevent punctual errors. Thus, we evaluate the final estimation through a set of predictions by applying a most-voted item strategy.

Data gathering

Two datasets, available online 47 , have been compiled. First, we gathered data in the mock-up apartment at Leon@Home Testbed described in “ Leon@Testbed ” section. Data was collected from the Laser Imaging Detection and Ranging (LIDAR) sensor onboard the Orbi-One robot mentioned in “ Orbi-One robot ” section. Both datasets consist of Rosbag files, a Robot Operating System (ROS) feature that allows for recording data during a time interval and playing them later.

The first dataset ( $\mathscr {D}_1$ ) is composed of 90 five-second Rosbag files. Data recorded correspond to a person walking straight in front of the robot. Each Rosbag file contains the data gathered by the Laser Imaging Detection and Ranging (LIDAR) sensor onboard the robot. We recorded data from five different people at three locations—shown in Fig. 1 c on the testbed schema. We recorded six Rosbag files for each location and person.

The second dataset ( $\mathscr {D}_2$ ) is composed of 108 Rosbag files. We recorded data from six people walking straight in front of the robot at three locations—see Fig. 1 c. We recorded six Rosbag files for each location and person. Five out of those six people are the same in both datasets. The last one is not registered in the system. We use its data to evaluate the false-positive cases in the system.

Data curation

Once we have gait data recorded as Rosbag files, it is necessary to curate and tag them to use them for fitting the Convolutional Neural Network (CNN) in charge of people identification. $\mathscr {D}_1$ , depicted in “ Data gathering ” section, have been used to train the Convolutional Neural Network (CNN).

Since Biometric RecognITion Through gAit aNalYsis (BRITTANY) uses a Convolutional Neural Network (CNN) to identify people, we need to convert data from Rosbag files into images $\mathscr {I}$ that describe people’s gait. Image generation process is shown in Fig. 2 . The Fig. 2 a shows the Laser Imaging Detection and Ranging (LIDAR) data as yellow points for the real scene shown in Fig. 2 b. The red arrow shows the location and orientation of the robot. First, we play each Rosbag file obtaining occupancy maps for each Laser Imaging Detection and Ranging (LIDAR) reading—see Fig. 2 c. From the above occupancy maps, People Tracking (PeTra) provides a second occupancy map segmenting the points belonging to people—see Fig. 2 d. Finally, People Tracking (PeTra)’s occupancy maps are aggregated. Such aggregation allows for depicting people’s gait—see Fig. 2 e.

Data curation process: ( a ) Laser Imaging Detection and Ranging (LIDAR) and People Tracking (PeTra) data visualized on Rviz—The red arrow shows the robot’s location an orientation, the yellow points show Laser Imaging Detection and Ranging (LIDAR) readings; ( b ) snapshot from Orbi-One robot camera; ( c ) occupancy map computed from Laser Imaging Detection and Ranging (LIDAR) data; ( d ) occupancy map computed by People Tracking (PeTra); and ( e ) occupancy map aggregation.

A individual occupancy map is represented as an image ( I ), see Fig. 2 d. In these images, white pixels represent a Laser Imaging Detection and Ranging (LIDAR) point belonging to a person, and black pixels represent 1) Laser Imaging Detection and Ranging (LIDAR) points belonging to objects, or 2) points where the Laser Imaging Detection and Ranging (LIDAR) sensor has not detected collisions. The final image ( $\mathscr {I}$ ) is created by concatenating individual occupancy maps, see Fig. 2 e. These occupancy maps are concatenated using the logical OR (||) operation. Equation ( 1 ) represents the operation performed to create the final aggregate occupancy maps. In the Equation, $\mathscr {I}$ represents the final image obtained from the concatenation of individual images ( I ), n is the number of I used in the aggregation, and s is the number of steps between successive images ( I ).

Different aggregation settings have been evaluated to select the best one. They are named as $\mathscr {C}_{n \times s}$ , where $n \in \langle 5,10 \rangle$ is the number of Laser Imaging Detection and Ranging (LIDAR) readings used in the aggregation, and $s \in \langle 0,1,2 \rangle$ is the number of steps between successive readings. For instance, in $\mathscr {C}_{5 \times 2}$ , five Laser Imaging Detection and Ranging (LIDAR) readings are used by considering one out of three successive Laser Imaging Detection and Ranging (LIDAR) readings. Figure 3 show a sample of the resulting aggregated occupancy map for each setting schema.

Aggregated occupancy maps for ( a ) $\mathscr {C}_{5 \times 0}$ , ( b ) $\mathscr {C}_{5 \times 1}$ , ( c ) $\mathscr {C}_{5 \times 2}$ , ( d ) $\mathscr {C}_{10 \times 0}$ , ( e ) $\mathscr {C}_{10 \times 1}$ , and ( f ) $\mathscr {C}_{10 \times 2}$ .

Convolutional neural network design

Convolutional Neural Network (CNN) usage has increased in recent years since a large number of systems integrate them 48 . Convolutional Neural Network (CNN) identify features from tagged datasets. Such features are often imperceptible to humans. Biometric RecognITion Through gAit aNalYsis (BRITTANY) uses a classification Convolutional Neural Network (CNN) to identify people. It receives an occupancy map ( $\mathscr {I}$ ) as input and returns a success rate for each person it was trained. To define the Convolutional Neural Network (CNN), Keras API 49 have been used using TensorFlow 50 as back-end.

To carry out the image classification to be performed by BRITTANY, we propose a new neural network architecture, hereafter ”custom”. To evaluate its performance, we have selected two other well-known architectures to perform image classification LeNet-5 51 and AlexNet 52 . These two architectures have been selected because they are the first well-known architectures that solved image classification problems. Both architectures have been modified to process as input a 256x256 image and as output a Dense layer of 5. Other architectures such as VGG16 53 have been tested, but due to the complexity in the deep layers, the model created did not generalize correctly to perform the image classification used by Biometric RecognITion Through gAit aNalYsis (BRITTANY).

Figure 4 , generated with Net2vis tool 54 , illustrates the three architectures, Fig. 4 a represents the custom architecture proposed in this work, and Fig. 4 b and c represents the architecture of LeNet and Alexnet respectively. In addition, Table 1 shows the number of trainable and non-trainable parameters for each proposed architectures. As can be seen, the custom architecture has a much lower number of trainable parameters than the other two well-known models.

CNN architectures of ( a ) Custom, ( b ) LeNet, ( c ) AlexNet.

The Convolutional Neural Networks (CNNs) models were trained on Caléndula the parallel computing cluster of Supercomputación Castilla y León (SCAYLE) 55 . Supercomputación Castilla y León (SCAYLE) is a public research centre dependent on the Community of Castilla y León (Spain) whose main activity is to support the improvement of R & D & I tasks. Six Convolutional Neural Network (CNN) models have been created for each architecture, one for each out of 6 aggregation settings ( $\mathscr {C}_{5 \times 0}$ , $\mathscr {C}_{5 \times 1}$ , $\mathscr {C}_{5 \times 2}$ , $\mathscr {C}_{10 \times 0}$ , $\mathscr {C}_{10 \times 1}$ , and $\mathscr {C}_{10 \times 2}$ ) depicted in “ Data curation ” section. In this way, a total of 18 Convolutional Neural Network (CNN) models have been trained.

The evaluation was carried out using the $\mathscr {D}_2$ —see “ Data gathering ” section. We played each Rosbag file using Biometric RecognITion Through gAit aNalYsis (BRITTANY) with different setting schemas to evaluate the accuracy. We need to know whether or not the user has been recognized properly for each run. Such data allow us to obtain the confusion matrix that allows visualization of the performance of our tool with the six setting schemas defined and the three Convolutional Neural Network (CNN) architectures used, described in “ Convolutional neural network design ” section.

Moreover, to evaluate the overall Biometric RecognITion Through gAit aNalYsis (BRITTANY)’s performance the following Key Performance Indicators (KPI)’s obtained through the confusion matrix are considered: Accuracy ( $\mathscr {A}$ ), Precision ( $\mathscr {P}$ ), Recall ( $\mathscr {R}$ ), and F $_1$ -score ( $\mathscr {F}$ ). As the proposed method is a multi-class classification, it is necessary to calculate Key Performance Indicators (KPI)’s for each class. Then an arithmetic average is calculated to obtain the Key Performance Indicators (KPI)’s of the global system 56 $\mathscr {A}$ —see Eq. ( 6 )—measures the proportion of correct predictions, both positive and negative cases, among the total number of cases examined. $\mathscr {A}_{k}$ is calculated as shown Eq. ( 2 ). The $\mathscr {P}$ score—see Eq. ( 7 )—shows the fraction of relevant instances among the retrieved instances. $\mathscr {P}_{k}$ is calculated as shown Eq. ( 3 ). The $\mathscr {R}$ score—see Eq. ( 8 )—shows the rate of positive cases that were correctly identified by the algorithm. $\mathscr {R}_{k}$ is calculated as shown Eq. ( 4 ). Finally, the $\mathscr {F}$ score—see Eq. ( 9 ) is the harmonic mean of precision and recall. $\mathscr {F}_{k}$ is calculated as shown Eq. ( 5 ). In equations, TP represents the true-positive rate, TN is the true-negative rate, FP is the false-positive rate, FN is the false-negative rate and K the number of classes into which the model classifies.

To evaluate the tradeoff between the TP and FP rates of each class, we have computed the Receiver Operating Characteristic (ROC) curve 57 for each one of the six setting schemas defined and the three proposed architectures. Besides, the Area Under the Curve (AUC) has been calculated to depict how much the models can distinguish between classes.

Results and discussion

Key Performance Indicators (KPI)’s: Accuracy, Precision, Recall and F1-score, for each of the trained models and each configuration proposed ( $\mathscr {C}_{5 \times 0}$ , $\mathscr {C}_{5 \times 1}$ , $\mathscr {C}_{5 \times 2}$ , $\mathscr {C}_{10 \times 0}$ , $\mathscr {C}_{10 \times 1}$ , and $\mathscr {C}_{10 \times 2}$ ).

Biometric RecognITion Through gAit aNalYsis (BRITTANY)’s evaluation was done as described in “ Evaluation ” section. We obtained a confusion matrix for each model (custom, LeNet and AlexNet) and each setting schema ( $\mathscr {C}_{5 \times 0}$ , $\mathscr {C}_{5 \times 1}$ , $\mathscr {C}_{5 \times 2}$ , $\mathscr {C}_{10 \times 0}$ , $\mathscr {C}_{10 \times 1}$ , and $\mathscr {C}_{10 \times 2}$ ). Figure 6 , shows the confusion matrices for each model and setting schema. In the first row are the confusion matrices, in blue, for the custom models, in brown (second row) the confusion matrices for LeNet and in the third row, in green, the confusion matrices for AlexNet. From left ( $\mathscr {C}_{5 \times 0}$ ) to right ( $\mathscr {C}_{10 \times 2}$ ) are the different configurations used. Every confusion matrix consists of rows and columns representing user identifiers. The matrices check situations where BRITTANY provided a correct or wrong outcome using each model respectively. A perfect system would have all the values on the main diagonal. According to the values shown in Fig. 6 , the Key Performance Indicators (KPI)’s, the Receiver Operating Characteristic (ROC) curve and the Area Under the Curve (AUC) have been calculated for each of the architectures and setting schemas. These Key Performance Indicators (KPI)’s are presented in Fig. 5 that show the Accuracy ( $\mathscr {A}$ )—Fig. 5 a, Precision ( $\mathscr {P}$ )—Fig. 5 b, Recall ( $\mathscr {R}$ )—Fig. 5 c, and F $_1$ -score ( $\mathscr {F}$ )—Fig. 5 d, for each model proposed and each setting schema. Moreover in Fig. 7 are presented the Receiver Operating Characteristic (ROC) curves for each model and settting schema. In the first row Receiver Operating Characteristic (ROC) curves for the custom models, in the second row Receiver Operating Characteristic (ROC) curves for LeNet and in the third row Receiver Operating Characteristic (ROC) curves for AlexNet. From left ( $\mathscr {C}_{5 \times 0}$ ) to right ( $\mathscr {C}_{10 \times 2}$ ) are the different configurations used. The Receiver Operating Characteristic (ROC) curves have been created for each user in the system (U0 - U4) and the not registered user (!U), in this way, it is also calculated the Area Under the Curve (AUC) for each user, model and setting schema.

Confusion matrices for each architecture: Custom in blue (first row), LeNet in brown (second row) and AlexNet in green (third row). From left to right the different configurations used: ( a ) $\mathscr {C}_{5 \times 0}$ , ( b ) $\mathscr {C}_{5 \times 1}$ , ( c ) $\mathscr {C}_{5 \times 2}$ , ( d ) $\mathscr {C}_{10 \times 0}$ , ( e ) $\mathscr {C}_{10 \times 1}$ , and ( f ) $\mathscr {C}_{10 \times 2}$ .

Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) for custom (first row), LeNet (second row) and AlexNet (third row). From left to right the different configurations used: ( a ) $\mathscr {C}_{5 \times 0}$ , ( b ) $\mathscr {C}_{5 \times 1}$ , ( c ) $\mathscr {C}_{5 \times 2}$ , ( d ) $\mathscr {C}_{10 \times 0}$ , ( e ) $\mathscr {C}_{10 \times 1}$ , and ( f ) $\mathscr {C}_{10 \times 2}$ .

Focusing on Fig. 5 a, we see that accuracy scores that measure the proportion of correct (both positive and negative) predictions among the total number of cases examined. The accuracy score is only higher than 0.8, specifically 88%, with the $\mathscr {C}_{10 \times 1}$ schema using the custom model. The accuracy scores in the schemas for the custom model are between 47% and 57%. The accuracy of LeNet and AlexNet for each schema is much lower than that obtained with the custom model, obtaining values between 19% and 30%. Except for the $\mathscr {C}_{10 \times 2}$ schema where LeNet obtains the same value as custom (53%).

The precision ( $\mathscr {P}$ ), see Fig. 5 b, represents the fraction of relevant instances among the retrieved samples. In this case, most of the values are higher than 50% for all the models; only the model AlexNet with schema $\mathscr {C}_{10 \times 2}$ reports a precision of 30%. The two models with a better precision are LeNet with the schema $\mathscr {C}_{5 \times 0}$ and custom with the schema $\mathscr {C}_{10 \times 1}$ , for both models, the precision score is 88%.

The recall score ( $\mathscr {R}$ )—also called sensitivity or true positive rate—, see Fig. 5 c is the ratio of positive instances correctly detected by the algorithm. In this score, the maximum value, 88%, is again for the custom model and $\mathscr {C}_{10 \times 1}$ schema. The recall scores in the schemas for the custom model are between 47% and 57%. The recall of LeNet and AlexNet for each schema report values between 19% and 53%.

It is often convenient to combine precision and recall into a single metric, see Fig. 5 d. The $F_1$ score ( $\mathscr {F}$ ) is the harmonic mean of $\mathscr {P}$ and $\mathscr {R}$ . Whereas the regular mean treats all values equally, the harmonic mean gives much more weight to low values. As a result, the classifier will only get a high $F_1$ score if both recall and precision are high. The model with the best $F_1$ score is the custom using the schema $\mathscr {C}_{10 \times 1}$ , specifically 88%. The remaining schemas for the custom model have a $F_1$ score between 48% and 59%. The $F_1$ score of LeNet and AlexNet for each schema is much lower than that obtained with the custom model, obtaining values between 22% and 35%. Except for the $\mathscr {C}_{10 \times 2}$ schema where LeNet obtains the same value as custom (53%).

The model with the best Key Performance Indicators (KPI)’s is custom using the schema $\mathscr {C}_{10 \times 1}$ , all the Key Performance Indicators (KPI)’s reports a score of 88%. Then, focusing on the confusion matrix of the custom model and the best schema, $\mathscr {C}_{10 \times 1}$ , shown in Fig. 6 (first row, fifth column), we see that most of the values are in the main diagonal. However, there are some errors. U1 was the only correctly identified user in all cases—we have 18 cases, six possible users at three different locations. U0 user was recognised as U0 in 17 cases. However, there is one case where U0 was not recognised as a registered user (!U). U2 was identified correctly in 16 cases, and he was not recognised (!U) in 2. U3 was correctly identified in all cases but one, where he was recognised as U4. It is important to point out that this is the only case in the evaluation of the $\mathscr {C}_{10 \times 1}$ schema and the custom model where two registered users were mistaken. The results for U4 are the worst in the evaluation. In 15 cases, the user was correctly identified, but in 3 cases, he was not recognised (!U). Finally, a non-registered user was used to test the Biometric RecognITion Through gAit aNalYsis (BRITTANY)’s robustness in front of unknown users. Such a user was wrongly recognised as U3 in 1 case. In the remaining 17 cases, he was correctly identified as non-registered.

Focusing on the Receiver Operating Characteristic (ROC) curves for each of the models obtained, see Fig. 7 . It can be seen that the best representation of the Receiver Operating Characteristic (ROC) curves corresponds to the custom model and the $\mathscr {C}_{10 \times 1}$ schema (first row, fifth column). The quality of the model increases when the curve moves towards the upper left corner of the graph. This is because it improves its TP rate, also minimising the FP rate. Moreover, the Area Under the Curve (AUC) values are used as a summary of the model’s performance. The more curve moves towards the upper left corner of the graph, the more area is contained under it and therefore, the classifier is better. A perfect classifier has an Area Under the Curve (AUC) of 1. The Area Under the Curve (AUC)’s for each user obtained from BRITTANY using the custom model and the $\mathscr {C}_{10 \times 1}$ schema are (0.97, 1.00, 0.94, 0.97, 0.91 and 0.94) respectively, all of them higher than 90%.

Finally, the accuracy obtained from Biometric RecognITion Through gAit aNalYsis (BRITTANY) using the custom model and the $\mathscr {C}_{10 \times 1}$ schema is 88%. Focussing on works presented in the Introduction that applied biometry technologies, the work 34 creates a silhouette from the point cloud extracted from the multi-line LIDAR and processes it through an LTSM-based CNN network obtaining an accuracy of 60%. The work 35 uses gait analysis to do person re-identification, using the silhouette obtained from the projection of the 3D point cloud of a person to a 2D image. This method obtains an accuracy of 87%. It should be noted that those works use a 3D LIDAR instead of a 2D LIDAR. The number of values provided by a 2D LIDAR sensor is much smaller than the ones provided by a 3D LIDAR, so it is easier to process data. This fact facilitates the use of the final system in real-time. Moreover, the models used in those works have not been evaluated against users outside the system. Therefore, people who are not registered in the system could be classified as legitimate users of it.

This paper presents Biometric RecognITion Through gAit aNalYsis (BRITTANY), a system that identifies people by analyzing their gait. Thus, the system is based on behavioural biometric technologies characterized for analyzing the features of a specific action performed by a person. This system processes sensor data obtained in real-time. A Laser Imaging Detection and Ranging (LIDAR) sensor was chosen because of its low computational demand as well as the privacy it provides. The Laser Imaging Detection and Ranging (LIDAR) readings are processed by People Tracking (PeTra) to create an occupational map segmenting the points that belong to people. Then, such segmented occupational maps are aggregated to build an image of people’s gait. Aggregated occupancy maps are processed by a Convolutional Neural Network (CNN) model that outcomes the user identifier that corresponds to the person in front of the robot. The final prediction depends on several estimations by applying a voting strategy to prevent errors. Biometric RecognITion Through gAit aNalYsis (BRITTANY) can be used in several applications on cooperative robotics and raises Human-Robot Interaction (HRI) since it allows the robot to “know” the people around him. Moreover, at indoor environments, the robot could detect foreigners and alert their presence. Besides, it also can be used to control access to critical infrastructures, making it more difficult for cybercriminals to carry out tailgating attacks.

The evaluation was done by analyzing several setting schemas named $\mathscr {C}_{n \times s}$ with three Convolutional Neural Network (CNN) architectures, one of them is proposed in this work, called ”custom” and the other two are well-known architectures (LeNet and AlexNet). The dataset $\mathscr {D}_2$ was used to measure the performance. It compiles 108 Rosbag files—Six recordings for six users at three different locations. Five users (U0–4) are well-known people. The last one (!U) is unknown to the system and was used to evaluate the system’s performance in front of strange people. The $\mathscr {C}_{10 \times 1}$ schema using the custom model gets the best results in all the Key Performance Indicators (KPI)’s computed. Such schema builds images by aggregating ten segmented occupancy maps taking one out of two. This schema provides an accuracy score of 88%.

As previously mentioned, this work is a Proof of Concept (PoC) for indoor environments with few people. We aim to determine if it is possible to identify people by their gait using a 2D LIDAR sensor to maintain the user’s privacy and reduce computational load. In future work, we propose to collect data from more users to evaluate Biometric RecognITion Through gAit aNalYsis (BRITTANY) as an authentication tool.

We want to point out that all the datasets generated are available online 47 . Besides, The source code developed during the current study is available under an open-source license in the GitHub repository 58 . Finally, a docker image with all required software to double-check the evaluation posed in this paper is also available online 59 .

Data availability

The datasets generated and/or analysed during the current study are available in the “Dataset for train and test BRITTANY (Biometric RecognITion Through gAit aNalYsis)” repository, 10.5281/zenodo.5825885.

Ulsch, M. Cyber Threat!: How to Manage the Growing Risk of Cyber Attacks (Wiley Online Library, 2014).

Book Google Scholar

Raiyn, J. et al. A survey of cyber attack detection strategies. Int. J. Secur. Appl. 8 , 247–256 (2014).

Google Scholar

Ogie, R. I. Cyber security incidents on critical infrastructure and industrial networks, in Proceedings of the 9th International Conference on Computer and Automation Engineering , 254–258 (2017).

Salahdine, F. & Kaabouch, N. Social engineering attacks: A survey. Future Internet 11 , 89 (2019).

Article Google Scholar

Maan, P. & Sharma, M. Social engineering: A partial technical attack. Int. J. Comput. Sci. Issues 9 , 557–559 (2012).

Panchpor, A. A., Shue, S. & Conrad, J. M. A survey of methods for mobile robot localization and mapping in dynamic indoor environments, in 2018 Conference on Signal Processing And Communication Engineering Systems (SPACES) , 138–144 (IEEE, 2018).

Cheng, J., Cheng, H., Meng, M. Q.-H. & Zhang, H. Autonomous navigation by mobile robots in human environments: A survey, in 2018 IEEE International Conference on Robotics and Biomimetics (ROBIO) , 1981–1986 (IEEE, 2018).

Richert, A., Müller, S., Schröder, S. & Jeschke, S. Anthropomorphism in social robotics: Empirical results on human–robot interaction in hybrid production workplaces. AI Soc. 33 , 413–424 (2018).

Fox, J. & Gambino, A. Relationship development with humanoid social robots: Applying interpersonal theories to human–robot interaction. Cyberpsychol. Behav. Soc. Netw. 24 , 294–299 (2021).

Lobato, C., Vega-Magro, A., Núñez, P. & Manso, L. J. Human-robot dialogue and collaboration for social navigation in crowded environments, in 2019 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC) , 1–6 (IEEE, 2019).

Wang, Y., Wei, X., Shen, H., Ding, L. & Wan, J. Robust fusion for RGB-D tracking using CNN features. Appl. Soft Comput. 92 , 106302 (2020).

Matti, D., Ekenel, H. K. & Thiran, J.-P. Combining lidar space clustering and convolutional neural networks for pedestrian detection, in 2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS) , 1–6 (IEEE, 2017).

Bu, F., Le, T., Du, X., Vasudevan, R. & Johnson-Roberson, M. Pedestrian planar lidar pose (PPLP) network for oriented pedestrian detection based on planar lidar and monocular images. IEEE Robot. Autom. Lett. 5 , 1626–1633 (2019).

Linder, T. & Arras, K. O. People detection, tracking and visualization using ROS on a mobile service robot. In Robot Operating System (ROS) (ed. Koubaa, A.) 187–213 (Springer, 2016).

Chapter Google Scholar

Rios-Martinez, J., Spalanzani, A. & Laugier, C. From proxemics theory to socially-aware navigation: A survey. Int. J. Soc. Robot. 7 , 137–153. https://doi.org/10.1007/s12369-014-0251-1 (2015).

Duong, H. T. & Suh, Y. S. Human gait tracking for normal people and walker users using a 2D lidar. IEEE Sens. J. 20 , 6191–6199 (2020).

Article ADS Google Scholar

Guerrero-Higueras, Á. M. et al. Tracking people in a mobile robot from 2D lidar scans using full convolutional neural networks for security in cluttered environments. Front. Neurorobot. 12 , 85 (2018).

Álvarez-Aparicio, C. et al. People detection and tracking using LIDAR sensors. Robotics 8 , 75 (2019).

Álvarez-Aparicio, C. et al. Convolutional neural networks refitting by bootstrapping for tracking people in a mobile robot. Appl. Sci. 11 , 10043 (2021).

Jain, A., Hong, L. & Pankanti, S. Biometric identification. Commun. ACM 43 , 90–98 (2000).

Adeoye, O. S. A survey of emerging biometric technologies. Int. J. Comput. Appl. 9 , 1–5 (2010).

Valdes-Ramirez, D. et al. A review of fingerprint feature representations and their applications for latent fingerprint identification: Trends and evaluation. IEEE Access 7 , 48484–48499 (2019).

Mane, S. & Shah, G. Facial recognition, expression recognition, and gender identification. In Data Management, Analytics and Innovation (eds Balas, V. et al. ) 275–290 (Springer, 2019).

Patil, C. M. & Gowda, S. An approach for secure identification and authentication for biometrics using iris, in 2017 International Conference on Current Trends in Computer, Electrical, Electronics and Communication (CTCEEC) , 421–424 (IEEE, 2017).

Obaidat, M. S., Traore, I. & Woungang, I. Biometric-Based Physical and Cybersecurity Systems Vol. 368 (Springer, 2019).

Xiao, Q. Security issues in biometric authentication, in Proceedings from the Sixth Annual IEEE SMC Information Assurance Workshop , 8–13 (IEEE, 2005).

Menshakou, P. & Murashko, I. Voice user identification in access control systems, in Proceedings of International Conference on Open Semantic Technologies for Intelligent Systems (OSTIS-2017). Minsk: BSUIR , 175–178 (2017).

Mohammed, A. A., Abdul-Hassan, A. K. & Mahdi, B. S. Authentication system based on hand writing recognition, in 2019 2nd Scientific Conference of Computer Sciences (SCCS) , 138–142 (IEEE, 2019).

Connor, P. & Ross, A. Biometric recognition by gait: A survey of modalities and features. Comput. Vis. Image Underst. 167 , 1–27 (2018).

Petrescu, R. V. Face recognition as a biometric application. J. Mechatron. Robot. 3 , 237–257 (2019).

Ouellet, S., Grondin, F., Leconte, F. & Michaud, F. Multimodal biometric identification system for mobile robots combining human metrology to face recognition and speaker identification, in The 23rd IEEE International Symposium on Robot and Human Interactive Communication , 323–328 (2014). https://doi.org/10.1109/ROMAN.2014.6926273 .

Al-Qaderi, M. K. & Rad, A. B. A multi-modal person recognition system for social robots. Appl. Sci. 8 , 387 (2018).

Karri, C., Cheikhrouhou, O., Harbaoui, A., Zaguia, A. & Hamam, H. Privacy preserving face recognition in cloud robotics: A comparative study. Appl. Sci. 11 , 6522 (2021).

Article CAS Google Scholar

Yamada, H., Ahn, J., Mozos, O. M., Iwashita, Y. & Kurazume, R. Gait-based person identification using 3D lidar and long short-term memory deep networks. Adv. Robot. 34 , 1201–1211 (2020).

Benedek, C., Gálai, B., Nagy, B. & Jankó, Z. Lidar-based gait analysis and activity recognition in a 4D surveillance system. IEEE Trans. Circuits Syst. Video Technol. 28 , 101–113 (2016).

Hasan, M. et al. Tracking people using ankle-level 2D lidar for gait analysis, in International Conference on Applied Human Factors and Ergonomics , 40–46 (Springer, 2020).

Robotics Group of Universidad de León. Leon@home Testbed (accessed 28 April 2022). https://robotica.unileon.es/index.php?title=Testbed (2022).

EU Robotics. ERL Certified Test Beds (accessed 28 April 2022). https://www.eu-robotics.net/robotics_league/erl-service/certified-test-beds/index.html (2022).

Robotnik. Robotnik homepage (accessed 28 April 2022). https://robotnik.eu/es/ (2022).

Quigley, M. et al. ROS: An open-source robot operating system, in ICRA Workshop on Open Source Software , vol. 3, 5 (Kobe, Japan, 2009).

Robotics Group of Universidad de Léon. Robotics Group homepage (accessed 28 April 2022). https://robotica.unileon.es/ (2022).

Álvarez-Aparicio, C. et al. Benchmark dataset for evaluation of range-based people tracker classifiers in mobile robots. Front. Neurorobot. 11 , 72 (2018).

Ronneberger, O., Fischer, P. & Brox, T. U-net: Convolutional networks for biomedical image segmentation, in Medical Image Computing and Computer-Assisted Intervention (MICCAI) , vol. 9351 of LNCS , 234–241 (Springer, 2015).

Long, J., Shelhamer, E. & Darrell, T. Fully convolutional networks for semantic segmentation, in IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 3431–3440 (2015). https://doi.org/10.1109/CVPR.2015.7298965 .

Álvarez-Aparicio, C. et al. Lidar-based people detection and tracking for home competitions, in 2019 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC) , 1–6 (IEEE, 2019).

Alvarez-Aparicio, C., Guerrero-Higueras, A. M., Rodr, F. J. & Mart, F. Detección y seguimiento de personas utilizando sensores lidar, in Jornadas Nacionales de Robótica Spanish Robotics Conference , 234–240 (2019).

Dataset 10.5281/zenodo.5825885 (accessed 12 January 2022). https://doi.org/10.5281/zenodo.5825885 (2022).

Mane, D. & Kulkarni, U. V. A survey on supervised convolutional neural network and its major applications, in Deep Learning and Neural Networks: Concepts, Methodologies, Tools, and Applications , 1058–1071 (IGI Global, 2020).

Keras. Keras homepage (accessed 28 April 2022). https://keras.io/ (2022).

TensorFlow. TensorFlow homepage (accessed 28 April 2022). https://www.tensorflow.org/ (2022).

LeCun, Y. et al. Lenet-5, convolutional neural networks. http://yann.lecun.com/exdb/lenet 20 , 14 (2015).

Krizhevsky, A., Sutskever, I. & Hinton, G. E. Imagenet classification with deep convolutional neural networks. Adv. Neural Inf. Process. Syst. 25 , 1–9 (2012).

Simonyan, K. & Zisserman, A. Very deep convolutional networks for large-scale image recognition . arXiv preprint arXiv:1409.1556 (2014).

Bäuerle, A., van Onzenoodt, C. & Ropinski, T. Net2vis—A visual grammar for automatically generating publication-tailored CNN architecture visualizations. IEEE Trans. Visual Comput. Graph. 27 , 2980–2991. https://doi.org/10.1109/TVCG.2021.3057483 (2021).

SCAYLE. Scayle (accessed 28 April 2022) (2022). https://www.scayle.es/ .

Grandini, M., Bagli, E. & Visani, G. Metrics for multi-class classification: An overview. arXiv preprint arXiv:2008.05756 (2020).

Fawcett, T. An introduction to roc analysis. Pattern Recogn. Lett. 27 , 861–874 (2006).

C.Álvarez-Aparicio. BRITTANY (Biometric RecognITion Through gAitaNalYsis) (accessed 28 April 2022) (2022). https://github.com/ClaudiaAlvarezAparicio/brittany .

C.Álvarez-Aparicio. BRITTANY Docker image (accessed 28 April 2022) (2022). https://hub.docker.com/r/claudiaalvarezaparicio/brittany .

Download references

Acknowledgements

The research described in this article has been funded by the Instituto Nacional de Ciberseguridad de España (INCIBE), under the grant ”ADENDA 4: Detección de nuevas amenazas y patrones desconocidos (Red Regional de Ciencia y Tecnología)”, addendum to the framework agreement INCIBE-Universidad de León, 2019-2021. Miguel Ángel González-Santamarta would like to thank Universidad de León for its funding support for his doctoral studies.

Author information

Authors and affiliations.

Department of Mechanical, Computer Science and Aerospace Engineering, University of León, 24071, León, Spain

Claudia Álvarez-Aparicio, Ángel Manuel Guerrero-Higueras, Miguel Ángel González-Santamarta, Adrián Campazas-Vega, Vicente Matellán & Camino Fernández-Llamas

You can also search for this author in PubMed Google Scholar

Contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by C.Á-A, Á.M.G-H, M.Á.G-S and A.C-V. The first draft of the manuscript was written by C.Á-A and Á.M.G-H and all authors commented on previous versions of the manuscript. The supervision was done by V.M. The project management and the funding acquisition were supplied by C.F-L. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Claudia Álvarez-Aparicio .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Álvarez-Aparicio, C., Guerrero-Higueras, Á.M., González-Santamarta, M.Á. et al. Biometric recognition through gait analysis. Sci Rep 12 , 14530 (2022). https://doi.org/10.1038/s41598-022-18806-4

Download citation

Received : 03 May 2022

Accepted : 19 August 2022

Published : 25 August 2022

DOI : https://doi.org/10.1038/s41598-022-18806-4

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

This article is cited by

Gait signature changes with walking speed are similar among able-bodied young adults despite persistent individual-specific differences.

Taniel S. Winner
Michael C. Rosenberg
Lena H. Ting

Scientific Reports (2024)

Deep learning-based route reconfigurability for intelligent vehicle networks to improve power-constrained using energy-efficient geographic routing protocol

Liyakathunisa Syed
P. Sathyaprakash
R. Premalatha

Wireless Networks (2024)

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

Explore articles by subject
Guide to authors
Editorial policies

Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Publications
Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

Advanced Search
Journal List

Wearables for Running Gait Analysis: A Systematic Review

Rachel mason.

1 Department of Sport, Exercise and Rehabilitation, Faculty of Health and Life Sciences, Northumbria University, Newcastle upon Tyne, UK

Liam T. Pearson

Gillian barry, fraser young.

3 Department of Computer and Information Sciences, Northumbria University, Newcastle upon Tyne, UK

Oisin Lennon

2 DANU Sports Ltd., Dublin, Ireland

Alan Godfrey

Samuel stuart.

4 Northumbria Healthcare NHS Foundation Trust, Newcastle upon Tyne, UK

Associated Data

Running gait assessment has traditionally been performed using subjective observation or expensive laboratory-based objective technologies, such as three-dimensional motion capture or force plates. However, recent developments in wearable devices allow for continuous monitoring and analysis of running mechanics in any environment. Objective measurement of running gait is an important (clinical) tool for injury assessment and provides measures that can be used to enhance performance.

We aimed to systematically review the available literature investigating how wearable technology is being used for running gait analysis in adults.

A systematic search of the literature was conducted in the following scientific databases: PubMed, Scopus, Web of Science and SPORTDiscus. Information was extracted from each included article regarding the type of study, participants, protocol, wearable device(s), main outcomes/measures, analysis and key findings.

A total of 131 articles were reviewed: 56 investigated the validity of wearable technology, 22 examined the reliability and 77 focused on applied use. Most studies used inertial measurement units ( n = 62) [i.e. a combination of accelerometers, gyroscopes and magnetometers in a single unit] or solely accelerometers ( n = 40), with one using gyroscopes alone and 31 using pressure sensors. On average, studies used one wearable device to examine running gait. Wearable locations were distributed among the shank, shoe and waist. The mean number of participants was 26 (± 27), with an average age of 28.3 (± 7.0) years. Most studies took place indoors ( n = 93), using a treadmill ( n = 62), with the main aims seeking to identify running gait outcomes or investigate the effects of injury, fatigue, intrinsic factors (e.g. age, sex, morphology) or footwear on running gait outcomes. Generally, wearables were found to be valid and reliable tools for assessing running gait compared to reference standards.

Conclusions

This comprehensive review highlighted that most studies that have examined running gait using wearable sensors have done so with young adult recreational runners, using one inertial measurement unit sensor, with participants running on a treadmill and reporting outcomes of ground contact time, stride length, stride frequency and tibial acceleration. Future studies are required to obtain consensus regarding terminology, protocols for testing validity and the reliability of devices and suitability of gait outcomes.

Clinical Trial Registration

CRD42021235527.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40279-022-01760-6.

The majority of studies tested young adult recreational runners, with an average sample size of < 30.

Most studies used one wearable (on shoe or tibia), typically an inertial measurement unit with a sampling rate of 100 Hz, with ground contact time, stride length, stride frequency and tibial acceleration outcomes most reported.

Most studies tested participants indoors, using a treadmill for a set duration or distance at a controlled speed.

Introduction

Running is one of the most popular sport and recreational activities worldwide as well as being a core component of many sports [ 1 ]. In addition to its beneficial effects on health, the prevalence and cumulative incidence proportions of running-related injuries (RRI) are high, which results in participation cessation [ 2 ]. It is well established that a contributing factor to RRI is abnormal running gait, meaning early detection of potentially harmful running gait pathologies is essential. Where biomechanics have been implicated, clinical running analysis has largely been limited to the use of subjective clinical observation or rating scales (e.g. the High-Level Mobility and Assessment tool), which may not be sensitive to subtle changes in performance with training or injury [ 3 – 5 ].

Quantitative running gait analysis, as a clinical tool for minimising injury risk and as a performance measure, has been well documented in the literature [ 6 – 8 ]. However, quantification of running beyond clinical observation has largely been performed using a two-dimensional video analysis [ 3 , 5 ] (particularly in commercial settings, such as running shoe stores), but this is limited to certain gait outcomes (i.e. foot strike patterns [FSP]) and still requires subjective visual/manual inspection and analysis of videos. To analyse more advanced measures, such as spatiotemporal (e.g. stride length [SL], stride time, step frequency [SF], speed), kinematic (e.g. angular velocity and joint angles) and kinetic (e.g. ground reaction forces [GRF]) measures, more cumbersome and expensive traditional (reference/gold-standard) gait laboratory measures are required (e.g. three-dimensional [3D] motion capture, force plate equipment, instrumented treadmills). However, use of gait laboratories for running gait assessment is limited because of the expense of equipment, the need for trained practitioners to collect and analyse data, and the requirement to attend a laboratory setting. Therefore, those traditional techniques are not readily available to performance or clinical settings and provide a limited understanding of running in ‘real-world’ environments [ 9 – 11 ]. Furthermore, laboratory-based testing often uses constrained protocols that may not represent usual running behaviour, such as assessing single foot strikes, unnatural force platform targeting and limited numbers of consecutive steps [ 12 ]. Numerous studies have sought to overcome this issue by using instrumented treadmills; however, further studies demonstrate the inconsistencies in running gait between over-ground and treadmill running [ 13 ]. In order to enhance understanding of running gait, further research in a natural running environment is required [ 12 ].

Wearable technology offers an alternative to overcome traditional assessment limitations and is becoming increasingly accepted by runners, coaches and clinicians [ 14 ]. Wearables utilising accelerometers, gyroscopes and magnetometers, applied individually or in combination as an inertial measurement unit (IMU), and ‘pressure-sensitive’ insoles allow quantification of a combination of spatiotemporal, kinetic and kinematic variables and have become a viable alternative owing to their portability and affordability [ 15 ]. Evidently, wearable devices can quantify various running gait outcomes in any setting (i.e. laboratory or outdoor/real world), which may enhance understanding of running performance, fatigue and injury mechanisms. Although research in this area is emerging, there have been some interesting developments. For example, previous studies have only been able to assess discrete timepoints (‘snap-shots’) throughout a run because of the use of force platforms and video analysis [ 16 – 18 ], whereas with recent improvements in accuracy, sensitivity and computing power, wearables have the potential to be an effective tool to measure the effects of fatigue on running biomechanics in the field, capturing the full duration of a run [ 19 , 20 ].

Studies have also explored the use of wearable technology to quantify running gait patterns [ 19 – 21 ]. Within those studies, a wide range of protocols have been used indicating a lack of standardised methodology, and it is unclear whether the various wearables are valid or reliable for running gait assessment, which limits running gait interpretation. Coaches, researchers, clinicians or athletes who want to conduct similar running gait assessments or research are left with a choice of numerous protocols, which differ in many aspects. In the process of developing robust protocols, it is often helpful to have evidence-based recommendations. Therefore, the purpose of this review is to provide a comprehensive overview of studies that have used wearable technology for a running gait analysis, in order to provide some guidance regarding the selection of appropriate methodologies. We focused the review on the following: (1) methodologies employed to assess the validity and reliability of wearables for running gait assessment; (2) the application of wearables to assess running gait (i.e. aims, participants, environment, sensor type/location, protocol); (3) commonly reported running gait outcomes and findings; and (4) recommendations for future protocols and research. For the purposes of this review, when reporting our findings, we first provide a comprehensive description of all reviewed studies and then group the reviewed articles into two areas: (A) those that purely examined the validity and reliability of wearables for running gait assessment and (B) application of wearable sensors to assess running gait in different populations to inform performance or clinical outcomes.

The protocol was prospectively registered on the PROSPERO International Prospective Register for Systematic Reviews website (registration no. CRD42021235527) in February 2021. Design and reporting of this review have followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement [ 22 ].

Search Strategy and Study Selection Process

A systematic search was conducted to identify potentially relevant papers in the following scientific databases: PubMed, Scopus, Web of Science and SPORTDiscus. The focus of this review was on journal articles published in English that described the use of wearable technology to assess natural running gait in adults. The general search strategy and search terms are described in Table Table1. 1 . Articles published up to 4 May, 2022 were reviewed. Thereafter, the article selection process consisted of the following steps using the PRISMA guidelines (Fig. 1 ): (1) an initial title screen for relevant articles was performed by independent authors (SS, RM), once the searched database results had been combined and duplicates had been removed; (2) both the titles and abstracts of the selected articles were reviewed (SS, RM) [a review of the full text was completed if it was not clear from the title or abstract whether the study met the review criteria]; and (3) the authors (SS, RM) read the full texts and selected articles based on the inclusion/exclusion criteria (Table (Table2). 2 ). Additionally, the references of all included studies were checked for additional publications that could be included in this review. At all stages of the study selection process, decisions regarding inclusion or exclusion were made by two authors (SS and RM), with a third author (GB) consulted to resolve discrepancies (Table 1 of the Electronic Supplementary Material [ESM]).

Systematic search strategy key terms

Wearable technology

“Wearable*” OR "Wearable Technology" OR "Wearable Devices" OR "Wearable Sensors" OR “IMU” OR “Inertial Sensor" OR "Inertial Measurement Unit" OR "Gyroscope" OR "Magnetometer" OR Acceleromet* OR "Force Plate" OR "Pressure Plate" OR "Pressure Sensor"

TITLE-ABS-KEY

Running gait

“Running” OR “Jogging” OR “Run” OR “Jog” OR “Sprint” OR “Sprinting” OR “Sprints” OR “Runners” OR “Joggers” OR “Athletics” TITLE-ABS-KEY

* indicates a wildcard, that the search term can have any ending, TITLE-ABS-KEY indicates a title, abstract and keyword search

An external file that holds a picture, illustration, etc.
Object name is 40279_2022_1760_Fig1_HTML.jpg

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 flow diagram. *An in-depth list of excluded articles and the reasons can be found in Table 1 of the ESM

Eligibility criteria

The articles contain a system for running gait analysis using wearable technology

Sensing modality used was a wearable accelerometer, gyroscope, magnetometer or a combination of those (IMU), or pressure insoles

Included at least one clearly defined running gait outcome measure, for example:

Spatiotemporal (global outcomes of the running gait cycle): e.g. running velocity, acceleration of centre of mass, distance, ground contact time, step length, step frequency (cadence), stance time and flight time

Kinematics (description of segmental or joint movement, generally in the three cardinal planes: sagittal, coronal [frontal], transverse planes, without consideration for forces): e.g. ankle dorsiflexion angle, ankle angular velocity or ankle angular acceleration

Kinetic (the action of forces in producing or changing motion): e.g. GRF, peak pressure, centre of pressure, braking, impulse, time to peak pressure, pressure time integral, loads, force time integral, contact area and peak tibial acceleration

Articles were written in the English language

Book chapters, review papers, case studies (i.e. a study examining one individual), letters, short communications, technical notes, conference proceedings and other non-peer-reviewed literature

Studies evaluating the use of wearable technology for determination of step counts, distance, level of physical activity, classification or recognition of types of physical activity

Studies focusing on the estimation of physiological measures (e.g. metabolic equivalents), maximal oxygen consumption, examination of external or neuromuscular load, stiffness, vibration and shock absorption of lower limbs

Studies aiming to determine running power, stability or economy

Studies investigating walking gait variability or regularity

Studies not evaluating straight running (e.g. change in direction tasks or cutting manoeuvres)

Studies investigating the use of biofeedback or gait retraining (i.e. non-natural running gait)

Studies involving use of altered weight conditions (e.g. wearable resistance, anti-gravity treadmills or water-based protocols)

Aims to evaluate only computer algorithms, machine learning or statistical approaches

Studies evaluating robotic systems, exoskeletons, prosthetics, virtual reality environments and simulated data or models

Study involves participants < 8 years of age

Study concerns non-human animal subjects

GRF ground reaction force, IMU inertial measurement unit

Data Extraction

Data were extracted by the author (RM) using a custom form to support standardised extraction (Appendix 1). Data were synthesised into a table format by the author (RM) and a second author (SS) confirmed data entry. Studies were divided into two categories based on the aims of this review: validity and reliability and application. Information extracted from each article included participants, sensor(s), study protocol, reference/additional measure, analysis, outcome measures and key findings.

Search Results

From the 7643 articles identified through the database search, 122 papers met the inclusion criteria. An additional nine articles were identified through a search of reference lists. The complete flow diagram of the screening procedure is shown in Fig. 1 . A total of 131 articles were reviewed, with overlapping reports on several topics; specifically, 56 examined validity, 22 examined reliability and 77 investigated the application of wearable technology for a running gait analysis (Fig. 2 ). Table 2 of the ESM provides key details about each article.

An external file that holds a picture, illustration, etc.
Object name is 40279_2022_1760_Fig2_HTML.jpg

Summary of types of studies ( a ) and participant type ( b ) included in the review

Participant Characteristics

Overall, studies included between three [ 19 ] and 187 [ 23 ] participants, with the average number of participants being 26 (± 27). The mean age of participants was 28.3 (± 7.0) years. Two studies did not provide any age-related details [ 21 , 24 ], with three studies providing age ranges only [ 25 – 27 ]. Three studies investigated running gait in participants with an average age over 50 years, none of which performed a comparison of gait patterns across age groups that included older adults [ 28 – 30 ]. Most of the reviewed studies ( n = 82) included both male and female participants, with eight examining differences between male and female participants [ 31 – 38 ] and three of these studies finding significant differences between sexes [ 32 , 36 , 37 ]. Thirty-nine studies had male participants only, while only three studies solely examined female participants [ 39 – 41 ], seven studies did not report the sex of participants [ 21 , 42 – 47 ] and two studies did not provide a breakdown of the sexes [ 26 , 48 ]. The primary group of interest was healthy young adults who were recreationally active (Fig. 2 ), with only six studies investigating injured runners [ 39 , 48 – 52 ] (Table 2 of the ESM). Twenty-four studies commented on the FSP of the participants: 18 of these investigated rear-foot strikers [ 34 , 43 , 52 – 70 ], one study examined rear-foot strikers or neutral FSP [ 71 ], and five studies compared running gait parameters between FSPs [ 23 , 40 , 52 , 68 , 69 ].

Wearable Instrumentation

Inertial measurement units.

Sixty-two articles stated that they used IMUs; however, 14 of these studies only used the accelerometer capabilities within the IMU [ 23 , 24 , 38 , 52 , 56 , 57 , 62 , 72 – 78 ] and 20 studies stated they used the accelerometer and gyroscope components for the data analysis [ 26 , 33 , 43 , 49 , 50 , 54 , 59 , 61 , 79 – 90 ]. The remaining 27 studies either did not comment on components used [ 70 , 91 – 98 ] or implied they used all accelerometer, gyroscope and magnetometer components for the data analysis [ 19 – 21 , 28 , 29 , 36 , 42 , 99 – 109 ]. One study used an IMU and a separate one-dimensional accelerometer [ 61 ]. One study solely used the gyroscope housed within the IMU, using a sampling frequency of 102.4 Hz and not commenting on the gyroscope range [ 27 ]. Across these studies, the most common sampling frequency was 100 Hz ( n = 12) [ 19 , 21 , 28 , 29 , 38 , 42 , 73 , 75 , 76 , 78 , 82 , 94 ], but included use of 10 Hz [ 36 ] and 2000 Hz [ 43 ], and the range of the accelerometers was between ± 2.0 g [ 36 ] and ± 200 g [ 56 , 57 ], with 16 g being the most frequently used ( n = 14) [ 24 , 26 , 36 , 59 , 83 , 85 – 89 , 99 , 100 , 102 , 103 ]. The gyroscope ranges (± °/s) used were 1200 ( n = 7) [ 19 , 20 , 59 , 86 – 89 ], 2000 ( n = 10) [ 21 , 26 , 42 , 61 , 83 , 85 , 92 , 100 , 102 , 103 ], 4000 [ 43 ] and a variety ( n = 1) [ 36 ]. A variety of sampling frequencies (4–1000 Hz), accelerometer (2, 4, 6, 8, 16 g) and gyroscope ranges (250, 500, 1000, 2000°/s) were used in one study that used an IMU [ 36 ]. Twenty-eight studies reported the weight and/or size of the IMU used, with a large range. IMUs were as small as 6.0 × 1.85 × 0.5 cm [ 83 ] up to 8.8 × 5.0 × 1.9 cm [ 38 ], and the weight of the IMUs ranged from 4 g [ 43 ] to 550 g [ 92 ] (Table 2 of the ESM).

Accelerometers Only

Of the 40 studies that stated they used single accelerometer configurations in their methodology, notably, 13 studies did not comment on the dimensions [ 31 , 41 , 51 , 58 , 60 , 63 , 64 , 110 – 115 ], one-dimensional accelerometers were used exclusively in four studies [ 44 , 48 , 61 , 116 ], one study featured a two-dimensional accelerometer [ 53 ], one study used both one-dimensional and 3D accelerometers [ 117 ], and 21 studies used 3D accelerometers only (Table 2 of the ESM). Where reported, sampling frequency was between 30 Hz [ 117 ] and 1667 Hz [ 111 ], with 1000 Hz being the most common ( n = 14) and the range of the accelerometers was between ± 0.05–2.0 g [ 117 ] and ± 50 g [ 118 ], with 16 g being the most frequently used ( n = 8). There was a large range in reported sizes of accelerometers from as small as 4.0 × 2.2 × 1.2 cm [ 119 ] up to 5.42 × 10.25 × 1.7 cm [ 120 ], weighing between 2.5 g [ 119 ] and 67 g [ 121 ] (Table (Table3 3 and Table 2 of the ESM).

Type of sensor used within reviewed studies

Type of wearable technology used		References
IMU (a combination of sensors in one unit; accelerometer, gyroscope, magnetometer)	61	[ – , , , – , , , , , , , , , , , , , , , – , – , ]
Accelerometer only	37	[ – , , , , , , , , , , , , – , – ]
Pressure sensor/insole	27	[ , , , , , , , – , , , , – ]
Pressure sensor and accelerometer	2	[ , ]
Pressure sensor and IMU	2	[ , ]
IMU and separate accelerometer	1	[ ]
Gyroscope	1	[ ]

Note: Table refers to how each study classified the technology used, rather than the components used for analysis (i.e. some studies used an IMU, but only analysed data from one element of the unit)

IMU inertial measurement unit

Pressure Sensors/Insoles

Of the 131 articles reviewed, 31 studies focused on pressure or force-sensitive insoles; two of those 31 studies investigated the use of a combined pressure insole and an IMU [ 98 , 122 ] and a further two studies utilised a pressure insole alongside accelerometers [ 53 , 123 ]. Of the studies that used pressure insoles, the lowest sampling frequency was 50 Hz [ 39 , 124 , 125 ] and the highest was 1029 Hz [ 123 ]; 100 Hz was the most common sampling frequency ( n = 13). Seven studies commented on the dimensions of the insoles/sensors [ 25 , 53 , 65 , 66 , 71 , 122 , 126 ], with the dimension range from 0.6 × 0.4 × 0.12 cm [ 65 ] to 2.55 cm [ 66 ] (Table (Table3 3 and Table 2 of the ESM).

Gyroscope Only

One study solely used a gyroscope (not encompassed in an IMU), with a sampling frequency of 1500 Hz and a gyroscope range of 250°/s [ 127 ] (Table (Table3 3 and Table 2 of the ESM).

Number of Sensors

In the reviewed studies that used IMUs, accelerometers or gyroscopes, most studies used one ( n = 56) or two ( n = 30) sensors. Few studies used more than two sensors, for example, others used three [ 77 , 97 ], four [ 62 , 74 , 117 ], five [ 106 ], seven [ 103 , 109 ], eight [ 19 , 20 , 85 ], nine [ 105 ], 12 [ 21 ] or 17 sensors [ 108 ]. Where studies used more than one sensor, they were not necessarily the same type of sensor (e.g. one IMU and one accelerometer). For example, two studies sought to compare multiple and single sensor units [ 93 , 130 ]. Notably, of the studies that used multiple sensors, six sought to investigate the influence of sensor location on outcome measures [ 74 , 85 , 93 , 101 , 128 , 130 ] (Table 2 of the ESM).

The most common inertial wearable locations were the tibia ( n = 42), mostly located at the distal anteromedial aspect; shoe ( n = 38), varying locations of dorsal aspect/shoelaces/instep, cavity, ankle, heel and fifth metatarsal; and lower back [including sacrum] ( n = 24). One study used instrumented earbuds [ 135 ], and a further four studies placed wearables on the sternum/chest and these were always in combination with a lower body sensor placement [ 19 , 20 , 89 , 93 ]. In the seven studies that used wearables on the upper back, five studies placed the sensor in a harness/vest [ 21 , 38 , 105 , 121 , 129 , 130 , 133 ]. Two studies located accelerometers on the wrist, housed in GPS watches [ 21 , 31 ] and one study mounted 17 sensors onto a lycra suit that participants wore [ 108 ] (Fig. 3 and Table 2 of the ESM).

An external file that holds a picture, illustration, etc.
Object name is 40279_2022_1760_Fig3_HTML.jpg

Frequency distribution of the body segments on which wearables were placed

Extracted Features/Outcome Measures

Table Table4 4 provides a full breakdown of reported outcome measures. Outcomes included spatiotemporal, kinematic and kinetic running gait parameters. Among the studies that investigated spatiotemporal parameters, measures of distance included SL ( n = 29) and less commonly, vertical oscillation ( n = 7), while ground contact time (GCT)/stance time ( n = 49), SF ( n = 36), and stride or step time ( n = 16) were the most frequently reported temporal measures. Measures of acceleration included peak or average acceleration of a particular body segment, most commonly the tibia ( n = 28). Where pressure insoles were used, plantar pressure ( n = 17), contact area ( n = 12) and pressure or force–time integral ( n = 10) were the most reported outcomes.

Outcome measures extracted from wearable technology within reviewed studies

Outcome	Definition	References

Ground contact time/stance time = 49	The time between initial foot contact and toe-off for the same foot	[ , – , , , , , – , – , , , , , , , , , , – , , , , , , , , , , , , – , , , , , , , , , ]
Cadence, step/stride frequency/rate = 36	The number of steps or strides taken during a given time	[ , , , , , – , , , , , , , , , , , , , , , , , , , , , , , , , , , , ]
Step/stride length = 29	The distance between successive points of initial contact of the same foot (stride) or opposite foot (step)	[ , – , , , , , , , , , , , – , , – , , , , , , , , ]
Step/stride time = 16	The time between two consecutive heel strikes of the same foot (stride) or opposite foot (step)	[ , , , , , , , , , , , , , , , , ]
Foot strike pattern/strike index/foot strike angle = 15	The moment, way or angle when the foot first makes contact with the ground	[ , , , , , , , , , – , , , ]
Flight time = 15	The time between toe-off from one foot to initial contact of the other foot	[ , , , , , , , , , , , , , , ]
Acceleration* = 15	The rate of change of the velocity of any segment (excluding tibia)	[ , , , , , , , , , , , , , , ]
Speed/velocity = 15	The rate of change of position (directional)	[ , , , , , , – , , , , , , ]
Gait events = 11	Identification of any of the key gait events, e.g. heel strike, toe-off, mid-stance	[ , , , , , , , , , , ]
Vertical oscillation = 9	The amount that the torso or COM moves vertically with each step or stride	[ , , , , , , , , ]
Swing time = 7	The period during which the foot is not in contact with the ground	[ , , , , , , ]
Cycle time = 4	The time taken to complete a single gait cycle (can be measured from any gait event to the same subsequent event on the same foot)	[ , , , ]
Acceleration of centre of mass = 4	The rate of change of the velocity of the centre of mass	[ , , , ]
Step/stride height/foot clearance = 2	The vertical distance obtained during the swing phase	[ , ]
Symmetry = 2	Any measure of imbalance between the right and left leg	[ , ]

Ankle/foot kinematics = 19	Description of ankle or foot movement in any of the three cardinal planes: sagittal, frontal, transverse planes, without consideration for forces	[ , , , , , , , , , , , , , , , – , ]
Hip/pelvis kinematics = 11	Description of hip or pelvis movement in any of the three cardinal planes: sagittal, frontal, transverse planes, without consideration for forces	[ , , , , , , , , , , ]
Knee kinematics = 10	Description of knee movement in any of the three cardinal planes: sagittal, frontal, transverse planes, without consideration for forces	[ – , , , , , , , ]
Trunk kinematics = 3	Description of trunk movement in any of the three cardinal planes: sagittal, frontal, transverse planes, without consideration for forces	[ , , ]

Derivatives of tibial acceleration = 28	Acceleration of the tibia, including peak positive acceleration, gradient, slope, magnitude, loading rate	[ , , , , , , , , , , – , , , , , – , , , , , , ]
Plantar pressure = 17	Pressure = force/area, where force describes the vGRF exerted and area describes the surface area of the foot that is in contact with the ground during running	[ , , , , , , , – , , , , , ]
Ground reaction force = 13	The force exerted by the ground on a body in contact with it	[ , , , – , , , , – ]
Contact area = 12	Area of contact for each or any plantar region	[ , , , – , , , , ]
Pressure or force–time integral = 10	The cumulative effect of pressure or force during a step cycle (area under the pressure–time or force–time curve)	[ , , , , , , , , , ]
Impact = 9	The vertical force observed during the initial contact	[ , , – , , , , ]
Force = 8	Including rate and magnitude of force development	[ , , , , , , , ]
Impulse = 8	A measure of the force applied for a specific time and distance	[ , , , , , , , ]
Braking = 7	The force applied to the body that causes it to slow down	[ , , , , , , ]
Load/loading rate = 5	The speed at which you apply forces to the body	[ , , , , ]
Centre of pressure = 2	The point of application of the GRF	[ , ]

Note: Some studies fall under more than one category

GRF ground reaction force, vGRF vertical ground reaction force

Environment

Figure 4 provides an overview of the environments used for running assessments. Most studies ( n = 93) used indoor facilities only that primarily involved treadmill running ( n = 62). Thirty-two of the reviewed studies investigated running gait in outdoor environments only, and six studies used a combination of both indoor and outdoor testing [ 21 , 53 , 55 , 60 , 91 , 143 ]. Eighteen studies examined running gait over more than one surface [ 21 , 36 , 46 , 53 – 55 , 60 , 66 , 67 , 80 , 87 , 91 , 105 , 118 , 125 , 131 , 143 , 151 ]. The most popular outdoor surface was a running track ( n = 16), followed by concrete ( n = 13). Five studies did not report the outdoor surface type where testing took place [ 28 , 33 , 50 , 59 , 89 ] (Fig. 4 and Table 2 of the ESM).

An external file that holds a picture, illustration, etc.
Object name is 40279_2022_1760_Fig4_HTML.jpg

Summary of environments that studies were conducted within ( a ) overall, ( b ) indoor environments and ( c ) outdoor environments

Running Gait Protocol

Duration/distance.

The duration or distance of the analysed running protocol varied greatly by study. One hundred and nine studies analysed running gait in a single day, while 22 studies tested running gait over 2 or more days (Table 2 of the ESM). Protocols were heterogeneous and consisted of:

Analysing a certain number of steps, strides or gait cycles ( n = 50) . For example, four stages of 100 strides [ 20 ], three different footwear types, and five trials each, analysing one right foot strike per trial [ 115 ].
Analysing running gait for less or equal to 60 s ( n = 42). For example, one 15-s run [ 56 , 74 ], and 30 trials lasting 30 s (five trials, six conditions, last 30 s of 3-min trials) [ 71 ].
Analysing running gait in trials lasting over 1 min and less than 5 min ( n = 17) [ 32 , 36 , 49 , 72 , 75 , 80 , 83 , 92 , 97 , 98 , 101 , 120 , 123 , 128 , 134 ]. For example, three sessions each consisting of three 5-min runs at varying speeds [ 75 ], seven 100-m runs (outdoor) and seven 60-s runs (treadmill) [ 91 ], or 3 min [ 36 ].
Analysing gait patterns over longer distances that were more representative of a typical run [i.e. more than 5 min] ( n = 22). For example, dissecting a 100-km (ultra-marathon) into ten 10-km segments to investigate the effects of fatigue [ 31 ], or analysing one 10-km segment and 15 2-km segments of a marathon race [ 29 ]. One study examined various distances; however, different participants were used for each distance [ 80 ].

There was variation in speed amongst the reviewed studies. Seventy-seven studies used controlled speeds (58 of these controlled at a set pace), with a range from 2 m/s [ 85 , 121 ] to 26 km/h [ 117 ]. Nine studies controlled speed based on individual performance; four of these studies used personal bests as the benchmark [ 42 , 48 , 87 , 108 ] and two studies controlled speed based on the participants’ preferred speed (e.g. 85 and 115% preferred speed) [ 21 , 98 ]. The remaining three studies used physiological measures to determine speed used [ 19 , 70 , 111 ], for example, one study controlled running speed at 2 mmol/L blood lactate [ 70 ] (Table 2 of the ESM).

Fifty-five studies examined running gait at self-selected speeds; amongst these studies there were large variations in instructing speed. For example, six studies used race scenarios [ 20 , 29 , 31 , 59 , 82 , 89 ], 14 studies asked participants to run based on perception (e.g. ‘easy run’/‘comfortable’, 75% maximum effort) [ 33 , 44 , 49 , 54 , 79 , 99 , 101 , 105 , 125 , 128 , 136 , 141 , 147 , 148 ], and a further 11 studies instructed participants to run at maximum effort/speed [ 74 , 86 , 88 , 90 , 99 , 100 , 104 , 105 , 129 , 132 , 133 ].

Eight studies combined controlled and self-selected speeds [ 35 , 62 , 76 , 116 , 118 , 127 , 132 , 135 ]. For example, Giandolini et al. examined participants at 10, 12, 14 (female) and 16 km/h (male), maximum aerobic speed and participant’s preferred speed [ 116 ]. Where speeds were reported, 46 studies included two or more speeds in their protocol.

Sixteen studies commented on the running gradient [ 28 , 34 – 36 , 76 , 80 , 90 , 101 , 108 , 109 , 125 , 128 , 131 , 144 , 147 , 149 ]. The majority of studies ( n = 5) used a 0% gradient [ 34 , 101 , 109 , 128 , 144 ] or a 1% gradient ( n = 4) on the treadmill [ 35 , 36 , 76 , 125 ]. Three studies analysed the effects of different gradients [ 28 , 80 , 90 ], and one study investigated the effects of low and high altitudes on running gait [ 82 ].

Forty-three studies required participants to wear standardised shod running shoes, of whom 42 utilised the participant’s own running shoes. Two studies tested participants in standardised footwear and in their own footwear [ 109 , 116 ]. One study tested participants in socks as participants wore the insoles seeking validation wearing tight-fitted socks without shoes to allow a more direct measurement comparison [ 150 ]. Lucas-Cuevas et al. used standardised shoes and participants’ own insoles inside the participants’ own running shoes [ 119 ]. Forty-six studies did not comment on the footwear used (Table 2 of the ESM).

Validity and Reliability Studies

Fifty-six studies focussed on the validation of wearables for running gait assessments, with 18 also examining the reliability of devices [ 47 , 98 , 99 , 103 , 104 , 110 , 117 , 120 , 130 , 134 , 136 , 140 , 144 , 149 ]. Eleven studies investigated between-day reliability [ 34 , 47 , 98 , 106 , 117 , 120 , 122 , 140 , 142 , 144 , 149 ], and three studies solely examined the reliability of wearable technology [ 87 , 134 , 138 ] (Table 2 of the ESM).

Protocols for Validity and Reliability

Participants.

Protocols to assess validity and reliability varied greatly. Overall, studies included between five [ 27 ] and 100 [ 95 , 96 ] participants, with the average number of participants being 22 (± 18). The mean age of participants was 26.8 (± 4.5) years. Two studies only provided age ranges [ 26 , 27 ] and one study did not report age [ 24 ]. Sixteen studies used male-only participants [ 61 , 73 , 84 , 92 , 94 , 97 , 98 , 110 , 115 , 117 , 121 , 127 , 129 , 130 , 133 , 138 ], two did not report or provide the breakdown of sex [ 26 , 45 ], and the remaining studies included both male and female participants. All studies included healthy participants and four studies commented on the FSP of the participants [ 34 , 61 – 63 ].

Environmental Control

Six validity and/or reliability studies used outdoor environments, with participants running on concrete [ 79 , 87 ], artificial turf [ 105 ] and track [ 102 , 105 , 120 , 152 ]. Of the remaining studies that used indoor environments, 31 ran on treadmills, 15 ran over-ground [ 45 , 63 , 78 , 85 , 94 , 104 , 110 , 115 , 121 , 124 , 127 , 130 , 141 , 142 , 150 ] and six ran on a track [ 61 , 99 , 100 , 129 , 133 , 136 ]. No studies used both indoor and outdoor testing or examined running gait over more than one surface. Seven studies commented on the treadmill gradient, one study set the treadmill at a 0, 10 incline and 10% decline [ 149 ], two studies used a 1% treadmill gradient [ 75 , 76 ] and the remaining study stated that no gradient was used (i.e. 0%) [ 34 , 101 , 128 , 144 ] (Table 2 of the ESM).

Distance/Time Control

Twenty-three studies focused on analysing a certain number of steps, strides or gait cycles, with the minimum being six foot strikes in total (three trials, two speeds) [ 127 ], and a maximum of 200 consecutive left and right steps of a 5-min run [ 140 ]. Thirty-three studies investigated running gait over particular distances or times whereby 23 studies analysed running gait for ≤ 60 s. Ten studies analysed running gait in trials lasting > 1 and < 5 min [ 36 , 75 , 92 , 97 , 98 , 101 , 106 , 120 , 123 , 128 ]. One study examined gait patterns over a long distance, i.e. up to 4 km [ 79 ], and another study did not comment on the number of steps or distance analysed [ 94 ]. Within reliability studies, ten analysed test-re-test reliability in a single day (i.e. two sessions in 1 day) [ 98 , 99 , 103 , 104 , 106 , 110 , 130 , 134 – 136 ] and 11 studies performed a test-re-test analysis on different days [ 34 , 47 , 87 , 117 , 120 , 122 , 138 , 140 , 142 , 144 , 149 ]. Those studies that assessed running gait on different days separated testing by a minimum of 24 h [ 34 , 140 , 144 ], and repeated testing within 1 week [ 120 , 149 ], 2 weeks [ 47 , 87 ] or 1 month [ 117 , 142 ], with one study repeating testing at 1 week and 6 months [ 138 ] (Table 2 of the ESM).

Speed Control

Thirty-one studies used controlled speeds, with the slowest speed set at 7 km/h [ 84 ] and the fastest speed set at 26 km/h [ 117 ]. Self-selected speeds were used in 21 studies, with a range from jogging [ 136 ] to maximum effort/sprint [ 86 , 99 , 100 , 102 , 104 , 105 , 129 , 133 ]. An additional five studies combined controlled and self-selected speeds [ 62 , 76 , 116 , 127 , 135 ]. One study did not comment on the treadmill speeds used [ 93 ]. Twenty-seven studies included more than one speed in their protocol; consequently 32 studies examined the effect of running speed on the validity and/or reliability of outcomes obtained (Table 2 of the ESM).

Footwear Control

Most studies did not comment on the footwear used. Thirteen studies standardised the footwear of participants [ 45 , 61 , 62 , 83 , 85 , 115 , 116 , 123 , 124 , 138 , 142 , 144 , 149 ], 17 allowed participants to wear their own running shoes [ 27 , 34 , 79 , 81 , 87 , 98 , 99 , 101 , 103 , 105 , 106 , 120 , 128 , 140 , 141 ] and one study required participants to run unshod while wearing insoles under socks [ 150 ].

Validation Reference Measures

Twenty-four studies used a laboratory reference of 3D motion capture, 14 used a two-dimensional video analysis [ 26 , 99 – 101 , 105 , 110 , 116 , 123 , 128 , 129 , 133 , 136 , 142 , 144 ], 17 used force plates [ 45 , 63 , 75 , 76 , 98 , 104 , 115 , 121 , 122 , 124 , 127 , 130 , 133 , 136 , 141 , 142 , 150 ], 17 used instrumented treadmills [ 38 , 62 , 81 , 84 , 93 , 95 – 97 , 103 , 106 , 123 , 135 , 140 , 144 , 149 ], one study compared measures to an accelerometery system implemented in the treadmill [ 34 ], 12 used timing gates/light barriers [ 61 , 63 , 86 , 99 , 100 , 102 , 104 , 105 , 110 , 115 , 127 , 129 ], five compared to other wearable technology [ 45 , 79 , 97 , 120 , 130 ] and one study used a practitioner observed step count [ 117 ] (Table 2 of the ESM).

Validity and Reliability Findings

Foot/shoe mounted devices.

Most validity studies ( n = 22) assessed shoe-mounted or foot-mounted devices. Reviewed studies showed that wearables could accurately measure stride time [ 85 ], speed, oscillation and GCT measures [ 79 , 86 , 134 ], step rate [ 93 ], FSP data [ 26 , 81 , 84 , 116 ] and SL [ 100 ] using shoe or foot mounted wearable technology. Conflicting findings regarding the validity of joint kinematics using shoe-mounted accelerometers were demonstrated [ 73 , 83 , 94 ].

Tibia-Mounted Devices

Fifteen studies showed that tibial-mounted devices are valid for the detection of gait events [ 63 , 127 ], step length [ 34 ], stride/step time [ 27 , 34 , 106 ], SF [ 34 , 93 , 106 ], tibial acceleration [ 34 , 115 ] and vertical GRF [ 136 ]. However, stance and swing times collected using a gyroscope yielded poor-to-moderate agreement with optical motion capture [ 27 ]. One study did not consider the validity; however, it demonstrated that an accelerometer had good-to-moderate reliability for peak tibial acceleration at 1 week and 6 months [ 138 ].

Lower Back and Waist Mounted Devices

Fifteen articles reported that wearables on the pelvis, waist or lower back are accurate for identifying stride, step, stance duration [ 106 ], centre of mass vertical acceleration [ 75 , 76 ], gait events [ 78 ], running speed, SL, SF [ 102 , 106 ] and kinetic measures [ 104 ]. Outcomes such as GCT, flight time and peak vertical GRF have conflicting evidence regarding accuracy and reliability [ 24 , 95 , 96 , 110 , 120 ].

Upper Back Mounted Devices

Six studies reported that wearables located on the upper back [ 38 , 105 , 121 , 129 , 130 , 133 ] had poor validity for examining gait symmetry [ 133 ], predicting GRF [ 121 , 130 ], measuring velocity [ 129 ] and peak or average accelerations [ 38 , 130 ], as well as poor reliability [ 130 ].

Multiple Device and Other Locations

Ten studies used more than one wearable in various locations and demonstrated good validity and reliability regarding spatiotemporal [ 106 , 117 ] and kinematic and kinetic measures [ 61 , 62 , 94 , 97 , 103 , 122 , 138 ]. However, the validity varied between outcome measures (i.e. good accuracy for knee kinematics but poor for ankle kinematics) [ 93 , 103 , 105 ]. Furthermore, the measurement of SF and GCT using an accelerometer embedded in a wireless earbud showed good test–retest reliability, face validity and concurrent validity [ 135 ].

Pressure Insole Devices

Eleven studies reported on pressure insoles, with most studies attempting to correlate plantar pressures with GRF [ 45 , 98 , 122 – 124 , 140 – 142 , 144 , 149 , 150 ]. Findings suggest that insoles are generally valid and reliable for measuring temporal measures [ 98 , 150 ] and kinetics, such as peak weight acceptance force, impulse and loading rate [ 124 , 140 , 142 , 150 ]. However, other studies suggest that the validity of the device is dependent upon the force outcome measure [ 123 , 149 , 150 ]. Overall, the validity and reliability of pressure insoles appears to be system [ 128 , 149 ], location [ 85 , 101 ] and speed dependent [ 27 , 99 , 102 , 127 ] (Table 2 of the ESM).

Application Studies

The aims of the applied use of wearable technology for running gait analysis fell into broad categories of footwear, clothing (e.g. compression socks, insoles), surface (as mentioned in Sect. 3.7.1 ), intrinsic factors (e.g. sex, FSP), performance (e.g. experience, speed), fatigue, detecting gait parameters (e.g. relationships between gait parameters) and running injuries (Table (Table5 5 ).

Summary of application of wearable technology

Application		References
Footwear and clothing	20	[ , , – , , , , , , , , , – , , , , ]
Surface	16	[ , , , – , , , , , , , , , , ]
Intrinsic factors	15	[ , , , , , , , , , , , , , , ]
Performance	17	[ – , , , , , , , , , , , , , ]
Fatigue	13	[ – , , , , , , , , , , ]
Detecting gait parameters	12	[ , , , , , , , , , , , ]
Running injuries	6	[ , – ]

Footwear and Clothing

Eighteen studies investigated the effects of footwear on running gait parameters (Table (Table5). 5 ). The majority of studies ( n = 17) investigated different types of footwear on spatiotemporal, kinematics and kinetics, and generally the studies were consistent in evidencing that footwear construction has a substantial influence on some running gait outcome measures obtained by wearable technology, for example, significant differences in tibial acceleration [ 44 , 64 , 113 , 114 ], SL [ 70 ] and loading parameters [ 37 , 43 , 45 , 62 , 65 , 71 , 148 ]. In contrast, other authors found no significant differences between shoe conditions [ 61 , 112 , 146 ]. In terms of clothing, Stickford et al. used wearable technology to examine whether wearing graduated lower-leg compression sleeves during exercise evokes changes in running biomechanics and Lucas-Cuevas et al. analysed the acute differences in stride parameters while running on a treadmill with custom-made and prefabricated insoles [ 119 , 139 ].

Intrinsic Factors

Results of the 15 studies that investigated characteristics of sub-groups or intrinsic factors relating to performance suggested that running patterns were likely individual and task specific (Table (Table5) 5 ) [ 29 , 32 , 80 ]. Of all the reviewed studies, five examined differences between male and female individuals [ 31 , 32 , 36 – 38 ], and three of these studies evidenced significant differences between sexes [ 32 , 36 , 37 ]. There were conflicting findings from the six studies that investigated the effects of FSP on running biomechanics [ 23 , 40 , 52 , 65 , 68 , 69 ]. Key findings argue that no significant differences existed for total maximum force, force–time integral, peak pressure and pressure–time integral, but the total contact area of rear foot strikers was higher than that of non-rear foot strikers [ 68 , 69 ]. In contrast, other studies demonstrated significant effects of the FSP on tibial acceleration, load rates and plantar pressure at varying plantar regions [ 23 , 40 , 52 , 65 ]. Two studies examined morphological differences of the foot and the influence on running gait [ 44 , 145 ]. Only one study examined the effects of age and anthropometric measures on running gait [ 31 ].

Performance

Of the applied studies that focused on performance aspects, 12 examined the influence of speed on running biomechanics [ 30 , 31 , 35 , 42 , 47 , 54 , 62 , 77 , 87 , 118 , 132 , 137 ], four investigated the experience of participants [ 30 , 32 , 42 , 118 ], one study examined the effects of altitude [ 82 ] and another study investigated gradient [ 90 ]. Associations of gait metrics with wellness and session perceived exertion was prospectively examined in one study [ 33 ] and specifically running kinematics in triathletes was investigated in another study [ 108 ].

Thirteen studies examined the effects of fatigue on running gait (Table (Table5). 5 ). The findings are conflicting regarding if changes in running gait are fatigue induced and if this is dependent on experience level. Some suggest that GCT, flight time, trunk anterior–posterior acceleration, peak impact acceleration swing time, swing velocity and foot strike angles show significant changes with fatigue [ 42 , 59 , 89 ]. In contrast, others indicate no changes in spatiotemporal or FSP throughout the run [ 42 , 88 , 132 ]. Burns et al. suggested that SF changes only with speed and not fatigue [ 31 ]. Studies suggest that fatigue-induced changes do occur but may be subject specific [ 19 – 21 , 111 , 143 ] and dependent on experience/skill level [ 21 , 29 , 72 ] or fatigue state [ 89 ].

Detecting Gait Parameters

Twelve studies sought to investigate methods that detect or influence running gait outcome measures (Table (Table5). 5 ). Studies sought to identify trends [ 25 , 28 ], examine relationships between running gait parameters [ 23 , 56 , 74 , 90 , 116 , 147 ] or investigate the effects of different methodologies on the outcome measures obtained [ 57 , 61 , 62 , 107 ].

Running Injuries

Applied articles focusing on running related injuries ( n = 6) sought to evaluate the effects of ankle taping, bracing and fibular reposition taping on running biomechanics [ 49 ], and to examine [ 52 ] and compare running gait parameters of injured and non-injured runners [ 39 , 48 , 50 , 51 ]. Table Table6 6 provides a summary of the most reported protocol features in the reviewed studies.

Summary of commonly reported details in reviewed studies

Protocol feature	Most reported
Participants	Young adults [average age of 28.3 (± 7.0) years]
Sample size	Average of 26 (± 27)
Experience	Recreational runner
Environment	Indoor on a treadmill
Run duration/distance	Set distance or duration
Speed/pace	Controlled speed
Gradient	0–1%
Footwear	Standardised shoes
Type of wearable	Inertial measurement unit at 100 Hz
Outcome measures	Ground contact time, stride or step length, stride or step frequency and tibial acceleration
Sensor location	Fixed to the tibia or shoe
Number of wearables	One

Only two studies sought to examine the usability, comfort or wearer’s perceptions of the device; both studies reported the wearables to be comfortable to wear and wearers did not feel affected in their movements [ 21 , 125 ].

This review examined 131 studies that examined the use of wearable technology for running gait analysis. Explicitly, this review reported on: (1) methodologies employed to assess validity and reliability of wearables for running gait assessment; (2) application of wearables to assess running gait; and (3) commonly reported running gait outcomes and findings. This review has demonstrated that the use of wearable technology for running gait assessment is emerging, but further work is required to establish a standardised methodology and the validity or reliability of instrumentation. We have provided a comprehensive overview of wearable technology used for a running gait assessment, and here we provide recommendations for future work.

Wearable accelerometers, gyroscopes, IMUs (combined accelerometer, gyroscope and magnetometer) and pressure insoles were used within the reviewed studies to examine running gait. There was generally a lack of consistency across the reviewed studies for several factors that may impact the accuracy of wearable technology used for a running gait assessment, which included the data acquisition rate, data analysis methods, and location and number of wearables. Our findings show that IMUs are the most used wearables for running gait assessments (closely followed by pressure insoles), but most studies have focused on analysing acceleration data only rather than gyroscope and/or magnetometer data [ 11 , 153 ]. However, evidence suggests that the use of all sensor data within a single IMU can improve the accuracy of movement quantification, particularly orientation [ 15 , 27 , 154 – 156 ]. Additionally, IMU accuracy for running gait assessments may have been impacted by the huge variation in sampling frequency and operating range between devices (4–1667 Hz, 2–70 g). For example, Mitschke et al. have shown that sampling frequency and operating range can influence the accuracy of outcome measures from IMUs, particularly when they are too low (e.g. < 100 Hz) to detect movement events [ 61 ]. Generally, wearables were deployed within the lower limb, with the tibia as the most common site (IMUs and pressure insoles) and most studies used one or two wearables, which may be because of the cost–benefit approach to the device set-up. For example, using multiple wearable technology inevitably costs more but there is a benefit of using multiple devices (that may be combined IMU and pressure insole systems), as more data acquisition allows for an increased accuracy of outcomes (e.g. gait events and spatiotemporal parameters) [ 157 ]. Most studies utilised only one wearable (IMU, accelerometer or gyroscope) to collect biomechanical data. However, it is important to consider the practicality and comfort of numerous wearables during natural running. Further research exploring the feasibility and necessity of utilising multiple wearables is required, or whether this can be condensed into one sensor, as this will enhance understanding of the optimal number and placement of wearables to deliver the most pertinent data while enabling a natural running gait.

A major issue in the approach to wearable instrument application is that only two studies examined the usability of the devices through engagement with end users. Wearable technology design and set-up can influence cost, usability and accuracy of the instruments, which may vary depending on the interests of different end users. Studies often lack considerations for the wearer’s physical, psychological and social preferences regarding the technology [ 158 ].

Outcome Measures

This review has highlighted that there is a need for a comprehensive assessment and reporting of running gait outcomes, which may require combined ‘multi-modal’ (e.g. combination of IMU and pressure insoles, or accelerometer and pressure insole) wearables to examine running gait. The reviewed studies primarily limited their assessments to only the examination of selective spatiotemporal or kinematic outcomes; specifically SF, SL, tibial acceleration and GCT were the most common outcomes reported. Despite numerous studies establishing that running biomechanics cannot be described based on a single parameter [ 159 – 162 ], most studies focused on singular (or a select few) running gait outcomes, for example, GCT [ 99 ], SF [ 31 , 117 ] or tibial acceleration [ 56 , 118 , 138 ]. Examination of selective parameters may explain in part the inconsistencies across study findings regarding the relationship between running biomechanics, performance and injury [ 161 , 163 – 166 ]. Furthermore, comprehensive reporting and consistency in the literature is hindered by the lack of consistent terminology for running gait outcomes, for example, vertical oscillation of COM and stance duration have no relation to RRI [ 14 , 163 ]. The lack of consensus is further impacted by the fact that there are no ‘gold-standard’ algorithms for the detection of running gait outcomes from wearable sensor set-ups, which likely explains the large volume of outcomes reported in the reviewed studies. In order to derive appropriate algorithms and report findings in a consistent manner, examination of multiple running gait outcomes (i.e. spatiotemporal, kinematic, kinetic) may require a combination of IMUs and pressure sensors, which allows for a comprehensive assessment and may improve outcome accuracy (e.g. vertical GRF is most accurate with the use of pressure sensors or multiple IMUs) [ 97 ], but the volume of outcomes may create other methodological issues when examining a finite number of individuals. Despite these limitations, it is pertinent to consider whether such idealist methodologies are clinically and practically feasible within a given context.

Outcomes obtained from small cohorts may not accurately represent the population being studied and may lead to poor statistical power or inconsistency across study findings. This was evidenced within the reviewed studies, as studies primarily investigated running gait in small sample sizes (i.e. n < 30) of young adults, which limits the generalisability of results. For example, Burns et al. demonstrated that the variability of an elite runner’s SF is linked to both speed and fatigue but not to any other characteristics of the runner [ 31 ]. In contrast, Reenalda et al. demonstrated that that changes in SF are dependent upon the individual; however, the authors were unable to perform an analysis at a group level because of their limited sample size ( n = 3), thus stating that the observed effects of fatigue on running mechanics are confined to the runners analysed only and may not be representative for other runners [ 20 ]. The small sample sizes of the reviewed studies are surprising considering there is evidence from walking studies that gait analyses in a natural environment can be conducted on larger scales owing to the advancements in wearable technology [ 153 , 167 , 168 ]. The inclusion of larger sample sizes would facilitate the identification of subgroups of running patterns and the generalisability of the findings into the populations being studied. With the portability and ease of use of wearable technology, future studies should consider monitoring the running gait patterns of larger samples as it will allow for prospective studies and subgroups to be identified. Furthermore, only three studies examined running gait with an average age of over 50 years. However, none of the studies that examined older adults compared outcome measures to younger adults. Burns et al. noted that SF was not related to age; however, their sample only consisted of 20 participants, with an age range of 26–56 years (average age 38.1 ± 6.4 years) [ 31 ].

Test Protocols

Differences among study protocols in running gait testing conditions, and the definition of outcome measures, limited the ability to directly compare outcomes across studies. Nonetheless these protocol differences highlight the versatility of wearables, proving they can provide data on realistic and spontaneous running scenarios. Treadmill running was the most common means to evaluate and quantify running gait. Use of a treadmill has the advantage of providing a standardised and reproducible environment where speed can be easily controlled and the required calibration volume for the optical system is considerably reduced. However, running speed is directly related to cardiovascular factors [ 169 ] and biomechanical factors [ 36 , 170 ], and therefore imposing a set speed through a treadmill, rather than allowing runners to select the speed at which they are comfortable running, may produce alterations in running gait. Indeed, Zamparo et al. and Lussiana and Gindre indicated that self-selected speed related to individual energy-saving strategies [ 170 , 171 ], and Kong et al. suggested that self-selected speeds may eliminate abnormal kinematic patterns [ 172 ]. Similarly, despite the known impact of the gradient on running gait, there were very few reviewed studies that examined this [ 173 – 175 ], but some studies did set the treadmill to 1% to compensate for the known differences between treadmill and over-ground running [ 176 ]. However, recent research has suggested that there may be more to consider than just the gradient when attempting to replicate over-ground running on the treadmill [ 177 – 179 ].

Protocols need to carefully consider where running is examined with wearables. Treadmill running may not truly reflect natural running behaviour, as Montgomery et al. demonstrated that non-motorised treadmills generate large reductions in peak tibial acceleration, large to very large increases in SF during running when compared to over-ground and motorised treadmills conditions [ 46 ]. Therefore, studies have moved beyond the laboratory to more natural running environments (i.e. indoor or outdoor running tracks, or sports venues), which has largely involved the examination of differences in running gait between different types of running surfaces [ 55 , 67 , 118 , 180 ]. For example, when Hong et al. compared plantar loads when running on a treadmill, concrete and natural grass, it was shown that running on a treadmill induced lower peak plantar pressure and longer contact time for the total foot and two toe regions [ 55 ]. Additionally, several other reviewed studies suggest that running on natural grass may reduce stress on the musculoskeletal system and alter gait compared with running on a more rigid surface such as concrete or asphalt [ 66 , 67 , 151 ]. Similarly, there may be differences in kinematic and kinetic patterns when running on a treadmill compared with over-ground running [ 14 , 53 , 55 , 67 ], which is not considered in running assessment protocols. Research has demonstrated that treadmill running may influence lower limb kinematic patterns, landing patterns and sagittal-plane foot strike angles when compared with over-ground running [ 166 ]. The differences exhibited can be attributed to several factors, such as treadmill running being unable to mimic instantaneous speed changes that inherently occur during over-ground running, as well as other environmental factors (i.e. irregular surfaces and gradient) [ 166 , 181 ]. However, some consider treadmill running can be comparable to over-ground running depending on the outcome measures examined [ 166 , 182 ], which highlights the need to carefully design protocols around specific running features of interest.

Most reviewed studies examined running over less than 1 min, but there was a lack of protocol consistency as studies varied in the number of steps, distance, number of trials and time of trials that they examined in runners, which made it difficult to generalise findings. Because of potential changes in running biomechanics over long runs, analysing an abundance of steps may be beneficial to gain consistency in outcomes [ 183 ]. Few authors have addressed a longitudinal running gait analysis, in terms of over an extended time period (e.g. training season) or over longer distances, using wearable technology [ 19 – 21 , 28 , 29 , 31 , 50 , 82 ]. However, the studies that examined longer runs assessed running in a more natural environment (i.e. on a running track or outside over-ground) that allowed for greater time and distances to be studied compared with treadmill studies. Examining more and longer runs would potentially help divulge data regarding injury mechanisms and performance measures, thus informing practice by determining typical healthy running patterns as well as atypical gait patterns. Similarly, moving towards more realistic running environments that may be expected for commercial wearables was also reflected in the fact that a third of the reviewed studies allowed participants to wear their own running shoes (with a third requiring standardised shoes and the rest not reporting their footwear) [ 116 , 119 , 150 ]. This may signify a move towards attempting to use wearable technology with any individual running footwear, which would replicate commercial use.

Validity and Reliability

Despite their widespread use, fewer than 10% of commercially available wearable technology are validated against an accepted ‘gold standard’ [ 184 ]. However, our review suggests that validation of research-grade (non-commercial) wearable technology for running gait assessment has been previously performed. Validity was performed by examining outcomes against ‘gold-standard’ reference measures (e.g. 3D motion capture, two-dimensional video capture, force plates, instrumented treadmills or timing gates). However, differences in laboratory references make it difficult to compare the validation of different wearable technology. For example, García-Pinillos et al. used a high-speed video analysis system (1000 Hz) as a laboratory reference [ 101 ], whereas the other studies have compared against the Optojump Next ® and video cameras [ 110 ], which is largely owing to the expense of laboratory references and the need for data capture in a more ‘natural’ setting (i.e. not in a gait laboratory). Photoelectric cell-based systems (i.e. Optojump Next ® ) and video measures were considered as adequate proxy systems given their demonstrated good validity in comparison to force platforms [ 185 , 186 ], but they may not be the best reference system available. Findings from this review would suggest that outcomes from wearable technology for running gait should be validated against a known and accepted laboratory standard reference, such as 3D motion capture and force plates, to establish validity. Wearable technology was generally found to be valid for examining most running gait outcomes, particularly spatiotemporal measures, compared to laboratory references; however, this appears to be dependent upon the location of the wearable, the system and testing protocol (e.g. speeds) used, as well as the gait characteristics obtained [ 74 , 85 , 101 , 130 ]. For example, accelerometers, gyroscopes or IMUs on the foot may provide the most accurate derivations of stride measures [ 99 , 101 , 128 ], but caution should be taken when using wearables located at the thoracic spine, as outcomes obtained from such placement appeared inadequate to predict gait symmetry, peak vertical and resultant GRF [ 38 , 121 , 129 , 130 , 133 ].

Reliability studies of wearables for running gait are less established, as the majority of studies included in this review used one experimental session, but there were several studies that performed test–retest runs within the same session [ 99 , 103 , 104 , 110 , 130 , 134 , 136 ] or two sessions on different days [ 47 , 98 , 117 , 120 , 138 , 140 , 144 , 149 ]. Results demonstrated that outcomes of GCT, flight time and SF are reliable from a foot or lumbar spine placement [ 110 ], while foot-worn IMUs can provide reproducible calculations of stride time and SL [ 61 ]. Furthermore, placement on the tibia and lumbar and thoracic spine had excellent reliability for determining vertical GRF from accelerometer data [ 136 ].

Application of Wearables

The reviewed studies of running gait measured with wearables focussed on several key areas of investigation, specifically injury, fatigue, performance, footwear/surface, methods for gait detection and intrinsic group factors. There were a range of differences in running gait outcomes with a group-based analysis of these factors. Despite differences being found, the specific spatiotemporal, kinematic and kinetic measures that could be used to best investigate certain aspects of running gait (e.g. fatigue, footwear) require further investigation. For example, while there were differences in running gait for those with current or previous injuries [ 48 , 50 , 52 ], there were no studies that examined outcomes for the risk of overuse running injuries.

Fatigue state was examined to understand changes in running mechanics with the potential for injury. However, few studies have exploited the benefits of wearable technology to explore real-world long-distance running sessions characterised by progressive fatigue [ 20 , 21 , 29 , 82 , 163 ]. Examining runners at varying stages or for the duration of a prolonged run in ecologically valid settings will add to the growing body of evidence using wearable technology to better understand the effects of training and fatigue on changes in running biomechanics [ 14 , 19 , 20 ]. These data can then be used to inform the runner of significant atypical changes in their running gait that may increase risk of RRIs. For example, it is well documented that running-related fatigue can affect running kinetics [ 187 ], kinematics [ 19 , 188 , 189 ] and certain spatiotemporal parameters [ 72 , 82 , 190 ]. Strohrmann et al. provides support for numerous cases, categorising changes into three groups: (1) changes that occurred for all runners (e.g. decrease of the heel lift); (2) changes that depended on the runner’s skill level (e.g. increase of foot contact duration); and (3) and changes that were highly dependent on the individual, (e.g. increase in shoulder rotation) [ 21 ].

Footwear was examined in a variety of studies, which primarily focussed on differences in running behaviour, with a suggestion that this may lead to injury. For example, Butler et al. evidenced that low-arch runners exhibited a reduction in peak tibial internal rotation in motion-controlled shoes compared with cushioned shoes, whereas high-arch runners experienced a lower peak positive acceleration in the cushioned shoe compared with the motion control shoe [ 44 ]. Similar to footwear, running surface has also been studied to examine the potential impact on performance and injury. For example, de Ruiter et al. demonstrated differences in running speed and GCT during outdoor over-ground running on flat terrain, and in varying weather conditions [ 79 ]. Studies have generally found that the footwear/surface can influence running gait characteristics, which needs to be carefully considered when making performance and injury risk/recovery decisions.

Intrinsic factors of runners may also impact running gait, with studies typically splitting cohorts into groups based on performance measures (amateur, elite), injury status (i.e. previously injured or not), age (young or old) or sex (male, female). The reviewed studies primarily assessed recreational runners, showing differences in running gait at different levels of performance [ 32 ]. For example, novice runners exhibit more pronounced changes in running kinematics in response to fatigue compared with elite runners [ 189 ]. Furthermore, Strohrmann et al. stratified runners based on their weekly mileage (experience), but did not find differences in mechanics across these groups [ 21 ]. However, not all studies have demonstrated differences between pre-determined intrinsic factor groups for certain outcome measures; for example, Burns et al. demonstrated that years of running experience did not significantly affect SF, and nor did sex [ 31 ]. There was a lack of sex-based analyses in the reviewed studies, which was surprising considering the established differences in running mechanics between male and female individuals [ 191 , 192 ]. For example, Moltó et al. observed no significant differences in pelvic tilt or obliquity between the sexes; however, they did find significant differences in the range of pelvic rotation, with female runners presenting a greater range [ 36 ]. Queen et al. also evidenced different loading patterns between sexes and significant differences existed for the foot contact area (middle forefoot), with a maximum force at the lateral forefoot dependent on the shoe type [ 37 ]. Findings from Clermont et al. support this, highlighting the importance of separating runners into sex-specific subgroups first when classifying runners based on performance in order to better reflect the kinematic differences between sexes, and this is consistent with previous research [ 32 , 193 , 194 ]. This further highlights the need for a comprehensive assessment of running gait outcomes in order to detect characteristics that may be impacted by intrinsic factors, which would aid performance enhancement and reduce injury risk/occurrence [ 29 , 72 , 189 ].

Practical Implications

This review provides insight into how wearable technology is used for investigating running biomechanics and there is an increasing body of evidence demonstrating its accuracy. Although beyond the scope of this review, with continued and improved use of wearables in runners, biomechanical data may be analysed using advanced techniques, such as machine learning and pattern recognition to enable identifying and tracking running demands without direct supervision. These predictive capabilities would be highly valuable to practitioners to monitor performance and fatigue measures in ecologically valid settings (Table (Table7 7 ).

Summary of directions for future research using wearable technology

Future research directions
Test the validity and reliability prior to performing clinical or applied studies Multimodal wearable technology may give more comprehensive assessment of running gait Studies require an appropriate sample size Using wearable technology during natural outdoor running over time would help confirm laboratory findings or expand upon our knowledge Examine effects of age and sex on running gait outcome measures Report outcome measures as comprehensively as possible Investigate the usability, comfort, as well as the wearer’s physical, psychological and social preferences regarding technology

Future research directions

Test the validity and reliability prior to performing clinical or applied studies

Multimodal wearable technology may give more comprehensive assessment of running gait

Studies require an appropriate sample size

Using wearable technology during natural outdoor running over time would help confirm laboratory findings or expand upon our knowledge

Examine effects of age and sex on running gait outcome measures

Report outcome measures as comprehensively as possible

Investigate the usability, comfort, as well as the wearer’s physical, psychological and social preferences regarding technology

Review Limitations

Several limitations of the review must be considered. The search was limited to four databases, albeit integrated by reference lists and hand searches to identify other relevant papers. The use of stringent exclusion criteria may lead to the omission of potentially relevant data. First, articles not published in English pose a language bias regarding article selection. Additionally, sensor modality was restricted to wearable accelerometers, gyroscopes, magnetometers or a combination of those (IMU), or pressure insoles, thus excluding GPS or mobile phone applications, which are common amongst runners [ 195 ]. Because of the varying definitions and methods of calculation, studies were also excluded if they focused solely on shock, stiffness or neuromuscular load. We excluded studies that applied interventions as this would influence the gait outcomes and may not be representative of a runner’s typical gait. Finally, because of the size and heterogeneity of the articles included within this review, no meta-analysis or formal quality assessment of each study was performed.

Wearable technology is rapidly becoming a feasible means to quantify running biomechanics in a more ecologically valid manner, with applications in sports medicine and sports performance. This review highlighted that most studies that have examined running gait using wearable sensors have done so with young adult recreational runners, using one IMU sensor (on shoe or tibia), with participants running on a treadmill and reporting outcomes of GCT, SL, SF and tibial acceleration. While this review comprehensively synthesised a large ( n = 131) number of previous studies, future studies are needed to determine optimal outcome definitions, sensor site, type of sensor and outcomes of interest for running gait.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Appendix 1 Data extraction form

Title, authors(s), year

Study type (i.e. validation, reliability, application)

Characteristics; age, sex, height, weight, numbers in each group, type (e.g. injured, RFS, recreationally active)

Type of device (i.e. IMU, accelerometer, pressure insole)

Model and brand of wearable technology (e.g. Shimmer3, Shimmer Inc.)

Number of devices used

Location of device(s)

Device characteristics; accelerometer range, gyroscope range, sampling frequency, dimensions, weight

Test protocol

Environment (e.g. indoor, outdoor), surface type (e.g. road, treadmill), gradient speed(s)

Analysed distance/time/steps

What was measured using wearable technology?

Reference measures or additional tools used (e.g. 3D motion capture, EMG)

Statistical techniques used

Main findings according to the study author(s)

Declarations

This project received collaborative funding from Northumbria University and DANU Sports Ltd. (Grant number 120162).

Rachel Mason, Liam T. Pearson, Gillian Barry, Fraser Young, Oisin Lennon, Alan Godfrey and Samuel Stuart have no conflicts of interest that are directly relevant to the content of this article.

As this study is a systematic review of publicly accessible information, no ethical approval was required.

Not applicable. Data included in this study were extracted from prior studies that obtained written prior consent for the publication of de-identified data.

Not applicable.

All data generated or analysed during this study are included in this published article (and its supplementary information files).

All authors contributed to the study conception and design. RM, GB, AG and SS drafted the manuscript and took part in formation of the search strategy. RM, LP and SS completed the data extraction. SS led the research area and supervised the completion of the manuscript. All authors reviewed and edited the final manuscript.

Animal Physiology
Human Physiology

A review of gait cycle and its parameters

January 2011
This person is not on ResearchGate, or hasn't claimed this research yet.

Northern India Engineering College

Abstract and Figures

Discover the world's research

25+ million members
160+ million publication pages
2.3+ billion citations

STRUCT CONTROL HLTH

Troilyn Jackson
Hao-Yuan Hsiao
Cameron Kyle Phan

Wenwen Zhang

Peyman Servati
NEUROREHABILITATION

Jaber Mohammed Fagehi
Muzaynah Aljosh
Adel A. Alhusaini
Raffaele Giannattasio
Stefano Maludrottu

SENSOR ACTUAT A-PHYS

Shutong Zhong
KNOWL-BASED SYST
Jianxin Lin

Chiraz Ben Abdelkader

Jorge O. Galante
Bull Prosthet Res
A H Burstein
A. BERNARD DROUGHT
ROSS C. KORY
Biomechanics
Pathokinesiology Department
Recruit researchers
Join for free
Login Email Tip: Most researchers use their institutional email address as their ResearchGate login Password Forgot password? Keep me logged in Log in or Continue with Google Welcome back! Please log in. Email · Hint Tip: Most researchers use their institutional email address as their ResearchGate login Password Forgot password? Keep me logged in Log in or Continue with Google No account? Sign up

IMAGES

Example of gait analysis report: (top) angle of hip, (middle) knee, and
Example of gait analysis report: gait parameters.
Gait analysis sample report
Biomechanical gait analysis
(PDF) A Practical Guide to Gait Analysis
(PDF) Gait analysis by high school students

VIDEO

Gait Analysis Example
Gait Analysis with RSscan Foot Pressure Plates 3D Simi Motion
Gait Analysis with synchronized electromyography (EMG)
BEHS 210: Assignment Analysis
Intro to Gait Analysis Webinar Preview
Camel pacing

COMMENTS

Present and future of gait assessment in clinical practice: Towards the application of novel trends and technologies
Instrumented gait analysis (IGA), which can provide accurate and precise quantitative measurement of gait patterns and characteristics, has long been the gold standard for gait assessment in research practice . IGA generally refers to the use of instrumentation to capture and analyze a variety of human gait parameters (spatiotemporal, kinematic ...
A comprehensive survey on gait analysis: History, parameters
Limited resources: Gait Analysis is a field having numerous applications, but still, there are limited resources available to learn and understand the techniques available for gait analysis. This research paper aims to provide an extensive review of this field. •
Gait Analysis Methods: An Overview of Wearable and Non-Wearable Systems
The present paper aims to provide a description of technologies and methods used for gait analysis, covering both semi-subjective and objective approaches. This section includes a discussion of the different methods. ... The latest research on gait analysis comparing the advantages and disadvantages of the different systems leads us to conclude ...
Gait Analysis in Neurorehabilitation: From Research to Clinical
Gait analysis can be defined as the set of procedures that are needed to observe, record, analyze, and interpret human locomotion [7]. In fact, digital-based technologies are fundamental to provide kinetic, kinematic, and muscle activation information that are not detectable by clinical observation alone [8].
Latest Research Trends in Gait Analysis Using Wearable Sensors and
Gait is the locomotion attained through the movement of limbs and gait analysis examines the patterns (normal/abnormal) depending on the gait cycle. It contributes to the development of various applications in the medical, security, sports, and fitness domains to improve the overall outcome. Among many available technologies, two emerging technologies that play a central role in modern day ...
Clinical gait analysis 1973-2023: Evaluating progress to guide the
A search on Google Scholar using the key phrase "gait analysis" revealed 28 results from the year 1973, of which only eight were relevant to human gait analysis, and of those, five were theses. A similar search for the year 2020 returned around 10,200 results. The aim of this paper is to review the current state of the art in gait analysis ...
Latest Research Trends in Gait Analysis Using Wearable Sensors and
PDF | Gait is the locomotion attained through the movement of limbs and gait analysis examines the patterns (normal/abnormal) depending on the gait... | Find, read and cite all the research you ...
(PDF) A Practical Guide to Gait Analysis
Gait analysis, accomplished by either simple observation or three-. dimensional analysis with measurement of joint angles (kinematics), joint forces. (kinetics), muscular activity, foot pressure ...
Gait analysis: clinical facts
There is a large volume of literature on the research use of gait analysis, but evidence on its clinical routine use supports a favorable cost-benefit ratio in a limited number of conditions. ... This paper provides an overview on guidelines for managing a clinical gait analysis service and on the principal clinical domains of its application ...
A comprehensive survey on gait analysis: History, parameters
This paper presents an elaborated schema, including gait analysis history, parameters, machine learning approaches for marker-based and marker-less analysis, applications, and performance measures. This paper also explores the pose estimation techniques for clinical gait analysis that open future research directions in this area.
Quantitative and Qualitative Running Gait Analysis through an
Abstract. Quantitative and qualitative running gait analysis allows the early identification and the longitudinal monitoring of gait abnormalities linked to running-related injuries. A promising calibration- and marker-less video sensor-based technology (i.e., Graal ), recently validated for walking gait, may also offer a time- and cost ...
Human Gait Analysis Using Machine Learning: A Review
The gait analysis is interpreted to include an overwhelming number of interrelated parameters, which, due to the high volume of data and their relationships and is difficult to implement. The integration of machine learning with biomechanics is a promising approach to simplify the evaluation. The aim of this paper is to educate readers about the key directions to implement the gait analysis ...
Gait Analysis in Neurorehabilitation: From Research to Clinical Practice
When brain damage occurs, gait and balance are often impaired. Evaluation of the gait cycle, therefore, has a pivotal role during the rehabilitation path of subjects who suffer from neurological disorders. Gait analysis can be performed through laboratory systems, non-wearable sensors (NWS), and/or wearable sensors (WS). Using these tools, physiotherapists and neurologists have more objective ...
GAIT ANALYSIS: SYSTEMS, TECHNOLOGIES, AND IMPORTANCE
Human gait is the identity of a person's style and quality of life. Reliable cognition of gait properties over time, continuous monitoring, accuracy of evaluation, and proper analysis of human ...
Integrating digital gait data with metabolomics and clinical data to
The analysis of gait- and mobility-related impairments in PD was a key focus of this study, as digital gait biomarker data is expected to be highly informative for classifying gait and movement ...
Gait & Posture
About the journal. Official Journal of: Gait and Clinical Movement Analysis Society (GCMAS), European Society of Movement Analysis in Adults and Children (ESMAC), Società Italiana di Analisi del Movimento in Clinica (SIAMOC), and the International Society for Posture and Gait Research (ISPGR) Gait & Posture …. View full aims & scope.
A comprehensive survey on gait analysis: History, parameters
This paper presents an elaborated schema, including gait analysis history, parameters, machine learning approaches for marker-based and marker-less analysis, applications, and performance measures. This paper also explores the pose estimation techniques for clinical gait analysis that open future research directions in this area.
Frontiers
This review paper demonstrated the success of ML techniques in detecting gait disorders, predicting rehabilitation length, and control of rehabilitation devices. ... Alam MM, Le Moullec Y, Niazi IK, et al. Latest research trends in gait analysis using wearable sensors and machine learning: a systematic review. IEEE Access. (2020) 8:167830-64 ...
Three-dimensional gait analysis of lower extremity gait parameters in
Research on gait using three-dimensional (3D) motion analysis has revealed that children aged 7-11 years exhibit adult-like gait patterns. These findings are supported by research on joint ...
Gait analysis methods in rehabilitation
Introduction. For the purposes of this paper gait analysis will be assumed to refer to the instrumented measurement of the movement patterns that make up walking and the associated interpretation of these.The core of most contemporary gait analysis is the measurement of joint kinematics and kinetics. Other measurements regularly made are electromyography (EMG), oxygen consumption and foot ...
(PDF) A Review on Clinical Gait Analysis
The aim of this research is to review various approaches for Gait Analysis and specifically clinical gait analysis.This paper includes the discussion on the background details of gait, related ...
Biometric recognition through gait analysis
A set of experiments were carried out to evaluate Biometric RecognITion Through gAit aNalYsis (BRITTANY). In this section, the main elements of the research are in-depth depicted.
Wearables for Running Gait Analysis: A Systematic Review
The complete flow diagram of the screening procedure is shown in Fig. 1. A total of 131 articles were reviewed, with overlapping reports on several topics; specifically, 56 examined validity, 22 examined reliability and 77 investigated the application of wearable technology for a running gait analysis (Fig. 2).
A review of gait cycle and its parameters
This. paper describes a brief of gait, gait cycle and its phases. Keywords: gait, stance, swing, gait cycle, stride, cycle time, step length. 1. INTRODUCTION. Walking [1 - 2] can be defined as ...

IEEE Account

Purchase Details

Profile Information

Information

Initiatives

Article Menu

JSmol Viewer

1. Introduction

Share and Cite

Article Metrics

Save citation to file

Add to My Bibliography

A comprehensive survey on gait analysis: History, parameters, approaches, pose estimation, and future work

Similar articles

Publication types

Related information

REVIEW article

1. Introduction

2.1. Search criteria

3. Clinical gait pathologies and parameters

3.2. Gait parameters associated with pathology

3.2.1. Neurological gait disorders in elderly people

3.2.1.1. Hypokinetic-rigid gait disorders

3.2.1.2. Cautious and careless gait

3.2.1.3. Dyskinetic gait or involuntary movements

3.21.1.4. Psychogenic gait disorders

3.2.1.5. Fluctuating or episodic gait disorders

3.2.2. Gait disorders in Parkinson's disease

3.2.3. Gait in diabetic peripheral neuropathy

3.2.4. Post stroke gait

3.2.5. Total hip arthroplasty (THA)

3.3. Clinical gait assessment measures and indices

4. Gait assessment technologies applicable to clinical settings

4.1. Imaging techniques for gait assessment

4.2. Portable wearable systems for gait assessment

4.3. Wearable-based gait computational algorithms

4.4. Data-driven gait rehabilitation in clinical settings

5. Mobile gait lab for clinical applications and beyond

6. Limitations

7. Conclusive remarks and future work

Author contributions

Conflict of interest

Publisher's note

Biometric recognition through gait analysis

Similar content being viewed by others

OPERAnet , a multimodal activity recognition dataset acquired from radio frequency and vision-based sensors

Enhancing gait recognition by multimodal fusion of mobilenetv1 and xception features via PCA for OaA-SVM classification

A semi-automatic toolbox for markerless effective semantic feature extraction

Materials and methods

Leon@Testbed

Orbi-One robot

Data gathering

Data curation

Convolutional neural network design

Results and discussion

Data availability

Acknowledgements

Author information

Contributions

Corresponding author

Ethics declarations

Additional information

Rights and permissions

About this article

Share this article

This article is cited by

Deep learning-based route reconfigurability for intelligent vehicle networks to improve power-constrained using energy-efficient geographic routing protocol

Quick links

Wearables for Running Gait Analysis: A Systematic Review

Liam T. Pearson

Oisin Lennon

Alan Godfrey

Associated Data

Conclusions

Clinical Trial Registration

Supplementary Information

Introduction

Search Strategy and Study Selection Process

Data Extraction

Search Results