Skip to Main content Skip to Navigation

Contribution à la capture du mouvement humain par stéréovision et machine learning pour l'analyse de la marche

Abstract : Gait analysis is the measurement of and ability evaluation of walking that can be used for fall risk assessment or as a diagnostic and prognostic tool for clinical applications. However, despite the clinical value, several difficulties attributed to the current established gold standard instrumentation, marker-based motion capture systems, limit the large-scale use in clinical applications. The current marker-based systems are costly and require a controlled laboratory environment. The test procedure is also time-consuming. Eliminating the markers would drastically shorten the patient preparatory time and would be more efficient. The objective of this study is to design a marker-less motion capture system for clinical applications. Recent advancements in computer vision and especially in convolutional neural networks, have provided the potential to pursue this objective. The designed system consists of four RGB cameras and can estimate the position of joint centers through a deep learning approach. For that purpose, a novel specific dataset has been collected including asymptomatic and pathologic subjects. To evaluate the validity of the developed system, its performance is assessed against a marker-based motion capture system in terms of joint position errors and clinically relevant gait parameters. The results demonstrate the high potential of the designed system for clinical applications.
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Thursday, October 21, 2021 - 2:12:11 PM
Last modification on : Friday, January 21, 2022 - 3:29:49 AM
Long-term archiving on: : Saturday, January 22, 2022 - 7:13:05 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03390546, version 1


Saman Vafadar. Contribution à la capture du mouvement humain par stéréovision et machine learning pour l'analyse de la marche. Biomécanique []. HESAM Université, 2020. Français. ⟨NNT : 2020HESAE069⟩. ⟨tel-03390546⟩



Record views


Files downloads