3D gait analysis is a crucial tool in assessing human movement patterns, aiding in the diagnosis, and monitoring of various neurological and musculoskeletal disorders. Traditional gait analysis systems often require cumbersome wearables or the use of a motion capture laboratory making them expensive and challenging to deploy outside of clinical settings.
In contrast, our unique approach leverages the recent advancements in Computer Vision and motion tracking resulting in the quantification of accurate gait biomarkers using a single off-the-shelf camera device
Our approach uses video footage of the subject's movements while s/he walks or performs a Timed Up and Go (TUG) test, in a single camera’s field of view. Using sophisticated deep learning algorithms, the two-dimensional image data is processed to locate the subject and identify key points on the body that are used to compute the pre-selected gait biomarkers.
We extract 3D joint trajectories, movement patterns, and segment individual gait cycles enabling the reconstruction of joint angles, step lengths, and other relevant gait parameters.
We compared the gait biomarkers obtained using our approach and using traditional motion capture that required multiple cameras and placement of markers on the body.
We use clinicians' feedback to improve the relevance and usefulness of the displayed results, which may improve diagnostic accuracy and patient care.
For angular joint measurements, the mean absolute error of our approach compared to motion capture is 4.60°.
License type:
Publisher Copyright
Funding Info:
This research is supported by core funding from: AERC
Grant Reference no. : SC20-RV230-HM