Marker-Less Vision System for Wheelchair Tennis Contact Detection on Treadmill

Marker-Less Vision System (MLVS) is a system that detects the start and the end of the contact phase between hand and hand-rim during wheelchair propulsion. It uses a camera to collect RGB and depth images, which are processed by MediaPipe, a software to identify hand keypoints, while a classifier determines contact on a frame-by-frame basis using the relative position between the hand and hand-rim in real-world coordinates. Previously, MLVS was trained and metrologically validated on stationary wheelchair ergometer data. This study extends its application to dynamic conditions by assessing its performance during treadmill-based wheelchair propulsion, where relative motion between the camera and wheelchair occurs. Six non-disabled participants propelled a wheelchair on a treadmill, and MLVS-detected contact initiation and termination times were compared to a reference obtained via manual annotation based on torque measurements and video inspection. The root mean square error (RMSE) was 1.9*102 ms for contact initiation and for contact termination, corresponding to approximately half the duration of an average contact duration of 350 ms. These results highlight the challenges introduced by relative motion and suggest potential areas for improvement in MLVS for dynamic propulsion scenarios.