Algorithm Development for Contact Identification during Wheelchair Tennis Propulsion using Marker-less Vision System

This study investigates marker-less vision systems for contact detection between the hand and hand-rim in wheelchair propulsion. The measurement system is based on a camera collecting RGB and depth images. The hand is detected through Mediapipe [1] and its position is expressed with respect to the wheel. Five features are extracted and given to a classifier able to determine when there is contact between hand and hand-rim from start to end. To validate this procedure, 17 able-bodied participants without prior experience in wheelchair propulsion pushed the wheelchair on an ergometer in six tests, given by the combinations of the presence/absence of a tennis racket in the hand of the participant and of three different speeds: 4 km/h, 5.4 km/h and maximal sprint. The results showed that the hand identification is not influenced by the presence of the racket, but it is heavily influenced by the speed of the hand. Moreover, the errors in contact detection were $-0.01\pm 0.12\mathrm{~s}$ for start and $0.00\pm 0.12\mathrm{~s}$ for end of contact (mean pm standard deviation), with a RMSE of 0.12s for both with a slight improvement when the racket is present and no significant differences between the tests executed at sub-maximal speed and maximal speed. The study highlights the potential use of marker-less vision systems for contact detection in wheelchair propulsion.