Comparing marker-less vision systems for contact detection in wheelchair propulsion: Evaluating the limitations of angular speed difference

The objective of this study is to compare the error in detecting the start and end times of hand-wheel contact during wheelchair propulsion using two marker-less vision systems: ASD and MLVS. The first system, based on angular speed difference (ASD), relies only on the difference in angular speed between the hand and the wheel, calculated from RGB videos collected at 240 Hz. The angular speed of the hand was assessed using MediaPipe, a marker-less detection system, while the angular speed of the wheel was computed using a visual encoder operating on coloured tape positioned on the wheel and hand-rim in two configurations: hexagonal and incremental. The second system, MLVS, uses RGB-depth videos collected at 60 Hz and incorporates additional parameters, such as radial distance and normal position, to evaluate contact. The reference for contact initiation and termination was obtained from the torque signal collected on an instrumented wheelchair ergometer, with subsequent corrections made through expert visual inspection of video recordings. Data were collected from two participants who propelled a wheelchair on a wheelchair ergometer at 1.11 m/s and 0.21 W/kg body mass. Various thresholds for the angular speed difference between the hand and the wheel were tested, and the root mean square error (RMSE) for contact initiation and termination times was evaluated. The results showed that the lowest RMSE obtained with ASD was approximately 155 ms, while the RMSE obtained with MLVS were approximately 90 ms for contact initiation and 60 ms for contact termination, comparable to what was obtained with MLVS in previous studies. Accordingly, the study concludes that using only the relative angular speed between hand and hand-rim is not a reliable predictor of contact initiation or termination within the assessed experimental setup.