Powered wheelchair maneuvering skill is essential for independence and safety in people with mobility impairment. However, real-time monitoring of maneuver compliance under training or assessment is challenging for therapists, affecting intervention effectiveness and assessment objectiveness. To overcome this and digitalize users' performance, rehabilitation engineers at the Hospital Authority Community Rehabilitation Service Support Centre (CRSSC) created a Boundary Detection System (BDS); this study assesses the effectiveness of the BDS, comparing system and manual counting.

BDS utilized wheelchair-clamped webcams recording real-time video of wheels and boundaries. Through computer vision algorithms, wheels (grey) and boundaries (yellow) were separated through color filtering, with contours identified using binary masking. Boundary violations were registered when wheel contours intersected a dilated boundary contour during indoor training within a training area. Subjects were asked to complete clockwise and counterclockwise circles for three laps without cues. Human ethics approval was acquired from the Central Institutional Review Board (Ref. No. KC/KE-23-0216/ER-1).

In total, 13 male and 13 female (mean age=67.3±10.2) wheelchair users were recruited. The system-detected boundary violations (8.74 ± 7.25) differ significantly (p < 0.001) from the manual counting method (7.22 ± 6.86), representing the high sensitivity of the proposed system. Overall, an average of 2.64 ± 2.48 boundary violations that lasted less than 0.5 seconds wascaptured by the system and validated by video inspection, which shows the system’s ability to eliminate human error.

This pilot study illustrates the viability of the BDS as a computer vision-based, scalable solution for objective wheelchair training monitoring. Future research will advance algorithmic accuracy and investigate integrations with clinical rehabilitation practice to modernize intervention effectiveness and objectiveness.