Ensuring the accurate tracking of hand and fingers movements is an ongoing challenge for upper limb rehabilitation assessment, as the high number of degrees of freedom and segments in the limited volume of the hand makes this a difficult task. The objective of this study is to evaluate the performance of two markerless approaches (the Leap Motion Controller and the Google MediaPipe API) in comparison to a marker-based one, and to improve the precision of the markerless methods by introducing additional data processing algorithms fusing multiple recording devices. Fifteen healthy participants were instructed to perform five distinct hand movements while being recorded by the three motion capture methods simultaneously. The captured movement data from each device was analyzed using a skeletal model of the hand through the inverse kinematics method of the OpenSim software. Finally, the root mean square errors of the angles formed by each finger segment were calculated for the markerless and marker-based motion capture methods to compare their accuracy. Our results indicate that the MediaPipe-based setup is more accurate than the Leap Motion Controller-based one (average root mean square error of 10.9° versus 14.7°), showing promising results for the use of markerless-based methods in clinical applications.
Maggioni et al. (Tue,) studied this question.