Real-Time Motion Analysis System Using Low-Cost Web Cameras and Wearable Skin Markers

Abstract

COVID-19 has restricted outdoor exercise and hospital visits for rehabilitation therapy. Home-based training and rehabilitation coaching systems have emerged as a way to overcome these circumstances. Conventional optical motion-capture systems, such as VICON, have been used for measuring precise movement and providing posture feedback during exercise or rehabilitation; however, its application is limited to professional facilities because of its high cost and space requirement. To extend the applicability to home-based use, we designed wearable skin markers (WSMs) with body segment-specific patterns that can be detected by low-cost web cameras. WSMs are band-shaped and stretchable and thus can be worn like cloth, with minimal effort for placement. The body segment-specific patterns enable real-time data processing, which reduces the marker data post-processing time. A 6-degree-of-freedom (DOF) pose for each WSM is obtained by recognizing the segment-specific patterns; the 3D configuration of the contoured corners of the patterns found by triangulation is then utilized to construct the coordinates of each WSM. The WSM system was validated via three experiments. The robustness of marker recognition was evaluated by measuring the false-positive and false-negative rates of WSM. For accuracy validation, the angle estimation results were obtained for the mechanical joint of a 3-DOF gimbal and lower-limb joints of a walking human subject and compared to the reference systems. The gimbal experiment was included to evaluate the accuracy of our system in the condition with no skin movement artifact. The maximum standard deviation of the difference between WSM and the encoder was 0.9 & DEG; for the gimbal experiment, and that between WSM and VICON was 5.0 & DEG; for the human experiment. The accuracy was comparable to the reference systems, making it suitable for home environment application.