Markerless Human Motion Analysis for Telerehabilitation: A Case Study on Squat

Rehabilitation is a crucial treatment process, normally in clinical settings, for patients recovering from surgery and those suffering from various illnesses. However, frequent hospital visits can be inconvenient, especially for those patients with travel or transportation difficulties. With the rapidly advancing healthcare artificial intelligence(AI) and technology, a few pioneering works have investigated remote therapy to provide wider medical access while reducing unnecessary travel. In remote settings, various technologies (e.g., inertial measurement unit, marker-based motion capture system, and markerless computer vision-based human pose estimation), which have different strengths and weaknesses, can be used to track patients' postures and movement. In this paper, we investigate the cost-efficient computer vision-based human motion analysis using OPENPOSE, which uses machine learning for pose estimation and human body parts detection. A dataset is collected when the subject performed deep squats at fast, normal, and slow speeds in home settings. Videos were recorded and analyzed for the computation of joint angles and joint angular velocities during a squat. These experimental results were evaluated by comparing them with data collected from a marker-based motion capture system. According to the results, the proposed markerless-based system provided comparable accurate joint angles and angular velocities estimation. While this study focuses on squatting, the findings have implications for home-based telerehabilitation in a smart city.