Accurate Shoulder Joint Angle Estimation Using Single RGB camera for Rehabilitation

Motion capture system offer the best interface to understand the trajectory development, which has benefited the research in animation, navigation, health care, etc. The existing human motion capture systems are based on lab-like environment requiring markers, which are computationally costly and time-consuming. MS Kinect is a marker-less technology introduced recently in the gaming industry. In this paper, we have undertaken a comparison study of accuracy in the estimation of shoulder joint angle between machine-learning-based RGB camera and MS Kinect. The results obtained are in time-domain plots for estimating shoulder angle. The results have shown a mean error of 9.29° and 5.3° (and RMS of 3.5° and 1°) in the sagittal and coronal planes respectively. The inaccuracy reported is in the range enough to assure its utility for use in stroke rehabilitation.