Hand acceleration measurement by Kinect for rehabilitation applications

Affordable motion sensors that are recently developed for video gaming have formed a budding line of research in the field of physical rehabilitation. These sensors have been used in many task-based applications to analyze the patients' status based on their completion of assigned tasks. However, as the accuracy of such sensors is lower than that of the clinical ones, their measured data has had very limited use in quantitative motion analysis to this date. The aim of this article is to determine Kinect's ability and accuracy in calculating higher-order kinematic parameters, such as velocity and acceleration, in hand movements. Four methods, i.e. moving average, Butterworth filter, B-spline, and Kalman filter, were proposed to calculate velocity and acceleration from Kinect's raw position data. The results were experimentally compared with two established motion capture systems, i.e. Vicon and Xsens, to analyze the strengths and weaknesses of each method. The results show that B-spline is the best method for calculating velocity and acceleration from Kinect's position data. Using this method, these parameters can be measured with an acceptable accuracy.