A Robust Hand Pose Estimation Algorithm for Hand Rehabilitation

During a rehabilitation session, patient activity should be continuously monitored in order to correct wrong movements and to follow patient improvements. Therefore, the application of human motion tracking techniques to rehabilitation is finding more and more consensus. The aim of this paper is to propose a novel, low-cost method for hand pose estimation by using a monocular motion sensing device and a robust marker-based pose estimation approach based on the Unscented Kalman Filter. The hand kinematics is used to enclose geometrical constraints in the estimation process. The approach is applied for evaluating some significant kinematic parameters necessary for understanding human hand motor improvements during rehabilitation. In particular, the parameters evaluated for the hand fingers are joint positions, angles, Range Of Motion and trajectory. Moreover, the position, orientation and velocity of the wrist are estimated.