Model based hand motion estimation through stereo vision

Recognition of hand movements plays a key role in both human computer interaction and rehabilitation activities. This paper focuses on hand pose estimation and motion tracking through a model-based stereo vision-based system. To allow for a complete 3D motion, initially a simple hand model with 20 landmark points was constructed and used to track its motion through a sequence of stereo images. Furthermore, a skeletal model representing the kinematical features of the hand was utilized to provide a meaningful hand motion and gesticulation. To evaluate and eventually recognize the performed hand motion, Kalman filter and Kalman smoother algorithms were implemented to evaluate the efficacy and assist in tracking of the proposed hand motion. These algorithms provided an estimate of the desired hand poses by minimizing the respective differences between the skeletal model and the 3D reconstructed model. As our proposed method required only 20 points, estimation results illustrate that the proposed approach is a cost effective and real time hand motion tracking approach, suitable for rehabilitation purposes.