Improving the performance of Kalman filter for hand tracking in Persian sign language video

Hand tracking is one of the most important phases of a sign language recognition system that affects the final recognition rate directly. Kalman filter is a well-known technique for object tracking. By minimizing the mean square error, this filter is able to estimate the past, present and future states in a process, even in systems that are inherently uncertain. Hand movement in sign language video is very complex. Hence, Kalman filter is a suitable estimator to predict the hands motion. In this paper, we present an approach to optimize the Kalman filter to track the movement of hands accurately. The modified Kalman filter is then compared with other tracking methods by testing on the Persian sign language video database made by authors.