Combined motion and region-based 3D tracking in active depth image sequence

Robustness needs to be specially emphasised in visual tracking based on active depth camera, due to its high-level noise and limited accuracy. In this paper, we present a robust 3D tracking approach that is able to extract object-independent motion trajectory under uncontrolled environment. Two novel algorithms are designed, including a motion-based tracking approach and a region-based approach, then a Kalman filter is employed to fuse their tracking results. This method is low-level and object-independent, we test its performance in tracking moving people and gesture. Experimental results shows that this method can tackle the situation of entry, exit, high noise, complex background, complex motion, and non-rigid motion.