Object tracking via Modified CamShift in Sequential Bayesian Filtering Framework

We present a robust object tracking algorithm which integrates Modified Continuous Adaptive Mean shift and Particle Filtering providing a framework for state estimation in nonlinear and non-Gaussian dynamic system. In order to overcome the various kinds of clutter and distracters problem, we employ a parameter associated with the similarity measurement to update window width adaptively via calculating histogram intersection between object and its background. Meanwhile, special morphological operations are adopted to improve the accuracy of object histogram back-projection. Experimental results show that the proposed algorithm is robust to partial occlusion, clutter and fast motion. Finally, we could obtain and analysis the target trajectory with fast motion as the basis for behavior analyze and understanding.