Occlusion detection is a difficult traffic in the occlusion occurs, the existing system has been blocked based on the regional and multi-block area is not proportional relationship between the division. meanshift algorithm is more popular, the algorithm is robust root tracking algorithm. However, over-reliance on the law target color information, so when multiple moving objects, the color looks similar to when this method often leads to failure of the root trace, In this paper, the object tracking using block and used to predict the kalman filter to speed up the convergence rate of block meanshift. We introduce a new outlook weights, less affected by background sub-block effect on the target track and improve robustness.