Abstract:Traditional particle filter algorithm needs a large number of particles to show posteriori probability density function of object state, the calculation of this algorithm is large, and the real-time of tracking is poor, so it is hard to track fast and sheltered object accurately. Considering above problems, this paper proposes a new algorithm that is inserting Mean Shift into particle filter algorithm, this method can make full use of clustering effect of Mean shift to make particles distributed more reasonably, which not only improves the diversity of the particles but also greatly reduces the number of particles used to describe object state. The experimental results show that the improved algorithm has stronger robustness and better real-time performance.