Mean-shift algorithm fusing multi feature

Abstract

The mean shift algorithm (Mean Shift, MS) has been widely used because of the advantages of fewer iteration times and better real-time performance. In the other hand, because of the use of single color histogram representation of the target feature, the MS algorithm can't always track well in complex condition. Aiming at the problem that the traditional Mean-Shift algorithm is unstable when the background color is similar to the target color or there is partial occlusion. In this paper, the original color features in traditional mean shift algorithm is transformed into HSV color feature and the textural feature is integrated into it to improve the tracking performance in the case of the background color is similar to the target color, and the four neighborhood search method (4 areas with the same size are expanded around the candidate area) is applied to solve the problem of partial occlusion. The comparisons of experiments show that the algorithm of this paper has higher accuracy than the traditional MS algorithm and the background weighted MS algorithm in the above complex environment, besides, the proposed algorithm has a good operating efficiency.