Abstract:

The MeanShift algorithm which is applied to tracking moving objects mainly uses a single histogram to describe the color characteristics of the object . This method obviously lacks spatial distribution information. As for this defect, Emilio Maggio et al have put forward an improved algorithm of blocking the object into regions. But the discriminant effect and the stability are not good enough in a complex environment. So this paper proposes a new method to improve it. On the one hand, reducing the number of human body regions is used to cut down the processing time without losing the space-related i  nformation. On the other hand, each sub-block is weighted by certain coefficients so as to improve the discriminant effect. The comparison experiment proves that the algorithm promotes the accuracy of identifying the moving people under a complicated environment and it has good stability.

Key words: MeanShift, people tracking, Bhattacharrya coefficient