An improved part-based compressive tracking method

摘要

In this paper, we propose an improved part-based tracking method based on compressive tracker. The traditional compressive tracking can hardly deal with occlusion and scale variation, which is not robust enough in these situations. In our part-based method, the occluded image patches which are selected by K-means are not able to participate in determining the location of the target in order to increase tracking accuracy. To solve the problem of scale variation, an algorithm is proposed to estimate the size of target with the information of the whole target. In addition, we also use the positions of every pail to help us to get a precise size of the target. With the combination of two methods, the scale of the target is obtained. Experiments compared with five state-of-the-arts have been done to prove the effectiveness and robustness of our method.