ABSTRACT
In this paper, two problems about the update rate and long-term tracking in target tracking model are discussed. Traditional correlation filter tracker only uses a fixed rate mechanism, so the target update rate is fixed. In this paper, we improved it so that it can adjust the update rate adaptively according to the similarity between different image sequences and first frame. Besides, in order to deal with more challenging scenarios and long-term tracking targets, we add a re-detection mechanism to the tracker. This method overcomes the limitation of the traditional correlation filter tracker using fixed update rate by studying the similarity between the frames of the image, and can adaptively change the update rate of the model. A large number of experimental results show the superiority of our improved tracker in accuracy and success rate.
