Object tracking with kernel correlation filters based on mean shift
Fei Feng, Xiao‐Jun Wu, Tianyang Xu · 2017
In order to enhance the robustness of kernel correlation filters (KCF) in complex-background environments, this paper proposes a mean shift method with an adaptive local object tracking algorithm. The KCF algorithm has a speed advantage by using a single template. We introduce the confidence map in the process of the tracking to determine the result of the current frame. If the result of the confidence map in the KCF algorithm is worse than the previous frame, then the block mean shift algorithm is introduced to track the target again. The local information is utilized to get the final position of the target. The experimental results show that the proposed algorithm has the ability of adaptive local information fusion. The research indicates that the proposed method not only retains the KCF calculation efficiency, but also uses the mean shift to correct potential error frames on video sequences. Compared with other traditional tracking algorithms, the proposed method shows a better performance.