Correlation Gaussian Particle Filter for Robust Visual Tracking

Abstract

Visual tracking plays a significant role in computer vision. Correlation filter (CF) based visual tracking algorithms have made great progress owing to their high efficiency. However, the traditional CF based tracking algorithm uses the fixed-size template, when the target changes in size, the tracker cannot track the object accurately. To address the issues, A novel accurate position and scale estimation algorithm that combination kernel correlation filter (KCF) with Gaussian particle filter (CGPF) for visual tracking has been proposed. The algorithm estimates the position of the target by KCF, and then corrects the position by a Gaussian particle filter algorithm. About the problem of scale variation information in the KCF, A scale correlation filter is constructed, and it is used to identify the scale with the maximum response to select the target size. Finally, The comprehensive experiments on OTB-50 visual tracking benchmark datasets are performed. The experiments results show that the proposed approach outperforms the KCF algorithm. And the results also prove availability of our method.