Tracking Algorithm Based on Dual Residual Network and Kernel Correlation Filters

Abstract

Visual target tracking is a target detection task for a period of time. During this period, the tracking target will undergo significant appearance changes due to deformation, sudden movement, complex background and occlusion. These changes make visual target tracking challenging. In this paper, a target tracking algorithm based on dual residual neural network and kernel correlation filters is proposed, which mainly solves the problems of inaccurate tracking. This method combines depth residual neural network with kernel correlation filter tracking algorithm. The template matches the depth residual feature to determine the location of the target and the kernel correlation filter based on residual feature is designed to detect the target, finally, the template matching result and kernel correlation filtering result are fused to determine the location of the target. The experimental results on a large-scale standard data set show that the proposed algorithm has the advantages of high accuracy. Compared with the previous algorithm, the proposed algorithm has good performance in tracking visual deformation, occlusion and fuzzy video objects.