Object Tracking Based on Kernel Correlation Filter and Multi-feature Fusion

Abstract

Target tracking is a hotpot issue in computer vision. This study proposed a robust tracking algorithm based on kernel correlation filter and multi-feature fusion. Firstly, an effective feature fusion strategy is designed which combines GRAY, HOG and LAB features for improving the robustness of the tracker. Secondly, we proposed a novel strategy to solve the occlusion challenge, which overcomes the shortcomings of KCF tracker in occlusion. Finally, a multi-scale filter is introduced in the model for solving the problem of scale change, and the target model is updated to alleviate tracking drift. The comprehensive evaluations are conducted, and the results show that the proposed tracker perform well against other algorithms.