A robust tracking method based on the correlation filter and correcting strategy

Abstract

Visual tracking integrates the technology of image processing and pattern recognition, etc., which has a lot of potential applications, such as automatic driving, safety monitoring, etc. This paper analyzes the advantages and disadvantages of the Kernelized Correlation Filter (KCF) and Tracking-Learning-Detection (TLD), which are two kinds of trackers. TLD tracker has correcting capability whereas its performance highly depends on the tracker, which is not robust to some cases, such as tracking non-grid objects. Inversely, KCF achieves good performance in tracking non-grid objects. However, KCF behaves badly in the presence of occlusion and out-of-view and it cannot correct errors during the tracking process. According to the characteristics of the KCF and TLD, this paper proposes a robust tracking method based on the correlation filter and correcting strategy. By using the advantages of the KCF and TLD, the proposed method achieves high tracking accuracy and correcting capability. Experimental results show that the proposed method outperforms other methods (KCF, TLD, Struck, SCM, ASLA, MTT and DFT) according to the success and precision plots of OPE, SRE, and TRE, respectively.