Improved Multi-sampling Kernelized Correlation Filter Target Tracking Algorithm

Abstract

In order to solve the tracking failure of kernelized correlation filter (KCF) tracking algorithm in the case of target fast motion and motion blur, proposing a multi-sampling tracking algorithm based on KCF. Firstly, a PSNR-based judgment mechanism is introduced to determine whether the current frame target is tracking errors. If the tracking error occurs, the search range is extended to a mutisampling search area. Finally re-detect the target of the current frame. The improved algorithm of this paper is compared with several classical correlation filter target tracking algorithms in the OTB video dataset. The experimental results show that the precision of this algorithm is 0.732 and the success rate is 0.575, ranking first, which is 5.3% and 4.3% higher than the KCF algorithm. Especially when the target has fast motion and motion blur, it has stronger tracking accuracy.