Target Tracking Algorithm Based on Kernel Correlation Filter with Anti-occlusion Mechanisms

Abstract

To solve the problem of target occlusion in kernel correlation filter, the paper proposes a target tracking algorithm based on kernel correlation filter with anti-occlusion mechanism. It combines histogram of oriented gradient, histogram of local intensities and color names to improve the tracking accuracy. Moreover, two anti-occlusion mechanisms are combined to deal with the problem of target occlusion. Firstly, the online support vector machine classifier is employed to re-detect the target. Secondly, the image correlation matching combined with color histogram matching are applied to locate the target lost for a long time due to the occlusion. Through an experiment on the OTB100 dataset, the results indicate that the algorithm proposed in this paper are superior to the other mainstream tracking algorithms in overlap precision, distance precision and center location error. Therefore, it has relatively high precision and robustness in the anti-occlusion performance.