An improved TLD algorithm with selective detection

This paper proposes an improved TLD algorithm with selective detection. In the tracking module, KCF is used as the short-term tracker, and the idea of backward tracking is introduced to judge the accuracy of the result, when the result is inaccurate, the proposed algorithm starts up the detection module; in the detection module, H component of the color image is used as the input, and the mean filter serves as the first layer of the cascaded classifier, the color histogram similarity measure method is used to judge the accuracy of the detection results; finally a fusion strategy is formulated according to the output results of the tracking module and the detection module. The experiments carried on the OTB-2013 data platform show that the tracking accuracy and success rate of the proposed algorithm are respectively 0.801 and 0.591, which are 20.4% and 16.5% higher than the TLD algorithm, besides the proposed algorithm have good adaptability in scenarios such as illumination change, occlusion, and scale change, which shows superior tracking robustness.