An extended KCF tracking algorithm based on TLD structure in low frame rate videos

The KCF (Kernelized Correlation Filter) algorithm achieved a good performance on target tracking challenges. However, it still has some defects and problems of false tracking in low frame rate (LFR) scenarios, target scale variation, occlusion and out of view target, that exists in the correlation filter based methods. In this paper, we overcome the shortcomings of KCF tracking algorithm based on Tracking-Learning-Detection (TLD) framework. The proposed algorithm trained two classifiers simultaneously, based on semi supervised co-training learning algorithm. Then, we comparatively evaluate the proposed method on TB-100 datasets by other trackers. The experimental results demonstrate that the precision and robustness of the improved tracking algorithm is higher than traditional KCF, TLD and the other top state-of-the-art tracking algorithms in LFR videos.