The Kernelized Correlation Filtering (KCF) tracking algorithm has no occlusion detection mechanism, and has a fixed learning rate.To solve the problems, a KCF tracking algorithm combined with the positive sample set is proposed.The mechanism determining target occlusion is set up by calculating the similarity between the positive sample set and the sample set to be tested, and thus the anti-occlusion ability of the algorithm is improved.As to parameter updating, the method of multi-step learning rate is adopted, which improves the accuracy of the target model.Experimental results show that, compared with that of the KCF tracking algorithm, the tracking accuracy of the proposed method is obviously improved.