结合正样本集的核相关滤波跟踪算法

A KCF Tracking Algorithm Combined with Positive Sample Set

针对核相关滤波(KCF)跟踪算法没有遮挡检测机制以及学习率固定的问题,提出了一种结合正样本集的核相关滤波跟踪算法。通过计算正样本集与待测样本集的相似度来建立目标遮挡判断机制,提高了算法的抗遮挡能力。在模型更新方面,采用了多段学习率的参数更新方式,提高了目标模型的准确性。实验结果表明,该算法与KCF跟踪算法比较,跟踪精度有明显提升。

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.