一种基于特征互补的新型长期跟踪算法

A new long-term tracking algorithm based on feature complementarity

针对特征互补学习跟踪算法(Staple)在长期目标跟踪时出现目标丢失的问题,设计了一种基于特征匹配的重检测算法.该算法通过相关滤波模型和颜色特征模型的互补学习所得到的最大响应值来判断目标是否丢失,并融合基于网格运动统计的特征匹配算法(GMS)构建重检测机制,对目标进行重定位,从而保持稳定地长期跟踪.实验结果显示:该算法一次通过评估的精确度为81.1%,相比改进前的Staple算法提升了17.5%;在目标丢失时,它还能够准确重定位到跟踪目标,且对目标遮挡有着较强的抗干扰能力.

To tackle the problem of abnormal target loss in long-term target tracking in feature complementary learning tracking algorithm(Staple),a re-detection algorithm based on feature matching is designed.The new algorithm uses the maximum response value obtained from complementary learning of the correlation filter model and the color feature model to determine whether the target is lost,and integrates the feature matching algorithm(GMS)based on grid motion statistics to construct a re-detection mechanism to relocate the target,so as to maintain stable long-term tracking.Experimental results show that the accuracy of the algorithm in this paper is 81.1%for one-pass evaluation,which is 17.5%higher than the previous Staple algorithm.It can accurately relocate the tracking target when the target is lost,and has a strong anti-jamming capability against target occlusion.