遮挡判别下基于特征匹配的相关滤波跟踪算法

Correlation Filter Tracking Algorithm Based on Feature Matching under Occlusion Discrimination

为解决目标跟踪过程中遮挡问题引起的跟踪丢失现象,提出一种遮挡判别下基于特征匹配的相关滤波跟踪算法。首先,通过分析输出滤波响应图,提出最大峰值波动和响应值波动两种计算模型判断目标是否发生遮挡。其次,引入结构相似性指标判断目标区域与检测区域的相似性,将遮挡进一步分为轻微遮挡与严重遮挡,并针对两种不同的遮挡程度,分别调整滤波器的更新方案。最后,采用SURF特征检测候选区域,实现严重遮挡情况下目标的重新定位。实验结果表明,与传统核相关滤波算法相比,所提算法在保证实时性的同时,能有效提高遮挡情况下目标跟踪的成功率和精确度,表现出更好的跟踪性能。

In order to solve the tracking loss caused by occlusion during target tracking,a correlation filtering tracking algorithm based on feature matching under occlusion discrimination is proposed.First,by analyzing the output filter response graph,two calculation models of maxi⁃mum peak fluctuation and response value fluctuation are proposed to determine whether the target is occlusion.Secondly,the introduction of structural similarity index determines the similarity with the target area detection region,the occlusion is further divided into slight oc⁃clusion and severe occlusion,and the updating scheme of filter is adjusted respectively according to the two different occlusion degrees.Finally,the SURF feature detection candidate area is used to achieve target relocation under severe occlusion.Experimental results show that,compared with the traditional Kernelized Correlation Filter algorithm,the proposed algorithm effectively improves the success rate and accuracy of target tracking under occlusion and show better tracking performance while ensuring real-time performance.