最佳伙伴相似性引导的核相关滤波跟踪算法

Best-buddies similarity-guided kernel correlation filter tracking algorithm

针对经典的核相关滤波(kernel correlation filter,KCF)算法在目标物体被遮挡、发生严重形变或旋转等复杂条件的跟踪失败问题,提出了一种利用最佳伙伴相似性引导的KCF算法。利用Edge boxes算法生成候选区域(region proposals,RP),并通过计算图像的尺寸、灰度和颜色相似度筛选RP。将筛选后的RP与模板图像进行最佳伙伴相似匹配,计算每个RP的最佳伙伴相似得分。融合最佳伙伴相似得分与KCF最大位置响应得分来预测目标可能位置。采用OTB100数据集评估算法性能,实验结果表明,与经典KCF算法和稀疏正则化判别式相关滤波算法相比,提出的算法可以在复杂条件下对目标有效跟踪,且精确度和成功率较高。

Aiming at the tracking failure of the classical kernel correlation filter(KCF)algorithm under complex conditions such as occlusion,serious deformation or rotation of the target,a KCF algorithm guided by the best-buddies similarity is proposed.Firstly,region proposals(RP)are generated by edge boxes algorithm,and RP is filtered by calculating image size,gray level and color similarity;Secondly,the best-buddies similarity score of each RP is calculated by matching the selected RP with template image;Finally,the best-buddies similarity score and KCF maximum position response score are combined to predict the possible position of the target.The OTB100 data set is used to evaluate the performance of the algorithm.The experimental results show that compared with the classical KCF algorithm and spatially regularized correlation filters algorithm,the proposed algorithm can effectively track the target under complex conditions,and has high accuracy and success rate.