一种抗遮挡核相关滤波目标跟踪算法

A kernelized correlaton filters with occlusion handling

针对核相关滤波器(KCF)算法在目标遮挡情况下易导致跟踪失效的问题,提出了一种抗遮挡核相关滤波(KCF)跟踪算法。该算法首先依据目标响应的峰值旁瓣比来确定滤波目标模型的更新方式,形成了目标模型自适应更新的KCF算法;其次,在目标的训练阶段分别提取目标的HOG和LBP特征直方图,并对其进行特征融合。对比试验结果表明,该算法针对OTB-2013测试数据集中的20组非遮挡运动目标视频帧来评估本文算法,并与原KCF算法进行对比,该算法的跟踪精度和成功率分别提升了9.5%、24%;选择测试数据集中29组被遮挡的运动视频帧来评估本文的抗遮挡性,该算法在目标遮挡情况下的跟踪精度和成功率分别提升了19.9%、8%。实验结果表明该算法具有明显的抗遮挡跟踪能力。

As the kernelized correlation filter （KCF） tracking algorithm has poor performance in occlusion handling, this paper proposes a occlusion handling tracking algorithm based on KCF. Firstly we design an adaptive model updating strategy by utilizing the peak to sidelobe ratio （PSR） of the response maps. Then we extract histogram of oriented gradient （HOG） and local binary pattern LBP features and fuse them in the training stage. To verify the effectiveness of the proposed algorithm, we use the OTB-2013 benchmark to obtain 20 groups of video sequences without occlusion and 29 groups of occlusion video se- quences. Experimental results demonstrate that the precision and success rate without occlusion have been promoted by 9.5 % and 24.0 % respectively. Especially in occlusion handling case, the two parame- ters have been by 19.9% and 8% ,respectively.