改进的核相关滤波器的目标跟踪算法

Improved kernelized correlation filter for target tracking

为降低目标跟踪过程中受遮挡、光照、尺寸变化等因素的影响,提出改进核相关滤波跟踪算法。HOG融合HIS空间颜色信息的多特征信息,采用尺度池因子方法自适应目标区域尺寸变化,通过比较初始目标特征与候选目标特征的巴氏相似度值与阈值大小,判断遮挡是否存在,若遮挡存在,进行遮挡处理,即停止对模型的更新,否则更新模型。与经典mean shift算法、Fragment算法以及经典KCF算法进行对比,对比结果表明,该算法处理目标跟踪过程中受遮挡、光照、尺寸等影响时,跟踪效果最佳,处理速度可达48fps。

To reduce the poor effects of occlusion,illumination,size change and other factors in the tracking process,and an improved kernel correlation filtering algorithm was proposed.Multi-feature combination was used which included color information in HIS spatial and HOG.The scale pool factor method was used to adapt to the size changed of the target.Whether the occlusion existed was determined by comparing the Bhattacharyya＇s distance of the initial target feature and the candidate target feature with the threshold.If the occlusion existed,the occlusion process was performed,namely the update of the model was stopped.Otherwise,the model was updated continuously.Compared with the classic mean shift algorithm,Fragment algorithm and classic KCF tracking algorithm,the proposed algorithm has best tracking effects and the processing speed is up to 48 fps when there are the effects of occlusion,illumination and size change during the tracking process.