融合卡尔曼滤波的抗遮挡目标跟踪KCF算法

KCF Algorithm for Anti-Occlusion Target Tracking Based on Kalman Filter

针对目标跟踪中的遮挡问题,在核相关滤波(Kernel Correlation Filter,KCF)基础上,融合卡尔曼滤波(Kalman Filtering,KF)和尺度不变特征变换(Scale Invariant Feature Transform,SIFT),提出一种新的抗遮挡目标跟踪方法。在KCF跟踪过程中,使用平均峰相关能量(Average Peak-to-Correlation Energy,APCE)值来检测目标是否被遮挡。若目标被遮挡,则使用KF预测目标的位置。当目标在后续帧中出现时,利用SIFT特征匹配定位目标的准确位置。最后利用KCF完成剩余帧的跟踪过程。在OTB100数据集上的实验结果表明,该方法对遮挡问题具有一定的鲁棒性。

Aiming at the occlusion problem in target tracking,based on Kernel Correlation Filter(KCF),Kalman Filter(KF)and Scale Invariant Feature Transform(SIFT)are combined,and an anti-occlusion target tracking method is proposed.During the KCF tracking process,the average peak correlation energy(APCE)value is used to detect whether the object is occluded.If the target is occluded,KF will be used to predict the position of the target.When the target appears in subsequent frames,the exact position of the target is located by using SIFT feature matching.Finally,KCF is used to complete the tracking process of the remaining frames.The experimental results on the OTB100 dataset show that the method is robust to the occlusion problem.