抗遮挡的相关滤波目标跟踪算法

Anti-Occlusion Object Tracking Algorithm Based on Kernelized Correlation Filters

针对传统的核相关滤波目标跟踪算法遮挡判断失败的问题,提出一种抗遮挡的核相关滤波目标跟踪算法.首先,在核相关滤波器框架上根据最小二乘分类器获得目标位置.然后,引入一个多尺度滤波器,并通过计算滤波器的响应最大值进行尺度预测.最后,在目标模型更新方面,根据目标位置置信图峰值尖锐度的差异性,正确更新模型.实验结果表明:文中算法的平均位置误差为6.18px,在阈值为20px时,平均距离精度为97.68%,平均帧率为30.8帧·s^(-1);其能在复杂背景下有效地解决目标尺度变化、完全遮挡等问题,具有更高的鲁棒性和精确性.

In order to solve the problem of wrong judgment of occlusion based on traditional kernelized correlation filters object tracking algorithm.An anti-occlusion object tracking algorithm based on kernelized correlation filters is proposed.Firstly,based on the framework of kernelized correlation filters,the object position is obtained by the regularized least-squares classifiers.Secondly,a multi-scale filter is introduced and scale estimation is obtained through calculating the maximum value response of the multi-scale filter.Finally,in terms of the target model updating,according to the difference of target position confidence map peak sharpness,the model can correctly updated.The experimental results demonstrate that the median center location error of the proposed algorithm is 6.18 px,the average distance precision is 97.68% when the threshold is set 20 px,and the average tracking speed is 30.8 frames·s^-1.The proposed algorithm can not only effectively solve target scale changes,complete occlusion and other issues in the complex background,but also has higher tracking robustness and accuracy.