一种改进的核相关滤波跟踪算法

An Improved Tracking Algorithm Based on Kernel Correlation Filter

在目标跟踪过程中,由于跟踪目标的移动、变形或者被遮挡,容易造成目标丢失。对KCF算法进行改进,增加目标丢失检测和运动轨迹估计的功能。根据响应峰值异常来检测目标丢失。若出现响应峰值异常,则中止对目标模板的更新和目标位置的检测,采用运动轨迹估计的方式来预测目标出现的新位置。实验结果表明,改进的KCF算法可以显著提高目标跟踪的精度和稳定性,并能及时对目标丢失进行判断和处理。

In the process of target tracking,the targets are easily lost due to movements,shape changes or occlusions.The KCF algorithm is improved by adding the functions of loss detection and trajectory estimation.Target loss is detected according to the abnormal response peaks.If the peak value of the response is abnormal,the update of the target template and the detection of the target position will be suspended,and the new location of the target will be predicted by trajectory estimation.The experimental results show that the improved algorithm can significantly improve the tracking accuracy and stability,and can judge and deal with target loss in time.