针对传统均值漂移算法利用固定核或对称核函数进行目标跟踪时出现目标跟踪丢失或 跟踪失败的问题,提出了基于各向异性核函数的自适应带宽均值漂移目标跟踪算法,以提高 目标跟踪的准确性、实时性。在符号距离核函数的基础上引入符号距离约束函数,构成各向 异性核函数,满足目标外部的区域函数值为零,为目标跟踪提供准确的跟踪窗。依据基于各 向异性核函数的均值漂移应用到目标跟踪中需满足跟踪窗内的样本点到中心点的向量权重之 和为0的思想,计算各向异性核函数模板的均值漂移窗口中心。利用相似度阈值对前后两帧 目标模板的变化情况进行限制,实现各向异性核函数模板的自适应更新及目标的准确实时跟踪。 实验结果表明所提出算法的准确性和实时性较高。

Aiming at the problem of target tracking lost or failure when the traditional mean shift algorithm uses fixed kernels or symmetric kernel function to track targets, an adaptive bandwidth mean shift target tracking algorithm based on anisotropic kernel function is proposed to improve the accuracy and real-time of target tracking. The signed distance constraint function is introduced based on the signed distance kernel function, and anisotropic kernel function is constituted, which meets that the function value is zero in external area of target, and provides accurate tracking window for target tracking. According to the fact that the mean shift based on anisotropic kernel function must meet the weights sum of the sample points to center point vectors in the tracking window is zero when applying to target tracking, the mean shift window centers of anisotropic kernel function templates are calculated. Restrictions on the target template change before and after are carried out by using similarity threshold, and the adaptive updating of the anisotropic kernel function template and accurate real-time tracking of the target are realized. Experimental results show that the proposed algorithm has higher accuracy and better real-time performance.