协方差矩阵具有融合多维特征,获得全局最优解的优点,在目标描述方面展现出了优秀的性能,然而传统的协方差匹配难以跟踪被严重遮挡的目标,且全局搜索易遭受相似背景的干扰。为了提高协方差跟踪的性能,提出了基于遗忘因子与卡尔曼滤波的协方差跟踪算法。利用协方差矩阵可实现多种特征的巧妙融合,采用基于遗忘因子的加权搜索策略,可以削弱窗口内相似目标的干扰,利用卡尔曼滤波预测目标运动轨迹并判断目标是否被严重遮挡,使遮挡消失后目标仍能被重新捕获。实验结果表明,该算法可在摄像机运动、目标旋转、缩放和被遮挡等情况下实现刚性与非刚性目标的稳健跟踪。

Covariance matrix is an excellent objet descriptor which can fuse multiple features and provide a global optimal resolution.Unfortunaely,it is hard for traditional covariance matching to track target when severe occlusion occurres.Further more,more similar background′s disturbances may be introduced by global search.In order to improve the performance of covariance tracking,a covariance tracking algorithm based on forgetting factor and Kalman filter is proposed.Multiple features can be skillfully fused using covariance matrix.In order to reduce the disturber from similar targets,weights on the distance function among covariance mattrixes is imposed using forgetting factor based on fuzzy membership.The Kalman filter is used to predict the trajectory of target and judge whether severe occlusions occurre which allow us to capture again when occlusions disappear.Experimental results show that the method can successfully cope with camera moving,clutter,occlusions,and target variations such as scale and rotation for tracking rigid and non-rigid targets.