鲁棒的基于均值漂移的自适应卡尔曼滤波目标跟踪  

Robust objective tracking using adaptive Kalman filter based on mean shift

作　　者：李小和[1] 屈展[1] 王魁生[1] 

机构地区：[1]西安石油大学计算机学院,陕西西安710065

出　　处：《西安石油大学学报（自然科学版）》2015年第5期106-110,12,共5页Journal of Xi’an Shiyou University（Natural Science Edition）

基　　金：国家自然科学基金(编号:51174162);陕西省教育厅专项科研计划(编号:14JK1584);西安市科技计划(编号:CXY1346(7));西安石油大学青年科技创新基金(编号:2013BS021)

摘　　要：提出了一种鲁棒的基于均值漂移的自适应卡尔曼滤波目标跟踪算法。首先建立卡尔曼滤波的系统模型,用卡尔曼滤波预测目标在当前帧的位置,并将该预测值作为初始值,用均值漂移算法搜索目标位置。然后将搜索结果作为观测值来修正预测值,并根据目标模型与由均值漂移算法搜索得到的候选目标模型及相应背景模型的Bhattacharyya系数自适应调整卡尔曼滤波的参数,从而提出了一种鲁棒的自适应卡尔曼滤波目标跟踪算法。仿真实验表明,该算法具有较好的跟踪精度,对遮挡具有较强的鲁棒性。A robust objective tracking algorithm using adaptive Kalman filter( KF) based on mean shift( MS) is proposed. Firstly,the system model of KF is constructed,and the center of objective predicted by KF is used as the initial value of the MS algorithm.The searching result of MS is fed back as the measurement of adaptive KF. At the same time,the parameters of KF are adjusted not only by the Bhattacharyya of objective model and candidate objective model,but also by the Bhattacharyya of objective model and background model adaptively. By this way,a robust objective tracking algorithm using adaptive KF based on MS is proposed. The experimental results show that the proposed tracking algorithm has good localization precision of tracking,and is robust against occlusion.

关 键 词：目标跟踪 卡尔曼滤波 均值漂移 AKF算法 鲁棒性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]