基于Mean Shift和自适应预测的非刚性目标跟踪算法  

Non-rigid object tracking algorithm based on Mean Shift and adaptive prediction

作　　者：常发亮[1] 赵瑶[1] 陈振学[1] 徐建光[1] 

机构地区：[1]山东大学控制科学与工程学院,济南250061

出　　处：《控制与决策》2009年第12期1821-1825,共5页Control and Decision

基　　金：国家自然科学基金项目(60775023;60975025);山东省自然科学基金项目(Z2005G03);山东省博士基金项目(2006BS01012)

摘　　要：传统Mean Shift跟踪算法在目标发生机动或存在遮挡的情况下跟踪效果不理想.对此,结合目标的形状特征和颜色的可区分度对传统的颜色直方图进行改进,给出了将Mean Shift和卡尔曼滤波器或粒子滤波器相结合的目标运动自适应跟踪算法,并针对粒子滤波器计算量大的问题,给出了运用两种不同运动模式粒子进行有效预测的方法.结果表明,该算法可实现快速的非刚性目标跟踪,对目标的不规则运动和严重遮挡具有很好的鲁棒性.Traditional Mean Shift tracker has poor performance when the object＇ s movement is maneuverable or occlusion occurs. Therefore, Mean Shift combined with Kalman filter or particle filter is proposed as a self adaptive tracking algorithm in this paper. The color histogram is improved, which combines the target shape information with color discrimination. Then, a self-adaptive tracking algorithm according to the object＇ s movement is proposed, which combines Mean Shift with Kalman filter or particle filter. Moreover, particles with two different motion-models are used to reduce the expensive computation of particle filter. The experimental results show the real-time performance of the algorithm in tracking non-rigid object as well as the robustness to irregular movement and heavy occlusion.

关 键 词：自适应预测 非刚性目标跟踪 Mean SHIFT 卡尔曼滤波器 粒子滤波器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]