基于均值漂移和卡尔曼滤波的目标跟踪方法  

Target Tracking Based on Mean-shift and Kalman Filter

作　　者：詹建平[1] 黄席樾[1] 沈志熙[1] 杜长海[1] 

机构地区：[1]重庆大学自动化学院,重庆400044

出　　处：《重庆理工大学学报（自然科学）》2010年第3期76-80,共5页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金资助项目(69674012);重庆市自然科学基金资助项目(2006BA6016)

摘　　要：分析了Mean-shift难以有效跟踪复杂背景下灰度运动目标的主要缺陷,提出了结合Mean-shift和强跟踪滤波器的目标跟踪方法。该方法利用强跟踪滤波器预测目标在当前时刻的起始位置,然后Mean-shift在该位置的邻域内寻找目标所处位置。同时,采用Bhattacharyya系数度量"目标模型"和"候选模型"相似程度,提出一种目标遮挡因子,作为目标被遮挡程度的判断根据,并由此确定"候选模型"是否更换为"目标模型",避免目标模型过度更新。对城区交通环境下的车辆目标进行跟踪。实验结果表明,该方法较原Mean-shift方法可明显提高阻挡情况下的目标跟踪稳定性。After analyzing the theoretic limitation of the Mean-shift to track gray background, a method, which combines Mean-shift and Kalman filter, is proposed. position of Mean-shift is predicted by Kalman filter at present, and then the Mean target in complex Firstly, the initial -shift is utilized to track the target position around the initial position. Meanwhile the same time, the Bhattacharyya coefficient is adopted to measure the comparability between the target model and the candidate model, An occlusion coefficient is proposed as the evidence of occlusion. Then determines whether or not the target model is replaced by the latter to avoid the target model being updated excessively. Experiments based on the vehicle objects in the city are carried out, and the simulation results show that the tracking stability and adaptability for the gray imaging target, even in occlusion, are improved significandy with the proposed method.

关 键 词：目标跟踪 Mean SHIFT算法 卡尔曼滤波 遮挡因子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]