融合角点特征与颜色特征的Mean-Shift目标跟踪算法  

Mean-Shift algorithm fused with corner feature and color feature for target tracking

作　　者：宋丹[1] 赵保军[1] 唐林波[1] 

机构地区：[1]北京理工大学信息与电子学院,北京100081

出　　处：《系统工程与电子技术》2012年第1期199-203,共5页Systems Engineering and Electronics

摘　　要：针对Mean-Shift算法稳定性差、无法适应目标遮挡的特点,提出了一种融合角点特征与颜色特征的目标跟踪算法。该算法利用Harris角点的特征不变性克服了Mean-Shift算法鲁棒性差的缺点,同时利用Mean-Shift算法中核概率密度估计特性克服了目标与背景角点难以区分的缺点。通过视频序列对该算法的跟踪稳定性与抗遮挡性能进行测试,结果表明,新算法的跟踪稳定性与抗遮挡能力优于基于单一角点或颜色特征的Mean-Shift算法。A novel target tracking algorithm fused with the corner feature and the color feature is proposed to solve the poor stability and the anti-blocking capability of the Mean-Shift algorithm.The invariance of Harris corner is used to solve the weak robustness of the Mean-Shift algorithm,and the kernel probability density estimation of the Mean-Shift algorithm is used to improve the ability of distinguishing target corners from background corners.Using a group of videos to test the proposed algorithm,the results show that the tracking stability and the anti-blocking capability of this algorithm are better than that of the Mean-Shift algorithm with single corner feature or color feature.

关 键 词：MEAN-SHIFT 特征融合 HARRIS角点 不变特征 BHATTACHARYYA系数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]