基于DESO的Mean Shift目标跟踪算法研究  

Study on Mean Shift tracking algorithm based on DESO

作　　者：王晓卫[1,2] 马晓军[2] 周启煌[2] 

机构地区：[1]陆军航空兵学院机载设备系,北京101123 [2]装甲兵工程学院控制工程系,北京100072

出　　处：《控制与决策》2009年第2期212-216,共5页Control and Decision

基　　金：国家自然科学基金项目(60175010);武器背景预研项目(10401025)

摘　　要：将基于DESO的运动预测算法和Mean Shift算法相结合,形成一种新的基于Mean Shift的快速目标跟踪算法.该算法以DESO预测位置作为MeanShift算法下一帧候选模型的计算中心,实现了对快速运动目标的跟踪,并通过DESO对目标运动轨迹进行预测,较好地解决了目标完全遮挡时的跟踪问题.实验结果表明,该算法具有预测精度高、实时性好、抗遮挡能力强的优点.The modified Mean Shift algorithm based differential extended state observer（DESO） is proposed by the combination of DESO and Mean Shift algorithm, which realizes tracking for target with high speed since it searches for target in the neighborhood of estimated position that DESO predicts. Moreover, the proposed algorithm uses target^s motion information that DESO predicts, which can solve target occlusion preferably. Experiment results show the proposed algorithm has superior features, such as higher prediction precision, stronger anti-occlusion performance, and can meet the practical requirement in real-time.

关 键 词：目标跟踪 运动预测 均值迁移算法 微分扩张状态观测器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]