均值漂移跟踪的双模板更新算法  

Dual template algorithm for Mean-Shift template update

作　　者：覃剑[1] 曾孝平[1] 曾浩[1] 

机构地区：[1]重庆大学通信工程学院,重庆400030

出　　处：《计算机应用研究》2009年第7期2771-2773,2795,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(30570473);重庆市信息产业发展基金资助项目(20051022)

摘　　要：针对均值漂移算法缺少必要的模板更新方法的缺点,提出了一种基于双模板判定的更新算法。该算法首先通过分析目标特征与背景特征的相对大小,设计了加权函数分别对前景和背景特征进行加权;然后在此基础上引入背景模板并构造双模板,通过对候选目标与双模板相似度系数的综合分析,可以准确判定跟踪状态及干扰产生的原因,以采取相应的模板更新策略。实验表明,该算法可以有效地增强均值漂移算法在目标姿态变化、前景遮挡等复杂条件下的跟踪效果,具有较好的跟踪稳健性。To improve the limitation of Mean Shift lacks method of template update, presented a dual template update algorithm. Firstly, through the comparison of features in target and background, designed the weighting function to weighted the feature distribution respectively. Then introduced a background template and the dual template was constructed based on it. Finally, after comprehensive analysis of similarity degree of candidate object and dual template, the proposed algorithm could accurately determine the tracking status and the cause of interference, and took corresponding template update strategy. Experimental results show that the proposed algorithm can effectively enhance the tracking effect under condition of changeable gesture of target and occlusion, and is more robust.

关 键 词：均值漂移 目标跟踪 模板更新 双模板 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]