一种融合改进Mean Shift和SIFT的跟踪算法  

An object tracking method by fusing improved Mean Shift and SIFT

作　　者：杜凯[1] 宋京妮 巨永锋[1] 田浩[1] 

机构地区：[1]长安大学电子与控制工程学院,陕西西安710064 [2]长安大学经济与管理学院,陕西西安710064

出　　处：《广西大学学报（自然科学版）》2014年第5期1102-1108,共7页Journal of Guangxi University（Natural Science Edition）

基　　金：陕西省科技工业攻关项目(2014K05-59);中央高校基本科研业务专项资金资助项目(2014G3322008)

摘　　要：针对传统Mean Shift(MS)跟踪算法易受遮挡,复杂背景和光照变化等因素影响,导致不准确跟踪或跟踪丢失,提出了一种融合改进MS和SIFT的跟踪算法。改进的MS算法根据目标对象的最新外接矩形尺寸,确定对象的分块方法,根据各块的Bhattacharyya系数值,确定各块的权重系数,获得初步的跟踪结果;采用SIFT特征匹配和校正的方法对其初步跟踪结果进行调整;采用线性加权的方法融合改进的MS跟踪结果和SIFT校正结果,得出最终的跟踪结果。实验表明,提出的方法比传统的MS算法和固定分块的MS算法都具有更好的跟踪性能。Traditional Mean Shift （ MS ） algorithm is easily vulnerable because of the influence of factors such as occlusion, complex background and illumination change, thus often causing inaccu-rate tracking or loss of the object.A novel tracking algorithm based on fusing improved MS and SIFT （ Scale invariant feature transform） is proposed.Firstly, an improved MS algorithm is utilized to ob-tain a preliminary tracking results and to determine the object block by the size of the latest enclosing rectangle and the weights of every block by their Bhattacharyya coefficients.Then SIFT feature matc-hing and correction is used to adjust the preliminary tracking results.Finally the tracking results of the improved MS and SIFT are fused by using linear weighted method.Experimental results show that the proposed algorithm is better than the traditional MS algorithm and the fixed block MS algo-rithm in tracking performance.

关 键 词：对象跟踪 Mean SHIFT SIFT 分块颜色直方图 BHATTACHARYYA系数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]