基于SIFT特征一致性的目标跟踪算法  

Object Tracking Algorithm Based on SIFT Features Correspondence

作　　者：黎云汉[1] 楼京京[1] 

机构地区：[1]义乌工商职业技术学院机电信息分院,浙江义乌322000

出　　处：《控制工程》2011年第6期966-969,共4页Control Engineering of China

基　　金：义乌市科技计划项目(09-1-31)

摘　　要：针对光照条件突然变化情况下混合目标模型Mean Shift算法无法准确跟踪目标的缺点,提出了一种基于SIFT特征一致性的目标跟踪算法。算法用SIFT特征来匹配帧间的感兴趣区域,同时使用包含初始帧信息和前一帧信息的混合目标模型Mean Shift算法计算帧间感兴趣区域的直方图,以直方图分布距离最小为原则计算Mean Shift向量,从而得到目标的真实位置。实验证明,该算法能在光照条件突然变化情况下对目标进行准确跟踪。To the problem that the mean shift algorithms with hybird target model can not track the object accurately when sudden changes in illumination, a tracking algorithm based on SIFT features correspondence is proposed. The algorithm uses SIFT features to match the region of interests across the frames. The mean shift algorithm of hybrid target model using the information of the first frame and the previous frame is applied to conduct the similarity search via color histograms simultaneously. The mean shift vector is calculated via the color histograms with the principle of distance minimization to get the target location. The experimental result shows that the proposed algorithm can track the object more accurately when sudden changes occurred in illumination.

关 键 词：mean SHIFT SIFT 目标跟踪 

分 类 号：TP27[自动化与计算机技术—检测技术与自动化装置]