基于直方图投影的质心跟踪算法  

A Centroid Tracking Algorithm Based on Histogram Back-Project

作　　者：宋涛[1] 马宇峰[1] 蔺海峰[1] 

机构地区：[1]西安通信学院,陕西西安710106

出　　处：《微计算机信息》2010年第26期205-207,共3页Control & Automation

基　　金：基金申请人:马宇峰;项目名称:远程数字视频监控系统的研究与实现;基金颁发部门:部委预研基金(9140A07051208JW0111)

摘　　要：针对传统mean-shift算法中核函数密度估计的缺陷,本文提出一种目标颜色直方图投影和mean-shift迭代算法相结合的质心跟踪方法,而直方图投影是一种非参非核的密度估计方法。目标的颜色直方图建立之后,将每一帧后续输入图像的象素值转化为该直方图分布下的概率值,利用mean-shift迭代在概率图中寻找密度分布的模式点,同时在mean-shift的迭代终止准则中提出平均灰度的匹配策略,来减小颜色相似物体对真实目标的干扰。在跟踪过程中,利用kalman预测目标的位置,设置目标邻域的投影范围。实验结果表明,与传统的mean-shift算法相比,该方法降低了计算复杂度,在复杂背景下有更好的跟踪效果。According to the fault of kernel function weight in traditional mean-shift, a centroid tracking method based on histogram back -project combined with mean-shift iterative algorithm is proposed. Histogram back -project is a method of estimating density, which is kernelless and nonparametric. A histogram model of the object is built in the initial frame and the value of pixels in the consecutive frames will be transformed into probability in the light of that model, meanwhile utilizing mean-shift to search the location of the mode and an matching strategy of mean gray is presented to terminate the iteration of mean-shift. In the course of target tracking, kalman filter is used to predict the position of object and searching scope is fixed in an area nearby the object. Experimental results show that compared with the traditional mean-shift algorithm, this method not only reduces the computation complexity, but also has better tracking effect in the complex background.

关 键 词：MEAN-SHIFT 直方图投影 平均灰度 卡尔曼滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]