一种新的基于分块直方图的Mean-Shift跟踪算法  

Improved Mean-Shift Tracking Algorithm Based on Block Histogram

作　　者：李群山[1] 张文[2] 

机构地区：[1]电子科技大学光电信息学院,四川成都610054 [2]内江师范学院工程技术学院,四川内江641112

出　　处：《电视技术》2012年第23期4-8,共5页Video Engineering

基　　金：四川省内江师范学院院级质量工程项目(XZY201103)

摘　　要：在经典的Mean-Shift算法中使用颜色直方图来描述目标,而这种颜色直方图中却存在大量空的颜色直方图区间,且没有融入图像的空间信息,在遇到与跟踪目标颜色分布相近的背景或目标时会造成跟踪丢失。针对颜色直方图的不足提出一种分块颜色直方图构建方法以融入空间信息,对得到的分块直方图进行简化,去除空闲的直方图区间。实验中对颜色分布相似的物体发生遮挡进行跟踪试验,并与经典Mean-Shift算法跟踪结果进行对比。实验结果表明,这种新算法能更稳定、准确地跟踪目标。Color histogram is usually used to describe the target in the classical Mean-Shift algorithm , but there are a large number of empty color his- togram interval . And the histogram is not integrated into the spatial information of the image,it is easy to lost tracking target when color distribution of tracking target is similar to the background or other target. To overcome this shortcomings, a new color block-histogram to integrate spatial information is presented in this paper. First target is to evenly divided into many subspace as rings, then calculate the color histogram of each sub-space and weight them , and finally remove free histogram to simplify block histogram interval. New method and the classic Mean-Shift tracking method were applied in the object tracking experiment. The experimental results show that the new algorithm can track the target more accurately and more stably.

关 键 词：目标跟踪 分块颜色直方图 Mean—Shift跟踪算法 直方图简化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]