基于分块背景更新的均值飘移跟踪算法  

Mean shift tracking algorithm based on block background update

作　　者：杨莉[1] 李茂林[1] 

机构地区：[1]南昌大学信息工程学院,江西南昌330031

出　　处：《计算机工程与设计》2015年第6期1571-1574,1613,共5页Computer Engineering and Design

基　　金：江西省科技支撑计划基金项目(20142BBE50037;20121BBE50023);江西省自然科学基金项目(20132BAB211039)

摘　　要：为提高融合背景信息均值飘移跟踪算法中的背景模型更新效率,提出基于分块背景更新的改进算法。将目标周围的背景区域划分成4块,分别以此4个子块的颜色直方图作为区域特征描述符进行计算,利用巴氏系数选择性地更新相似性较低的3个区域。在构建目标候选模型时,利用目标物体空间信息赋予候选目标区域像素点相应的距离权重。对整块背景信息更新策略和加权更新背景信息策略进行比较,比较结果表明,该算法具有更高的迭代收敛效率和更强的抗背景干扰能力。To improve the background model updated efficiency of the background information integrated mean shift tracking al‐gorithm ,a weighted improved algorithm based on the block background update was proposed .The background region of the tar‐get was divided into four sub‐blocks and the color histogram models of the four sub‐blocks were calculated ,respectively .The similarities of the lower three sub‐blocks model were selectively updated using Bhattacharyya coefficient .In the construction of the target candidate model ,the object spatial information was used to give the corresponding distance weight to pixels of the can‐didate target .The improved algorithm has stronger ability on iterative convergence and background interference resistance .

关 键 词：目标跟踪 均值飘移 分块背景 距离权重 颜色直方图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]