H.264压缩域中mean-shift聚类运动目标分割算法  

An algorithm of mean-shift clustering-based moving object segmentation in H.264 compression domain

作　　者：孙乐[1,2] 戴明[1] 李刚[1] 陈晓露[1,2] 王子辰[1,2] 宋悦铭[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,中国科学院航空光学成像与测量重点实验室,吉林长春130033 [2]中国科学院大学,北京100039

出　　处：《光电子．激光》2013年第11期2205-2211,共7页Journal of Optoelectronics·Laser

基　　金：国家"863"高技术研究发展计划(2008AA121803);国家"973"重点基础发展规划(2009CB72400603B)资助项目

摘　　要：针对视频监控、检索,提出了一种在H.264压缩域下运动对象分割的新算法。根据实际监控应用特点,算法主要利用H.264码流中提取的运动矢量(MV)、分块尺寸信息对原始的MV场(MVF)进行可靠性分析以及空间滤波、mean-shift聚类处理,从而得到可信度较高的MVF场。首先,从H.264码流中获取原始MV并进行归一化处理,同时对原始MVF进行两步空间滤波;其次,对不同的块大小分配相应的权值作为每个样本的权重系数,将处理之后的MVF作为样本空间,利用mean-shift聚类获取真实的MVF;最后,根据可靠的MVF标记运动目标。实验结果表明,本文提出的mean-shift聚类运动目标分割算法可以获得有效并可靠的分割结果。Video segmentation has played a critical role in most video processing fields.A new algorithm of moving object segmentation in the H.264compression domain for video surveillance and retrieva l is proposed.According to the practical application characteristics,this algorithm mainly utilizes motion vectors and partitioned bl ock size information that are directly extracted for H.264bit streams to analyze the reliability of original motion vector field （MVF）,and p rocesses the MVF using spatial filter and mean-shift clustering,thus gets reliable MVF.First,the original motion vectors are obt ained from the H.264stream and normalized after two steps of spatial filter.Then,appropriate weight is assigned for partitioned block size as the weight coefficient of each sample and then the processed MVF is used as the sample space for mean-shift clustering algorit hm to obtain the real MVF.Finally,the moving target will be segmented from the background according to the marked reliable motion vectors.T he experimental results demonstrate that the proposed mean-shift clustering-based moving object segmentation algorithm coul d achieve good performance and efficiency.

关 键 词：视频监控 运动对象分割 H 264 mean-shift聚类 

分 类 号：TN941.1[电子电信—信号与信息处理]