针对监控视频场景的压缩域运动对象分割方法  

Object Segmentation Method in Compressed Domain for Surveillance Video Scene

作　　者：王小龙[1] 梁久祯[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《电视技术》2014年第15期24-28,共5页Video Engineering

摘　　要：提出了一种压缩域背景建模与运动对象分割框架:首先提取H.264视频压缩码流中的MB-bits与4×4块残差系数,基于MB-bits场进行压缩域Vibe背景建模分割出宏块级运动对象区域,然后结合最大熵模型提取4×4边缘块进行运动对象轮廓细化最终分割出压缩域运动对象。试验对比分析表明,提出的算法能快速、准确地提取压缩域运动对象,系统具有一定鲁棒性。A compressed domain based scheme is proposed to model background and segment moving objects in this paper. First of all,the size in bits that a macroblok MB occupies（MB-bits） and 4 ×4 residual coefficients within such a MB are extracted from the compressed bitstream,based on MB-bits field in compressed domain Vibe background modeling is build to segment moving object in MB level. Then combining maximum entropy model to refine object contour by extracting edge of 4 × 4 block, and finally segment moving object in compressed domain. Experimental comparison and analysis show that the proposed algorithm can extract the compressed domain motion object rapidly and accurately with certain robustness.

关 键 词：压缩域 背景建模 运动对象分割 最大熵 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]