利用MPEG-4视频流的运动目标检测算法  

Motion Object Detection Based on MPEG-4 Video

作　　者：董文涛[1] 杜鹏[1] 田玉敏[1] 

机构地区：[1]西安电子科技大学计算机外部设备研究所,西安710071

出　　处：《西安交通大学学报》2008年第4期454-457,共4页Journal of Xi'an Jiaotong University

基　　金：陕西省自然科学基金资助项目(2006F48)

摘　　要：提出了一种基于MPEG-4压缩域运动目标检测算法.该算法直接在压缩码流中获取运动矢量,利用迭代滤波器的良好特性,尽可能还原真实运动矢量信息,并采用自适应阈值选取法的运动矢量标识出运动区域.在此基础上,计算滑动窗口内八邻域直流(DC)系数与中心块DC系数之差的模,根据选定的阈值来判断中心块的运动特征,然后扩展运动区域,计算新区域中背景块的平均DC系数,并与边缘块做差值计算,进而调整边缘块.算法充分利用了MPEG-4码流中蕴含的运动信息,克服了仅由运动矢量检测运动目标存在误差太大的缺点,适用于计算机数字视频技术领域.实验结果表明,所提算法不需要对压缩流完全解码,可大大提高检测效率,改善检测效果,对不同视频序列具有良好的普适性,特别是在光强、亮度等因素的影响下依然保持良好的鲁棒性.An algorithm to detect the moving object on MPEG-4 compressed streams is presented. The first step of the algorithm is to mark the moving regions based on the motion vectors, that have been filtered by cascade filter. Then a local object detecting is implemented by using DC coefficients of each marked block and calculating the difference between the central block and its neighbors＇. Finally, the border is rectified by extending the objects＇ region and comparing border blocks＇ DC coefficients with the average background＇s value of the new region. Testing results show that the minimal decoding on compressed bit stream is performed and the detecting effect is dramatically improved. The algorithm performs well on different videos and keeps its robust character despite of other influence, such as light intensity.

关 键 词：目标检测 运动矢量 直流系数 视频技术 

分 类 号：TN919.81[电子电信—通信与信息系统]