用于智能交通"车流量检测"的压缩域内运动目标检测  

Moving Objects Detection from MPEG-1 Video Stream

作　　者：黄若芸[1] 莫玉龙[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200072

出　　处：《上海大学学报（自然科学版）》2003年第2期100-104,共5页Journal of Shanghai University:Natural Science Edition

基　　金：国家自然科学基金(69875009)资助项目

摘　　要：该文提出了一种从MPEG-1视频流的压缩域中直接检测多运动目标的方法.利用MPEG-1位流中的运动矢量和DCT系数两类信息,依据运动目标在图像序列中的时空相关性,检测跟踪多运动目标.相比空间域的检测,压缩域在宏块层面上检测目标,并可以避开逆DCT的计算,因而提高了检测速度,可以应用在实时性强的场合.此文以智能交通车流量检测为平台,主要分析了相对固定摄像机远离和走近的多目标运动模型,并提出一种在压缩域内更新背景的算法.实验结果证明这种方法速度快,在车流量畅通时识别率较高.We present an algorithm that detects multiple moving objects directly from MPEG1 stream. To detect and track the objects, the system makes use of motion vector and DCT coefficients in the MPEG1 bit stream based on the temporalspatial correlation in image sequence. Since moving objects are detected on the macroblock level so that inverse DCT is not needed it improves detection speed compared to the spatial domain approach, this is suitable for realtime applications. On a vehicle flow survelliance platform, we study a model of multiple zoom in/out moving objects using a fixed camera, and present an algorithm of updating background in the compression domain. Experiment results confirm the fast processing speed and high detection rate in freeflowing traffic.

关 键 词：智能交通 MPEG-1视频流 运动目标检测 运动矢量 图像处理学 车流量检测 压缩域 

分 类 号：TN919.81[电子电信—通信与信息系统] U495[电子电信—信息与通信工程]