基于帧间差的区域光流分析及其应用

传统视频运动检测常用帧间差、背景差等方法检测图像变化的存在,它们能敏感地给出变化区域,在此基础上可方便提取运动对象的位置和轮廓等静态特性,但并不能直接得到运动对象的速度、方向等运动特性;光流法虽可进一步给出运动场中每个像素位置的运动特性,但它所涉及的计算量庞大,难以直接应用于有实时需求的场合。因此,文章提出了一种基于联合帧间差的区域光流分析方法,通过联合帧间差方法提取运动区域,针对运动区域进行光流计算,保持了实时处理所需的速度并降低了光流计算的计算代价,并讨论了其在视频监控、交通监管等场合的应用。

Temporal Difference Based Local Optical Flow and its Application

Temporal Difference and Background subtraction methods are always adopted in video motion detection.They can easily find the static features of object,such as position or outline of moving object,but they also have their limitation on abstracting motion features like velocity or direction.Optical flow can make up this shortcoming,while the high demand of computing has restricted its application in real-time situation.A temporal difference based local optical flow method is introduced here,which can combine the virtue of the two methods,with fast in speed and also low cost in computing.The use of such method in video surveillance or traffic monitoring is also discussed.