Method For Moving Object Detection In Compressed Video

Nowadays moving object detecting is widely used upon compressed video in severalareas, such as multimedia, human interface and etc,as its fast detecting speed and accuracy.There are two ways to do object detecting depending on the original video: pixel domainbased and compressed domain based. The method in pixel domain mainly utilizes featuresof image and correlation between images. Compared with pixel domain object detection,applications in compressed domain mainly take advantages of encoding context such asmotion and residual image in compressed video to extract object features. Thus it’sneedless to fully decode the video stream. There are various of algorithms about objectdetection for MPEG-X/H.26X compressed video.Based on previous object detecting methods in pixel and compressed domain, wedesign a uniform framework via encoding context which can do object detecting understatic background both for MPEG-X/H.26X compressed video. To changing background,we also develop a model to build the camera motion template. The contribution for thispaper include:We analyze the video coding standards of MPEG-X/H.26X,and reflect the syntax invideo coding to the corresponding feature in video frame. Provide solid foundation to doobject detecting.To static background video, we improve the method which makes use of the encodingfactors to do object detecting, import the edge detection to further locate object’s edge torefine the detection result. Improve the spatial-temporal filter and reduce the computingcomplexity.To changing background video, we discard the global motion estimation for camera,develop a sub-area camera motion template construction and a probability fieldaccumulation method,which is used to extract the moving object by matching the cameramotion template.