Moving Target Detection Algorithm Based On H.264 Interframe Prediction

The detection of moving objects from the video sequence is an important researchtheme in the computer vision field.In the life,abundant significant visual information is inthe movement.It's crucial for the farther image analyzing and processing to divide themoving objects quickly and exactly from the variational background in the real time.This paper is mainly on the research of the moving objects dejection technology fromthe video sequence.It involves not only simple background sequence in which the objectsare moving while the camera is static but also complex background sequence in which theobjects and the camera are moving together.The motion estimation and compensation technology in the video coding is to use theprevious frame as the predictor for the current frame.In the H.264 video codingstandards,important differences from earlier standards include the support for a range ofblock sizes(from 16×16 down to 4×4) , fine sub-sample motion vectors(quarter-sampleresolution in the luma component) and the prediction of motion vectors.These precisemotion vectors imply abundant image motion information.This paper extracts the precise motion vectors in the course of H.264 videocoding.Background area and some moving objects areas can be detected by statisticalanalyzing and clustering these fine motion vectors.Adopting this technique,the surveillanceclient can detect the moving objects after receiving the compressed bit-stream whichimplies abundant motion information instead of receiving the large raw video sequencedata.Then it can improve the performance of remote detection of moving objects.