Motion Object Detection Based On Mpeg-4 Video

Moving object detection is one of the active branches of imaging processing and computer vision. The testing results reflect the main information in the video stream, which take a very important role in many areas, such as video analysis, video search, as well as intelligent digital video surveillance and so on.This thesis does some research in moving object detection deeply, provides an overview of the commonly used moving object detection algorithms, and proposes ameliorated algorithms.Focusing on the problem of moving object detection on MPEG-4 compressed streams captured by a stationary camera, an algorithm to detect the moving object using DCT coefficients and motion vectors of the motion information in MPEG is presented. The algorithm is to firstly mark the moving regions based on the motion vectors, which has been filtered by using cascade filter. On this basis, the DC coefficient is used to rectify the moving regions. Testing results show that minimal decoding on compressed bit stream is performed by the method and the tracking speed can be dramatically improved; the computing time decreases much; the shortcomings of error have been overcome which due to the detection only using motion vector and maintains a good robustness in the factors of light intensity, brightness and so on.When moving object detection under moving camera, GME(Global Motion Estimation)is an important tool and widely used in moving object detection, and other fields.In this thesis, a new method for the estimation of global motion from compressed image sequences is presented. The algorithm first obtains the regional of moving background according to the distribution of the motion vector field. And then, with regard to global motion models, four-parameter estimation model is adopted. Finally, according to the estimated error, the moving object is extracted. Using this method, it is not necessary to fully decompress MPEG video sequence; the disturbance of local moving information is eliminated during the estimation of the parameters; and both the efficiency and detecting effect is improved dramatically.