Dissertation > Excellent graduate degree dissertation topics show

Research on Dense Estimation of Motion Vector Field and Motion Object Extraction in MPEG-2 Compressed Domain

Author: ChenZuoZuo
Tutor: YangGaoBo
School: Hunan University
Course: Communication Engineering
Keywords: Compressed Domain MPEG-2 Compressed domain optical flow estimation Moving Object Segmentation
CLC: TP391.41
Type: Master's thesis
Year: 2008
Downloads: 72
Quote: 0
Read: Download Dissertation

Abstract

The segmentation of the video object (VO) refers to extract meaningful moving object from a video sequence. It has broad application prospects in the field of intelligent video surveillance systems, target recognition, video summaries. The initial study was started from the pixel domain, despite the high accuracy of segmentation, pixel-domain video object segmentation there is time-consuming and large, it is difficult to meet the needs of real-time applications, defects. Video usually to the encoded video stream carried storage and transmission, for the encoded video stream, if restored to the pixel domain, and then uses the pixel domain video object extraction algorithm, will be significant with increasing the amount of computation, even more difficult to meet the requirements of real-time applications. Therefore, the research focus in recent years, video object segmentation from pixel domain over to compressed domain, ie directly in the compressed domain video object extraction. The movement information can be partially decoded video stream to obtain the motion vector of each macro block, which can avoid the time-consuming operation of the motion estimation / compensation, etc., there are conducive to extract real-time video object. However, the limited available information and the motion vector with the real object movement there is a certain error in the compressed domain video object extraction, segmentation accuracy is usually lower than the pixel domain. Compressed domain motion object segmentation lies compressed domain motion information densification and accurate, in order to improve the accuracy of segmentation. This paper studies compression domain of optical flow field approximation, the motion vector field that is accessible from the video stream approximate optical flow method, based on the MPEG compressed domain moving object extraction. The main work of the paper is as follows: first, a brief introduction of the MPEG-2 video coding standard, a comparative analysis of the estimated optical flow field and block-based motion estimation, summarized the basic principles of compressed domain optical flow estimation method. Second, an MPEG-2 compressed domain motion vector densification and accuracy of the method. It utilizes an easy to get the information of the MPEG-2 video stream - motion vectors and DCT coefficients, especially AC coefficients, one kind of MPEG-2 compression algorithm domain approximation optical flow field generator. First motion vector processing, it is estimated that the DCT coefficients of P and B frames, image reconstruction DC 2AC reuse Black optical flow estimation algorithm to preliminary estimates of the optical flow field Finally credibility retain high credibility The degree of the moving object edge optical flow estimation, generate MPEG-2 compressed domain approximate optical flow field. Simulation results show that it effectively solves the the aperture problem exist in the estimation of the optical flow field and provide a good motion analysis based, can be for moving object segmentation. Third, the optical flow field based on MPEG-2 approximate compressed domain moving object segmentation algorithm. The optical flow based on the MPEG-2 approximation, using the global motion estimation model, first rejected through iterative optical flow of the edge of the background removal, and then using the motion vector consistency model to extract the edge of the foreground moving object. Typical video test sequence Experimental results show that the method can be made close to the pixel-level segmentation accuracy.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile