A Study on Camera Motion Estimation in MPEG Video

• Main Authors: Chien-Ming Lai, 賴建明
• Other Authors: Ming-Shi Wang
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/97174028743945158542

碩士 === 國立成功大學 === 工程科學系碩博士班 === 95 === Camera motion estimation is very important from a video processing view. Video application, including object segmentation and tracking, mosaic and content-based video indexing and retrieval, based on the video content should represent motion efficiently. Since, to decode MPEG video and estimate camera motion is extremely time-consuming. To obtain efficient estimates of camera motion by directly processing data easily obtained from MPEG-compressed video becomes more important. In this thesis, two main concerns of camera motion estimation from motion vector are addressed. Firstly, we introduce the connection from a GOP to next GOP by B-frame bidirectional reference. Secondly, we propose a 2-pass camera motion estimation to cope with outliers caused by the uncertainty of motion vectors directly from MPEG video format. Experiment results show that the 2-pass camera motion estimation method is reliable. Finally, we have developed a mosaic system by applying the proposed method.