科研之友
微信
新浪微博
Facebook
分享链接
摘要
Many video surveillance systems are being done in the area of motion based video object segmentation an the pixel domain, which exploits visual attributes and motion information. Given most existing images and videos stored in compressed form, specific manipulation algorithms could be developed to deal with compressed streams without full decoding of compressed images/videos. This paper obtains motion masks from 1 and P frames in MPEG video. Then, motion vectors (MVs) decoded directly from P frames are used for moving object extraction By k-means clustering of MVs conforming to the same motion, connected or near blocks moving in similar directions that appeared to be of the same moving object could be grouped together These moving objects can then be tracked by following the A Vs in consecutive P frames A confidence value indicating tracking result is calculated from these matched blocks. Finally, three criteria involving only boolean operations for object behavior analysts are defined to determine intrasion, halt, or occlusion Experimental results demonstrate the feasibility of the proposed system
- 出版日期2010-10
