期刊名称：International Journal of Computer and Information Technology
印刷版ISSN：2279-0764
出版年度：2015
卷号：4
期号：5
页码：768
出版社：International Journal of Computer and Information Technology
摘要：The ability of accurately segmenting a video into events or scenes without any human intervention has been a challenge in computer vision. In this paper, we present a more accurate way of segmenting a video into events using trajectory discontinuities, which consist of two stages. In the first stage, we estimate possible regions where the scene changed using comparison of motion features we extracted by edge detection segmentation and the second stage, we use the estimated regions to estimate trajectory discontinuity using Large Displacement Optical flow (LDOF) which calculates the actual frame where the scene has changed. Our experiments have shown that this method have high accuracy rate and it is faster that the conventional way of event video segmentation
关键词：Large Displacement Optical flow; Video event ; segmentation; Edge detection; spatial information
