Abnormal Event Detection in Real Time Video
Sidi Ahmed Mahmoudi, Sharif Haidar, Nacim Ihaddadene, Chabane Djeraba · ORBi UMONS · 2008
This paper describes an approach to detect abnormal mo- tion in videos. The core of the approach detects portion of video that corresponds to sudden changes of motion vari- ations of a set of de¯ned points of interest. Optical °ow technique tracks those points of interest. There are su±- cient variations in the optical °ow patterns in a mob scene when there are cases those showing abnormalities. The geo- metric clustering algorithm, k-means, clusters the obtained optical °ow information to get the distance between two consecutive frames. In general, comparatively high distance indicates abnormal motion. To demonstrate the interest of the approach, we present the results based on the detection of abnormal motions in video, which consists of both normal and abnormal motions.