Abstract: With the need of security surveillance，the detection of abnormal events in video surveillance has been paid more and more attention.This paper proposes an approach of detecting gathering abnormal events.Its focus is optical flow changes of the points of interest.When foreground obtained by background subtracted is more than the setted threshold and the distance among objects in a certain range，the points of interest are extracted，using the optical flow method tracks the points of interest then caluates entropy，once the entropy is greater than the setting threshold，it is judged to report a gathering event and to give an alarm；otherwise，this frame’entropy is set a fixed value，continuing processing next frame of the background subtracted.Its effectiveness has been verified through the real shot of the video sequence.

Key words: background subtracted, points of interest, optical flow, entropy, gathering events

摘要： 随着人们对安全监控的需求，视频监控中的异常事件检测越来越引起人们的关注。介绍了一种聚众异常事件检测的新方法，其核心是感兴趣点光流信息的变化。当使用背景差分法所获得的前景大于设定阈值，并且目标之间相互距离在某一范围内时，提取感兴趣点，使用光流法跟踪感兴趣点，然后计算熵值，一旦熵值大于设定的阈值，则聚众事件发生，系统报警；否则，将该帧熵值设置为某一固定值，继续下一帧的背景差分处理。通过实拍视频序列验证了该算法的有效性。

关键词: 背景差分法, 感兴趣点, 光流法, 熵, 聚众事件