提出了一种基于监控视频的异常事件识别模型, 该模型可以实时监测视频中的前景目标, 并通过分析目标的运动信息判断是否有异常事件的发生。首先, 采用背景建模的混合高斯算法提取前景目标; 然后, 用金字塔迭代的 L.K特征点跟踪算法得到前景的光流运动信息, 并通过分析前景的面积比例、速度方差、整体熵判断视频中是否有异常事件的发生; 最后, 利用爆炸、人群短时聚集和分散两种异常事件做仿真实验。结果表明, 该模型可以准确提取前景目标区域, 并可以快速、精准地判断监控视频中的异常事件, 可以为管理部门及时发现和控制异常事件提供有效的帮助。

In this paper, we proposed an abnormal event recognition model based on surveillance video. The model can monitor the foreground target in video in real time and determine whether there is an abnormal event by analyzing the motion information of the target. The model utilizes the background model-based hybrid Gaussian algorithm to extract the foreground target. The L-K feature point tracking algorithm based on the gold tower iteration is subsequently adopted to obtain the foreground optical flow motion information. The abnormal event is judged by the analysis of the foreground area ratio, speed variance, and the overall entropy. Two kinds of abnormal events, such as explosions, short-time crowding and dispersion are chosen for simulation, the results show that the model can accurately extract the foreground target area and correctly determine the occurrence of abnormal events. Furthermore, the method can quickly and accurately identify abnormal events in the surveillance video, and can help the management department to find and control abnormal events in time.