基于监控视频的Farneback光流算法的人体异常行为检测

Human abnormal action detection based on the Farneback optical flow arithmetic of surveillance video

为了解决视频监控系统中异常行为检测精度不高,计算方法复杂的问题,提出一种基于监控视频的Farneback光流算法的人体异常行为检测方法。该方法选用密集光流算法提取运动区域内的光流信息,采用直方图方向熵和幅值熵的乘积计算出用于表示行为混乱程度的W值,以此来判断是否发生异常行为,且W值越大,说明当前运动越混乱,发生异常行为的可能性越大。实验结果表明,所提算法可以有效地检测指定监控区域内的人体异常行为,检测率可达到92.33%,该方法简单实用,为监控视频的智能分析提供技术支持。

For the purpose of settling a matter that the poor detection of degrees of abnormal action in video monitoring system and the intricate computing way,a detection approach of human abnormal action based on the Farneback optical flow arithmetic of surveillance video is come up with.The means uses the dense optical flow arithmetic to extract the optical flow information of the moving area,and the product of the histogram directional entropy and amplitude entropy is used to calculate the W value used to represent the degree of action disorder,so as to judge whether abnormal behavior occurs.Moreover,the larger the W value is,the extent of chaos in the current motion is to serve as a catalyst for the probability of abnormal action.Experiment consequences indicate that the mentioned arithmetic can validly detect abnormal human action in the specified monitoring region,and the detection rate can attain 92.33%.The method in this paper is simple and practical,and provides technical support for intelligent analysis of surveillance video.