基于运动差值熵的实时人群异常检测  

Real-time crowd anomaly detection based on motion difference entropy

作　　者：王振 王旭智[1,2] 万旺根[1,2] Wang Zhen;Wang Xuzhi;Wan Wanggen(School of Communication and Information Engineering,Shanghai University,Shanghai 200444,China;Institute of Smart City,Shanghai University,Shanghai 200444,China)

机构地区：[1]上海大学通信与信息工程学院,上海200444 [2]上海大学智慧城市研究院,上海200444

出　　处：《电子测量技术》2020年第10期119-124,共6页Electronic Measurement Technology

基　　金：上海市科委国际合作项目(18510760300)资助

摘　　要：检测拥挤场景中的人群异常情况作为近年来计算机视觉的热点,非常具有挑战性,在监控视频领域,人群异常检测对于公共安全意义重大。因此提出了一种基于运动差值熵的实时检测人群异常的方法,针对拥挤场景人群高密度高遮挡的问题,使用改进光流法来提取人群运动的轨迹,在得到人群运动的光流后,通过计算光流的幅度值大小来生成每帧视频中人群的运动图,然后通过相邻帧之间的人群运动差值来表示人群运动的变化,最后计算出人群运动差值熵来检测人群异常。通过在公共数据集上进行了相关的实验,实验结果表明,与最先进算法相比,在拥挤场景中所提出的方法具有较高的人群异常检测准确率。Detecting crowd anomaly in crowded scenes has become a hot spot in computer vision in recent years.It is very challenging.In the field of surveillance video,the detection of crowd anomaly is of great significance to public safety.A method for real-time detection of crowd anomaly based on motion difference entropy is proposed.Aiming at the problem of high density and high occlusion of crowd in crowded scenes,it used an improved optical flow method to extract the trajectory of crowd movement.After obtaining the optical flow of crowd motion,next step is calculating the magnitude of the optical flow to generate the crowd motion map in each frame of video,and then use the crowd motion difference between adjacent frames to represent the change in crowd motion.The last step is calculating the crowd motion difference entropy to detect crowd anomaly.Having performed relevant experiments on public datasets,and the experimental results show that compared with the most advanced algorithms,the proposed method in crowded scenes has a higher accuracy rate of crowd anomaly detection.

关 键 词：改进光流法 运动差值熵 自适应阈值 人群异常检测 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]