适用于密集人群的异常事件实时检测方法  

Real-time detection method of abnormal event in crowds

作　　者：潘磊[1,2] 周欢[2] 王明辉[2] 

机构地区：[1]中国民用航空飞行学院计算机学院,四川广汉618307 [2]四川大学计算机学院,成都610065

出　　处：《计算机应用》2016年第6期1719-1723,共5页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61071162);中国民航总局应用开发科技项目(MHRD20140212);四川省教育厅科研项目(16ZB0032);中国民用航空飞行学院面上基金资助项目(J2012-40)~~

摘　　要：在密集人群场景下,针对现有异常检测算法在实时性和适用性方面的不足,提出了一种基于光流特征和卡尔曼滤波的实时检测方法.该方法首先提取图像的全局光流强度作为运动特征;然后对全局光流值进行卡尔曼滤波,并对残差进行分析;假设残差在正常状态下服从高斯分布,利用假设检验加以验证;运用最大似然(ML)估计得到残差的概率分布;在一定置信度下,确定正常状态的可信区间和异常状态的判定公式,并以此判断异常事件是否发生.实验结果表明,该方法对尺寸为320×240的视频,平均检测时间低至0.023 s/frame,且准确率可达95%以上.因而,该方法在保证较高检测率的同时,还具有良好的实时性.In the field of dense crowd scene, in order to improve the defects of present anomaly detection methods in realtime performance and applicability, a real-time method was proposed based on the optical flow feature and Kalman filtering.Firstly, the global optical flow value was extracted as the movement feature. Then the Kalman filtering was used to process the global optical flow value. The residual was analyzed based on the assumption that the residual obeyed a Gauss distribution in normal condition which was validated by the hypothesis testing. Then the parameter of the residual probability distribution was calculated through the Maximum Likelihood（ ML） estimation. Finally, under a certain confidence coefficient level, the confidence interval of normal condition and the judgment formula of abnormal condition were obtained, which could be used to detect the abnormal events. The experimental result shows that, for the videos with the size of 320 × 240, the average detection time of the proposed method can be as low as 0. 023 s / frame and the accuracy can reach above 95%. As a result,the proposed method has high detection efficiency and good real-time performance.

关 键 词：智能视频监控 异常事件检测 光流法 卡尔曼滤波 残差分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]