计算机应用 ›› 2016, Vol. 36 ›› Issue (6): 1719-1723.DOI: 10.11772/j.issn.1001-9081.2016.06.1719

• 虚拟现实与数字媒体 • 上一篇    下一篇

适用于密集人群的异常事件实时检测方法

潘磊1,2, 周欢2, 王明辉2   

  1. 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
  • 收稿日期:2015-12-17 修回日期:2016-03-08 出版日期:2016-06-10 发布日期:2016-06-08
  • 通讯作者: 王明辉
  • 作者简介:潘磊(1982-),男,四川彭州人,讲师,博士研究生,CCF会员,主要研究方向:图像处理、模式识别;周欢(1990-),男,江西吉安人,博士研究生,主要研究方向:图像处理、计算机视觉;王明辉(1971-),男,山东青岛人,教授,博士生导师,博士,主要研究方向:高维复杂信息智能处理、图像处理、模式识别、多源信息融合。
  • 基金资助:
    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)。

Real-time detection method of abnormal event in crowds

PAN Lei1,2, ZHOU Huan2, WANG Minghui2   

  1. 1. College of Computer Science and Technology, Civil Aviation Flight University of China, Guanghan Sichuan 618307, China;
    2. College of Computer Science, Sichuan University, Chengdu Sichuan 610065, China
  • Received:2015-12-17 Revised:2016-03-08 Online:2016-06-10 Published:2016-06-08
  • Supported by:
    This work is partially supported by the National Natural Science Foundation of China (61071162), the Application and Development Technology Program of the Civil Aviation Administration of China (MHRD20140212), the Scientific Research Program of the Education Department of Sichuan Province (16ZB0032), the General Funds of the Civil Aviation Flight University of China (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。

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

Abstract: In the field of dense crowd scene, in order to improve the defects of present anomaly detection methods in real-time 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.

Key words: intelligent video surveillance, abnormal event detection, optical flow method, Kalman filtering, residual analysis

中图分类号: