Abstract: In the field of intelligent video surveillance, in order to improve the efficiency of abnormal event detection and the defects of present methods in poor real-time performance and applicability, this paper proposes a real-time and high efficiency method. This method firstly extracts the global optical flow value as the movement characters, and constructs the visualizing expression of global optical flow. Then the image entropy analysis is used to obtain the statistical parameter in normal conditions. Finally, the confidence interval in normal condition and the anomaly judgment formula are given, which can be used to detect the abnormal event. The experimental results show that, for the video size of 320×240, the average detection time can be as low as 0.031 s in each frame and the accuracy can reach above 96%. As a result, the method has high efficiency and good real-time.

Key words: intelligent video surveillance, dense crowd, abnormal event detection, global optical flow image, image entropy

摘要： 在智能视频监控领域，为了提高密集人群中异常事件的检测效率，改善已有算法在实时性和适用性方面的不足，提出了一种实时高效的检测方法。该方法首先提取图像的全局光流强度作为运动特征，并构造全局光流强度的图像化表达；然后利用图像熵进行分析，获取正常状态下图像熵的统计参数；最后确定正常状态的可信区间和自适应的异常判定公式，从而判断异常事件是否发生。实验结果表明，该算法对尺寸为320×240像素的视频，平均每帧的检测时间低至0.031 s，且准确率可达96%以上，具有较高的检测效率，且实时性较好。

关键词: 智能视频监控, 密集人群, 异常事件检测, 全局光流图, 图像熵