视频中异常行为自动检测技术研究

Research on Automatic Detection Technology of Abnormal Behavior in Video

为提升计算机对视频中异常行为的识别、判断能力,提供及时发现异常行为并阻止其造成更大损失的可能性,本文提出2种基于视觉低层特征设计的异常行为检测技术,利用混合高斯模型和区域像素灰度值判断运动目标是否进入危险高度,进而达到对攀高的检测;利用光流—聚类法和目标跟踪法实现初步、精确判断是否存在逆行行为。攀高实验中选取2个不同场景下的视频,逆行实验选取室外复杂环境中拍摄的视频进行检测,检测结果显示:攀高检测中,无误检;逆行检测中相较于传统光流法,误检率明显降低。

In order to improve the computer’s ability to recognize and judge abnormal behaviors in video, two abnormal behavior detection technologies based on low-level visual features are proposed to detect ascending and retrograde behaviors in abnormal behaviors respectively, so as to make it possible to detect abnormal behaviors in time and prevent them from causing greater losses. The climbing detection technology uses the mixed Gaussian model and the gray value of regional pixels to judge whether the moving target enters the dangerous height, so as to achieve the purpose of climbing detection. Retrograde detection technology uses optical flow-clustering method and target tracking method to determine whether retrograde behavior exists. In the climbing experiment, two videos in different scenes were selected for detection and no false detection were reported. The retrograde experiment selects the video shot in the complex outdoor environment for detection. Compared with the traditional optical flow method, the false detection rate is significantly reduced.