监控视频中异常事件检测方法研究

发布时间：2018-04-08 18:51

本文选题：智能监控　切入点：异常事件　出处：《华中科技大学》2007年硕士论文

【摘要】： 目前,大多数视频监视系统无法在无人值守情况下自动协助保安人员及时发现可疑或者异常的事件,被记录下来的视频一般被用作事后的证据。解决这个问题的方法之一就是采用智能视频监控系统,它从这些大量视频数据中检测出含有可疑或者异常的行为,并及时地将这些行为报告给安全人员。异常事件检测的实验程序正是基于这一目标而设计的。实验程序实现对人的徘徊事件与群殴事件这两类异常事件的检测。在银行、停车场等特殊场所,人的徘徊在一定程度上可以被认为可疑的事件,这类事件主要表现为目标的运动方向在一定的时间内发生多次变化。因此,在获得运动目标轨迹的基础上,通过分析其方向特征就可以确定人的徘徊是否属于可疑事件。由于群殴事件与其它事件在视觉上具有一定的可分性,理论上可以在非压缩视频中提取出能够区分群殴事件与其它事件的运动特征。实验程序中取每1秒内的帧序列为一个小片段,对这些小片段依次提取运动强度、运动方向直方图、行程长度、运动强度比例4类运动特征值,从这些片段的特征值中找出能够区分群殴事件与其它事件的阈值,根据特征阈值即可识别出监控视频中是否含有群殴事件。实验研究发现,基于运动轨迹分析的徘徊事件检测方法具有实用性,其中,实现的基于感兴趣区域的运动分割方法可以较好地提取出前景中的运动目标。基于运动活动性的方法可以适用于场景复杂的情况下对群殴事件的检测,并具有较高的查全率和查准率。
[Abstract]:At present, most of the video surveillance system cannot unattended automatically assist the security personnel to detect suspicious or abnormal events, the recorded video data are used after evidence. One way to solve this problem is to use intelligent video surveillance system, it is from these massive video data can detect abnormal behavior, and timely report to the security personnel. The experimental procedure to detect abnormal events is designed based on this goal.
Experimental procedures to detect people wandering events and events of these two types of abnormal events. In the bank, a special place for parking lot, people wandering in a certain extent can be considered suspicious events, this event mainly direction of movement target vary within a certain period of time. Therefore, in based on the moving target, through the analysis of the characteristics of direction can be determined whether it belongs to the people around the suspicious events.
The brawl and other events have certain separability in vision, the theory can extract motion features in non compression can distinguish the brawl with other events in the video. The experimental procedure in sequence frames per seconds for a small fragment of these small fragments according to the extraction of exercise intensity, motion direction histogram, length of stroke, characteristic of the motion of 4 kinds of motion intensity ratio, to distinguish between events and find other events from the threshold characteristics of these fragments of value, the threshold value can be identified according to the characteristics of monitoring whether the video contains a free event.
Experiment results show that the detection method of trajectory analysis of wandering events based on the practical, the implementation method of ROI based motion segmentation can effectively extract moving objects in the foreground. Based on the method of motion activity detection can be used on brawl in the scene of complex situations, and has high the recall and precision.

【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2007
【分类号】：TP391.41

【引证文献】