Research On Abnormal Behavior Detection Of Surveillance Video And Its Software Implementation

Intelligent video surveillance as the mainstream of security measure, through the cameras on all kinds of public places such as transportation areas and residential areas. In recent years, due to the lower hardware cost, the use of surveillance camera also increased gradually. Information also appear to come to the geometric model of growth. Under the condition of limited resources, how to detect abnormal behaviors in public places quickly and accurately is particularly important. Therefore, intelligent video surveillance technology got the attention of scholars from all sectors of society.Nowadays, most of the monitoring systems using manual basis, which requires the staffs closely monitor the equipment. But in this case, it will cause fatigue and tension easily. Especially in the workplace, single staff has to observe multiple scenes. Tend to be too many things to see, causing negligence easily and affect the monitoring results. How to obtain useful information in massive video is the focus of the security field.Although there are many scholars participated in this field, they still can’t meet the actual requirements. It requires monitoring system to determine the abnormal behaviors independently. But in the early years of the study, it need to predefine behaviors and couldn’t make its own understanding and judgment of the abnormal behavior. According to the above problem, this paper made the following work: 1. Aim at the problem of determining the abnormal behavior adaptively, it puts forward a kind of unsupervised method of judgment, this method has certain adaptability, it does not need to predefine behaviors. Determine whether there is abnormal behavior through the trajectory of moving targets in the public environment, especially the complex action appears to be more effective and flexible. 2. In order to solve the problem in the process of extracted initial foreground area which separate into a large number of fragments, this paper adopts the method of polygon fitting. According to merge them, it reduces the complexity of the outline which belongs to the same prospect area of fragments. In this way, it will be able to express the behavior of the target area effectively. 3. Preliminary moving object trajectory contains a lot of noise points, and the mixed and disorderly points inconsistent with the actual center position, which impact on subsequent trajectory clustering. This paper takes line fitting of the moving target trajectory to eliminate this kind of interference, and take line’s endpoint of the paragraph instead of the original trajectory. It can greatly reduce the computational complexity of trajectory clustering, and improve the processing speed.