Research On Video-based Traffic Parameters Detecting System

The extraction of traffic parameters plays a key role in urban transportation system , video detection is an important way of getting traffic parameters. There are several different types of detector including traditional magnetic loop, ultrasonic detector, infrared detector and so on, and compared with them, video detector utilizes technology such as image processing, computer vision, pattern recognition and has the advantage of low price, getting much more traffic parameters, detecting wide regions. The research on video detection in China is carried out gradually, and such research has attractive application.By analyzing and comparing detection based on virtual region and detection based on moving object detection and tracking, this paper adopts the latter.The main research work in this paper is listed as follows:①Through comparing several ways of moving region detection, this paper adopts moving region detection based on 3D OGHM which has good anti noise performance, can detects much more feature information such as vehicle texture. Aiming at the problem that frame difference has bad anti noise performance and the moving object obtained has serious problem of voids, this paper proposes a algorithm of 3D OGHM moving object detection based on Level Set segmentation. This algorithm uses a kind of improved Level Set algorithm to segment image processed by 3D OGHM. The results show that the algorithm of detecting moving objects proposed in this paper has strong anti-interference ability, can get more accurate and complete moving objects.②According to analysis on moving objects tracking, this paper adopts bidirectional matching to track moving objects. This paper proposed a system take moving objects detecting and tracking as core can be used to obtain traffic parameters. This system uses timing global detection, and local detecting between twice global detection and adopts bidirectional matching method to track moving objects. This system is applicable to multiple objects in fixed scene. Though testing the algorithms in this paper are proved to be effective.