Modeling And Research Of Vehicle Detection System Based On Traffic Video Surveillance

With the rapid development of the security industry, the video surveillance system has been widely applied in various fields. The traditional video surveillance system is inefficiency and requires a lot of manual operation. On the other hand, the intelligent video surveillance system has aroused general concern because of its high efficiency and unattended operation.The paper is focused on the intelligent traffic video surveillance system modeling. This system is capable of real-time monitoring of vehicles in the scene, and warning of vehicles entering the warning area, which greatly improve the detecting efficiency. This paper has done research and experiments on the key technologies of vehicle detecting and tracking, and also proposed some improved algorithms. The system is designed and optimized based on system design principles. Specifically, the paper has completed the following work:First, in moving target detection, the paper analyzes the optical flow, frame difference and background subtraction and other traditional methods, and then focus on the Gaussian mixture model. According to the characteristics of the application scene, a new algorithm for background modeling and updating is proposed. Experimental results show that the proposed algorithm has high efficiency. Second, in shadow filtering, this paper studies principles of the shadow and attributes in different color model. According to the differences between shadow and non-shadow regions in the outdoor single light source environment, this paper proposes an improved shadow detection algorithm based on normalized RGB color model. Third, in moving target location, this paper analyzes the camera model with nonlinear distortion and study on the classic Tsai two stage calibration method. Making full use of the existing traffic markings in the scene, we propose a fast camera calibration algorithm for the fast transformation between image coordinates geographical coordinates. Next, in vehicle recognition and tracking. According to the location information provided by the system, the paper improves vehicle identification and matching method based on regional features and realize vehicle tracking using Kalman filter. Finally, the software design has been completed and the paper analyzes the performance of the system with performance analysis tools, and uses GPU to accelerate for eliminating system bottlenecks, and makes the system meeting the real-time demand.