Intersection Traffic Conflicts Based On Video Image Processing, Automatic Detection Technology

With the people's growing emphasis on the traffic safety in recent years, the traffic safety assessment technique which bases on traffic conflict is attracting more and more researcher's attention, and is applied to the actual road works. But, in the practical application of traffic conflict technique, we still follow the traditional method of manual observation recognition. This method is very vulnerable to human factors, and the conflict information is not precise enough. As the development of the intelligent transportation techniques, the automation detection technology based on video images processing become new direction of the alternative to the manual detection.This paper firstly introduces this subject's background and the status in the aspect of conflict theory, conflict detection method and conflict recognition. After the analysis of related relevant literatures, founding that traffic conflicts automation detection based on video image processing is not sufficient, the existing researches mainly have following problems:unrealistic detected scene, not-accurate tracking results, single-parameter-based conflict identification. This paper uses video image processing to research traffic conflicts detecting, and puts the research keys on the vehicle detecting, vehicle tracking and conflict identification. The main work of this paper and the results are as follows:1)Aimed at the situation of the traditional Gaussian mixture background model easily susceptible to light suddenly change, a new background algorithm of combining the Gaussian mixture model and image edge information successfully solve the above mentioned problem. And, for the new problem caused by the improved Algorithm, this paper uses dilation, erosion and contour detection to solve them.2) For the complex situation of traffic flow within the intersection and no Gaussion, linear constraint of particle filter, in this paper, a new vehicle tracking algorithm based on particle filter making the vehicle foreground area feature in the binary image and the histogram in the HSV color image as particle weights. Compared with the traditional particle filter, the new algorithm improves tracking results. Through the above research, enhance the robustness of vehicle detection and gain more accurate vehicle information, and lay the foundation for traffic conflict identifying. 3) Taking account for the actual situation of traffic conflict detection in the intersection, this paper introduces the concept of field of vehicle conflict, and according to this, establishes a new conflict identification model based on the size and speed of the vehicle and uses simulation to confirm it.