Traffic Statistical Algorithm Research Based On Video Image

With the development of the city,the traffic flow has been increasing rapidly and the problems of traffic congestion have been very serious.Nowadays,social scientists generally believe that intelligent transportation system(ITS)is one of the effective ways to solve the problem of urban traffic congestion.In ITS,acquiring real-time traffic parameter information and distributing road resources effectively is the key to solve the congestion problem.Acquiring the traffic parameter information includes vehicle flow,vehicle speed,vehicle distance,etc.Based on the work of previous studies,this paper carries on the research on vehicle detection and statistical methods by means of image processing algorithms,mathematical morphology algorithm,and Kalman filtering algorithm.The main of research works include:This paper studied related algorithm of traffic flow statistics,and summarized three basically challenging problem in traffic flow statistics process.1)Noise points on images were increased due to environmental disturbance.2)The incompleteness of vehicle detection zone was caused by detection omission problem in stagnancy of vehicle.3)Traditional virtual wireframe detection accuracy was low and the traffic flow statistics accuracy was poor.Aiming above three problems,this paper took following studied.1)In vehicle detection,white noise points of image were more.It caused fuzziness of the detection picture and affected the vehicle detection effects.Aiming at the problem,this paper used erosion,dilation,opening/closing operation related algorithms of mathematical morphology,and selected appropriate structural elements to take erosion and filling for white noise points and eliminate the disturbance of noise points finally.2)Aiming at the detection omission and error detection problems appeared in vehicle stagnancy or slow moving,this paper introduced mixed gauss background model and proposed a background parameter update mode of improved model.According to the mean value of pixel points and real-time update model of variance,foreground region of the vehicle is judged with combination of vehicle detection results of previous frame.It solved the problem taking vehicle in stagnancy as the background.Contrast experiment indicated that the improved algorithm had improved the vehicle detection accuracy and avoided the appearance of detection omission and error detection.3)Traditional virtual wireframe algorithm has a low accuracy.In order to solve this problem,the thesis tracked and counted vehicles by Kalman filter algorithm,judged whether it's the same vehicle by matching weighted sum of different deviations according to the characteristic that testing areas of vehicles share similar deviations in center of mass,area and minimum wireframe area of outer boundary and recorded vehicle information for tracking and counting.The experimental results demonstrated that this algorithm was of higher accuracy in testing numbers of vehicle compared to virtual wireframe algorithm.