Study And Application Of Real-time Vehicle Monitoring At The Traffic Crossing

Vehicle Monitoring is a very important part in the Intelligent Transportation Systems which performs Real-time monitoring of intersection traffic condition, the dynamic traffic incident detection and traffic parameter extraction. Nowadays, it is very difficult during the process of vehicle monitoring to construct general-purpose background in a variety of environment, adapt the changes in the weather and shadow interference and perform vehicle identification. To solve these problems, this paper undertake a study around the vehicle detection and tracking in city traffic scenes, mainly concluding background modeling method, vehicle detection and identification, shadow removal and vehicle tracking. Then we apply these methods into snap system of red light runners. This paper includes the following four areas of content:(1)Secondary selected stratege based background modeling methodsAddressing difficulties of obtaining the background image during traffic jam, we construct background modeling method based on secondary selected stratege. We perform first selection of frames according to the degree of traffic congestion which is measured by the variance value of frames, and then perform secondary pixel-wise selection by frame difference method to exclude the sporting pixels of the candidate frames, finally utilize the medium method to estimate the background value of each pixel. This method could obtain a better background image in a short period of time even in traffic jam.(2)Vehicle detection and identificationAddressing the shadow interference and the geometry deformation of projection mapping, we propose the vehicle detection and identification methods at the complex traffic scenes. We utilize adaptive threshold to detect forground, contruct illumination model of vehicle to remove the shadow of foreground in RGB color space, eliminate the effection of geometry deformation by inverse projective mapping at the moment vehicle is driving to the detection region of manual separation, classify the vehicles into three categories: large, mid-size and small car. The detection methods can rapidly make judgement about the vehicle arrival and accurately identificate the categories of vehicle.(3)Mean shift based vehicle tracking technologyOn the basis of mean shift algorithm, we eliminate the impact of background pixels on the the color probability density of target vehicles combining with the method of background subtraction, which can speed up the iteration speed and improve the iteration precision. Firstly, we construct tracking objects of vehicles as a prerequisite during the detection period, then utilize improved mean shift algorithm to track. Finally, we predict vehicle trajectories by building the constant acceleration motion model of revised steering angle. This step realizes the real-time tracking of vehicles in complicated situation in a high accuracy.(4)Realization of monitoring system of vedio red light runnerWe design the hardware program which performs vehicles recorder system of red light runner based on tracking and apply it into realistic traffic intersection. The system can judge and record the action of red light runner according to the lane of the target vehicle and the traffic lights of the current lane.