Research On The Key Technology Of Intelligent Traffic Monitoring System Based On Vision

The intelligent traffic monitoring system based on vision has important function to maintain the traffic order, improve the driver's safety awareness, protect people life and property safety. This paper studies the key technology of intelligent traffic monitoring system based on vision.First of all, for the fluctuant interference problems such of the branches swing, this paper propose a moving target detection algorithm based on spatio-temporal information after the analysis of the typical wave jamming time characteristics and spatial characteristics.First,use the average background model to obtain the initial foreground image, and then use the fluctuant interference cycle shown in time, a small range of characteristics and space, build a model for the interference pixels, this model for further judgment of the foreground pixels, interference of prospects.Finally for the problem that the foreground contour information is not obvious, and the big empty area,we using the initial foreground image to fix it. Compared with the the existing algorithm, the algorithm this paper proposed has better robustness for fluctuant interference problems, and has a small amount of calculation.Compared with the GMM algorithm, calculation time of this algorithm increased by 16.8%, and the algorithom meet the real-time requirements of traffic scene.Secondly, the current image distance measurement algorithms usually require prior measurement of multiple physical, or complex calibration which is inconvenient to operate.And the security problem caused by the complex measurement in the traffic scene is great Aiming at this problem, this article we propose an image distance measurement algorithm based on imaging principle and data regression modeling.We do the measurement from vertical and broadwise. The camera imaging principle formula of longitudinal distance, and then combined with the characteristics of wide lane, using the method of data regression model, calculate the gain function and the transverse distance to calculate the longitudinal distance. The algorithm of horizontal and vertical measurement errors are within 10.1% and 14.5%, and only need to measure camera installation height and lane width and to measure the length of the distance to the scene, the measurement process is convenient and quick.Thirdly, a target tracking algorithm based on corner point and trajectory prediction is proposed. Firstly, the target vehicle is extracted from the corner point, and the optical flow method is used to track the corner points. For the corner point of drift, the corner classification algorithm based on Euclidean distance and angle similarity is proposed. Compared with the corner classification method in TLD, the proposed method can preserve more effective corner points and more robust. In the case of occlusion, a trajectory prediction algorithm based on vehicle foreground feedback is proposed, which can reflect the real position of the vehicle, and feedback and correct the results. Experiments show that this algorithm has good robustness and real-time performance in the target tracking of complex traffic conditions such as target connection and occlusion. Compared with the Camshift algorithm and particle filter algorithm, the tracking accuracy of the proposed algorithm is improved by 14.7% and 30.3% respectively, compared with the Camshift algorithm, the real-time performance is improved by 36.9%.In the end, aiming at the problem that the vehicle rolling line detection algorithm is not robust to the disturbance such as residual shadow and vehicle body cover, a rolling line detection algorithm based on vehicle centroid is proposed. Firstly, the foreground of the vehicle is obtained by moving target detection, and shadow removal algorithm is used to get the accurate vehicle foreground area. Then the vehicle rolling line is judged by the distance between the vehicle and the outside of the vehicle. For rolling line vehicles, further judgement is made by the distance of the vehicle and the traffic line. On the basis of this algorithm, this paper designs the software flow of the rolling line supervision system, and uses the system to carry out a large number of experiments,the result shows that this system has a strong robustness to the interference of residual shadow and vehicle body cover, and the precision of rolling line detection is more than 93.5%. Compared with traditional algorithm,the precision of the algorithm is improved by 21.4%.