Complete Extraction Of Various Vehicle Trajectories In Night City Traffic Monitoring

With the development of national economy,citizens' living quality is improved evidently.As the sharp rise in the number of private cars in urban transportation system,the situation of road traffic safety has been becoming more serious.Because of the constantly increasing of the urban traffic accident rate at nighttime,it has became especially important to get the information of the nighttime vehicles accurately.By the connection between the location information from the vehicles'trajectories and the information of vehicles and their owners provided by the vehicle license plate recognition system,the surveillance data is transmitted exactly between two adjacent monitoring points.Finally the data network of urban transportation monitoring system can be achieved.It has great social significance,such as reducing the accident rate,regulating traffic flow,optimizing driving directions and investigating wreckers' responsibility.The presented tracking system has combined the vehicle's type and trajectory feedback correction in order to obtain the complete trajectory in nighttime traffic surveillance system.After filtering the restored headlight in the distant,middle and close regions respectively,the initial trajectories of headlights are obtained by area overlapping and trajectory feedback is then applied for correcting mismatching and fitting incomplete trajectories.Vehicles are classified according to the actual vehicle's width calculated by homography matrix and the number of headlights.Based on the temporal and spatial similarity,headlights are paired.The obtained trajectories of vehicles are finally corrected and optimized.The complete extraction of vehicles'trajectories between two adjacent monitoring points are achieved.The stable features such as the actual width of vehicles,displacement and velocity can be computed by building the transformation relationship between the image coordinate system and the world coordinate system based on homography matrix,in order to reduce the disturbance from perspective transformation and light noise on the headlights' pairing.Other noises can be deleted by the filtering according to different region and stability evaluation,and then feedback correction is used to solve the problem of objects'disconnected trajectories which is caused by the conglutination and occlusion in the process of headlights' pairing.The challenge of chaotic and incomplete trajectories of the vehicles can be settled by classifying vehicles according to the width of vehicles and the number of headlights.12 videos including cars,trucks,buses,motorbikes,cars with a single headlight,cars with two different power headlights and cars with four headlights are processed by our presented system.Compared with the existing advanced algorithms,experimental results show that the headlights' correct matching rate and vehicles'tracking rate are improved,and vehicles'long tracking trajectories can be extracted completely between the two adjacent monitoring points by the presented tracking system.