Research On Intelligent Monitoring System Of Road Network Operating State Based On Traffic Video

With the rapid development of the transportation industry and the vigorous promotion of intelligent transportation construction,the intelligent traffic monitoring system has been widely concerned by researchers at home and abroad.By using visual perception technology to extract the effective information of traffic video,the traditional method of "people looking for information" is transformed into the method of "information looking for people" to obtain the traffic information automatically,so as to give full play to the advantages of the intelligent traffic management system.This thesis designs an intelligent monitoring system which can collect the operating state of the road network from the real-time traffic video.The system can collect the information of vehicle driving and traffic flow,monitor the abnormal traffic events such as illegal parking,vehicle retrograde,illegal lane change,pedestrian intrusion and traffic congestion.Once an abnormal traffic incident occurs,the system will alarm immediately,so as to ensure the safe driving of vehicles and reduce the economic losses and casualties caused by traffic accidents.The main work of this thesis is as follows:(1)For real-time access to the location information of vehicles,pedestrians and nonmotorized vehicles on the road,this thesis compares and analyzes the performance of three object detection algorithms of Faster R-CNN?YOLO v3 and Tiny YOLO v3.On the premise of satisfying the detection accuracy of the system,Tiny YOLO v3,which has faster detection speed and lighter weight,is selected to detect the objects in the traffic video.(2)In order to verify the performance of the algorithm designed in this thesis in traffic video,after collecting the real scene monitoring videos of urban roads,highways,tunnels and other places in different periods of time,this thesis screens and labels them to establish the traffic monitoring database of the road network with more than 35000 samples.(3)In order to obtain the vehicle trajectory,the positions of the same vehicle in different video frames need to be linked to realize the vehicle tracking in the video sequence.This thesis analyzes and compares the principles and performances of three multi-object tracking algorithms based on optical flow feature,motion information and fusion of appearance feature and motion information.(4)This thesis analyzes how the monitoring system realizes various monitoring tasks in the traffic video,and on this basis,the object detection algorithm and the object tracking algorithm are combined.After comparing various algorithms,the author chooses the joint algorithm combined by the object detection algorithm and the multi-object tracking algorithm which fused with appearance feature and motion information,to achieve the acquisition of traffic information and the monitoring of traffic events.After testing,the accuracy of the joint algorithm on various tasks can reach more than 91%.On this basis,the algorithm can monitor the multi-channel traffic video in turn,so as to track the vehicles on the road when there is an interval between the video frames,and realize intelligent monitoring of the operating state of the road network.In conclusion,based on the image visual perception technology,this thesis uses the object detection and multi-object tracking technology to obtain the vehicle driving information in the traffic video and monitor the abnormal traffic events,and designs an intelligent traffic monitoring system that can monitor the operating state of the road network in real-time.44 Figures,20 Tables and 63 References.