Study On Video-based Traffic Network Status Recognition

In recent years,the number of automobiles in my country has increased year by year,various traffic problems have become increasingly serious,and problems such as traffic congestion,traffic accidents and traffic pollution have become increasingly serious.In order to solve the increasingly serious traffic problems and improve people’s travel experience,it is not only necessary to strengthen the construction of road infrastructure to increase the traffic capacity of the road network,but also to use intelligent transportation technology to manage people,vehicles,roads and the environment as a whole,so as to achieve more Good traffic control,and real-time acquisition of traffic flow information is the basis for the realization of intelligent traffic systems and effective traffic control.Video has the advantages of visualization,large amount of information,wide coverage,convenient installation and maintenance,etc.How to extract traffic flow information based on video has important application value and social significance.This dissertation mainly studies the video-based information collection technology and application,focusing on video vehicle detection technology,traffic congestion status recognition and display methods.The main work of the thesis includes:(1)For video vehicle detection,a YOLOF-based single-stage feature detector network is proposed to extract traffic flow information.The existing target feature extraction network mainly uses pyramid representation to expand the receptive field of semantic features,and multi-scale feature fusion will cause a lot of resource consumption,resulting in a large delay in massive video detection.In this paper,a simplified feature encoder is used to replace the existing feature fusion module with complex structure,which improves the response time and maintains better accuracy.The model is trained using COCO2017,and compared and analyzed with other detection algorithms,and good performance results are obtained.(2)In order to identify the state of the traffic network,this section proposes a calculation index based on the traffic flow to identify the state of the traffic network,which can realize the dynamic evaluation of the traffic state of the traffic network including several junctions.Firstly,YOLOF-SORT traffic flow information extraction network is proposed based on motion tracking to extract traffic from traffic video sets and compare with other traffic statistics methods.Then,a traffic state discrimination index Kme based on vehicle occupancy and movement information is proposed,and the traffic state is calibrated by using the collected road vehicle traffic video data set,and compared with the road service level K based on traffic saturation.(3)Complete the design and implementation of the regional real-time traffic congestion status visualization application subsystem.The subsystem includes:video collection,video object detection,front-end visual display,data storage and update and maintenance modules.The real-time traffic status information of the road section can be uploaded to the front-end,and the front-end can perform color matching in different states to obtain the real-time traffic state color scale.