Research On Urban Road Congestion Detection Based On YOLOv7

Urban traffic is an important part of modern urban development,and with the rapid economic development in China,the number of private car ownership is increasing year by year,and traffic congestion is happening more and more frequently.Since the congestion determination method of the existing map software is based on the number of navigation software installed in the user’s vehicle or cell phone,there is a defect that the prediction result is inaccurate,and the cameras of the traffic control department cannot completely cover all the road sections,so it cannot provide comprehensive and accurate feedback of the road condition information.In response to this situation,this paper proposes a real-time congestion determination method combined with the driving recorder,which plays the advantage of the feedback of the driving recorder on the real scene of the road and makes up for the defects of the navigation software’s determination;in addition,the images recorded by the driving recorder are continuous,which can also solve the shortage of the traffic control department’s camera range that cannot cover all road condition information.The main work of this paper is as follows:(1)To make a comparative analysis for the current mainstream object detection models,summarize the object detection dataset,mainstream deep learning network models and frameworks,and make a quantitative comparison analysis of these models and frameworks,and finally select YOLOv7 as the algorithm for vehicle detection.(2)Taking the high-tech zone and the administrative center area on the north bank of the Weihe River in Baoji,Shaanxi Province as the study area,the main roads and bridges in this area are subjected to field video capture by the driving recorder,and the images in the video are intercepted as the datasets for vehicle detection and lane line detection.YOLOv7 is used to detect vehicles on congested roads and determine the number of vehicles on congested roads.Traditional and deep learning-based lane line detection methods are introduced,and Lane Net,Canny Edge Detection,Hough Transform,Inverse Perspective Mapping and other lane line detection algorithms are tested on the local dataset and Tu Simple dataset to verify the lane line extraction effect.Finally,YOLOv7 is selected as the lane line detection algorithm by combining the characteristics of the local dataset and research needs,and the images in the video are extracted as the deep learning lane line dataset,and the deep learning model is trained with YOLOv7.(3)Combine the driving recorder and GPS device to detect road congestion in real time by detecting vehicles on congested road sections while driving,and mark the congested road sections in red on the map;propose a road congestion determination method combining vehicle and lane line information,based on the average speed,assisted by the number of vehicles and lane lines detected on congested road sections,and propose the calculation formula of Congestion Index and Congestion Standard.The vector map of major roads and bridges in the study area is drawn and analyzed,and the congested road sections during the peak commuting period are analyzed in the form of a table based on the collected data from driving recorder,and the congestion is finally reflected on the map.According to the new determination method,the congestion status of bridges and roads in the road network is updated,and the updated congestion status of bridges and roads in the road network is visualized in the map.(4)According to the different congestion levels of roads and bridges in the road network,different congestion coefficients are assigned to different roads and bridges in the road network,and the congestion coefficients are fed back on the map.Based on the travel routes of Baidu map,new travel routes are proposed,and the final congestion coefficient of each route is calculated by combining the length of road sections,so that the travel route with the lowest congestion degree can be analyzed.(5)The Passenger Car Unit is selected as the measure of traffic status in the field,and eight points are selected in the road network,where the number of vehicles driving through is observed and counted in the field,based on four inlets at the intersection,southeast,northwest,and three driving directions,left-turn,straight-turn and right-turn,as the auxiliary statistics;different conversion factors are assigned to different vehicles,and the Passenger Car Unit of each inlet and each direction is calculated and summarized in The table verifies the reliability of the road congestion detection method based on the driving recorder,and provides more accurate information about the road congestion in the road network.