Design Of Traffic Flow Monitoring System Based On Road Video

With the continuous growth of the number of motor vehicles in the current society,traffic problems such as road congestion are becoming more and more serious.As one of the solutions to this problem,the intelligent transportation system has attracted extensive attention from researchers in various countries.Real-time traffic flow data is indispensable for intelligent transportation systems.Among all traffic flow monitoring solutions,video-based traffic flow monitoring solutions have their unique advantages.Based on the above information considerations,designs a traffic flow monitoring system based on road video,the main contributions includes the following contents:(1)Regarding the problem of vehicle target detection,compares various current solutions and selects the target detection algorithm Retina Net based on deep learning as the solution to this part of the problem.The UA-DETRAC data set is selected as the training data to train the vehicle target detection algorithm.By comparing with similar algorithms,it proves the advantages of this detection algorithm.(2)Regarding the problem of vehicle number statistics,the multi-target tracking algorithm Deep SORT is adopted as the solution to the problem.Use the VeRi vehicle re-identification data set to train and test the deep feature extraction network.Analyzes the shortcomings of the deep feature extraction network and makes related improvements.Without significantly changing the calculation time of the algorithm,the number of algorithm parameters is greatly reduced,the model volume is compressed,and accuracy of the algorithm is improved on the VeRi data set.(3)Modify the tracking trajectory number generation mechanism in the DeepSORT algorithm to implement the vehicle number statistics function.This solution extends the statistical area of the number of vehicles to the entire camera field of view,which has certain advantages.(4)Aiming at the possible statistical deviation of the number of vehicle statistical algorithms in some situations,relevant improvement measures are proposed and verified.