Research On Two-way Traffic Flow Statistics Algorithm In Highway Section Scenari

In recent years,with the continuous increase of traffic demand,road traffic problems have become increasingly serious.Efficient and accurate traffic statistics can provide traffic departments and drivers with real-time traffic information,which is of great significance for alleviating and unblocking traffic congestion and improving road utilization.At present,videobased traffic flow statistics methods have obvious advantages over traditional methods,and have broad application prospects.In order to efficiently obtain traffic flow information from videos,this dissertation proposes a traffic flow statistics algorithm based on improved YOLOv5 and ByteTrack.The detailed work is as follows:(1)Improvement of vehicle detection algorithm.Aiming at the problem that the YOLOv5 algorithm has poor recognition effect on small-scale vehicles at the far end of the video in complex traffic scenes,an improved YOLOv5 algorithm was proposed.Improvements are mainly divided into three aspects.Firstly,a long-distance dependent attention mechanism was introduced in the backbone network to capture effective context information,enhanced the feature extraction ability of the model and weakened noise interference.Secondly,during the training process,the α-CIo U loss function was used to replace the CIo U loss function as the localization loss to improve the bounding box regression accuracy.Finally,the Mixup data enhancement method was used in the data preprocessing stage to increase the diversity of sample data sets and improve the robustness and generalization ability of the model.The experimental results show that the improved YOLOv5 algorithm has an average accuracy of 79.2% on the UA-DETRAC test set,which is 3.8 percentage points higher than the original YOLOv5 algorithm,and the detection speed reaches 93 FPS,which shows the effectiveness of the improved YOLOv5 algorithm.(2)Implemented a vehicle tracking algorithm based on improved YOLOv5 and ByteTrack.The ByteTrack algorithm uses the Byte data association method to mine real vehicle targets in the low-confidence detection frame.In the traffic rush scene,it can effectively reduce vehicle ID recognition errors and switching.However,the original detector of the ByteTrack tracking algorithm is YOLOX,and the parameters and weight of the YOLOX model are too large,which is not conducive to deployment on edge devices.In this dissertation,the improved YOLOv5 algorithm was used to replace the YOLOX algorithm for vehicle detection.Experimental results show that this algorithm has fewer identity switching times than SORT and Deep SORT algorithms,and it can track vehicles stably and continuously.(3)A virtual detection frame counting algorithm was proposed.The virtual detection line algorithm is prone to the problem of missing vehicles in vehicle counting.In this dissertation,a virtual detection frame with a certain width was set in the center of the video frame to increase the vehicle detection range and reduce statistical errors,thereby improving the accuracy of traffic flow statistics.Finally,combined with the improved ByteTrack algorithm and the virtual detection frame algorithm,the traffic flow statistics were carried out.Tests were carried out under two different traffic scenarios: flat peak and peak traffic.Compared with the original ByteTrack algorithm,the improved ByteTrack algorithm improved the accuracy by 2.2 percentage points and 1.5 percentage points respectively in the flat peak and peak traffic scenarios,demonstrating the effectiveness of the proposed algorithm for traffic flow statistics.