Research On Identification Of Typical Traffic Safety Violations Of Motor Vehicles Based On Deep Learning

The traffic management departments of the public security organs in my country are still using traditional manual supervision methods to the violations of motor vehicle traffic safety on the road,which are time-consuming and labor-intensive and prone to supervision omissions that need to be solved urgently.At present,intelligent transportation systems are widely used,and intelligent transportation systems based on artificial intelligence,machine vision and other fields are gradually replacing tedious manual operations.Based on the analysis of domestic and foreign target detection and tracking technologies,this paper proposes a machine tool for motor vehicle detection and tracking based on deep learning,and innovatively propose a deep learning-based identification model for typical motor vehicle traffic safety violations,which greatly reduces the cost of identification of violations and provides a basis for traffic management departments to enforce the law.The main research contents are as follows:Firstly,analyze the research background and significance,evaluate the research status of target detection,target tracking and illegal behavior identification technology,determine the research content and organizational structure of the paper,and make an overview of related theories.Secondly,the motor vehicle detection algorithm is designed based on the improved YOLOv5,the YOLOv5 backbone network is lightweight and improved by Ghost Net,the SE and CBAM attention mechanism modules are added to improve the detection accuracy,and the preprocessed KITTI and UA-DETRAC data sets are used for detection training.Build the YOLOv5-Ghost vehicle detection algorithm.Thirdly,the vehicle tracking algorithm is designed based on DeepSort,the target detector in the original algorithm is replaced with the YOLOv5-Ghost algorithm,the YOLOv5-DeepSort vehicle detection and tracking algorithm is constructed,and the preprocessed Veri-776 data set is used for tracking training.Realize vehicle detection and multi-target continuous tracking.Finally,based on OpenCV,the identification algorithm of typical motor vehicle traffic safety violations is designed,the YOLOv5-DeepSort motor vehicle detection and tracking algorithm is cascaded,and the identification model of typical motor vehicle traffic safety violations is constructed.Shooting video automatically recognizes motor vehicle traffic safety violations and extracts key frames for background storage,and synchronizes video for evidence collection.The constructed YOLOv5-Ghost vehicle detection algorithm has a m AP of 99.8%,an Fps of 58 f/s,and a model size of 48.4MB.Compared with the same type of YOLOv5 algorithm,it improves the detection accuracy and reduces the proportion of the model.The recognition effect of the model is verified by collecting urban road surveillance video and drone shooting video.The results of experiments show that the recognition rate of illegal behavior of motor vehicle compaction line changing lanes reaches 98.2% to 98.7% depends on the type of lines;the recognition rate of illegal U-turn of motor vehicles is 97.9%.The model can detect motor vehicles and identify typical traffic safety violations quickly and accurately,effectively alleviate the problems of fixed video forensics locations and high labor costs,and make up for the shortcomings of insufficient off-site law enforcement evidence in the field of traffic management,which provides practical management methods for supervising motor vehicle traffic safety violations to the traffic management department.