Recognition Of Vehicle Lane Changing Behavior Based On Driving Recorder

Illegal lane changes mainly include behaviors such as changing lanes with solid lines and changing lanes without turning lights on dotted lines.On the one hand,changing lanes in violation of regulations leads to traffic congestion and reduced road capacity;on the other hand,it may also cause major traffic accidents,leading to serious consequences such as car accidents and casualties.In response to vehicle lane changing behavior,the traditional solutions mainly include on-site law enforcement,physical contact detection,and manual video capture,but there are problems such as low efficiency,high cost,and susceptibility to interference.Violation detection methods based on surveillance video and driving recorders are fully automatic,efficient,and all-weather.Deep learning promotes the development of target detection,behavior recognition and other technologies,making video-based automatic lane change detection methods widely available application.However,the fixed surveillance video has a small coverage area and poor operability,while the driving recorder has the characteristics of mobility,wide coverage,and low cost.Therefore,this article uses the driving recorder as the video source to detect the violation of the vehicle when changing lanes with solid lines and changing lanes without turning lights,which is of great significance to standardizing bad driving behavior and reducing the occurrence of traffic accidents.The main research work of this article is summarized as follows:(1)Designed a method for judging the vehicle lane change geometry.Analyze the characteristics of the video image of the driving recorder,and convert the lanechanging process of the vehicle into the change of the geometric position relationship between the lane and the vehicle,so as to determine the lane-change behavior and direction of the vehicle.(2)Realize the acquisition of vehicle driving status information.Including lane line extraction,vehicle detection and tracking,vehicle turn signal recognition,license plate recognition.Based on the U-Net model,the lane lines are divided to obtain dashed and solid lines,and double solid lines,dashed solid lines,and solid dashed lines are obtained through a combined algorithm;vehicle detection and tracking according to YOLOv3(You Only Look Once vision3,YOLOv3)and Deep SORT(Deep Simple Online and Realtime Tracking,Deep SORT);YOLOv3-tiny model is used to identify left and right turns Turn signal;license plate detection based on YOLOv3-tiny model,and identification of illegal license plate numbers based on the trained character model.(3)Two automatic identification methods for illegal lane change behaviors,including solid line lane change and dashed lane change without lighting,are constructed,and the license plate number of the offending vehicle and the location of the offending vehicle are obtained,and the offending vehicle is automatically captured.A prototype system for identifying violations of lane changes has been developed,which can show the specific process of vehicle lane change identification and save vehicle violation information.