Research On Fine-grained Recognition And Risk Assessment Of Abnormal Driving Behavior Based On Multi-source Data

With the rapid growth of car ownership and the increasing number of drivers,the road traffic safety situation in China has become more and more severe.Statistics show that more than 90% of traffic accidents are directly or indirectly related to drivers.Therefore,in order to improve road traffic safety and reduce the occurrence of road traffic accidents,it’s necessary to identify and evaluate the driving behavior of drivers.On the basis of reviewing the literature on driving behavior at home and abroad,in order to solve the deficiencies of existing research,this paper proposes an abnormal driving behavior identification method based on multi-source natural driving data,and constructs the corresponding driving risk.The evaluation system enables the model to identify fine-grained abnormal driving behaviors,and generate drivers’ real-time driving risks as well as cumulative driving risks based on the frequency,duration and degree of danger of the abnormal driving behaviors.The main research contents are as follows:First of all,based on mobile phone sensor data,this article chooses a variety of traditional machine learning algorithms and deep learning algorithms to identify the driver’s sudden acceleration,sudden braking and sudden steering behavior,and compares the results of the model evaluation.The evaluation results show that the time series model in deep learning has the best recognition effect,and can reach an accuracy and recall rate of more than 75%.Next,based on the video data and GPS data of forward driving records,this paper uses video recognition algorithms to detect lane lines and calculate the distance between the vehicle’s central axis and the center line of the lane.The driver’s sudden lane change and dangerous overtaking behavior are then identified.The evaluation results show that the method has a good recognition effect and can reach an F1-score value above 0.8;the YOLOv5 object detection algorithm and the single camera ranging algorithm are used to calculate the distance of the vehicle in front,combined with real-time GPS speed data to identify the driver’s behavior of following the car too close;Through GPS speed data and speed limit information obtained from Open Steet Map,the driver’s overspeeding behavior can be identified.Finally,based on the recognition results of various abnormal driving behaviors,this paper constructs a risk assessment index system for abnormal driving behaviors,and further proposes real-time risk and cumulative risk assessment methods for abnormal driving behaviors of drivers.The experimental results show that these two methods can effectively evaluate the realtime risk and cumulative risk of drivers in different driving styles and different road types,and achieve the expected results.In summary,this article fully explores the entire research process from the recognition of abnormal driving behavior to risk assessment.Through the combination of multi-source natural driving data and multiple recognition algorithms,the fine-grained recognition of abnormal driving behavior is realized,and finally the assessment of the driver’s abnormal driving risk has been completed.The results can serve scenarios such as real-time reminder and guidance of abnormal driving behaviors of drivers,real-time risk control of the future urban traffic brain,fleet management,driver qualification review,targeted driving training,and differentiated insurance pricing.