Prediction Of Vehicle Driving Status On Approach Of Signalized Intersection Based On Trajectory Data

In recent years,with the development and improvement of social and economic levels,the number of cars has increased rapidly,and the number of traffic accidents in urban areas has also increased,among them,the number of traffic accidents at and around signalized intersections accounts for a large proportion.Research on dangerous behaviors at signalized intersections is becoming increasingly important.The dangerous behaviors of signalized intersections mainly include the study of driver’s driving behavior based on dilemma,the impact of different types of driver’s personal characteristics on driving behavior,and the research of driver’s driving behavior and decision-making under different signal control methods.This article takes the risk caused by the dilemma of choice behavior at the signalized intersections as a starting point and re-expands the definition of the dilemma.In the past,the study of the dilemma focused on the driver’s driving behavior during the yellow light,but in the actual observation,the vehicle during the last period of green light may also fall into a dilemma.In order to fully show the influence of signal control on the dilemma of intersections and the driving behavior of vehicles at signalized intersections,this paper studies the driver ’s decision-making behavior at intersections by analyzing the driving process of vehicles at the end phase of green and yellow to improve the driving safety of vehicles at intersections.Due to the development of the intelligent city,in the research of the transportation field,there are many ways to obtain traffic flow parameters,such as the development of the earlier coil detector,infrared,and the floating car method which has appeared in recent years,video monitoring and movement sensors can obtain vehicle trajectory data,which provide basic data support in theoretical research.In this paper,the first-to-stop and the last-to-go vehicles of free driving during the last period of green light and yellow light are taken as the research object.Through the information of the vehicle trajectory,the operating characteristics of the vehicle at the approach of the signalized intersection and the changing law of the vehicle’s driving status are analyzed.Firstly,through the analysis of the vehicle trajectory information,it is learned that the vehicles driving at the approach of the signalized intersections have a great randomness during the signal transition period.With the change of countdown time and space of the vehicles,the driving status of the vehicle may transform,and under the different countdown time of the signal,there are significant differences in the speed,acceleration,and spatial position of the first-to-stop and the last-to-go vehicles at the approach.Secondly,based on the definition of the dilemma,according to the real-time trajectory information of the vehicle,the vehicle trajectory characteristics under different signal lights are extracted,and the driver’s decision behavior model during the signal transition period is established,and the driver’s decision behavior model under different signal light countdown conditions is obtained.Finally,to predict the first-to-stop and the last-to-go vehicles trajectory information.In this paper,the HMM widely used by scholars in the past and the recurrent neural network(RNN)emerging in recent years are selected to identify and predict vehicle driving status.The vehicle trajectory is modeled by the continuous hidden Markov model to obtain the hidden status of the trajectory data,and the status transition matrix at the adjacent time is calculated to predict the driving status of the vehicle at the next moment.Through the LSTM network,a variant of the recurrent neural network,the time series prediction model and the driving status classification of vehicles are established to predict the vehicle trajectory characteristics at multiple times in the future,including the speed and acceleration of the vehicle at different signal countdown times,and the distance to the stop line,the prediction effect of the LSTM model is significantly better than the hidden Markov model,and the accuracy of the time series prediction and classification model reaches more than 95%.The difference between this article and previous research is that it focuses more on the vehicle trajectory pattern of signalized intersections,which not only predicts the position of the vehicle at the next moment,but also predicts the status information of the vehicle,including vehicle’s speed,acceleration,etc.,which can provide real-time vehicle information It is more conducive to intelligent traffic management and control,and also provides theoretical support for intelligent vehicle-road collaboration systems.