Characteristic Analysis And Modeling Of Dangerous Driving Behavior At The Entrance And Exit Of Urban Expressway

Urban expressways as the main skeleton of urban roads with heavy traffic and high speed features.And the entrances and exits of expressway as an important urban road network facilities to achieve steering,complex traffic environment and frequent lane change behavior within its scope,it is easy to cause the driver to make operation that cause traffic conflicts and even traffic accidents,and once a traffic accident occurs in the entrance and exit of the expressway can easily cause serious accidents and traffic congestion.Based on this,This paper aims to reveal the formation mechanism of dangerous driving behaviors and traffic conflicts,and explores traffic conflicts as metrics to measure dangerous driving behavior to reveal the driving behavior characteristic factors that affect traffic conflicts,so that the identification of dangerous driving behaviors at urban freeway entrances and exits model will be established.The study will provide timely and reliable traffic safety warning information for policy makers,managers and travelers,which is of great significance to improving the service level of urban expressway entrances and exits and the level of active safety management.Using drone aerial video,this paper analyzes the types of dangerous driving behaviors at the entrances and exits of urban expressways and studies the formation process and types of traffic conflicts,using traffic conflicts as an entry point,total 17 driving behavior characteristics including the characteristic of speed,acceleration,lateral acceleration,lane-changing position,head spacing and offsets the lane centerline with and without traffic conflicts in the first half of the process of lane change at the entrance and exit of urban expressways are extracted and analyzed respectively.Significance analysis is carried out to obtain the characteristic parameters of dangerous driving behavior.The results show that,under the conflict and non-conflict situations,most lane changing driving behavior indicators have the significant difference.The severity of traffic conflict is used as a measure of dangerous driving behavior,and three risk levels are determined by this index: no risk of conflict,general risk of conflict and severe risk of conflict;The multinomial logit model is used to construct risk of dangerous driving behavioral model based on traffic conflict,this model takes into account the differences in various driving behavior characterization parameters of the vehicle under different lane changing behaviors,the interaction terms between the lane change type and other driving behavior characterization parameters are constructed and integrated into the model to analysis the driving behavior factors that affect the severity of traffic conflicts.The results show that: the behavior characteristic including the position of longitudinal lane changing,the position lateral lane changing,duration time,speed at starting point,maximum speed,maximum longitudinal acceleration,lateral acceleration at starting point,distance of offset centerline,distance between the behind vehicles,and distance between the head vehicles,which under different types of lane changing have a significant impact on the marginal effect of each types of conflict risk.To compare the effect of different characteristic parameters of dangerous driving behavior on the recognition of dangerous driving behaviors,four different characteristic parameters of dangerous driving behaviors are constructed,namely full factor characteristic parameters,significant factor characteristic parameters,partial factor characteristic parameters and principal component characteristic parameters,as input dataset.The conflict risk level index is used to divide the driving behavior into no conflict behavior,general conflict behavior and severe conflict behavior as output dataset.Methods such as SVM and BP neural network are used to identify and classify dangerous driving behaviors at entrances and exits of urban expressways.The model is evaluated with accuracy,recall,F value and ROC curve.The final result shows that the classification performance of the model which uses significant factor characteristic parameters including interaction terms that have a significant impact on the severity of traffic conflicts as characteristic parameters of dangerous driving behavior for general conflict and severe conflict categories is better than that of models using other dangerous driving behavior characteristic parameters,and the classification performance by SVM for general conflict category and severe conflict category is better than neural network.