Research On Spatiotemporal Transmission Of Traffic Conflict Risk At Urban Road Intersections

At intersections,a diverse range of road users converge and share road space,which can easily lead to traffic conflicts and a high accident rate.Currently,urban traffic management departments often conduct post-event evaluations on accident data after traffic accidents occur.However,this accident handling mode has several disadvantages,such as difficult data acquisition,strong randomness,and lack of management initiative,making it challenging to improve management efficiency and advance accident prevention.To address this issue,traffic conflict technology is considered a safety analysis and evaluation method that can replace accident data.By collecting traffic conflict at intersections,safety assessments of intersections can be realized.The trajectory data of road users record their spatio-temporal activity,and contain important information that can be used to identify traffic conflicts.The risk of traffic conflict exists in the interaction of road users’ trajectories,which is essentially a type of spatiotemporal long series data,complex and variable,but its evolution law can be followed.The Transformer model is an encoder decoder structure proposed by Google,which has advantages in processing long sequence data.Quantifying the traffic conflict risk at intersections based on trajectory data and analyzing the spatiotemporal evolution of conflict risk using Transformer models can achieve proactive accident prevention work.This study focuses on the "trajectory data-traffic conflict identification method-traffic conflict risk measurement-traffic conflict risk evolution model." Firstly,machine vision technology is used to automatically extract trajectory data of road users in the video,and a new identification method for traffic conflict events,called the regional post-intrusion time,is proposed.This method grids the intersection,takes a specific grid as the object,calculates the moment when the road user arrives and leaves each grid,and outputs the post-invasion time of each grid at a specific time.This batch identification method for various types of traffic conflict events at intersections is different from traditional identification methods.Secondly,the risk of traffic conflicts is defined to quantify the level of traffic safety at intersections,considering the severity of multiple types of traffic conflicts.Finally,a traffic conflict risk evolution model is built based on Transformer,which has a built-in attention mechanism for learning and understanding the temporal and spatial evolution of traffic conflict risk at intersections.The model predicts the global traffic conflict risk at target intersections in the future time window.Case analysis of three intersections in Xi’an City verified the effectiveness of the proposed model.This study provides key technical support for quantifying and predicting the traffic safety level of major urban intersections.It provides a theoretical and practical basis for improving the comprehensive analysis of urban traffic situations,management and control capabilities,and building an accurate and efficient smart traffic management system,especially under limited management resources.This approach can improve the efficiency of urban traffic management and advance accident prevention gateways.