In-depth Analysis of Traffic Conflicts at Interchange Merging Areas Based on Bayesian Network and Conflict chain

Abstract

Frequent acceleration and deceleration behavior and the var-ious merging speed of different vehicles have caused the merging area to become a high-risk section. To investigate the influencing factors and the underlying relationships on the traffic conflicts, an in-depth analysis of traffic conflicts at merging areas is conducted. UAV video data is used to extract vehicle trajectories to collect microscopic influencing factors. A Bayesian network model is proposed to explore the in-ner-relationship among influencing factors. An infor-mation-entropy and decision-tree based discretization method (IEDT) is used to improve the performance of the Bayesian network model. Key traffic conflict chain are then identified based on the Bayesian network. The results show that IEDT can improve the prediction accuracy of Bayesian network model about 9% than K-means, and 24% than equal distance method. Unstable lane-keeping ability, too small or too large merging duration etc. are significant factors that will increase the merging conflicts. The speed of on-ramp vehicle is the key factor that have inner-relationship with other factors. These results provide further insight to the mechanism of merging traffic conflicts, and can be helpful for traffic engineers and traffic enforcement officers take countermeasures to improve traffic safety.