Conflict Risk Analysis on Motor and Non-motor Vehicles at Non-signalized Intersection Based on Computer Vision

Abstract:

Non-signalized intersections serve as critical nodes in urban mobility systems. In the absence of traffic signal control， these intersections often experience mixed flows of motorized and non-motorized vehicles， increasing the likelihood of vehicle–bicycle conflicts. Such conflicts frequently lead to abrupt braking and evasive maneuvers by drivers， ultimately reducing the efficiency of urban mobility. Therefore， it is essential to analyze the conflict risks at non-signalized intersections to inform intersection design improvements and enhance overall urban mobility performance. This study proposes a trajectory extraction framework based on computer vision techniques to analyze traffic interactions at non-signalized intersections. Utilizing over 2，000 minutes of video footage， we examine various traffic behaviors and conflict processes among different types of road users. A random-parameter ordered probit model is employed to capture the coupling relationships between driver behavior， environmental factors， and the severity of traffic conflicts. The results reveal that the sequence of intrusion during conflict events—combined with the type of road user involved—is highly correlated with the severity of the conflict. Notably， the involvement of electric bicycles significantly increases the risk level. The proposed trajectory extraction framework offers a valuable tool for researchers studying traffic conflict dynamics， while the findings provide theoretical guidance for enhancing urban traffic efficiency and informing practical traffic management strategies.