Classifying Pedestrian Crossing Flows: A Data-Driven Approach Using Fundamental Diagrams and Machine Learning.

This article investigates pedestrian dynamics in crossing flows at varying crossing angles (α), analyzing velocity (v), density (ρ), avoidance number (Av), and intrusion number (In) to characterize behavior and classify crossing scenarios. Using experimental trajectory data across seven angles from 0° to 180°, the study constructs velocity–density fundamental diagrams revealing distinct functional relationships—power law at 0°, linear at intermediate angles (30°–120°), and logarithmic at obtuse angles (150° and 180°)—reflecting different interaction patterns such as unidirectional flow, counterflow lane formation, and complex intermediate crossings. Dimensionless metrics Av and In quantify collision anticipation and personal space intrusion, with Av strongly influenced by crossing angle while In shows negligible dependence. Classification of crossing scenarios using logistic regression and random forest models demonstrates that combining all four features yields robust performance, with velocity and avoidance number identified as the most influential variables. These findings offer practical insights for crowd management and pedestrian infrastructure design, emphasizing the importance of crossing angle in flow efficiency and safety.