Analyzing pedestrian-vehicle interaction dynamics at unsignalized crosswalks: a game mode analysis through MCMC and BN modeling

Abstract

Insufficient traffic control at unsignalized crosswalks causes pedestrians and vehicles operations to rely on their perception of road conditions, environment, and potential risks, resulting in game-like interactions with strategic decisions and behavioral adjustments. However, the characteristics and mechanisms of these behaviors remain poorly understood. This study analyzes pedestrian-vehicle interaction patterns at unsignalized crosswalks using Unmanned Aerial Vehicle (UAV) data. Ten game types are identified and classified into no-game, single-game, and multi-game interactions. A Pedestrian-Vehicle Game Index (PVGI) is proposed to quantify interaction dynamics by integrating speed, acceleration, and distance. Markov-Chain Monte Carlo (MCMC) simulation determines the PVGI domain as [−4.0, 2.0], distinguishing pedestrian-yield [−4.0, 0] and vehicle-yield [0, 2.0]. A Bayesian Network (BN) model, combined with the Gaussian Mixture Model (GMM) and Expectation-Maximum (EM) algorithm predicts second-round game patterns with an accuracy of 83.78%. These findings provide actionable insights for improving traffic safety and efficiency at unsignalized crosswalks.