Probability estimation for pedestrian crossing intention at signalized crosswalks

Abstract

With the rapid development of the techniques for autonomous driving and ADAS in the last decade, more advanced methods to understand pedestrian behavior are required. Crosswalks at intersections are the one of most hazardous where many accidents between turning-vehicles and pedestrians occur. In this paper, we present a method for estimating the pedestrian's intention to cross a signalized crosswalk or stop in front of it. The intention is crucial to not only the collision avoidance but also smooth traffic in the context of autonomous driving by reducing unnecessary risk margins. Regarding the behavioral flow of pedestrian: assessment, decision-making and physical movement, as a stochastic process, we construct a probabilistic model with the Dynamic Bayesian Network. It takes account of not only pedestrian physical states but also contextual information and integrates the relationship among them. By employing the particle filter as a Bayesian filtering framework, the model estimates the pedestrian state from signal information and pedestrian position measurements. Evaluation using experimental data collected in real traffic scene shows that the proposed model has an ability to detect the pedestrian intention to cross a crosswalk even when he/she is far from it.