Bayesian Network Model of Decision-Making for Pedestrians' Crossing Behavior Considering Gaze Information at Unsignalized Intersection

Understanding pedestrians' behavior is a crucial challenge for achieving fully autonomous driving because it has a high degree of freedom and varies from person to person. This paper particularly focuses on the decision-making of the pedestrian, and aims to build a mathematical model of it. Generally, pedestrians make decisions considering other traffic participants and the traffic environment. Consequently, in a complex situation with multiple traffic participants, the mechanism of decision-making becomes intricate, and this makes it difficult to obtain an accurate mathematical model. To address this issue, this paper proposes a structured probabilistic model for the decision-making considering gaze information. In the proposed model, pedestrians' overall decision-making is divided into individual decision-making for each traffic participant considering gaze information, and expressed by a logistic regression model. These divided decision-makings are then integrated using a Bayesian network. By introducing this structure for the decision-making, the expansion of the state space of the model can be suppressed, and a reduction of the amount of training data can be achieved. Finally, to validate the proposed method, simulator experiments are conducted focusing on the crossing behavior at unsignalized intersection, and the performance of the model is confirmed.