1. Introduction
The emergence of autonomous driving technology [12] has led to a growing focus on pedestrian safety and transportation convenience. One of the key technologies to achieving these goals is the ability to predict whether pedestrians will cross or not cross. Using the prediction results, the autonomous vehicle can slow down or stop to prevent any accidents related to pedestrians [27], [35]. However, predicting pedestrian behavior is not easy because the intention of humans is unclear, and there are many external factors [3], [41], [42] that influence their behavior, such as interaction with other pedestrians, traffic signs, road congestion, and vehicle speed. These external factors may affect the future actions of pedestrians in the traffic environment. As a result, anticipating crossing intentions in advance is significantly challenging.