Pedestrian Action Classification from a Vehicle's Perspective

Jaime Villa, Arturo de la Escalera, José María Armingol

Abstract: In the field of autonomous driving, understanding contextual information about the environment is essential for ensuring safety and efficiency. In a complex setting such as urban roads, it is not enough to know the position of vehicles and pedestrians; it is also crucial to understand the risks associated with these elements and their actions. This article implements various techniques that not only allow for the localization of pedestrians but also interpret their actions from a vehicle's perspective, analyzing their individual movements and how these are influenced by the vehicle's own motion. To accomplish this, the TITAN dataset is used, which undergoes data post-processing before training various models.

External IDs:dblp:conf/robot/VillaEA24