Citation

Zhang C, Berger C. Transp. Res. Proc. 2023; 72: 1653-1660.

Copyright

(Copyright © 2023, Elsevier Publications)

DOI

10.1016/j.trpro.2023.11.637

PMID

unavailable

Abstract

Pedestrians are the most vulnerable road users in urban traffic scenarios and need to be protected from potentially hazardous situations. It is essential for automated vehicles and modern driver-assistance systems to better predict pedestrians' behavior to prevent road crashes. Predicting pedestrian behavior is challenging because their behavior can be influenced by many factors. In recent years, deep learning (DL) methods as powerful tools have been utilized by many researchers to improve such predictions, but few researchers have analyzed the factors that influence pedestrian behavior prediction in DL. This paper uses DL to predict and analyze the factors that influence pedestrian behavior, especially the interactions between pedestrians and other road users. We focus on real-world urban traffic and use the publicly available Waymo Open Dataset for training, testing, and analyzing.

Language: en

Keywords

automated vehicles; behavior analysis; Deep learning; pedestrian behavior prediction; pedestrian interactions; trajectory prediction