Citation

Yang Z, Gong Z, Zhang Q, Wang J. Int. J. Transp. Sci. Technol. 2023; 12(4): 1052-1063.

Copyright

(Copyright © 2023, Elsevier Publishing)

DOI

10.1016/j.ijtst.2022.12.003

PMID

unavailable

Abstract

To study the intention behind pedestrian crossing behavior, this study extracts the trajectory data of vehicles and pedestrians from intersection videos. Based on the classic traffic conflict theory, TAdv is selected as the primary indicator to describe the pedestrian-vehicle conflict, and pedestrian crossing events are defined to represent the interaction state of pedestrians and conflicting objects at a certain time in the crossing conflict. This paper proposes a Kalman filter-based crossing event recognition method, and then uses the topic model in natural language processing technology to mine pedestrian behavior "topics" behind different crossing events, and obtains an LDA-based pedestrian crossing description model. The results show that: on the whole, pedestrians have high requirements for the right of way and will not easily change their behavior. Pedestrians have higher speeds in conflicts with non-motorized vehicles than motorized vehicles and have greater expectations of victory in conflict games. Pedestrians often adopt conservative behaviors at low risk and choose other strategies after the conflict has evolved to a certain degree (high risk). There are two types of pedestrians with the highest demand for the right of way. One is the aggressive pedestrians, who will adopt aggressive rushing strategies when facing non-motor vehicles while adopting the most conservative avoidance strategies when facing motor vehicles. The other is the pedestrians with small impacts from the outside world, whose crossing state will not easily be affected by vehicles and changes in traffic.

Language: en

Keywords

Intersection; Latent Dirichlet Allocation; Pedestrian crossing behavior; Traffic safety