Citation

Hnoohom N, Chotivatunyu P, Maitrichit N, Nilsumrit C, Iamtrakul P. Transp. Res. Interdiscip. Persp. 2024; 24: e101036.

Copyright

(Copyright © 2024, Elsevier Publishing)

DOI

10.1016/j.trip.2024.101036

PMID

unavailable

Abstract

To progressively analyze the urban factors on different levels of crosswalk road safety, the study focused on developing a video-based safety methodology using a pedestrian and vehicle tracking approach for calculating time-to-collision (TTC) and post-encroachment time (PET) to improve safety in crosswalk areas. The study area is limited to Thammasat University in an urban environment where pedestrian and vehicle traffic across the crosswalk is simulated in controlled scenarios. The dataset was collected using closed-circuit television (CCTV) cameras installed at locations in the study area. The YOLOv8x with BOT-SORT was used to track pedestrians and vehicles from the CCTV videos. The overall mAP was 0.861 with a precision of 0.899, a recall of 0.785, and the objects classifier's average accuracy of 0.786. The TTC and PET values were calculated based on the collected data. These metrics can be used to determine the time required for a vehicle or pedestrian to come to a stop or avoid a collision, as well as the time it takes a vehicle or pedestrian to recover from an evasive movement or when the pedestrian has been hit by the vehicle. The 30 videos of near-miss accident scenarios, where 15 videos of pedestrians and vehicles and 15 videos from disordered persons and vehicles, were used to analyze the TTC and PET values to identify areas for further improvement. From the results, the TTC calculation achieved an accuracy of high risk at 90 percent in all scenarios, while PET values higher than 1.0 second resulted in 18 scenarios.

Language: en

Keywords

Crosswalk; Object detection; Object tracking; Pedestrian; Road safety; Urban development