Citation

Wang Y, Xu W, Zhang W, Zhao JL. IEEE Trans. Intel. Transp. Syst. 2022; 23(3): 2116-2129.

Copyright

(Copyright © 2022, IEEE (Institute of Electrical and Electronics Engineers))

DOI

10.1109/TITS.2020.3033276

PMID

unavailable

Abstract

Driving risk evaluation is a critical issue in driving safety analysis. In traditional driving risk evaluation models, vehicles are analyzed as isolated units. Nevertheless, vehicles are surrounded by other vehicles during driving in a real setting, and therefore, the driving patterns of target vehicles are inevitably affected by surrounding vehicles. In this paper, we proposed a new driving risk evaluation model incorporating the driving patterns of the target vehicle, the driving patterns of surrounding vehicles, and the interactions between the target vehicle and surrounding vehicles to improve driving safety in situations like car-following and merging situations. According to experiments on real data, our proposed method outperformed the state-of-the-art methods by at least 14.2% in terms of precision. Our work verified that factors like the driving patterns of surrounding vehicles and the interactions between a target vehicle and surrounding vehicles can enhance the performance of driving risk evaluation. The experimental results also showed that the driving patterns of surrounding vehicles in different positions vary in evaluating the driving risk of the target vehicle. More specifically, the surrounding vehicles in cross positions casted the strongest influence, followed by the surrounding vehicles in the diagonal cross positions. In summary, our study provided a novel way to enhance driving risk evaluation performance for driving safety improvement.

Language: en

Keywords

Acceleration; Accidents; Analytical models; deep learning; Driving risk; Logistics; Neural networks; region connection calculus; Safety; safety space; Vehicles