Citation

Fourkiotis M, Kazaklari C, Kopsacheilis A, Politis I. Transp. Res. Proc. 2022; 60: 536-543.

Copyright

(Copyright © 2022, Elsevier Publications)

DOI

10.1016/j.trpro.2021.12.069

PMID

unavailable

Abstract

The improvement of road safety is considered as a top priority on the agendas of governments and transport policy making stakeholders worldwide, with the 'Vision zero' target being the pinnacle of the European Commission's road safety strategy. Increased attention has been given to pedestrians, since crashes involving this vulnerable user group, have a higher mortality rate. As a result, research focusing on the behaviour of pedestrians and on the application of Intelligent Transport Systems that will assist pedestrians, is of increased importance. This paper attempts to predict pedestrian behaviour on crossings with a Countdown Signal Timer (CST), through the application of two machine learning algorithms. In the frame of the case study presented, an intersection in the Kalamaria city of Thessaloniki, Greece, where countdown signal timers have been installed on the pedestrian traffic lights, is analysed. For the needs of the analysis two models were implemented, a Deep Neural Network (DNN) and a Logistic Regression model.

RESULTS of both models indicated a satisfactory performance. In detail, the DNN model managed to estimate the pedestrians' crossing speed with a Mean Squared Error value of 0.0647 (km/h)2, while the Logistic Regression model, which classified pedestrians based on their behaviour, achieved an accuracy of 97%. The accurate determination of pedestrians' crossing behaviour could not only underline the influence of countdown signal timers, but also highlight appropriate countermeasures that can make infrastructure safer for this user group.

FINDINGS of this research can prove beneficial for researchers, infrastructure operators and policy makers alike, in their effort to improve the safety level of pedestrians.

Language: en

Keywords

artificial neural network; countdown signal timer; crossing behavior; machine learning; pedestrians; road safety