Work

An analysis of pedestrian crossings through deep learning models and crash data

Year: 2025

Type: article

Abstract: This study focusses on pedestrian crossing (crosswalk) analysis and classification in urban contexts. Deep learning models and graph theory tools serve as the foundation for the proposed approach. Deep learning models are used to identify and regress crosswalk images based on specific indices that assess pedestrian exposure to crash events, as well as raw crashes. The crosswalk images were captured using Google Earth’s capabilities and are from the entire set in Città Studi, Milan, Italy. Additionally, 5-year pedestrian crash data are evaluated and linked to crosswalks when applicable. Classification produces good results, with an accuracy of approximately 60–70%. Regression models work well with exposure indices but poorly with raw crashes. Correlations between exposure indices and crash data are negative and very low.

Source: Transportation Research Interdisciplinary Perspectives

DOI: 10.1016/j.trip.2025.101449

Authors: Lorenzo Mussone, Omar el Hassan

Institution: Politecnico di Milano

Language: English

Topic: Traffic and Road Safety

Subfield: Safety, Risk, Reliability and Quality

Field: Engineering

Domain: Physical Sciences

SDGs: Sustainable cities and communities, Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $2,400