Understanding cycling route choice behaviour through street-level images and computer vision-enriched discrete choice models

Abstract

This study investigates cyclists’ preferences for cycling environments, such as lane type, pavement type, and lane width. We conducted a stated route choice experiment where respondents evaluated alternatives varying in cycling environment, travel time and the number of traffic lights. Unlike previous studies, we used thousands of street-level images to represent the cycling environment, providing rich information on safety and quality that text or numbers cannot convey. We analysed the data using recently proposed computer vision-enriched discrete choice models, which integrate computer vision into traditional discrete choice models. Thereby, we can infer cyclist trade-offs and estimate willingness-to-pay values. Results show that cycling environments strongly influence route choice, with cyclists preferring green areas and separated cycling lanes. For an 11-minute trip, cyclists are willing to take a 1.5-minute detour for a separate lane instead of a mixed-traffic road. These findings offer insights for designing cycling environments that align with cyclists’ preferences.