Computer vision-based assessment of cyclist-tram track interactions for predictive modelling of crossing success

• Published in: Journal of safety research Vol. 87; pp. 202 - 216
• Main Authors: Gildea, Kevin, Hall, Daniel, Mercadal-Baudart, Clara, Caulfield, Brian, Simms, Ciaran
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.12.2023
• Subjects: Civil Engineering; Computer vision; Engineering and Technology; Infrastructure Engineering; Infrastrukturteknik; Samhällsbyggnadsteknik; Single bicycle crashes; Surrogate measures of safety; Teknik; Tram tracks; Video analysis; Single bicycle crashes; Computer vision; Surrogate measures of safety; Tram tracks; Video analysis
• ISSN: 0022-4375; 1879-1247; 1879-1247
• DOI: 10.1016/j.jsr.2023.09.017

•A high incidence of single bicycle crashes on tram tracks were observed in Dublin city over a short study period.•Unsuccessful crossings were more common at specific locations with environmental constraints such as nearby kerbs, or traffic pressures, i.e., nearby/passing traffic.•Predictive modelling illustrates the strong relationship between crossing angle and crossing success, indicating the necessity of allowing for safe crossing angles.•Recommended interventions include track realignments, implementation of jughandle cycle lane designs, and measures to reduce vehicle traffic volumes and speeds.•This study demonstrates the potential of video-based analyses in understanding causative factors for single bicycle crashes. Single cyclist/bicycle crashes (SBCs) are common, and underreported in official statistics. In urban environments, light rail tram tracks are a frequent factor, however, they have not yet been the subject of engineering analysis. The prevalence of traffic camera footage in urban environments presents an opportunity for detailed site-specific safety insights for these cases. In this study, we present a video-based frequency and risk analysis for unsuccessful crossings (UCs) on tram tracks in wet road conditions at 9 locations around Dublin city centre, Ireland. We also devise a predictive model for crossing success as a function of crossing angle for use in a Surrogate Measure of Safety (SMoS) framework, and investigate the use of convolutional neural networks (CNNs) for semi-automatic estimation of cyclist crossing angles. Modelling results indicate that cyclist crossing angle is a strong predictor of crossing success, and that cyclist velocity is not. Findings suggest that in general, infrastructural planners should design for crossing angles of no less than 30° (ideally 90°). Findings also highlight the prevalence of external factors which limit crossing angles for cyclists. In particular, kerbs are a common factor, along with passing/approaching vehicles or other cyclists. A SMoS framework is introduced for cyclist interactions with tram tracks, and an open-source application is provided (SafeCross1https://anonymous.github.io/1). Furthermore, results indicate that further training on a relatively small sample of 100 domain-specific examples can achieve substantial accuracy improvements for cyclist detection (from 0.31AP0.5 to 0.98AP0.5), and crossing angle inference from traffic camera footage.