Do visual attributes of streetscapes affect car crashes? Applications of computer vision techniques and Machine learning

Abstract

This study examines the relationships between visual attributes of streetscapes and car crashes by quantifying the visual characteristics of urban road landscapes from a driver's perspective. Utilizing street panoramic images, advanced computer vision, and interpretable machine learning techniques, the research identifies key visual factors impacting traffic safety. The findings reveal that green spaces in urban areas can reduce traffic accidents, supporting the idea that natural elements calm drivers and enhance safety. Conversely, excessive signage and high visual complexity increase accident rates due to cognitive overload and distractions. These insights have significant implications for urban planning and traffic safety policies. By pinpointing specific visual features that influence car crashes, urban planners and transportation engineers can design interventions to modify these elements, ultimately enhancing road safety.