Unveiling the impact of micro-level visual features on urban congestion in Chicago

Urban traffic congestion remains a persistent challenge in contemporary cities, with most existing research focusing on macro-scale built environment features such as land use and transport network and overlooking the visual and perceptual cues embedded in micro-scale streetscapes that influence real-time driving behavior. To address this gap, this study investigates the relationship between street-level visual features—captured from Google Street View imagery—and traffic congestion at the road segment scale in Chicago, U.S.A. Leveraging semantic segmentation and geographically weighted regression, we extract seven key visual features and examine their spatially varying associations with congestion intensity. We further propose a dual-pathway interpretive framework, distinguishing a physical-functional path (where visual features act as proxies for traffic-generating land use and street function) and a perceptual-behavioral path (where visual complexity and natural elements shape driver cognition and behavior). Our findings demonstrate that incorporating visual features significantly improves model performance, increasing the explanatory power by approximately 19 % compared to models using only macro-level variables. Visual elements such as buildings and vehicles are found positively associated with congestion in high-demand corridors, while features like sky visibility, trees, and sidewalks exhibit congestion—mitigating effects—not universally, but in specific visual and functional contexts as experienced by drivers. Importantly, these effects are spatially heterogeneous, reflecting variations in local land use patterns, street hierarchy, and perceptual environments, as captured from a driver’s viewpoint. This study highlights the value of integrating visual-perceptual attributes into urban mobility analysis and calls for context-sensitive transport planning that considers both structural and cognitive dimensions of the streetscape.