Understanding perception of cycling safety from street-view images: uncovering non-linear effects of urban factors

Abstract

Cycling is critical in transitioning to more sustainable transportation systems for shorter trips, including firstand-last-mile links to transit. To support this transition, cities need to provide cyclists with environments where they feel safe and comfortable. Thus, analyzing cyclists' safety perceptions is critical for planners and decision-makers to improve cycling uptake, as this is the main deterrent for individuals to cycle. If cyclists perceive the network as unsafe, they will prefer different modes for their regular trips. Yet, capturing and understanding how individuals perceive cycling risk is complex and often slow, with researchers defaulting to traditional surveys and in-loco interviews. In this study, we tackle this problem by understanding the perception of cycling safety from real-world images, together with imagery data, mapping data, and data processing tools. We use an Explainable Boosting Machine (EBM), a glassbox machine learning algorithm, to analyze the impact of image characteristics and other mapping information (i.e., urban elements) on individuals' perceptions. Insights are captured directly from street-view images and the surrounding built environment to allow researchers to individually analyze non-linear impacts of each characteristic on each cycling environment. Our results show how this approach facilitates the continuous assessment of changing cycling environments and its use in efficiently assessing different locations with the growing number of openly available street-view images across cities. In turn, this can help evaluate more effectively how road environment features relate to cyclists' risk perception on the road.

Keywords: Perception of Safety, Cycling, Subjective Safety, Explainable Boosting Machine, Berlin (Germany)