Pedestrian Perceived Risk of Construction Obstructions and Barriers Identified via Image Segmentation

Pedestrian safety near construction sites is increasingly threatened by sidewalk obstructions such as materials and equipment. Despite growing attention to pedestrian-friendly environments, few studies have quantified how such visual encroachments affect the perceived risk. This study introduces a novel image-based framework that integrates the Segment Anything Model (SAM) for the object-level segmentation of sidewalks, obstructions, and barriers in 61 real-world street images. The results revealed significant differences in the perceived risk across four sidewalk environment types, defined by the presence or absence of obstructions and barriers. In addition, the proportion of sidewalk occupied by obstructions was strongly correlated with the perceived risk, whereas the relative size of barriers had no significant effect. This study provides a practical understanding of how temporary obstructions and barriers affect pedestrians’ perceived risk, supporting more effective safety management in urban construction environments. It also contributes to advances in research by providing an objective method for assessing visible hazards using advanced image segmentation techniques.