From Visual Evidence to Road-Level Screening of Cycling-Environment Risks: A Cyclist-Perspective Mobile-Sensing Study in a Historic Urban District

Safe, usable cycling environments support sustainable mobility, yet road surveys and street-view imagery often miss temporary encroachments and dynamic traffic interactions. This study developed a cyclist-perspective mobile-sensing workflow for road-level risk screening. Object detection and short-term association supplied visual and temporal evidence. Object positions relative to cycling space and the temporal persistence of those relationships were interpreted using coding rules and manual verification to rate applicable factors within a ten-factor framework. Weighted sums of standardized factor ratings yielded sample-level Composite Risk Index (CRI) values, subsequently aggregated to the road level. Fieldwork in Qinhuai District, Nanjing, yielded 339.51 km of trajectories, 16,732 valid location–passage-time samples, and 388 directed road evaluation units. Insufficient effective width for non-motorized vehicles, mixed motor-vehicle and non-motorized-vehicle traffic, and missing or discontinuous non-motorized-vehicle lanes were widespread; inadequate lateral buffers, parking encroachment, and visibility blind spots were localized. Rankings remained stable under alternative weighting and repeated-observation treatments. Independent fixed-camera observations showed a significant positive correlation between road-level CRI and confirmed near-conflicts per 100 non-motorized-vehicle passages (Spearman’s ρ = 0.904, permutation p < 0.001). The workflow converts cyclist-perspective observations into comparable road-level screening results. These results can guide field-verification priorities and targeted cycling-space management on surveyed roads.