Abstract

Urban parks are vital for public health and social cohesion, yet access remains profoundly unequal for people with mobility impairments. Traditional studies focus on distance- or time-based accessibility metrics, overlooking experiential barriers. This research develops a novel AI‑enhanced framework to assess equitable access to urban parks in Shenzhen, China, comparing general mobility individuals with those who have mobility impairments. We generate regular and accessible routes using navigation APIs and analyze street‑view imagery along these paths through a deep learning pipeline that quantifies visual elements, classifies scenes, and groups them into broader categories for comparison. Results reveal a hierarchy of exclusion. General mobility individuals face moderate spatial inequality, while mobility-impaired individuals confront severe network fragmentation. The median 15-minute park accessibility score for wheelchair users is zero across all districts, and the reachable percentage (i.e., the proportion of housing points with at least one park within 15 min) for manual wheelchair users is below 16%. For visual entropy, a measure of how varied and unpredictable the streetscape is along a route, accessible routes show significantly higher values than regular routes (e.g., mean entropy of 1.203 versus 0.8 for the general mobility group). Spatial analysis shows clustered access in the urban core versus systemic deprivation in suburbs. The findings demonstrate that equitable park access requires designing for the most constrained users, moving beyond proximity to ensure continuous, barrier-free networks. This approach serves as a replicable framework for assessing and promoting inclusive urban design.