Abstract
Rapid transit expansion across Latin American metropolitan areas has significantly improved regional mobility; however, infrastructural gaps persistently exclude persons with disabilities from equitable urban participation. Traditional municipal transit assessments often rely on top-down, coarse audits that overlook localized, micro-scale physical barriers, while purely computational approaches frequently disregard the lived experiences of marginalized transit users. This study proposes an interdisciplinary framework integrating participatory co-design methodologies with artificial intelligence (AI)-driven predictive spatial analytics to evaluate and enhance transit accessibility in two dense Latin American urban corridors (Lima and Medellín). Working alongside local disability advocacy organizations, we conducted participatory mapping sessions to categorize critical pedestrian-to-transit friction points. These qualitative insights guided the deployment and fine-tuning of a computer vision workflow designed to audit street-level imagery for infrastructure deficits, including degraded sidewalks, missing tactile paving, and absent curb ramps. The hybrid model achieved an 88.4% mean average precision in identifying accessibility micro-barriers and revealed pronounced socio-spatial disparities between central business districts and peripheral feeder routes. By closing the feedback loop between machine learning models and lived experiences, the framework provides urban planners with an actionable, citizen-centered decision-support tool for prioritized infrastructure remediation.
