"Where Can I Park?" Understanding Human Perspectives and Scalably Detecting Disability Parking from Aerial Imagery
Jared Hwang, Chu Li, Hanbyul Kang, Maryam Hosseini, Jon E. Froehlich · 2025 · ASSETS '25: Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility · doi:10.1145/3663547.3746377
Summary
This mixed-methods paper addresses the critical gap in understanding and assessing disability parking in the United States. Despite ADA mandates requiring 4-8% of public parking spaces to be accessible, there has been no large-scale investigation of the quality or allocation of disability parking, nor significant research into how people with disabilities (PwDs) actually experience and use these spaces. The research has two components: a qualitative interview study with 11 PwDs exploring their disability parking experiences, and the development of AccessParkCV, a novel deep learning pipeline for automatically detecting and characterizing disability parking from orthorectified aerial imagery. The interview study reveals that PwDs approach parking in highly personalized ways based on their mobility needs, vehicle type, and prior experiences. Participants reported persistent challenges including inadequate space design, public misuse of accessible spots, and a constant "mental calculus" weighing access needs against effort. The technical component introduces a two-stage pipeline: a CoDETR-based locator model that detects parking spaces from 512x512 aerial image tiles, and a YOLOv11-based oriented bounding box (OBB) characterizer that estimates space width. The pipeline was trained on a new open dataset of 11,762 labeled parking objects across 5,125 images from Seattle, Washington D.C., and Spring Hill, Tennessee, covering seven classes of parking configurations.
Key findings
The interview study found that PwDs adapt their parking strategies significantly based on personal mobility needs, often developing creative workarounds like carrying signs or cones to protect their space, parking across multiple spots, or relying on family members as drivers. Participants strongly desired real-time, reliable information about parking availability, space characteristics (width, access aisles), and accessible routes from parking to building entrances. The AccessParkCV pipeline achieves a micro-F1 score of 0.89 for parking detection and an average width estimation error of only 5.40% compared to human annotators (whose inter-annotator error was 1.77%). Cross-city evaluation showed strong performance in regions similar to training data (above 90% recall in Seattle and DC) but drops in unfamiliar cities like Los Angeles (64-72% recall), suggesting per-city training data may be needed. The authors demonstrate two applications: a personalized parking search app allowing users to filter by space width and access aisle presence, and an urban analytics visualization showing disability parking density by census tract. Six design recommendations emerged: adequate spacing, visible signage, maintenance, safe location, connectivity to building entrances, and compliance auditing.
Relevance
This paper makes important contributions to physical-world accessibility infrastructure assessment at scale. The combination of qualitative research with PwDs and scalable computer vision techniques addresses a real gap — while digital accessibility is increasingly well-studied, the physical built environment remains difficult to audit systematically. The open-source pipeline and dataset provide practical tools for municipalities, disability advocates, and urban planners to assess ADA compliance across entire cities. The interview findings highlight that accessible parking is not just about space availability but about the entire journey from car to destination, including space width, access aisle presence, route accessibility, and real-time occupancy information. The work connects to broader themes of crowdsourced accessibility data and demonstrates how CV can complement human auditing efforts for physical accessibility compliance.
Tags: disability parking · computer vision · urban planning · aerial imagery · object detection · mobility · wayfinding
Standards referenced: ADA · ADA Standards for Accessible Design