Accessibility Scout: Personalized Accessibility Scans of Built Environments
summary
An LLM-based system called Accessibility Scout analyzes photos to produce personalized accessibility scans by building a user model, iteratively updating via user feedback, and explaining what actions a feature hinders. The team evaluates 500 images (technical eval) and runs a user study with 10 participants plus a six-person formative study; results show personalization yields more useful concerns (higher median for 9/10, significant for 5). The system reports a false positive rate 6.63%, cost 2 cents per image, and average latency 10.74 seconds, and demonstrates scalable, adaptive assessments that go beyond ADA checklists.
AI-written summary · about
chapters (12)
- 0:00Introduction and setup
- 1:10Motivation and personal needs
- 2:55Current accessibility assessment limits
- 3:37Introducing Accessibility Scout
- 5:02System design and workflow
- 6:23Live demo with Anna
- 7:26Technical evaluation results
- 8:27User study findings
- 9:32Personalization examples
- 10:13User feedback and benefits
- 11:57Future work and applications
- 13:05Q&A on confidence and limitations
