Exploration of AI in Vision Accessibility

Explore AI for vision accessibility, from screen readers to a Braille OCR pipeline. See real-world workflows and learn how AI can address accessibility challenges.

Overview

This demo illustrates the current assistive technology landscape and a yolo-based OCR pipeline for Braille. As a student developer with severe visual impairment, James showcases sample workflows for computer use and navigation, from screen readers to vision language models, while discussing potential promising applications of AI in accessibility for the visually impaired.

James also showcases a deep learning Braille-to-print transcription pipeline supporting English and Cantonese. The demo goes through the pipeline architecture, models used, training techniques addressing low-resource Braille data, and disambiguating between homophones, prevalent in Cantonese.

Links

https://github.com/endernoke/braille-ocr
YOLO-based mobile web app translates photographed Braille to Unicode.

Tech stack

Exploration of AI in Vision Accessibility

Related talks