BrailleVision

BrailleVision reads any embossed Braille page with a phone camera, instantly translates it to English, corrects errors via a server‑less LLM, and speaks the text without installation in browser.

Comment

Inspiration

Over 253 million people live with vision impairment worldwide, yet less than 1% of published books exist in Braille. Even those who can read Braille face a slow, tactile-only experience with no way to quickly scan a full page. We asked: what if a smartphone camera could read Braille for you, instantly?

What it does

BrailleVision is a real-time, browser-native Braille reader. Point your phone camera at any physical embossed Braille page — the app detects the raised dot patterns using computer vision, translates each 2×3 Braille cell to English, and speaks the result aloud using text-to-speech. A Bulk Translate mode captures entire pages and builds a full readable passage. No app download. No server. Just a URL.

How we built it

  • Zero-dependency CV Pipeline: Instead of loading the 8MB OpenCV.js library (15s load time), we wrote a pure JavaScript Canvas 2D pixel-analysis pipeline that runs at 30+ FPS with sub-1s startup.
  • Braille Tracker: A temporal majority-vote stabilizer smooths out noisy frames so the translation doesn't flicker.
  • LLM Post-Correction: A Vercel serverless function calls GPT-4o-mini to fix detection errors and produce clean English output.
  • Web Speech API: Stabilized TTS reads translations aloud naturally.
  • Bulk Translate Mode: Aggregates frame-by-frame translations into a full passage with copy-to-clipboard support.
  • Accessibility-first: ARIA live regions, high-contrast mode, keyboard navigation, and screen reader support baked in.

Challenges we ran into

  • Physical embossed Braille produces only shadows, not color contrast — detecting these reliably across different lighting conditions required careful pixel-mean threshold tuning.
  • Replacing OpenCV.js with a lightweight pure-Canvas pipeline while maintaining detection accuracy was a significant engineering challenge.
  • Stabilizing real-time translation to avoid character flickering required building a custom temporal voting tracker.

Accomplishments that we're proud of

  • Sub-1-second load time with zero external CV dependencies
  • Fully serverless — zero infrastructure cost
  • Works on any device with a browser and camera
  • Complete accessibility compliance (WCAG AA)

What we learned

Lightweight, purpose-built pipelines consistently outperform general-purpose libraries for constrained use cases. Browser APIs (Canvas, Web Speech, Clipboard) are powerful enough to build real assistive technology without any backend.

What's next for BrailleVision

  • Multi-line Braille page scanning with automatic line detection
  • Support for Grade 2 contracted Braille
  • Offline PWA mode (no internet required)
  • Export translated passages as audio files

Built With

Share this project:

Updates

Submission history