BrailleVision 👁️

Real-Time Physical Braille to English Converter

BrailleVision Hackathon 2026 Submission

What It Does

Detects real physical embossed Braille from a camera feed and converts it to English text and speech in real time.

Pipeline

Camera → YOLOv8 Detection → Letter Recognition → English Text → Text-to-Speech

Tech Stack

• Python
• YOLOv8 (Ultralytics)
• OpenCV
• Streamlit
• pyttsx3 (TTS)
• Roboflow Dataset

How to Run

pip install -r requirements.txt
python3 -m streamlit run app.py

Dataset

• Source: Roboflow Universe
• 1016 training images
• 26 classes (A-Z)
• Format: YOLOv8

Model

• Base: YOLOv8 nano
• Fine-tuned on physical Braille dataset
• Output: best.pt

AI Tools Used

See ai_tools_disclosure.md

Future Improvements

• Grade 2 Braille support (contractions)
• Mobile app version (Flutter/Android)
• Multi-language Braille support
• Improved handling of angled/curved Braille
• Word-level confidence scoring
• Offline embedded deployment