BrailleVision

• 

  Real-time Braille cell detection with YOLOv8 — bounding boxes and confidence scores on physical Braille paper

• 

  Line-by-line English text output with Text-to-Speech accessibility feature

👁️ BrailleVision — Real-Time Physical Braille to English Converter

Inspiration

Millions of visually impaired people rely on Braille daily, but Caregivers, teachers, and volunteers often cannot read Braille. I wanted to build a bridge—a tool that lets anyone instantly understand physical Braille using just a camera.

What I Built

BrailleVision is a real-time assistive technology app that uses a camera to scan physical embossed Braille and converts it to English text and speech instantly.

Pipeline: Camera → OpenCV Preprocessing → YOLOv8 Detection → Letter Recognition → English Text → Text-to-Speech

How I Built It

• Fine-tuned YOLOv8 nano on 1016 physical Braille images
• Built real-time Streamlit web app with live camera feed
• Implemented line-by-line reading with dynamic cell grouping
• Added pyttsx3 TTS for audio accessibility output
• Trained for 100 epochs achieving 92.2% mAP50 accuracy
• Built entirely solo in under 25 hours

Challenges

• Handwritten Braille has irregular dot spacing
• Lighting variations affect dot visibility
• Multi-line Braille requires accurate row segmentation
• Real-time performance on Mac CPU

What I Learned

• YOLOv8 object detection for small object recognition
• Physical Braille cell structure and Grade 1 encoding
• Accessibility-first UI/UX design
• Real-time computer vision pipeline optimization

Built With

• numpy
• opencv
• python
• pyttsx3
• roboflow
• scipy
• streamlit
• ultralytics
• yolov8