Drishti360

Dhrishti360: Turning Braille into understanding. Capture a photo, and our AI instantly translates full-page Braille documents into readable, context-aware digital text.

Comment

Inspiration

Millions of visually impaired individuals rely on Braille for reading, yet accessing printed Braille content often requires specialized devices or manual transcription. We wanted to create an AI-powered solution that can instantly recognize Braille from real-world images and convert it into readable text, making information more accessible to everyone. Drishti360 was inspired by the vision of bridging the communication gap between Braille users and non-Braille readers through computer vision and deep learning.

What it does

Drishti360 is an end-to-end Braille-to-English translation system that detects physical Braille cells from images and converts them into readable text in real time.

Key capabilities:

  • Detects Braille cells on books, labels, signs, packaging, and documents.
  • Uses YOLOv8-based object detection to identify all 64 possible 6-dot Braille patterns.
  • Applies multi-pass image preprocessing for challenging lighting conditions.
  • Automatically groups detected cells into lines and paragraphs.
  • Decodes Braille into English text with support for number and capitalization modes.
  • Provides an interactive GUI with ROI selection, confidence control, detection visualization, and TXT/PDF export.
  • Works on both CPU and GPU environments.

How we built it

  • Trained and optimized a YOLOv8-m model on multiple Braille datasets containing both document and real-world scene images.
  • Developed a multi-pass preprocessing engine using OpenCV to improve detection under varying lighting and contrast conditions.
  • Implemented box normalization and deduplication algorithms to improve alignment and eliminate duplicate detections.
  • Built a Braille decoding engine that translates binary dot patterns into English characters while handling Braille control symbols.
  • Created a complete Python desktop application using Tkinter for image loading, ROI selection, visualization, and export.
  • Integrated the entire pipeline into a seamless workflow from image input to text output.

Challenges we ran into

  • Detecting tiny Braille dots under poor lighting and noisy backgrounds.
  • Handling curved surfaces and perspective distortions in real-world images.
  • Eliminating duplicate detections caused by overlapping bounding boxes.
  • Maintaining correct reading order across multiple lines and paragraphs.
  • Decoding Braille control symbols such as number and capitalization indicators accurately.
  • Balancing model accuracy with real-time performance on CPU devices.

Accomplishments that we're proud of

  • Successfully built a complete Braille reading pipeline from detection to translation.
  • Achieved robust detection across multiple datasets and diverse environments.
  • Added innovative features beyond the original model, including preprocessing optimization, ROI selection, box normalization, and intelligent line grouping.
  • Created a user-friendly interface that requires no technical expertise.
  • Enabled export of translated content into accessible formats.
  • Designed a scalable architecture that can be extended to mobile and assistive technologies.

What we learned

  • Real-world accessibility problems require more than just accurate AI models; usability and workflow matter equally.
  • Image preprocessing can significantly improve object detection performance.
  • Braille decoding involves contextual state management, not just symbol translation.
  • Dataset diversity is critical for building robust computer vision systems.
  • Accessibility-focused AI can create meaningful social impact when combined with practical user experiences.

What's next for Drishti360

  • Real-time camera-based Braille reading for smartphones.
  • Offline mobile deployment using quantized YOLOv8n models.
  • Text-to-Speech integration for instant audio feedback.
  • Support for Hindi, Tamil, Bengali, and multilingual Braille systems.
  • 8-dot Computer Braille recognition.
  • Cloud-based API for integration with assistive applications.
  • Smart wearable support for visually impaired users.
  • Generative AI-powered explanations and summarization of detected content.
  • Continuous learning pipeline with user-contributed Braille datasets.

Drishti360's vision: Transform any Braille text in the physical world into instantly accessible digital information, empowering inclusive communication through AI.

Built With

Share this project:

Updates