BrailleVision
Inspiration
Braille is one of the most important communication systems for visually impaired individuals, yet many caregivers, teachers, volunteers, and even family members cannot read it. We wanted to build a tool that could bridge this communication gap using AI and computer vision. Our goal was to create a system that can instantly interpret real physical Braille from paper and convert it into readable English text and speech in real time.
What it does
BrailleVision is an AI-powered accessibility system that uses a camera to detect physical Braille dots from paper, segment them into Braille cells, decode them into English text, and read the result aloud using text-to-speech. The platform supports image uploads and real-time webcam scanning through a responsive web interface designed for accessibility and usability.
The system also provides:
- Real-time confidence scoring
- Camera quality feedback
- Speech playback
- Accessibility-focused UI
- Detection overlays and visual debugging
How we built it
We built BrailleVision using Python, OpenCV, FastAPI, HTML, CSS, and JavaScript.
The backend pipeline includes:
- Image preprocessing using grayscale conversion, CLAHE enhancement, Gaussian blur, and adaptive thresholding
- Braille dot detection using blob detection and contour analysis
- Braille cell segmentation using spatial clustering and dynamic spacing estimation
- Braille pattern decoding using Grade 1 Braille mapping
- Text-to-speech output using offline speech synthesis
We wrapped the AI pipeline inside a FastAPI backend and connected it to a responsive frontend that allows users to upload images or scan Braille through a camera interface.
Challenges we ran into
One of the biggest challenges was detecting real embossed Braille dots under inconsistent lighting conditions and different camera angles. Unlike digital Braille symbols, physical Braille introduces shadows, noise, blur, and uneven spacing that make detection difficult.
Another major challenge was grouping detected dots into accurate 2×3 Braille cells while minimizing false positives. We also had to redesign our confidence system because early versions produced high confidence scores even when decoded text was incorrect.
Building a stable real-time pipeline while maintaining smooth UI performance was also a significant challenge.
Accomplishments that we're proud of
We are proud of building a full end-to-end AI accessibility system capable of:
- Detecting real physical Braille from camera input
- Decoding Braille into English text
- Providing real-time speech output
- Running in a responsive web interface
- Supporting accessibility-focused interactions
We are also proud of implementing:
- Real-time webcam mode
- Stability tracking
- Confidence scoring
- Camera guidance assistance
- Before/after visual overlays
Most importantly, we transformed a complex computer vision problem into a usable assistive technology prototype.
What we learned
This project taught us how challenging real-world computer vision can be compared to controlled datasets. We learned the importance of preprocessing, geometric consistency, segmentation stability, and user-centered accessibility design.
We also gained experience in:
- OpenCV pipelines
- Real-time AI systems
- FastAPI backend development
- Frontend integration
- Accessibility-focused product design
What's next for BrailleVision
We plan to improve decoding accuracy using more advanced computer vision techniques and larger real-world Braille datasets. Future versions may include:
- Mobile deployment
- Multi-language Braille support
- Offline edge AI support
- Better real-time performance
- Automatic document scanning
- AI-assisted correction for uncertain Braille patterns
Our long-term vision is to make BrailleVision a practical assistive tool for schools, caregivers, accessibility workers, and visually impaired communities worldwide.
Log in or sign up for Devpost to join the conversation.