AI-Powered Reader for Visually Impaired Users

The clinical picture

Most printed text is not accessible to someone who cannot see it. Not books particularly, which are often available in other formats, but the ordinary printed material that structures a day: a letter, a form, a label, a notice on a wall.

Dedicated reading devices exist. They tend to be expensive, which puts them out of reach in exactly the places where alternative support is thinnest.

Why it's hard

Building a cheap version is not simply a matter of running optical character recognition on a photograph.

The two things that decide whether a device is actually usable are accuracy and latency. Text that is transcribed wrongly is worse than no text at all, because the user has no way to know it is wrong. And a device that takes several seconds to respond stops feeling like reading and starts feeling like waiting.

The idea

A small, low-cost device built from commodity parts: a single-board computer, a camera, and cloud recognition services, with the engineering effort spent on making the loop from capture to speech fast enough to feel natural.

How it works

The device pairs a Raspberry Pi with an HD camera for text capture, using Microsoft Azure Cognitive Services for optical character recognition and text-to-speech, reaching 98 percent reading accuracy.

The responsiveness work was in the software rather than the hardware. Restructuring the data structures and text-processing algorithms cut system latency by 40 percent, which is the difference between a device that feels responsive and one that feels like it is thinking.

What happened

The device works as a self-contained reader and was awarded Best Assistive Project by the Faculty of Engineering at the Federal University of Technology, Minna, for innovation and impact.