OCR and Text2Speech for blind people

Hello,

I'm trying to build a light portable device designed for blind or visually impaired people. The device is capable of recognizing text in front of the person using a camera. The text is converted to speech and the person will be able to hear the synthesized voice that pronounces the words in earphones.

The camera is mounted on special glasses with attached earphones. I think this is a good location for the camera because all humans are capable of precise head-hand coordination. Even without seeing, a human is able to point its head in the direction of his hand. This means that if a blind person is holding a book, that person is capable of pointing the glasses straight at the book.

Here are my thoughts so far:

The image sensor will be small, similar to this:
http://www.raspberrypi.org/camera
Sensors from Omnivision are small and I think they are great for this application.

I think ARM running Linux is a good option, because many image processing applications and speech synthesizers exist for this platform.

Overall, I want this device to look similar to this (of course it has other functionality):
http://forum.allaboutcircuits.com/attachment.php?attachmentid=60717&d=1382902520

I need help on build an embedded system capable of running Linux, do image processing and text2speech. The electronics involved will be quite complex, so any thoughts on which microcontroller and memory to use are welcome. The software is, perhaps, even more complex.

Anybody who is interested in this project and wants to help, please post here, or write me an email: xkyve1 [at] gmail.com

I will update this post with my progress. So far I've started on PC and a Raspberry PI.

Thank you

I would say that you need to use the most powerful processor you can to do the image processor and speech functions. In fact you may want to have multiple processors.

The ARM is good. But if you truly want to finish it, just write the code in a portable way and you can upgrade to a more powerful plateform once you do the proof of concept, if needed.

First find some OCR software that works with randomly presented text (i.e not neatly lined up like a printed page in a scanner). This is a case where if you can't find the software you need, you might as well not start.

Agreed! OCR'ing text from a photo (especially a less than perfect photo) is one of the most difficult tasks a processor can do.

Even the big developers with million dollar budgets and large teams have not got very far with that kind of task!

Every time you go to register at a web site and it uses Recaptcha (the one where you enter two words) you are helping to read a word that OCR has failed to recognize.