Hello,

I'm very interested in building a device with an image sensor that is able to do OCR (Optical Character Recognition). What image sensors do you recommend me for this task. I'm hoping to spend no more than 30-40 dollars on the image sensor.

I want to interface the image sensor to a microcontroller and do one of the following:

- process stream of images directly
- send stream of images to PC/smartphone for processing

Basically, I want to point the image sensor at a book for example and recognize the characters. Also, it needs to be small.

I'm from Europe and I'm having a hard time trying to locate good online shops that have these kind of parts.

Any suggestions, help on the image sensor and lens required would be appreciated.

Later edit:

I'm trying to build a device for blind people. The mini-camera will be mounted on some glasses. Earplugs are attached to the glasses so that the person may hear what the device "sees" using OCR. After that, text to speech must take care of the audio part.

This is the reason I'm trying to build my own camera module, so it has to be small, because I want to integrate it somehow into glasses, maybe behind one of the lenses, or right between the lenses, so that I can integrate other circuits behind the lenses.

Thanks

 

Any decent digital camera can be a "image sensor". There are lots many OCR software that can do the conversion.

I generally use Irfan View.

Ramesh

 

The Raspberry Pi now has a camera module, it could probably do what you want...

 

Ummmm...a smart phone already has an "image sensor". In many cases, two. Why add a third?

 

I'm trying to build a device for blind people. The mini-camera will be mounted on some glasses. Earplugs are attached to the glasses so that the person may hear what the device "sees" using OCR. After that, text to speech must take care of the audio part.

@joeyd999
This is the reason I'm trying to build my own camera module, so it has to be small, because I want to integrate it somehow into glasses, maybe behind one of the lenses, or right between the lenses, so that I can integrate other circuits behind the lenses.

Also I want to put some small rechargeable batteries. Wireless recharging would be awesome, so that the blind person only has to put the glasses on a surface that is able to recharge the glasses through induction. But these are just details.

Anyhow, I image something like this: a blind person picks up a book and is able to "hear" it by wearing those special glasses. Or maybe he walks on the street and is able to hear directions from the glasses like: "careful, steps in front". Unlimited possibilities.

For now I'm trying to figure out the image capture part. Any help appreciated. Thanks

@sirch2
Found this https://www.modmypi.com/raspberry-pi-camera-board
Thank you for the tip. Omnivision image sensors seem the best way to go. Pixy uses them also:
http://www.kickstarter.com/projects/254449872/pixy-cmucam5-a-fast-easy-to-use-vision-sensor

Now I have to find a way getting my hands on an image sensor module from Omnivision.

@Ramussons
Didn't know that Irfan View has OCR. Nice.

 

Anyhow, I image something like this: a blind person picks up a book and is able to "hear" it by wearing those special glasses.
Or maybe he walks on the street and is able to hear directions from the glasses like: "careful, steps in front".
Unlimited possibilities.


@Ramussons
Didn't know that Irfan View has OCR. Nice.

This part is possible.
This part would require some creativity. One can Either compare the captured image with a (large) database to Identify It, or by invoking Artificial Intelligence with a Learning Capability.
In either case, I think, you would need a 3D imaging.

Ramesh

 

Thanks for the input. I will focus for now on OCR + Text2Speech. Other features would require some degree of machine learning as you've stated.

I just have to find a good image sensor available as a module on an online shop.
Omnivision sensors are hard to find. Where do I get some samples in Europe? I don't think large distributors will take small orders from me.

 

I suggest that you get started on a PC, using Jpeg images as an input, and with whatever software you're able to find or write. That way you will be able to get some kind of OCR software running without having to develop the hardware first, and then you can say "OK, now I want to get this software running on a portable setup" and start developing the system to do it. You'd have the opportunity of an intermediate step which would still use the PC-based software but could run off a webcam or whatever camera you cared to attach.

 

This is the reason I'm trying to build my own camera module, so it has to be small, because I want to integrate it somehow into glasses, maybe behind one of the lenses, or right between the lenses, so that I can integrate other circuits behind the lenses.

Blind people (most) still have eyeballs, which take up physical space. Not sure if there will be room inside the glasses for circuits.

I've never been blind so I can't speak with any certainty, but from watching blind people, they seem to be more "in tune" with the position of their hands than with the position of their head. Haven't you ever seen a blind person walking with a dog or with a stick, and their face is directed in some random direction; maybe to the side and up into the sky? I think that it's similar to when a person goes deaf; at first, they can speak normally, but as time goes on longer and longer and they can't hear their own speech, they forget how to pronounce words properly and develop strange speech patterns. I suspect that after a person goes blind, they probably keep their head pointed in a logical direction for a while, but as time goes on and they don't have anything to look at, their head just settles in whatever direction the muscles & tendons pull it in. (all this is conjecture based on observation; I am no expert on the subject). So how would a blind person know if he/she is pointing his/her head in the right direction? Wouldn't it make more sense to let them use the tools that they're most comfortable with (their hands)? It would be much easier too, to just write a program in a phone than designing all that hardware.

 

@strantor: I have considered different places to mount the device, even the one you suggest, on the hand.

After talking to a friend who is in med school, I think the final choice would be on the head of the blind person. Why? Because of the head-hand coordination we all have. Even if you can not see, you can still align your head with you hand. Try following a finger with your head while keeping your eyes closed. After you open your eyes you will see that you are perfectly aligned with the finger.

I'm thinking also on the scenario where a blind person walks on the street. He doesn't have to keep his hand raised all the time. Two free hands are always good.

The "glasses" are convenient because electronic boards and housing can be mounted for example instead of lenses. The components involved will be thin. Electronics could be mounted on the side shields, also. I've attached a picture with some spy glasses. Plenty of electronics there.

Now, for the technical part. I was thinking on a microcontroller capable of running Embedded Linux, an ARM core perhaps. External RAM and Flash is a must, I guess.

The image sensor will be an Omnivision. Something similar to what's on this board:
http://www.raspberrypi.org/camera

For OCR I found this: https://code.google.com/p/tesseract-ocr/
It has Linux support and can be compiled on an ARM platform apparently.

Any suggestions for a beginner in Embedded Linux? Pleas post here because the subject changed:
http://forum.allaboutcircuits.com/showthread.php?p=662094#post662094

Thank you,
This topic can be closed

 

@strantor: I have considered different places to mount the device, even the one you suggest, on the hand.

After talking to a friend who is in med school, I think the final choice would be on the head of the blind person. Why? Because of the head-hand coordination we all have. Even if you can not see, you can still align your head with you hand. Try following a finger with your head while keeping your eyes closed. After you open your eyes you will see that you are perfectly aligned with the finger.

I'm thinking also on the scenario where a blind person walks on the street. He doesn't have to keep his hand raised all the time. Two free hands are always good.

The "glasses" are convenient because electronic boards and housing can be mounted for example instead of lenses. The components involved will be thin. Electronics could be mounted on the side shields, also. I've attached a picture with some spy glasses. Plenty of electronics there.

Now, for the technical part. I was thinking on a microcontroller capable of running Embedded Linux, an ARM core perhaps. External RAM and Flash is a must, I guess.

The image sensor will be an Omnivision. Something similar to what's on this board:
http://www.raspberrypi.org/camera

Fee OCR I found this: https://code.google.com/p/tesseract-ocr/
It has Linux support and can be compiled on an ARM platform apparently.

Any suggestions for a beginner in Embedded Linux? Pleas post here because the subject changed:
http://forum.allaboutcircuits.com/showthread.php?p=662094#post662094

Thank you,
This topic can be closed

thanks, I test the tesseract ocr engine, it good and simple to use.