Portable Banknote Recognizer

[atmega64]

Dear dougy83 and ronsimpson

Thank you for your really good helps.

Here you can see Iranian money:
http://banknote.ws/COLLECTION/countries/ASI/IRN/IRN-CBIRI.htm

ronsimpson, Could you please give me more information about "color scanner in a pin"? I wondering what exactly do you mean by "average color"? I am not so familiar with image processing.

Unfortunately, I could not summarized different methods yet, but I think using of a color sensor is a good idea. User can sweep the money with a small portable device (similar to a pen that has a color sensor on top of it). At the moment, I am just looking for an algorithm to find the dominant color. There are different methods i.e. neural networks but I prefer simple ideas.

 

[Misterbenn]

image processing

we're not suggesting image processing, we're not using a camera to take a picture and then analyzing it. So no need for complex image processing or neural networks.
For color sensing I would use something like this https://www.adafruit.com/products/1334?gclid=CKX6397YqMQCFYQlgQodsmcAug
Which gives a digital output of color and will interface easily with development boards such as an arduino.

 

[KeepItSimpleStupid]

FWIW: Could you insert the bill into something and look for a particular color at one to 4 spots? Maybe one or two denominations might get the same color and then you might be "prompted" to insert the other side? Or you insert the bill on two edges. These are just ideas.

Anyway: Averaging is just that. e.g. Red intensity divided by number of pixels, but it's possible that a "histogram: would be a better choice. e.g. the % of samples that have similar RGB values.

Calibration is always an issue.

 

[ronsimpson]

I think the finished product is like a pen (small hand held) with no moving parts.
The prototype can be bigger.
A color sensor is small.
I see a color sensor to look at money. + White LED to light the money. + Switch to know when the product is pushed down on the money.
Push down on the money, LED turns on, sensor starts reading. Move across the money. Lift up to stop.
Measure the color maybe 50 time/second. 50 to 500 readings.
Most of the color sensors have Red, Green, Blue outputs. If we have 100 readings, add up and divide/100 to find the average.
Now we might have numbers like R=123, G=234, B=54. But very old dirty money might read 1/2 as bright. R=61, G=118. B=27.
So maybe we need to find the ratio between R, G, and B. R=1, G=2, B=0.5

For testing the sensor can be connected to a PC. Many scans must be taken. The question is can you get good data.
Do all "1" give numbers that are close together and different than "2" and different than "5".

This is a very simple test. It you do not get good data I can find better tests that require more computer power.

 

[atmega64]

Dear KeepItSimpleStupid and ronsimpson,

I am very appreciate for your helps.

ronsimpson, Actually your approach is very close to mine. I think in this stage, It is a good idea to start building a prototype (primary version) and as you said, we should connected it to a PC for collecting data. Then we can analysis them to find a good decision making algorithm.

Misterbenn, Could you please give me a little more information about the logic behind of your device? Is that something similar to ronsimpson (post #24 of this page)?

 

[Misterbenn]

Misterbenn, Could you please give me a little more information about the logic behind of your device? Is that something similar to ronsimpson (post #24 of this page)?

Yes its we are talking about the same idea.

The device i posted gives you the intensity of colors, Red, Green, Blue. This will interface with Arduino https://www.arduino.cc/ development boards so you could build a prototype very quickly

 

[ronsimpson]

Post #15 and #22 show color detectors.
Any digital sensor can be connected to a pc.
It might be easier to use a arduino in between the sensor and the PC.
If you are looking at the arduino you should also look at the raspberry pi. There are several versions. The lowest cost is best in your case. The raspberry pi can use a small camera to look at the money.