Machine vision: shadow measurement technique using a LED

[hassou7]

Dear all,

I have been working on a small project in machine vision. From a telecentric collimated illumination using a high power LED with a small etendue, a round object to measure covers an amount of this light and is projected on a CCD camera creating a shadow. From the pattern of intensity distribution on the CCD, I should have the dimensions of my object knowing the size of the pixels, and the distance from object to CCD.
The problem is the transition form dark to white on the CCD: in fact, to detect the edges of the piece I want to measure, I should compute a threshold based on Fresnel diffraction. This threshold is really hard to determine since it is based on partially coherent light (LED).

Is there anyone that have been working with this kind of technique? That would be really nice. Well thanks in advance.

I am new in this forum, I will try and have a look around to help as much as I can do.

Best,

 

[IRstuff]

Seems to me that you're way overthinking the problem. If you need that much fidelity in your model, you're using the wrong tool. A measurement of the light to dark transition of a known dimensioned object should be adequate.

TTFN
faq731-376

 

[hassou7]

Thanks for the answer. In the attachment, you will find a draw of the method. As I am looking for an edge detection to measure my piece, the pattern of the projected light on the linear CCD is shown in the same attachment. This pattern is clearly a Fresnel diffraction one.
Normally, with a coherent light, the threshold is always 0,25. With a totally incoherent bulb illumination, I got a value of 0,5.
Now, since that I am using an LED, a partially coherent light source, the threshold value is osciallating between 0.25 and 0.5.. And the hardest part is how to compute it exactly to not have major errors. I understand that a proper processing for edge detection could give some results, but I will need a threshold value anyway.

 

[DanStro]

I can't say that I've ever had to do this but maybe this article will help.

http://www.radiantzemax.com/kb-en/Knowledgebase/Performing-Partially-Coherent-Image-Analysis

HTH,
Dan

Han primo incensus

 

[hassou7]

Dan,
Thank your for this article, it is really good. However, I still have the same problem in getting the exact value of the threshold.

 

[IRstuff]

You've not stated how big this object is or how accurately you need to make this measurement. If this approach is so difficult, why are you using it? I've made submicron measurements with a standard optical microscope, without have to deal with shadowing, so unless you are making nanometer level measurements, this approach bites.

TTFN
faq731-376

 

[hassou7]

Well the idea is to have a system that measures diameters from 100 microns to 10mm , it should be costless. Considering this method, it is the easier way to achieve if I deal with a halogen lamp (threshold will be 50% in that case): a telecentric backlight illumination which is simple to calibrate. An object space telecentric imaging system and a CCD camera. The only problem left is to compute the transition Black-White on the CCD with a LED. The camera has 7450 pixels of 4.7 microns each. Having a precision near the micron would be a very good thing (subpixels). That is why the threshold is the most important part.

I hope that it is better explained that way. Thank you by the way.