Several distortion questions

Main Page | Report this Page

Science Forum Index » Optics Forum » Several distortion questions

Page 1 of 1

Author

Message

aplus

Posted: Wed Nov 05, 2003 10:25 am

Guest

Hi all,

We have to measure object shape, object window 6x4 mm, with an accuracy of
+/- 1µ.

We tried two approaches:
1&2 ) Lighting with collimated light
1 ) Camera with telecentric lens (high unknown distortion)
2 ) Camera without lens (blurred image )
1&2 ) Sub pixel computation.

Approach # 1 (TELECENTRIC LENS ON CAMERA)
--------------
What can we expect, in term of distortion figure, from a telecentric lens
magnification 1x.

Edmundoptics offers a telecentric lens with 0.5% max distortion.
Can we expect lower distortion at an affordable price?

If no, we have to works the problem the other way : distortion correction.

Again, edmundsoptics offers distortion targets but the accuracy is only 4 µ.
Is it possible to find distortion targets with accuracies better than 1 µ?

Approach # 2 (NO LENS ON CAMERA)
--------------
What can we expect in term of pixel center accuracy on a CCD sensor?

We used collimated light from a LED? Do someone know some tricks to limit
blurr, polarization, monochromatic light with interfernce filters...?

We tried collimated light from a laser diode using home made optics. we get
mostly speckles. Can we expect good images from a high quality collimated
laser diode?

Regards,

Alain Riaud

Phil Hobbs

Posted: Wed Nov 05, 2003 11:47 am

Guest

aplus wrote:

Quote:

We have to measure object shape, object window 6x4 mm, with an accuracy of
+/- 1µ.

We tried two approaches:
1&2 ) Lighting with collimated light
1 ) Camera with telecentric lens (high unknown distortion)
2 ) Camera without lens (blurred image )
1&2 ) Sub pixel computation.

Approach # 1 (TELECENTRIC LENS ON CAMERA)
--------------
What can we expect, in term of distortion figure, from a telecentric lens
magnification 1x.

Edmundoptics offers a telecentric lens with 0.5% max distortion.
Can we expect lower distortion at an affordable price?

0.5% is pretty good. You can do better, but you'd have to decide on
what's affordable.

Quote:

If no, we have to works the problem the other way : distortion correction.

Again, edmundsoptics offers distortion targets but the accuracy is only 4 µ.
Is it possible to find distortion targets with accuracies better than 1 µ?

Sure. But you're talking about a lot of pixels--do you have enough?

Quote:

Approach # 2 (NO LENS ON CAMERA)
--------------
What can we expect in term of pixel center accuracy on a CCD sensor?

It's normally excellent, being lithographically defined. CCDs with
microlenses may not be quite as good. But your measurement isn't well
thought out (see below).

Quote:

We used collimated light from a LED? Do someone know some tricks to limit
blurr, polarization, monochromatic light with interfernce filters...?

Trying to do this by detecting the shadow is going to be very hard.

Small changes to the shape of the corners and edges, not affecting the
outer dimensions, will change the edge diffraction patterns
dramatically, and screw up your measurement.

Quote:

We tried collimated light from a laser diode using home made optics. we get
mostly speckles. Can we expect good images from a high quality collimated
laser diode?

Speckles are a consequence of the high coherence of the laser, and will
always appear when a laser beam is scattered off a rough surface. The
only exception is when the beam is focused very tightly (NA > 0.7 or so).

Trying to get 1 part in 6000 accuracy by number-crunching on a
garden-variety CCD image is a complete waste of time, on a par with
trying to learn to fly by flapping your arms. CCD pixels are usually
irregularly shaped, and have nonuniform sensitivity across the pixel
area. Thus most sub-pixel interpolation is doomed, unless the camera is
*very* carefully characterized and the image is sampled at several
points in each resolvable spot, as is done in astronomy. Just the moire
pattern between the grid and the pixels will make your calibration as
flaky as can be. The magnification of your lens won't be athermalized
to this level, either.

A thinned, back-illuminated CCD is one way to handle this, but there's a
class of much easier methods. The easiest of all is to get a decent
encoded XY stage, which can do a lot better than 1 um, and attach the
CCD to a microscope lens. That way, you can look just at the corners of
the window, one at a time, and between the video measurement and the
known stage position, you can effectively get 100,000 pixels per line
instead of 1000 (on a good day).

Another approach, if moving the sample isn't allowed, is to mount four
lenses where you expect the corners to be, and (with folding mirrors, so
you can get enough room) attach four cameras, one to each corner.
Because the relative separations of the lenses can be calibrated
accurately, you can get high accuracy in the same way as above.

Personally, I'd use the stage scanning method if I possibly could.

Of course, if your budget is small, you probably aren't making that
many, in which case a good micrometer should do the job just fine.

Cheers,

Phil Hobbs

Helpful person

Posted: Wed Nov 05, 2003 1:18 pm

Guest

"aplus" <aplusVOTREDENTIER@wanadoo.fr> wrote in message news:<bob4ss$qnp$1@news-reader5.wanadoo.fr>...

Quote:

Due to symmetry a lens operating at unit magnififcation will have zero
distortion if it is a symmetric design. (A symmetric design is nearly
always advantage for a 1:1 lens.) For your application the lens needs
to be telecentric. Otherwise any small amount of defocus will affect
image size.

The accuracy you are looking for of +/- 1 micron over a field diagonal
of 7.2mm requires a resolution of 1000 line pairs per mm and the
equivalent of 14400 resolution points. I'm not sure how you achieve
this resolution in a practical configuration with a suggested 1:1
lens.

You may want to look closely at sub pixel computation or other
solutions.

One method that may work is to use a confocal microscope with stage
scanning. Sub micron resolution then becomes easy. You can then use
a two axis stage with the sub micron accuracy you need.

Michael

Posted: Wed Nov 05, 2003 2:09 pm

Guest

within my machine vision experience, using sub-pixel methods, it's only
feasible to get about +/-0.2 pixel resolution --and that assumes good
lighting & lenses, etc.

I'd suggest that what you need isn't going to happen with a "standard"
camera. Even using sub-pixel methods, to get .1micrometer resolution in a
6mm field, you'd need something on the order of 3000 pixels -- which is
significantly better than the resolution of the best commercially available
machine vision cameras. You may be able to source something rare with that
big a pixel count, but I've never heard of it--and it certainly won't come
cheap. It's conceivable that you might be able to get into the right
ball-park with one of the new 4 mega-pixel still cameras, but it'd involve a
non-trivial amount of custom engineering.

You're going to have to come up with something fancy to make this work...

"aplus" <aplusVOTREDENTIER@wanadoo.fr> wrote in message
news:bob4ss$qnp$1@news-reader5.wanadoo.fr...

Quote:

Hi all,

We have to measure object shape, object window 6x4 mm, with an accuracy of
+/- 1µ.

We tried two approaches:
1&2 ) Lighting with collimated light
1 ) Camera with telecentric lens (high unknown distortion)
2 ) Camera without lens (blurred image )
1&2 ) Sub pixel computation.

Approach # 1 (TELECENTRIC LENS ON CAMERA)
--------------
What can we expect, in term of distortion figure, from a telecentric lens
magnification 1x.

Edmundoptics offers a telecentric lens with 0.5% max distortion.
Can we expect lower distortion at an affordable price?

If no, we have to works the problem the other way : distortion correction.

Again, edmundsoptics offers distortion targets but the accuracy is only 4
µ.
Is it possible to find distortion targets with accuracies better than 1
µ?

Approach # 2 (NO LENS ON CAMERA)
--------------
What can we expect in term of pixel center accuracy on a CCD sensor?

We used collimated light from a LED? Do someone know some tricks to limit
blurr, polarization, monochromatic light with interfernce filters...?

We tried collimated light from a laser diode using home made optics. we
get
mostly speckles. Can we expect good images from a high quality collimated
laser diode?

Regards,

Alain Riaud

Ron Gibbs

Posted: Thu Nov 06, 2003 8:24 am

Guest

"aplus" <aplusVOTREDENTIER@wanadoo.fr> wrote in message
news:bob4ss$qnp$1@news-reader5.wanadoo.fr...

Quote:

Cameras have insufficient resolution for the task; scanning solutions may be
possible. But don't expect cheap - your resolution requirement is
uncomfortably close to the wavelight of light!

regards,
Ron Gibbs

******************************************************
Gibbs Associates
The Optical and Laser Scanning Consultancy
www.gibbsassociates.freeserve.co.uk
******************************************************

Page 1 of 1 All times are GMT - 5 Hours The time now is Fri Jan 09, 2009 8:02 am Science Forum Index