Page 34 of 36

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 21 May 2011, 13:49
by Shaknum
I can certainly understand the imposition of life. I too have been doing just the bare minimum of improvements with my book scanner lately, I've got bills to pay and comps to take...

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 16 Aug 2011, 08:55
by eslavko
Hi..

I just wonder if there is possibility to get height from two cameras. Ie to take stereo photo of same page with 2 cameras and then calculate height from image's diferencies. The angle betwen camera is known (or paralel). Can be that used to 'dewarp' paper?

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 16 Aug 2011, 10:28
by daniel_reetz
Yes, you can use stereo depth to dewarp a book, but unless each page has a lot of features to match, it can be tough because you'll have to recalibrate for every book. Most vision-based dewarping methods fail because there isn't consistently tons of text on each page.

Said more comprehensibly, to determine depth from stereo you need to be able to match a point in image A with a point in image B, and text can be those points. Without text, the dewarping will fail.

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 16 Aug 2011, 11:36
by eslavko
so empty pages will fail but loaded one not. This is not so bad.

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 05 Jan 2012, 00:12
by mazaheris
daniel_reetz wrote:The laser diode, mounted with zip ties, and shining on my hand. These are focusable diodes - the beam can be made to be very sharp, and it's bright enough to be usable. I'm impressed that they only cost $3 and I'm going to order a dozen more.
Daniel, could you please tell us where we can buy laser diodes like that? Are they available from any online shopping site? I have found some, but their price are really high, and I'm not sure about their performance too.

Thank you.

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 05 Jan 2012, 00:44
by daniel_reetz
Yep, sure - I bought the red ones at DealExtreme, they were great quality:

http://www.dealextreme.com/p/red-laser- ... m-5mw-5928

They no longer have the green modules listed... but in general I have had good luck buying from DealExtreme, and the prices are hard to beat.

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 16 Jan 2012, 19:54
by TomHorsley
I had a thought the other day: Why lasers? What if you had a nice physically stable glass slide with colored lines on it and just projected it on the book with a slide projector? It would be easy to make a slide with a grid for picking up finer details of the page geometry rather than just having a couple of laser lines.

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 16 Jan 2012, 19:59
by daniel_reetz
When there is a cylinder lens in front of the laser, it forms a more-or-less perfect plane in space, which makes the math simple and the physical setup simple. Whenever you involve a lens or projector, you have to deal with abberations and lens-induced distortion, which might not be bad but is definitely another level of complication.

Still, we've done basically no structured light scanning with a plain old grid; there's definitely room for it.

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 23 Jan 2012, 14:18
by HenryD
Hi I am new to this but would projecting a grid with a Microvision SHOWWX Laser Pico Projector work as a way to Sense the 3D Surface/Structure of a Book? I have one and could help test it but I don't know much about programing so would need help with that part!
anyway I hope this idea helps thanks Henry

Re: Methods To Sense The 3D Surface/Structure Of A Book

Posted: 23 Jan 2012, 15:06
by Anonymous2
I think Dan settled on two lasers because most book deformations happen near the top and bottom, where the center is linearly interpolated. The 3D grid might be a bit of an overkill in terms of processing power and resources.

The dewarping code/logic that we came up with works with anything that can map 2D image coordinates to 3D world coordinates, which your method does as well. You'll have to create code that will accurately extract the coordinates of the lines, though, which might be tricky.

Basically, as long as you can extract a 3D model or height map with your technique, dewarping will work.

We have a somewhat solid foundation to work with in terms of basic theory and code structure, so you can try working with it if you want. The project sort of abruptly died last year and hasn't had much input since, so you're welcome to reboot it again!