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

[daniel_reetz]

Anonymous2 is exactly right.

There's every reason to try out new methods, but we do have a method working that requires no projector and has working code. The cool thing about the laser diodes is that they are so cheap and don't require any input from any device (like a projector needs a computer attached).

I would love to see some images of the showWX projecting lines on the surface of a book, just to see how, or if, the laser speckle causes problems or not. Also a shot of a completely white projection for book illumination would be most awesome.

I've always wanted one of those projectors, for a few different reasons. I really hope laser powered projection gets taken up a notch sometime soon. They're all still very tiny and low power.

[HenryD]

Thanks for the fast reply I will post some links to photos tomorrow evning after work.
were is the code hosted for the 2 laser diode setup and is there a compiled softwere?

[daniel_reetz]

I think Anonymous2 had his own software, and the rest is built into Book Scan Wizard, though I'm not sure it's available in the latest release.

[Anonymous2]

My code is on GitHub, but is written in Python and requires a few modules to run.

Just as a warning, my code is somewhat functional. It isn't pretty nor will it work for arbitrary images.I might clean it up later this week, as the project is well over a year old, but I make no promises...

[CaptOn]

I'm working on my own C# command line program.

Any hints from those that have been there before?

Is Steve D gonna release the laser line detection version / code any time soon? would send a few buck his way if it would help...

[steve1066d]

I never got beyond the experimental stage. It did appear that the quality I could get from the laser approach wasn't as good as could be done with dewarping in ScanTailor.. but perhaps another set of eyes would help. I don't think I quite perfected the math to do the dewarping.. it was close, but I think it wasn't quite right. .

However, all the experimental code is checked into BSW, and released with a GPL license.

The laser code is in here:
http://bookscanwizard.svn.sourceforge.n ... ard/unwarp

Take a look at the code.. Its not the best documented code, but at least there's a few comments in it to give an idea on what I was doing.

If you have questions about it, let me know.

[CaptOn]

..

[spamsickle]

I haven't done any work at all on dewarping, but I have done away with the platen and contented myself with scanning "warped" pages for the past couple of years.

If I've said this before I apologize for repeating myself, and for letting the answer from someone who knows better slip my mind, but it seems to me that the two laser line scanning assumes the page is "cylindrically" warped, i.e., that the variance from flat is constant from the top of the page to the bottom. That may be a reasonable assumption in some cases -- it looked like Steve's video had him holding the pages at the sides, while the book was flat on the table, and that probably makes the warp pretty cylindrical.

I've noticed however that many times the warped pages I create are more conical than cylindrical, and I'm wondering if this method would still be workable. I still use a 90-degree cradle to hold my books, and two cameras to shoot, but I hold the pages down at the bottom of the book with one or two fingers. Typically, this means that the top of the book bows out more than the bottom, and the shape of the warped page would be more like a cone than a cylinder. It doesn't really matter, because I'm not shooting with lasers anyway, but I wondered what kind of results people are getting by generating a height map using two laser lines. If you calibrate from a flat table, maybe this technique could still handle more conical warping, but it seems like it would be a lot more difficult to calculate than (I assume) the height as a function of the distance between the two lines.

But what I really returned to this thread to seek, and since I don't see it, to mention, is that LeapMotion device. Apparently there was a lot of buzz about it last summer, but I only heard about it recently, which is just as well since I'm not the most patient person. It's not set to ship until May 13, 2013, so being out of the loop hasn't cost me much.

The specs say it has 0.01 mm precision, and senses over a 2-foot radius, so it would almost appear to be designed for scanning small 3D shapes like books. I'm not sure what flexibility the SDK will provide, because I just applied and who knows where that will go. I'm sure their drivers are written to detect and track hands, fingers, and motion, but it should be possible to get 3D information somehow before all that post-processing happens; at least, I hope so.

[Shaknum]

I have used the Leapmotion and unfortunately it won't presently work for our needs. Its main purpose is to track the movement of objects (fingers, hands, etc...) in 3D space. The SDK provides not access to point cloud data, and the demo with something looking like a point cloud seems to be something else. So for now we don't know that this will ever be able to provide point cloud data, but it is very fast and accurate for tracking motion.

[Shaknum]

By the way, this new product called the Duo, http://www.kickstarter.com/projects/codelabs/duo-the-worlds-first-diy-3d-sensor, may be more suited to our usage since it is DIY and very open. I am not sure if it will provide point cloud data, but some statements on the kickstarter page sound like it does/will. Also, we don't know if it can be modded to work with higher resolution cameras.