Quote:Original post by angrytofu
You seem to be into this subject yourself. Im not that smart to go create my own program to do this unfortunatly.

Laser scanning using the method I implemented is easy... you need to understand trigonometry since everything is calculated with a right triangle. You also need to know how to open an image and apply some kind of threshold to find where the laser is in the image.

Quote:Original post by Talroth
I'm also sure you could get some rather high res meshes if you used the same concept for the modeling as David, but place your model on a precision turn table, and give it a 1degree turn each time you rescan.

That's exactly the problems with laser scanners [wink]. All calculations can easily be simplified to simple triangulation. Once you get the 3D data you might need some filtering and others things like that. My code is about 600 lines (using FreeImage and Glut32, counting all the blanks, comments, debug output, and 3D rendering of the point cloud) and it completely reconstructs a point cloud from the images (I even built it multi threaded). The real problem with laser scanning is the "hardware" part.

Lasers might be expensive, I tried using a laser level but the laser line is just too large (will try to reduce it manually) and in most cases I lose precision and my reconstruction looks more like noise.

Precision in all things you measure is necessary otherwise you end up with a precision of 1-2mm easily (which is not good for small detailed surfaces).

Here is a non-detailed surface a "curvy" wood cylinder (I don't know better word for that in English, sorry) on a "cool whip" plastic bowl with a 1mm precision, image every 5 degrees:



And here's the reconstruction of an angel figurine that covered by small flowers and many other small details (sitting on the same kind of plastic bowl):


I guess that my next step after getting a thinner laser will be precision [smile] in my measurements and also in the rotation. I've been thinking about using a target in the background for calibration (like the DAVID application uses) to make everything more exact.

JFF

[Edited by - jff_f on February 27, 2008 5:17:51 PM]

Quote:Original post by Talroth
jff_f, first, fix your quote. I hate being quoted for something I didn't say, especially when the words Not and Smart are in close proximity.

Sorry for that. I didn't mean to... it was a copy-paste error...

Quote:Original post by Talroth
Second, do you have time to post the image of a single scan of the angel? Then post another image with four or five passes of the scanner? (and put each pass as a different color?) Might help confirm where your problems are.

Well, first that angel is 800 km where I am [smile] so I can't make better pictures with it.

I have an exam tomorrow morning I still got a lot to do since then. I'll put that on top of my TODO list since it will take couple of seconds (maybe I'll take a break from reading later and do it then). I have 72 passes, I'll try to find a good way to show them. Each pass represents the data that could be extracted from a laser line.

Also because the laser line is too large I got other precision problems that I haven't mentioned. You should find the laser line only once in every line of the image but with the complex object (angel, specially with the wings, see image below) we can see it twice in the image. That gave me problems with the reconstruction as I'm taking the middle value for now.



Anyway, like I said I'll post better screen shots later when I have more time.

JFF

Well... trying to put colors showed a bug! Wasn't doing the rotation correctly... I was using sin() and cos() with an angle in degrees! Didn't expect such a mistake from me!

Now the whole thing looks a lot better! I can even distinguish the wings now [wink]. And I remember that the laser wasn't going up to the head of the angel.



Can see the wings:


Details on the flowers covering the base of the object:


A picture of the original thing (the images used for the reconstruction we not taken in the same conditions):


Now we can see that this has "correctly" reconstructed except for some scaling (because the calibration data had to be estimated since I accidentally lost it).

I hope you can "see" the 3D correctly... showing a point cloud in images is hard. Also hope that the colored passes helps you (I have 6 colors and they loop over the whole 360 view).

In the last picture you can also see the way things are setup. Camera mounted at the base of a T. Laser and objects would be on the top line. Laser and camera pointing to the object. Now I need something I can rely on for the object rotation.

JFF