Slashdot Mirror
Build a Rotisserie Scanner With Legos
WalkingBear writes "All you 3d geeks out there should take a look at this. This guy has built a 3d scanner (scans 3d objects resulting in a 2d cylindrical image map) out of a flat bed scanner and Lego. Also has a turntable style for use with digital cameras."
That inner page of his is like 660KB all in, so I can see this guy's server taking a crunch real soon. His JPEGs were uber-unoptimized, so I've optimized them and put up a temporary mirror so you can all see the joy that is the rotissery scanner :-) It will disappear in a few hours or so.
Rotissery Scanner Mirror
The funny thing is, if you read the article you'll see that he's gotten better results by just stitching digital photos together. The scanner has actually given him rather poor images (he's got a nasty light leak), and you need to be able to put the thing you're scanning on a spit...
There is some other software available, which I can't be bothered to look for right now. It works like this: say you have a few pictures of an object, from several known angles. Then the program allows you to define common points and edges on each photo. Since the program knows what angle each photo was taken from, it can do the math to build a 3D mesh of the object.
Don't remember what it was called, but I used it a few years ago. Someone else get some karma and Google up a link.
...
That's the only one I know of.
...could deproject the scanned image to represent the object from an arbitrary angle, or create an ultra-smooth animation of it rotating about the scanned axis. It's the exact analog of the way a QTVR cylindrical panorama is rendered into a window, only there you're inside the object instead of outside.
This technique could be incredibly useful for creating photorealistic views of 3D objects from any angle about one axis.
Wish I'd thought of it. Now how long until IPIX patents it?
There are two that I know of:
Photomodeller [www.photomodeller.com] - where you specify verticies and edges on a series of photos taken from different perspectives. It will generate the 3d mesh and map the images onto the faces -- good for regularly shaped objects in particular.
The other is D Scupltor from D Vision Works [www.d-vw.com] where you take photos from regular angles of an object (using a rotating turntable such as he has built) and reference markers in the image. The software will then generate a complete mesh and map the images to it.
Well, it's not just in the labs anymore. Some Canadian guys over at Braintech have been able to accomplish 3d robot vision, and it's being used largely by car manufacturers.
http://www.braintech.com/sc3d_overview.html
// file: mice.h
#include "frickin_lasers.h"
at this link. Check his mathematical lego sculptures. It was covered in a slashdot article too.
Sure you could wrap the image around a perfect cylinder, but without knowing the "stetch factor" at the top and bottom, you wouldn't know how close to the center of the cylinder to render that point.
:)
Think of it like rendering a sky in games like doom/quake/half-life/etc. The sky is really just a very large box with a sky texture applied to the inside. You as a player are inside the box so you don't realize it's a box. Once you go outside the box, it becomes painfully obvious that there is no sky, just a box.
QTVR is inside the cylinder, but with the scanned image, you're outside
no comment
If you look all the way to the bottom, he takes that creepy warped skull image and wraps it on a 3D model to make something that almost looks like a 3D skull. It looks kind of crummy, but it's decent for a start, and much better than anything I've done in the same field (absolutely nothing).
He never explicitly says what his purpose in all this was (although he claims inspiration from the Matrix), but I guess he's wanting to use this to make it easier to get textures for 3D objects based on actual objects.
The scanner in no way reads depths though... the 3D model he uses in the end is NOT determined by the scanner. That would take a hell of a lot more work, and probably lots of those 2x1 blocks that always seem to run out.
-"One machine can do the work of fifty ordinary men. No machine can do the work of one extraordinary man." -EH
Actually, couldn't you use Gimp's Map to Object -> Sphere thing to make it look like the object is spinning? You would have to make a square selection, map it to the sphere, save it as a layer, then move on to the next square selection. Though you'd have to somehow fix the distortion caused when stretching the image to map it to the sphere...
You'd probably want to make yourself a script for Script-Fu to do it for you.
There is REALVIZ ImageModeler.
The technology of deriving 3d geometry information from 2d images is typically known as image-based rendering. Gabe Newell (head of Valve Software) said that this is one of the technologies they didn't have enough time to get into Half-Life 2. It was first demonstrated in The Campanile Movie and has since been used for special effects purposes in movies (including the Matrix). It is very impressive, not only for its ability to generate the geometry from the images, but also its way of generating perfect distortionless texture mapping for said geometry.
Such software exists. It is called Rotomapper, and can be found here, hosted on my website. I did not make the original application, and google turns up nothing on the filenames or application title. I just uploaded it now. I don't remember where I found it.
A home-made 3d scanner does not seem unfeasible. It would require having a computer-controllable laser range finder mounted on a vertical post, which is connected to a horizontal post. The laser could slide vertically on its post while the vertical post could slide horizontally on the horizontal post. Motors/pulleys/pistons would be integrated where necessary. In this way the laser would be able to move in a local XY plane.
All of these electronic components would be computer controlled, with the software controlling the laser's XY position. The software would increment the laser's position in both the X and Y directions, and find the distance (using the laser range finder) between the XY position and a point perpindicular to the XY plane on the object being scanned. That distance would provide the Z coordinate. The computer would store each set of XYZ coordinates and then generate a mesh from it.