13,400,969 members (41,086 online)
See more:
I'm working on my WirmForms project.
I have 2 2D images from different angles how to reconstruct it in 3 dimensions?
Posted 4-Feb-13 2:38am
Updated 4-Feb-13 3:26am
v2
aspnet_regiis -i 4-Feb-13 9:05am

Thats a very easy question... But the answer is tooo difficult... You will have to write a complete project in image processing to do that....
Marcus Kramer 4-Feb-13 9:45am

If you succeed at this project and if you can share algorithms and such, this sounds like it has the potential for a fantastic article.
Mich_90 4-Feb-13 10:12am

Marcus ,if I will solve the problem.I will share solution.
Marcus Kramer 4-Feb-13 10:17am

Thanks. I'm sure many of the geeksmembers here will be pleased with that. Cheers.

Good luck, Marcus, but I'm afraid you might need to wait for long. :-)
Especially taking into account that such problem is theoretically proven unsolvable in general case, and where it is "solvable", it does not mean "solution" in traditional mathematical sense of this word.

This is a very interesting topic. I've put some information in my answer. Please take a look, this is really interesting... :-)
—SA
Mich_90 10-Feb-13 12:03pm

To start with: this problem does not have a general solution. Strictly speaking, it is not solvable at all. It might only have so called "quasi-solutions" which are only possible if you have some powerful preexisting assumption on the expected shape of the body.
—SA

## Solution 1

Please see my comment to the question.

It is well-known that such problem does not have a general solution at all, at least, not a solution in the traditional mathematical sense of this word.

There is a branch of mathematics devoted to some classes of so called ill-posed problems. Most typically, ill-posed problems appears from the formulation of different kinds of inverse problems:
http://en.wikipedia.org/wiki/Ill-posed_problem[^].

The model problem of this class, closest to the problem which OP is interested in, is the problem of tomography:
http://en.wikipedia.org/wiki/Tomography[^],
http://www2.imm.dtu.dk/~pcha/Tomo.html[^].

Even though the problem of tomography looks like and extreme hi-tech (and it actually is), the seemingly easy problem of reconstruction of the 3D model by 2D projections is not easier at all.

The formulation for a wide class of ill-posed problem was formulated by Jacques Hadamard and the approach to such problems was developed by A.N.Tikhonov. According to this approach, the ideas of traditional "solution" is given up, and the concept of "quasi-solution" is introduced. For example, if some quasi-solution is found and describes some 3D shape which really produces the correct set of projections used as the input data of the problem, some other 3D shape can be found, which produces exactly the same set of projections. None of such solution can be called a "real" solution, because an ambiguous is normally required. In practical applications of the quasi-solution, some preexisting knowledge of the expected shape is used. For example, a set of tomographic images of a brain can be resolved in a set of different 3D images, but the software produces only the 3D model of the brain which conforms to preexisting knowledge of how a human brain is generally shaped.

You can get better ideal on this extremely difficult topic if you read some books/articles. Please see:

Most article needs special academic or personal subscription, so they are no easy to get.
Try to find some: http://bit.ly/11KWV51[^].

Yes, this is an adequate search criteria. If you try to find something using your formulation of the problem, you won't find anything.

—SA

## Solution 2

Here is the solution on open cv site:
http://opencv.jp/opencv-2svn_org/c/calib3d_camera_calibration_and_3d_reconstruction.html[^]

http://sourceforge.net/projects/emgucv/[^]
the name of the project is Simple3DReconstruction.

Top Experts
Last 24hrsThis month
 OriginalGriff 180 Maciej Los 160 phil.o 139 CHill60 105 John Simmons / outlaw programmer 100
 OriginalGriff 5,079 Maciej Los 2,800 Richard MacCutchan 1,783 CPallini 1,685 RickZeeland 1,085