Jump to content
  • Advertisement

Archived

This topic is now archived and is closed to further replies.

Lab Rat

Stereoscopy

This topic is 5546 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

This may not be the perfect place, but I know most of the people that post here are pretty damn smart... so. I''m working on heightmap generation through stereoscopic images. This means that I have 2 images (camera centered on same point in space) with approxiamately 3.5 degrees of offset from normal (total of 7 degrees of difference). From these images, I have successfully extracted edge information and have mapped relevant points in the images to vectors in 3space (transformed by the appropriate plane of projection). I''m now trying to analyze the lighting circumstances to determine proper (relative) depth. Is there a good way to do this? Right now I assume that an edge denotes a change in sign of slope of the depth, i.e. dL/d(x,y) = 0. This is fine for extremely simple geometry, but for curved objects with poorly defined edges, I can see this becoming an incredibly difficult problem. Also, I''m trying to use a linear scale of depth vs. light intensity, but that doesn''t seem to pan out properly mathematically. Any insight you guys might be able to give would be GREATLY appreciated. P.S. The end-goal of this is to be able to construct high poly, accurate models of complex scenes based on a pair of proper images. This means that 3-D terrain can be accurately constructed from something like satellite photos. Anyway, thanks again.

Share this post


Link to post
Share on other sites
Advertisement
You're not going to get a neat mapping of intensity to depth unless the object you're looking at is all the same material.

If you've got a stereo photograph, you might want to think about identifying a feature in each image and comparing their positions in the photo. The amount of displacement will to relate to the depth. This is generally much easier. You can read about this at http://link.springer-ny.com/link/service/journals/00138/bibs/8010005/80100280.htm [EDIT: bah, you need a login for that. Anyway, you should be able to find the journal in the periodical section of any decent college or university.]

If you want to continue with shading info (which is a very difficult problem) check the machine vision literature. 'Shape from shading' is the keyword you want to search for.

[edited by - cheesegrater on April 18, 2003 1:50:16 PM]

Share this post


Link to post
Share on other sites
For now, the assumption that the material is constant is valid. Until I'm comfortable that it can be done quickly and accurately enough, I'm using simulated data (using OpenGL to render a pair of images with tight control over variables). After that, I will be generating the images via grayscale photography using sculpted models.

I don't know if this will ever actually be able to do multiple material processing... or if it will ever need to.


[edited by - Lab Rat on April 18, 2003 1:58:34 PM]

[edited by - Lab Rat on April 18, 2003 1:59:17 PM]

Share this post


Link to post
Share on other sites

Ah, a lab setup. That''s different, then.

In that case, you want to get yourself a nice close to lambertian material (minimal specular). I don''t know much about sculpting materials, but AFAIK something like play-dough ought to work. Light your scene with a single light. That''ll simplify things immensely.

The intensity will directly map to the dot product of the light vector and the normal. By comparing neighboring normals you can come up with curvature, and using a little basic calculus or trig you should be able to come up with a depth map from there.

Definately look up shape from shading papers. They''re solving the problem in more complex setups than you seem to be, but the basic ideas are the same.

Share this post


Link to post
Share on other sites
You can find an interesting paper on shape from shading at this address :

Shape from shading and viscosity solutions :
http://www-sop.inria.fr/odyssee/research/1/index.en.html

Shape from shading is often used when you have only one image from a scene, but if you have 2 or more images, there are others methods to fully reconstruct a 3d model from those images.

Several methods are presented in this research project :
http://www-sop.inria.fr/robotvis/robotvis-eng.html

You can look at these pages for demos :

3-D Shapes from Stereoscopic Images :
http://www.inria.fr/robotvis/demo/diffprop/

PDE''s and the Stereo Problem :
http://cermics.enpc.fr/~keriven/stereo.html

Share this post


Link to post
Share on other sites

  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

Participate in the game development conversation and more when you create an account on GameDev.net!

Sign me up!