3D-Coordinates from depth maps

[bgo]

Please help. I'm a researcher and want to use Maya to generate test sequences with ground-truth depth maps. They are only useful when I can compute the exact location of a 3D point from pixel coordinates + depth. Additionally, they must depend only on camera parameters (focal length, fov, far&near clip plane).

I now have the following problem: The depth maps Maya writes are always greyscale images covering all values from 0 to 255 (at least after I convert them from bloody IFF to some format I can read with QT). Conclusion: One has to know the location of the nearest/most far away object point of the scene which is visible in a camera image in order to calibrate the mapping which computes coordinates from depth. The fact that the way depth maps are constructed depends on the objects in the scene is totally absurd IMO. It is certainly quite different from the way the OpenGL Z-Buffer behaves.

Does anybody know how to create more sensible Depth Maps with Maya? I have a deadline for a paper in one week and would really be grateful for any advice. Also, a formula how to compute the exact Z-Coordinate (in camera coordinates) from the depth map value would be quite welcome.

Thanks in advance,
Bastian.

[dobieag01]

A little tip on depth maps is that you need to specify your near and far clipping planes. An easy way to do this is to have 2 locators one for your near and one for your far clipping planes. Then specify that the Z value (or whatever direction is normal to the camera view you are using) of your near clipping locator specifies the value of the camera near clipping plane and similar for your far clipping plane. An easy way to do that is to write an expression such as :
camera1.nearClippingPlane = locator1.translateZ;
camera1.farClippingPlane = locator2.translateZ;

Now you would be able to calculate the height of the terrain from the image b/c you can start with the Z value of the far clipping locator to be 0 and black on the image. Also you can say that the near clipping locator Z value is your heighest terrain height and white on the image.
I should have specified that you're going to set the far clipping locator to the lowest point of the terrain and the near clipping locator is the heigest point of the terrain. The closer you have these locators to the min and max of the terrain, the better the image will be. Remember that since the image only holds values per pixel, you're going to get some interpolation errors depending on how high resolution you're working with.

Another thing you need to do is open the attributes for the camera you're rendering w/ and turn off the auto clipping plane determination.

I hope I answered your question, or at least help you understand an easy way of specifying your z-depth settings.

Later,
Dobie

[joojaa]

Actualy the auto clipping planes are realy usefull most of the time. But not in your case fortunately you can disable it! under the camera settings as above.

but the idea behind auto clipping planes is to utilize the z buffer to its fullest so that mnimal errors appear!

use 16 bit files! also the dpth maps can be querried as floats. And cam cordinates can be querried rendertime using sampler info. And if you make a custom format you can output directly the info you want with internal precision!

but what do you do with this info-