IBMR/Photorealistic 3D models from Photography

I’ve been thinking about Paul Debevec’s
Campanile Movie and the Fascade software that they wrote to extract 3-d models
from conventional still photographs. I read some of his papers and did a
bunch of back research on related things like
Canoma, Image Modeler, Photomodeler and ArcheoVIZ.

It seems to me that the Fasade software is pretty spectacular in
what it can do for modeling of environments or urban landscapes.
Few of the related softwares I mentioned above do quite what
Fascade does. It also seems to me that it is not-unthinkable that
one could jerry-rig Blender to behave like Fascade with appropriate
plugins and modifications.

The core mod would need to be “image warping,” the projection of a
2-D image back on to a 3-D model to use still photographs to paint
textures onto the 3-D elements. There are references to this in all of
the image based model rendering (IBMR) sites. It’s sort of like regular
rendering, only backwards. You use the photographs of an object and
a 3-D model to find the textures on an object rather than using
textures and a 3-D model to calculate an image of the object.

That would be a major piece of 3-D rendering type mathematics
that I am probably not up to. The ArcheoVIZ program has source
code for it, though. There are something like camera calibration
and fitting a constrained set of 3-D model parameters to the
images that I might be up to. There would need to be some
GUI enhancements because you need to be able to mark off
points and edges in all of the still images. The main problem
would be the image warping, though.

Does anyone have any thoughts or hacks?

Free photogrammetry in Blender? That would be great! :smiley: :smiley:

Especially since Photomodeler Lite is no longer available. :frowning:

Koba.

If anyone is still interested in developing this idea for Blender you should take a look at this…

[>] http://www.vreel-3d.de/plugins/PhotoMatch/links.html

I think it uses Cinema4D’s coffee scripting language to find the vanishing points of a photo. Placing the ends of 4 splines helps find the location of the camera.

When I saw their documents (tutorial/userguide) it became clear that this should be possible in Blender with python - unfortunately I’m no programmer. Just a lowly designer. But it would be great for the Blender community to have this feature - even tho it may only work with buildings or other objects with some right angles.

Any takers?

Look at this!

:o
THAT would be cool.