[photo-3d] 3D and stereocorrelation

["Greg Camilli" <gac@xxxxxxxxx>]

Some time back I posted a query about stereo correlation, and its use in
deriving 3D coordinates from stereo pairs.  A few people emailed me
privately for information, and here is the result of my investigations.
First, standard photogrammetry techniques can be used to compute
z-coordinates from registered photographs providing a suitable array of
visible ground control points.  Thus, it would seem that in theory, 3D
models for close-range photographs could be derived from stereo that
"simulate" the kind of overlap used in aerial photography -- about 60%.

However, as the curvature of the surfaces increases, it seems doubtful
whether this approach by iteself would lead to accurate models; mapping is
related to but very different from modeling. What may be possible, however,
is to take a number of stero pairs that cover the surface of the object in
question, derive 3D point clouds, and then stitch them together.  As a
final, step, one would have to compute surfaces over the point clouds.  At
least one high-resolution laser scanner (using time-of-flight, not stereo
corelation) comes complete with the software to capture point clouds , stick
them together, create surfaces, and integrate color information (or digital
images) with the surfaces.  Sounds good?  Yes, but how many of us have a
spare 130K-160K?  There is another option that appears to be of interest.  A
company in Vancouver called Point Grey has come out with a camera they call
the Triclops -- you got it: three lenses on a camera that are calibrated
with respect to each other.  After snapping the 3-part image (left, right,
top), it is tranferred to a laptop and the sterocorrelation algorithm begins
producing a 3D point cloud.  The cost of this system (called Digiclops) is
about 5K-6K, but we have not yet tried it out -- so I cannot relate
information on how well it works under various circumstances.

I hope this is helpful to those who inquired, and I would appreciate
receiving any further information that you might have.

Sincerely,

Gregory Camilli