[photo-3d] Re: Algorithm wanted

[Bruce Springsteen <bsspringsteen@xxxxxxxxx>]

I like Ken Luker's algorithm, mostly because I really dig numbered lists!

> 1. Find the most distant element that is in both photos (infinity if
possible).   Use that
> point as a common rotation center. 

This would be an excellent beginning, however I'm not assuming that we
know which points are more or less distant, only that we can recognize
homologous points in the two prints, and that we know which print is
which.  We can't free-view to find out, because that would necessitate
knowing how to orient the two images, and that begs the question! ;-)

> 2. Fix one photo in place and put a straightedge on that photo passing
through the chosen > distant point.   Note the point where the left side
of that first photo intersects the
> straightedge. 
> 3. Put a straightedge on the second photo, passing through the same two
points (the
> rotation center and the same left point as on the first photo's left
edge).

We can't assume a reliable edge, or that the two prints are trimmed the
same, or what the left edge of either print would be, I'm afraid.  Sorry.
:-(
 
Steps 4 through 7 are really cool, and I want to keep them if possible!

Step 8 has a nice method of establishing which are left and right prints,
but I think it assumes we know left and right *withinin* the prints again,
which I didn't assume.  We know which print is the left and which the
right, so step 8 does both more and less than I need, I think. ;-)

Could Ken's method be changed to use any three non-colinear points and
their homologues, instead of edge-intersections, which I've disallowed in
my conditions?  That might be the ticket!

Ken, you might enjoy this scenario for the same problem:

Two prints of star fields, taken from two extremely remote (from one
another) planets, (the prints labelled Planet Left and Planet Right)
promising when viewed stereoscopically to show the stars in their relative
distances.  We can identify each star in each print, to establish
homologues, but we don't know what the two planets we shot from were, so
we don't know how the "stereo horizon" is oriented in the pair of images. 
Each image is perfectly round, having been taken through identical
telescopes on each planet.  How can we efficiently use only the star
points themselves in the images to correctly arrange the stereo pair?

Bruce ("Give me a slide bar long anough, and a place to stand, and I will
stereograph the world.") Springsteen



__________________________________________________
Do You Yahoo!?
Yahoo! Mail - Free email you can access from anywhere!
http://mail.yahoo.com/