Re: Ortho and Hyperstereo

[P3D John Bercovitz <bercov@xxxxxxxxxxx>]

First let me say that it is _incredibly_ heartening to see
that SO many people on this list understand the fundamental 
principles of the geometric reconstruction which is the basis 
of stereoscopy.  It sure wasn't always like that on this list.
For that matter, it hasn't been that good in the literature - 
it's full of errors.  If I were to croak now, I'd croak happy. 
This bodes well for the future of stereo photography because
people will be less likely to make errors that they can't quite
put their finger on.  And no, I'm not talking about creative
use of stretch or squash or hyper or hypo.  That's fun.
I might even celebrate by posting 3D messages to the copyright
mailing list.  888-)  [No, I'm not a moderator anymore; I'm
speaking (and gently teasing) _only_ for myself here.]
 
To go to specific discussions, I agree that neither 10% mismatch 
in stereobase nor 10% mismatch in focal length is going to be 
observable except under the best of conditions.  However, a 
mismatch of a factor of 2 is almost always detectable if there's 
anything the least bit familiar in the scene and if the nearest 
object isn't a long way away.
 
> Say the object your are trying to photograph has nothing of 
> significance in the foreground or background and these areas 
> will be ignored by the viewer, but the object itself has features 
> best displayed by 3d.
 
I'd say that the viewer ignoring any part of the scene is 
debateable.  It seems like if there's something wrong in a 
scene, why that's what the eyes gravitate to.  8-(
 
> Suppose the apparent viewing distance you wish is five feet but 
> you can't get closer than 10 feet.  Can you just double the lens 
> separation (ie. from 62mm to 124mm) to achieve the desired effect, 
> or is there a whole lot more I have to know?
 
I guess it depends on what you mean by "desired effect".  If you want
it to look like you're five feet away, you're going to have to put the
whole thing in a computer and have it recompute the relative sizes and
positions of all the objects and then spit out a stereo pair.  The 
relative sizes of objects change with the perspective point (point of 
view) and there isn't much you can do about that short of a massive 
effort such as I've just suggested.
 
If you just want as much binocular disparity as you'd get at five feet
so that the depth is well-defined, you're on the right track here.
 
John B


------------------------------