From: dither@vms.macc.wisc.edu (Dennis Adams)
Subject: Re: Viewing 3D images in a Head Mounted Display
Date: 26 JAN 92 17:06:31    
Organization:  University of Wisconsin Academic Computing Center



StereoGraphics corporation (makers of CrystalEyes stereo viewing
glasses) suggests _parallel_ cameras, and then horizontal translation
of the resultant images (2D translate).  The distance between the
cameras can be adjusted depending on the world you are viewing (ie:
very close if your object is a molecule, 6.3 cm or so for "real" worlds,
wider for hyperstereo).  The image shift was used to set the "depth" of
the result, and mainly used to compensate for the fact that the images
were superimposed on the monitor and some leakage occured in the glasses.
I would assume that you would want to set it up so with a head-mounted
display, a point at "infinity" would appear at the same place on
each retina (no retinal disparity).  Closer objects would increase
retinal disparity.  A previous poster had an important point: NEVER
let there be negative retinal disparity!  This never happens in the
real world, and makes more people explode (ok, well, at least feel
bad).  If you don't know the user's interoculer distance, assume it is
on the small side for safety.

The reason parallel cameras are used is because if you "toe in" the
cameras you also get vertical disparity, which is real bad, too.

I think the reason people complain of "double images" in computer
graphics stereo projections (and never in the real world, where they
happen all the time) is because in the real world, the double images
are typically out of focus, while in CG, everything is always crisp.
This is a problem with planer-stereo methods, as opposed to "real" 3D.

Disclaimer: It's been a while since I messed with the stereo stuff.  I
do not represent StereoGraphics or my employer.