I experimented with this method a decade ago when I was working
on a stereo 16mm film for the late Kent Wilson. The parallel
line-of-sight method was described in a popular book on stereo methods
.. sorry I don't remember the reference and it would take a lot of
digging to find it.
To be honest, I have never liked this parallel method.
I feel that it simply fails to give as good a stereo effect as
the traditional "toe-in" method (in my subjective opinion).
As the apparent object comes closer to the face, it seems to me that
the approximation of parallel lines of sight separated by a small
constant simply breaks down. Why not try both and judge for yourself.
The fact that points on the object that
are very close to the viewer or very far away appear double (and out of
focus) is also exactly what happens with real vision. Here's an
experiment to try:
Put the index finger of one hand a few inches from your nose and
place the index finger of your other hand about a foot away.
Now keep your vision focused on the far finger. The near finger,
which should remain out of focus, will in fact appear double.
Why then is it a problem?
In real vision we can change our focal point (the place
where the image is in focus and the lines of sight of our two
eyes converge) at will. In a fixed stereo image (not dynamic as in VR),
we are stuck with the point the artist chooses. When we are distracted
to look at another part of the object, it looks wrong. This effect is
minor when objects are far away.
There is also another phenomenon at work here too, which I learned about
when I worked in virtual reality. If you project a stereo
image on a wall (say 5 feet away from the viewer) but depict an object
that is very close to the viewer (or visa versa), it will appear out of
focus when
viewed with both eyes and in focus when viewed with one eye. The brain
takes its cue for focus from the degree of convergence of the eyes.
Sorry, I offer no fixes here.
There is a very cool stereo supply company worth looking at however:
http://www.stereoscopy.com/reel3d/
Also be sure to check out
Richard Gillilan
MacCHESS, Cornell
This method of generating stereo images is correct, but also leads to
quite a bit of vertical parallax (the so-called "toe-in" projection) -
this is why many stereoscopic images are hard to view properly
(usually,
edges of the image are out of focus). What you really want is a
non-symmetric camera frustrum (dunno how hard this is to do in
pymol....)
where the two images should look along parallel vectors separated by
some
distance (something like 1/20 the focal length). Check out Paul
Bourke's
page for all the details:
http://astronomy.swin.edu.au/~pbourke/stereographics/
Hope this helps.
Regards,
Tim F
