[Sidefx-houdini-list] Stereo Cameras

Andrew D Lyons tstexture at gmail.com
Fri Oct 27 12:32:19 EDT 2006


Hi Mark,

I'm no expert - but I've done a little stereo over the years. The
toe-in method is one I've used, and thought well enough of. The
problem is that you need to render each object seperately and
composite afterwards. This is so that you can have both cameras
converging on a "lookat" object that matches the z distance (distance
from cameras) of each object you are trying to render. This is fine in
pre-rendered stereo because we are accustomed to rendering objects
seperately and compositing layers. However it doesn't work for
realtime applications as easily, when you might have multiple objects
in the scene at different z-depths. In scenes like this, if your
cameras converge on the foreground object, you get terrible sight line
crossing behind it, and the distant objects diverge and disassociate
from each other. This is part of the reason why I think the paper you
refer us to has deprecated the "toe-in" method. This paper is
concerned primarily with Open GL methods - in which the toe-in method
would cause bad eye-crossing, and be difficult to work with - unless
you defined a seperate camera for every object and it's varying depth
and somehow combined image buffers afterwards. If you think about how
we look at things, our eyes do in fact converge on objects, so in a
way toe-in is correct - unless you are trying to implement stereo in
open GL - in which it is difficult to the point of being impossible.

If you are set on the so called "correct" method, I'd suggest
experimenting with the Houdini camera "crop" settings. It looks as if
you could keep your camera's sight lines parallel, and just crop off
the outer side of each camera to get an equivalent viewing frustum to
the one they describe.

If this so called "correct method" is what you are doing though, you
are proabably better off doing what most big stereo productions seem
to do these days, and render with one camera, but store depth
information for every pass. You can replicate the stereo effect in
COPs by splitting off two image streams from the original render, and
then use the depth information to shift each one away from the center
as it gets closer to your cameras. You would need to devise an
expression that calculates exactly what that divergence should be for
your camera, but this will save you a lot of time compared to
rendering every layer with two cameras and trying to manage that.

Finally, if you don't already have some workstation based way of
previewing your stuff in stereo, I'd suggest the red/blue anaglyph
technique. Get a pair of red-blue 3D glasses from somewhere, (the web)
and use a channel copy COP to split your color planes out correctly
for each camera so you can preview your stereo effects on your
computer. Merge the red layer from the left camera, and cyan (which is
equal parts blue and green) from the right camera and you'll be in
business. (Works better with low-chroma or grayscale images.)

http://en.wikipedia.org/wiki/Anaglyph_image
http://www.stereoscopy.com/reel3d/anaglyph-glasses.html

Enjoy!

Cheers





On 27/10/06, Dan Seddon <dan.seddon at framestore-cfc.com> wrote:
> Mark A. Bolstad wrote:
> >
> > Searching the mailing list, I saw a couple of previous postings are
> > creating stereo images, but none of the methods actually accomplish
> > what I need. We're going to a show with a large projection system that
> > is running a passive stereo system (very cool system that uses color
> > shifting instead of polorization).
> >
> > What I need is a way to compute a pair of asymmetrically-skewed
> > cameras. I have parented a left and right eye camera with what appears
> > to be the correct eye separation. What I need is a way to set up the
> > cameras similar to this (see "Correct method):
> >
> > http://local.wasp.uwa.edu.au/~pbourke/stereographics/stereogl/
> >
> > It's easy to setup the toe in method, but I would like the method above.
> >
> > Thanks,
> >
> > Mark
> >
> Hi Mark
>
> You could use the Window X/Y offset to do what you want - I think it's
> more correct than using the crop as it's post-perspective in terms of
> it's transform (which I think is more correct - but could be completely
> wrong). You'll have to mess around with numbers to get it right though,
> if you're not sure about this perhaps your could create a number of
> reference objects to define your field of views (a bit like a look at
> null) and mess around with the numbers until they fit.
>
> Cheers
>
> Dan
> _______________________________________________
> Sidefx-houdini-list mailing list
> Sidefx-houdini-list at sidefx.com
> https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
>


-- 
=======================================
Andrew D Lyons | Digital Artist | http://www.tstex.com
=======================================



More information about the Sidefx-houdini-list mailing list