Wolfgang 

Regarding your proposal to "toe in" your cameras for stereo. Sure it will work
but try it and you will find that as you move out from the center of the image
you will get increasing vertical parallax introduced. Our visual system can cope
with that easily but if you plan to create long stereo sequences then your viewers
will experience increased eye strain. What you propose is quite common in filming
with commodity cameras because of the low resolution, it depends on whether
you want to achieve a uptimal result or not.

ps: Can I suggest you think in terms of distance to zero parallax instead of
"window distance". Make your eye separation 1/30 of this zero parallax distance.
Zero parallax being the distance at which things will appear to at the screen,
this way you can deal with arbitrary units rather than using our human eye
separation and scaling everything to our human scale. There are times when real
eye separations are important, the main ones are
1. multiple or very wide angle screens
2. attempting to create real world 1:1 scale impressions, eg it was important
   for this project
   http://astronomy.swin.edu.au/~pbourke/stereographics/vroom/
   http://www.vroom.org

-- 
Paul Bourke
pdb_NOSPAMswin.edu.au

Post a reply to this message

In article <pdb_NOSPAM-174B1A.18301401042004@news.povray.org>,
 Paul Bourke <pdb### [at] swineduau> wrote:

> Regarding your proposal to "toe in" your cameras for stereo. Sure it 
> will work but try it and you will find that as you move out from the 
> center of the image you will get increasing vertical parallax 
> introduced.

Well, if the images are displayed to each eye separately, on screens 
oriented to each eye, it would be best to render exactly what each eye 
would see. Two ordinary perspective cameras, rotated to focus on some 
point in the scene.

However, when viewing on screen, you are viewing two projections of the 
scene on a single flat surface. My results with parallel cameras have 
been very difficult to view, but translating and shearing the cameras by 
adjusting their directions seems to produce very good results.

-- 
Christopher James Huff <cja### [at] earthlinknet>
http://home.earthlink.net/~cjameshuff/
POV-Ray TAG: <chr### [at] tagpovrayorg>
http://tag.povray.org/

Post a reply to this message

Only if you have limited range of depth. Toe-in will produce
keystone distortion that will make the images difficult to
view.

There is a physiological basis for the 30:1 rule. At about
30 times the distance between our eyes (2.5 inches * 30
= 70 inches) our eyes cease to converge (cross as you say)
and are essentially parallel.

All known viewing systems use images on the same plane, so images
produce through rotation of the cameras/convergence/toe-in will conflict
with the viewing system. It would be interesting to imagine a stereo
viewing system where the image planes rotate to stay oriented with
the eye as they converge (cross) to look at objects closer than the
virtual 70 inches.

Harolddd

"Christopher James Huff" <cja### [at] earthlinknet> wrote in message
news:cjameshuff-7D27A4.13453501042004@news.povray.org...
> In article <pdb_NOSPAM-174B1A.18301401042004@news.povray.org>,
>  Paul Bourke <pdb### [at] swineduau> wrote:
>
> > Regarding your proposal to "toe in" your cameras for stereo. Sure it
> > will work but try it and you will find that as you move out from the
> > center of the image you will get increasing vertical parallax
> > introduced.
>
> Well, if the images are displayed to each eye separately, on screens
> oriented to each eye, it would be best to render exactly what each eye
> would see. Two ordinary perspective cameras, rotated to focus on some
> point in the scene.
>
> However, when viewing on screen, you are viewing two projections of the
> scene on a single flat surface. My results with parallel cameras have
> been very difficult to view, but translating and shearing the cameras by
> adjusting their directions seems to produce very good results.
>
> -- 
> Christopher James Huff <cja### [at] earthlinknet>
> http://home.earthlink.net/~cjameshuff/
> POV-Ray TAG: <chr### [at] tagpovrayorg>
> http://tag.povray.org/

Post a reply to this message

In article <406c9f6e$1@news.povray.org>, "Harold" <bai### [at] 3dculturecom> 
wrote:

> Only if you have limited range of depth. Toe-in will produce
> keystone distortion that will make the images difficult to
> view.

What I'm doing does not produce keystone distortion. I'm talking about 
modifying the direction vector, while leaving the right vector alone. 
The result is basically the same as cropping images from parallel 
cameras, and seems to be a lot easier to focus on, as well as making 
better use of the image space.


> There is a physiological basis for the 30:1 rule. At about
> 30 times the distance between our eyes (2.5 inches * 30
> = 70 inches) our eyes cease to converge (cross as you say)
> and are essentially parallel.

And if two cameras are rotated to converge at a point at a distance 
greater than 30 times the separation of the cameras, the keystone 
distortion will be very small, because the cameras will be essentially 
parallel. For most such scenes, rotating the cameras will neither give a 
benefit nor cause a problem. When focusing on nearer objects, rotating 
cameras should simulate what the eyes really see more closely.


> It would be interesting to imagine a stereo
> viewing system where the image planes rotate to stay oriented with
> the eye as they converge (cross) to look at objects closer than the
> virtual 70 inches.

Yes, it would. The system would have to either track the eyes or rely on 
head movement to adjust the display to what the user is focusing on. It 
wouldn't necessarily need to physically move the displays, it could 
compensate for the display perspective in software. Ignoring this is 
obviously "good enough" for the brain, but it might slightly reduce 
strain.

-- 
Christopher James Huff <cja### [at] earthlinknet>
http://home.earthlink.net/~cjameshuff/
POV-Ray TAG: <chr### [at] tagpovrayorg>
http://tag.povray.org/

Post a reply to this message

"Christopher James Huff" <cja### [at] earthlinknet> wrote in message
news:cjameshuff-9B6DD8.13311402042004@news.povray.org...
>
> What I'm doing does not produce keystone distortion. I'm talking about
> modifying the direction vector, while leaving the right vector alone.
> The result is basically the same as cropping images from parallel
> cameras, and seems to be a lot easier to focus on, as well as making
> better use of the image space.

Okay, I have to admit my ignorance. I don't know how modifying
the direction vector affects the image.

> When focusing on nearer objects, rotating
> cameras should simulate what the eyes really see more closely.

Yes, but like I said, there is no viewing system that takes
convergence into account. What you end up with is images
with angular distortion because they are rendered from the
perspective of converging eyes, but they are viewed in
parallel.

Harolddd

Post a reply to this message

In article <406dbcda$1@news.povray.org>, "Harold" <bai### [at] 3dculturecom> 
wrote:

> Okay, I have to admit my ignorance. I don't know how modifying
> the direction vector affects the image.

It basically makes the image projection asymmetrical. A typical 
perspective camera puts the image plane centered on and perpendicular to 
the view axis. Rays come from the camera and pass through a point in 
this plane corresponding to that point in the image being rendered. 
Shearing the direction vector is like pushing the view plane to one 
side, it is the same as rendering a larger image and cropping part off 
one edge.

Hmm, while looking for a graphic to illustrate this, I found a page 
which describes this technique, calling it "off-axis projection":
http://www.cs.wpi.edu/~matt/courses/cs563/talks/stereohtml/stereo.html

And another page, comparing the pros and cons:
http://www-lsi.upc.es/~lucac/projects/stereopsis/

-- 
Christopher James Huff <cja### [at] earthlinknet>
http://home.earthlink.net/~cjameshuff/
POV-Ray TAG: <chr### [at] tagpovrayorg>
http://tag.povray.org/

Post a reply to this message