Hi all, 

I want to set up a nice stereoscopic camera for anaglyphs, cross-eyed 
view and parallel view images. [Results will be made available publically 
when I succeed...]

I know, there is StereoPOV but I need to look more precisely at the issue: 
IIRC, StereoPOV uses a non-perpendicular camera which results when 
simply doing: 




(stereo_win_distance is the distance where the left and right rays 
cross each other and hence the plane which seems to be at the depth of 
the projection screen plane when viewed stereoscopically.) 

I do not understand why it is correct to use such a non-perpendicular 
camera. 

According to my understanding, the camera actually is generally NOT 
perpendicular but the angle between right and direction (=delta) depends 
not only on the ratio of eye_distance to stereo_win_distance but also on 
the actual eye ray crossing distance when viewing the image. 
This means that 
(a) cross-eyed view and parallel view require different delta-angles. 
(b) the above description (*) is incorrect; e.g. for anaglyphs one would 




I hope anyone here has enough clue to point out exactly what is going on. 

Regards,
Wolfgang

Post a reply to this message

> I want to set up a nice stereoscopic camera for anaglyphs, cross-eyed 
> view and parallel view images. [Results will be made available publically 
> when I succeed...]

High quality stereo is one of my main activities, I use OpenGL, POVray,
3dStudioMax, and a bunch of other rendering tools. I would be happy to
answer any specific questions you might have although only in regard
to creating stereo pairs that are to be presented to each eye 
independently, that is, I don't deal in any of the cross eye business.

To you question about camera position, yes parallel cameras is correct,
toe in camera introduce vertical parallax and hence increased eye strain.

I have tools that implement the following is a range of packages that
don't have builtin stereoscopic support
    http://astronomy.swin.edu.au/~pbourke/povray/raystereo/
-- 
Paul Bourke
pdb_NOSPAMswin.edu.au

Post a reply to this message

I agree with Paul Bourke, cameras should always be
parallel. Converging (what you refer to as cross-eyed
view) cameras introduce keystone distortion. Stereo
set ups based on converging cameras only work with
a short depth of field. Converging cameras do set the
stereo window at the convergence point, which is
convenient, but you can set the stereo window of
parallel views by trimming the excess.

Correctly rendered stereo pairs can be viewed by
the cross-eye method, but not created by toe-in of
the cameras.

"Wolfgang Wieser" <wwi### [at] gmxde> wrote in message
news:40670613@news.povray.org...
> Hi all,
>
> I want to set up a nice stereoscopic camera for anaglyphs, cross-eyed
> view and parallel view images. [Results will be made available publically
> when I succeed...]
>
> I know, there is StereoPOV but I need to look more precisely at the issue:
> IIRC, StereoPOV uses a non-perpendicular camera which results when
> simply doing:
>
> location = location +- eye_distance/2 * right;
> direction = direction -+ eye_distance/(2*stereo_win_distance) * right (*)
>
> (stereo_win_distance is the distance where the left and right rays
> cross each other and hence the plane which seems to be at the depth of
> the projection screen plane when viewed stereoscopically.)
>
> I do not understand why it is correct to use such a non-perpendicular
> camera.
>
> According to my understanding, the camera actually is generally NOT
> perpendicular but the angle between right and direction (=delta) depends
> not only on the ratio of eye_distance to stereo_win_distance but also on
> the actual eye ray crossing distance when viewing the image.
> This means that
> (a) cross-eyed view and parallel view require different delta-angles.
> (b) the above description (*) is incorrect; e.g. for anaglyphs one would
> need to compensate for eye ray crossing at the image "center" (focal
> point) depth rather than at the stereoscopic window depth.
> My experiments, however, did not yet confirm that.
>
> I hope anyone here has enough clue to point out exactly what is going on.
>
> Regards,
> Wolfgang

Post a reply to this message

Paul Bourke wrote:
> High quality stereo is one of my main activities, I use OpenGL, POVray,
> 3dStudioMax, and a bunch of other rendering tools. I would be happy to
> answer any specific questions you might have although only in regard
> to creating stereo pairs that are to be presented to each eye
> independently, that is, I don't deal in any of the cross eye business.
> 
I am sorry for not answering more quickly.
Okay, your web page made things a bit clearer. But still...

It seems that creating left/right stereo pairs for independent 
projection onto the eyes (using a stereo helmet with two screens 
or whatever) can be done using for the camera: 

location = primary_location +/- eye_distance/2*right
right, up and direction stay unchanged from primary parameters. 

(where "primary" means: "of the corresponding non-stereoscopic camera")

Okay, this, however, requires you to cut off a certain part of 
the left and right image. Now, if one uses a non-perpendicular 
camera: 

location = primary_location +/- eye_distance/2 * right
direction = primary_direction -/+ eye_distanc/(2*windepth) * right
right and up stay unchanged from primary parameters.

...it seems that one can create the required left/right image 
pairs without cutting off parts of the image. 
[Can anyone confirm that?]

But that is just the first half; in case the above was correct, 
here comes my actual problem: If one views the stereo pairs using 
the anaglyph method or cross-eyed, the eye directions are not 
perpendicular to the projection screen because they cross 
each other before (cross-eyed) or somewhere behind the screen. 

Now, the problem is: How can one compensate for that?
(I.e. which direction and right vectors need to be chosen?)

Wolfgang

P.S: I like your homepage!

Post a reply to this message

This isn't theoretical.

When I want a pair of pictures
for a stereogram,

I render the scene and move the file
then rotate the objects or the camera
10 - 12 degrees on only one axis of
choice.

The picture pair can be viewed via
a viewer or the cross - eyed method.

Post a reply to this message

In article <407319af@news.povray.org>,
 "john.s<replacewithdot>grant" 
 <"john.s<replacewithdot>grant"@sym### [at] icoca> wrote:

> This isn't theoretical.
> 
> When I want a pair of pictures for a stereogram, I render the scene 
> and move the file then rotate the objects or the camera 10 - 12 
> degrees on only one axis of choice.

You might want to read some of the other messages in this thread. Though 
the obvious method, this way has some problems. Basically, it leaves a 
greater non-overlapping image area, making the stereo view of the scene 
smaller, and can cause eye strain (the images are exactly what the eye 
would see, but they are projected onto a flat surface in front of the 
eye). If you compensate for that effect, it might be more useful...

-- 
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