Re: [Plib-devel] off-axis stereo projection with plib

[<p.l...@ci...>]

Scrive Bram Stolk <br...@sa...>:

> steve wrote:
> > Bram Stolk wrote:
> >
> >>ssgHello(ALL)
> >>
> >>
> >>I'm trying to add a stereo option to a PLIB application.
> >>Now, I want to do the stereo properly.
> >>This means not using the 'toe-in' method.
> >
> >
> > No!  Just render the left eye, move the camera about 6cm to the right
> > and render the right eye.  Don't change anything else!
>
> I'm pretty sure this is not enough.
> Your method may create some illusion of some depth, but it is
> not proper stereo.
>
> Especially with large screens, you want objects to appear
> on the screen, in front of the screen, and behind the screen.
>
> This means that for an object that travels from eye point to
> infinity, this will happen:
>
> - first, the projection for R is of the left of projection L
> - at screen distance, the objects for L and R coincide in
>   their projection
> - after that, the projection for R is on the right of the
>   projection for L.
>
> Your method will not create negative,positive,zero parallax.
> Only negative parallax.

Ok, wait. We have to questions to answer.

First is: on- or off-axis projection? well, it mostly depends on whether the eye
is on- or off-axis wrt the center of the projection screen. Not always off-axis
projection is the case with stereo, even when passive (think of the HMD case,
where the eye is always on-axis), and it's a real-world reference system fact
(i.e. measured in meters, go to measure screen corners and where your eye is,
IPD is ~.06m). I have some Plib-ready code for general projection if someone
wants.
Anyway off-axis projection helps rendering a more correct stereo pair, as well
as any other even non-stereo multiple-projector large screens.

Second is: what modelview/camera model to adopt for stereo pair rendering?
I haven't Paul Bourke's code under my eyes, yet I remember that it had two
concerns from my point of view:

- he wants to avoid toe-in effect, yet adopts the glLookAt, which actually
resembles a rotation;

- it implies the choice of a zero-parallax plane (can't remember the actual lens
parameter, something like focal...) - this can be a trouble to do, and to
combine world and real-world spaces.

I vote (and I do in my stereo Plib-based application) for Steve's translation
along original x-axis of the half-IPD

> Read Paul Bourke's info on off-axis projection again, it
> has some informational figures.

I'll go refreshing it.

>    bram

Greetings,

Paolo