[sldev] Nvidia 3D Goggles?

[Celierra Darling]

I've been thinking about this.  I think that the method with the
smallest barrier to entry is via a webcam and detecting eyes in its
images.  A quick search turns up 2004 D'Orazio, "An Algorithm for Real
Time Eye Detection in Face Images" [1].  They used 640x480 images, and
with a paltry Pentium 3 CPU, they report 300 ms per image.  I have to
assume a modern GPU implementation should be much faster, and the
300ms is without context between images; perhaps a Kalman filter[2]
would be nice to narrow the search?

Then the only things needed from the user are, I think, the physical
screen size, and a calibration picture with the person a known
distance away (to figure out the webcam's field of view).

One thing I'm wondering about: if I understand it right, Johnny Lee's
solution *should* make things look as if the monitor were a window
into SL.  But the placement of the monitor could be problematic - the
view might end up seeming awfully small if the person's far away from
the monitor.  (Personally, my vertical field of view would be just 16
degrees, and 24.8 degrees horizontally.)  I wonder if this would be
okay?  Would scaling up/down the head movement be better, or would it
become more unnatural?

[1] http://ieeexplore.ieee.org/search/wrapper.jsp?arnumber=1334521
[2] http://en.wikipedia.org/wiki/Kalman_filter#Example

~Celi


On Fri, Sep 19, 2008 at 4:23 PM, Philip Rosedale <philip at lindenlab.com> wrote:
> I remain skeptical that people will ever want to suffer a large performance
> cost in visual acuity or frame rate for the benefit delivered from binocular
> disparity.  I think that using a camera to detect eyepoint and shifting the
> view frustrum in the manner demonstrated by Johnny Chung Lee with his Wii
> experiments is probably a more compelling way of enhancing the sense of
> depth, and this approach imposes no additional rendering costs.  While some
> people are not even sensitive to the information conveyed by a stereo pair,
> everyone can clearly see the effects of properly moving the eyepoint.
>
> Philip
>