From: jpc@tauon.ph.unimelb.edu.au (John Costella)
Subject: Re: TECH: Eye-gaze systems
Date: Sat, 5 Dec 92 21:13:57 EST


> Extract from New Scientist, 28 November 1992, page 20.
> 
>         The Japanese company Canon recently launched an autofocus
>         camera which senses where the photographers eye is looking,
>         and adjusts the lens to focus on whatever is in that part of
>         the field of view. But patent records of the European patent
> 
> This immediately suggested to me that with the use of an eye-gaze
> detection system in a HMD, the amount of information that needs to be
> represented on the displays might be reduced. It may be possible to
> use higher resolution where the eye is focused, and lower resolution

There has been a fair amount of work on this. It's not all plain 
sailing though, because the eye tends to take in the significant
detail in *all* the scene (given enough time) by flicking around
the various areas, then going back to those interesting bits for
a second look. You can end up playing "fovea-tag" if you're not
careful. 

You also have to be careful about relying too much on foveal
resolution arguments to reduce peripheral display resolution.
While the *static* resolution does rapidly degrades with angle 
from fovea, the sensitivity to *motion* increases. Stick your right
arm straight out to the right, and wave your index finger just a
teency amount. Although you would do well to recognise the finger 
from a pencil in terms of static resolution, you can see it *move* 
even a millimetre no problems at all. Degrade display resolution 
out there and you are killing off the peripheral detection of 
motion.

But the foveal tracking *can* be useful by averaging out the amount
of time spent in various areas, to get an idea for which broad
areas are more "important" to the viewer. You then don't get the
huge increase in effective resolution that the naive argument
suggests, but you can "tweak" the rendering model you're using for
certain objects, so the viewer thinks the overall scene looks more
realistic. It sounds like the Nikon people are using the tracking 
exactly in this "importance-measuring" way: 

>         The camera focuses on, and sets the exposure for, the area on 
>         which the eye has settled longest.

>         Nikon suggests a grid of 96 blocks, 8 high and 12 across with
>         500 readings taken at intervals 5 milliseconds. The camera
>         focuses on, and sets the exposure for, the area on which the
>         eye has settled longest.

500 x 5 ms = 2.5 seconds, so we *are* talking about average long-term
"importance" measurements here. A nice application.

It's interesting to see foveal tracking hitting the streets in any 
form... means there will be $$ reasons for them to continue the
research :). 

John

----------------------------------------------------------------------------
John P. Costella              School of Physics, The University of Melbourne
jpc@tauon.ph.unimelb.edu.au         Tel: +61 3 543-7795, Fax: +61 3 347-4783
----------------------------------------------------------------------------