From: brucec@phoebus.labs.tek.com (Bruce Cohen)
Subject: Re: Eye-tracking system proposed
Date: 1 Jul 91 18:07:04 GMT
Organization: Tektronix Inc.



> Perhaps one could explore using the
> infrared lumination method (already somewhat common for eye-tracking) to
> center the entire head position?  Comments?  -- Bob Jacobson]

Some off-the-top-of-the-head thoughts:

The biggest drawback to current eye-tracking systems that I know about
is the relatively small volume in which the eyes can be tracked.  I
think this is partly because the tracking degrades as the line to the
eye from the sensor moves off the normal to the face at the eye, and
also simply because there are alot of angles of head orientation which
block the sensor from seeing the eye.  Tracking the entire head and
using the position and orientation data to aim the gaze detection
sensors could make that problem easier.

There are some tradeoffs involved:
    1) Do you want to use more than one set of sensors, or are you
       willing to accept a limited set of head orientations?
    2) Is a target on the head such as a headband acceptable?  Not
       having one means you need to deal with variables like differences
       in hair color and skin pigment between users in sensing the head
       orientation.
    3) Head-tracking needs to take place at frame-rate, or possibly
       less; 15-30 fps might be fine.  Eye-tracking has occasional need
       for low-millisecond response times, e.g. to detect and determine
       the motion parameters of saccidic eye-motions.  Motion-blur and
       frame-to-frame delay may mean that eye motions are hard to track
       with camera images, so two sensors may still be necessary.

The key question has to be: what do you want to do with the gaze data?
If you're only interested in knowing the average point of view over some
hundreds of milliseconds, so you know where to compute the image to high
resolution for foveal vision, I think that's practical with a single
type sensor; if you want to track current eye-position and motion, that
may be quite another problem.

--
------------------------------------------------------------------------
Speaker-to-managers, aka
Bruce Cohen, Computer Research Lab        email: brucec@crl.labs.tek.com
Tektronix Laboratories, Tektronix, Inc.                phone: (503)627-5241
M/S 50-662, P.O. Box 500, Beaverton, OR  97077