From: dstamp@watserv1.waterloo.edu (Dave Stampe-Psy+Eng)
Subject: Re: Eye-movement monitoring
Date: Thu, 16 Jan 1992 00:20:29 GMT
Message-ID: <1992Jan16.002029.16184@watserv1.waterloo.edu>
Organization: University of Waterloo



james@dgbt.doc.ca (James Tam) writes:

>With respect to queries concerning eye-movement "interfaces", I have come 
>across a brochure describing a system that uses the pupil-center/corneal-
>reflection method but that doesn't require attachments to the head.  
>
>The eye-tracking component tracks the pupil center and the corneal 
>reflection of an image generated by a video camera.  The head-tracking 
>component tracks the subject's eye as he moves his head by using a 
>motorized, rotating mirror to maintain the eyes within the view of the camera.
>The control of the mirror is based on information from the camera image.  
>Accuracy is claimed to be less than a degree.  


When I started researching eye tracking systems 18 months ago for
a new lab, I looked at this and other systems.  I'd like to summarize
the problems with this, and some other systems that utilize corneal
reflection in addition to pupil tracking.

First, these systems tend to be rather pricey.  I expect that mirror-
stablized system is over $40000.  ISCAN also nearly doubles the price
of their pupil-tracking system when you ask for the corneal-tracking
option.

Second, any system that does not rely on head mounted cameras will have
lower resolution.  Why?  Because the image of the eye must be smaller
in the camera's FOV, to allow for head motions.  This is also true for
the mirror systems, as they have a finite response time, and I suspect 
they do not move the mirror until the eye is fairly far off-center in
the camera image (that's what I'd do to stablize the feedback loop).

Now ISCAN quotes their pupil-imager's resolution as 512Hx256V.  Having 
used this device, I know that the usable pupil-movement range is
about 140Hx70V, as the pupil must fit entirely in the TV image, and
the pupil image gets larger as you increase the image magnification
to increase the resolution.  Figure in a maximum eye-movement range
of +/- 15 degrees before the pupil gets occluded by eyelids or the
image becomes nonspherical (causing errors in center location) and
your resolution is about 1/4 degree H by 1/2 degree V.

I have found big advantages to keeping the pupil movement resolution
as high as possible, such as allowing simple noise reduction and 
saccade detection.  If the resolution drops below 0.6 degree, it
becomes much more difficult to cleanly extract eye movement "events".

The problem with corneal reflection plus pupil image tracking is tha
it effctively halves you resolution while doubleing the noise level.
The idea behind this method is that pupil image and corneal reflections
both move with the head, but corneal reflections only move half as
far with head movements.  Thus subtracting reflection position from
pupil position should result in 1/2 of the eye movement, with head
motoin cancelled out.

Unfortunately, both the pupil and corneal reflection data are quantized,
and also have high (+/- 2 units) noise levels.  These tend to add,
and the result is 4x the noise (well over 2 degrees).

With my own system, the head-mounted tracking rig is very stable
with relation to the eye (<1 degree of drift) and head motion is
compensated for within 1/6 of a degree by a video tracker.  This
 along with novel noise-reduction techniques, has resulted in
the detection of new phenomena in gaze direction control, etc.
The system data rate is 60 Hz, and I have seen no need yet for
higher rates.  That might help predict saccade endpoints, but
from what I've seen published, that doesn't work well anyway.  So
unless you're researching eye-movement dynamics, the extra cost
isn't worth it.  And the ISCAN high-speed tracker decreases the
already marginal vertical resolution by a factor of 3.

In conclusion, I'm not convinced that corneal reflection/pupil
combinations are the top-end systems they're made out to be.  Of
course, I'm biased toward my own developments in the field, but
I believe that the ongoing tests will prove the superiority of
the concept.


--------------------------------------------------------------------------
| My life is Hardware,                    |                              | 
| my destiny is Software,                 |         Dave Stampe          |
| my CPU is Wetware...                    |                              | 
| Anybody got a SDB I can borrow?         | dstamp@watserv1.uwaterloo.ca |
__________________________________________________________________________