Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!mcsun!ukc!edcastle!aipdc
From: aipdc@castle.ed.ac.uk (Paul Crowley)
Newsgroups: comp.human-factors
Subject: Re: Eye Movement Trackers
Message-ID: <11078@castle.ed.ac.uk>
Date: 15 Jun 91 12:52:58 GMT
References: <91Jun14.160659edt.6227@neat.cs.toronto.edu>
Organization: Put your analyst on danger money, baby!
Lines: 43

In article <91Jun14.160659edt.6227@neat.cs.toronto.edu> mgreen@cs.toronto.edu (Marc Green) writes:
>All trackers require careful and frequent calibration.  Further, It is
>not easy to tell where a person is looking, even if you know the
>position of the pupil; there is a big difference between knowing the
>position of the eye and the locus of gaze. People also make many
>involuntary eye movements, so a tracker would unintentional cause
>actions to occur. I don't think that eye trackers will ever become
>popular. Just too many problems.

I've come up with an idea for an eye tracker that's probably impractical
but appeals to me because if it worked it would be simple and cheap and
work better than the other options.  It goes something like this: 
attach electrodes to the back of the brain (where the vision processing
goes on) and momentarily disturb one half of the screen, then the other
half.  The response will be markedly stronger in the half in which the
user is looking (I have seen this, and it's true).  Split that half into
halves again and repeat until a small box has been marked off.

There are three serious problems with this.  One, looking at a display
that flickers over and over to find out where you're looking would be a
pain in the neck.  Second, it might be hard to measure the response,
although it's so obvious to look at that it might not be too hard. 
Third, the halving process might take too long.  It can be made slightly
faster by using thirds instead of halves, but even then fourteen tests
are needed, meaning 42 disturbances to get your viewpoint down to 1 in a
million pixels.  I don't know how long it would take to disturb the
screen and measure the response, so this could take anywhere between a
tenth of a second and a minute.  Oh, and you have to attach electrodes
to the back of your neck, although this is not as much hassle as you
might think - just don't do it if you need to look your best... 

I think that eye trackers have enormous potential if combined with
rudimentary speech recognition.  Speech recognition software exists that
can accept simple commands with close on 100% reliability, and we can
pick the right word faster than we can pick the right mouse button, and
with a greater range.  If you have an Athena-Widget style "highlight
when the eye is in the box" then people will usually do what they meant
to, and you have a natural and _very_ fast user interface.
                                         ____
\/ o\ Paul Crowley aipdc@castle.ed.ac.uk \  /
/\__/ Part straight. Part gay. All queer. \/
"I say we kill him and eat his brain."
"That's not the solution to _every_ problem, you know!" -- Rudy Rucker