[tt] practical eyetracking

Mon Sep 3 19:05:52 UTC 2007

http://www.technologyreview.com/Infotech/18254/

Friday, March 02, 2007

An Alternative to the Computer Mouse

A user interface that tracks eye movement may provide an alternate way to use
a computer.

By Kate Greene

A researcher at Stanford has created an alternative to the mouse that allows
a person using a computer to click links, highlight text, and scroll simply
by looking at the screen and tapping a key on the keyboard. By using standard
eye-tracking hardware--a specialized computer screen with a high-definition
camera and infrared lights--Manu Kumar, a doctoral student who works with
computer-science professor Terry Winograd, has developed a novel user
interface that is easy to operate.

"Eye-tracking technology was developed for disabled users," Kumar explains,
"but the work that we're doing here is trying to get it to a point where it
becomes more useful for able-bodied users." He says that nondisabled users
tend to have a higher standard for easy-to-use interfaces, and previously,
eye-tracking technology that disabled people use hasn't appealed to them.

At the heart of Kumar's technology is software called EyePoint that works
with standard eye-tracking hardware. The software uses an approach that
requires that a person look at a Web link, for instance, and hold a "hot key"
on the keyboard (usually found on the number pad on the right) as she is
looking. The area of the screen that's being looked at becomes magnified.
Then, the person pinpoints her focus within the magnified region and releases
the hot key, effectively clicking through to the link.

Kumar's approach could take eye-tracking user interfaces in the right
direction. Instead of designing a common type of gaze-based interface that is
controlled completely by the eyes--for instance, a system in which a user
gazes at a given link, then blinks in order to click through--he has involved
the hand, which makes the interaction more natural. "He's got the right idea
to let the eye augment the hand," says Robert Jacob, professor of computer
science at Tufts University, in Medford, MA.

Rudimentary eye-tracking technology dates back to the early 1900s. Using
photographic film, researchers captured reflected light from subjects' eyes
and used the information to study how people read and look at pictures. But
today's technology involves a high-resolution camera and a series of infrared
light-emitting diodes. This hardware is embedded into the bezel of expensive
monitors; the one Kumar uses cost $25,000. The camera picks up the movement
of the pupil and the reflection of the infrared light off the cornea, which
is used as a reference point because it doesn't move.

Even the best eye tracker isn't perfect, however. "The eye is not really very
stable," says Kumar. Even when a person is fixated on a point, the pupil
jitters. So he wrote an algorithm that allows the computer to smooth out the
eye jitters in real time. The rest of the research, says Kumar, involves
studying how people look at a screen and figuring out a way to build an
interface that "does not overload the visual channel." In other words, he
wanted to make its use feel natural to the user.

One of the important features of the interface, says Kumar, is that it works
without a person needing to control a cursor. Unlike the mouse-based system
in ubiquitous use today, EyePoint provides no feedback on where a person is
looking. Previous studies have shown that it is distracting to a person when
she is aware of her gaze because she consciously tries to control its
location. In the usability studies that Kumar conducted, he found that
people's performance dropped when he implemented a blue dot that followed
their eyes.

In his studies of 20 people, he found that participants that needed to type
and point could point faster using the gaze-based appraoch than using a
mouse, although the error rate--20 percent--was fairly high. But overall,
about 90 percent of participants reported that they preferred using EyePoint
to the mouse.

It's the 20 percent error rate that could cause some problems, says Ted
Selker, professor at the MIT Media and Arts Technology Laboratory. "[It's] a
huge amount," he says, "because a person can notice a significant decline in
accuracy at just 5 percent." Selker adds that the low accuracy could make
text editing a challenge.

Kumar concedes that the system isn't perfect, but he contends that many of
the errors came from people, who due to lack of practice, clicked links that
they thought they had looked at but were only in their peripheral vision.
Indeed, he says, trackpads, trackpoints, trackballs do not perform as well as
a mouse either but are still viable input devices. Kumar says he's been
working on algorithms that show promise for making EyePoint more accurate by
accounting for peripheral vision related errors. Still, he allows that
EyePoint might work poorly for certain people, such as those with thick
glasses, special contact lenses, or lazy eyes.

Even so, Kumar is confident in the technology and its development as a tool
for the general population. To that end, he has tested a number of different
interface schemes, all under a project called Gaze-enhanced User Interface
Design (GUIDe). Another application, called EyeExposé, is made for Apple's OS
X feature called Exposé, in which a person can hit the F11 key to miniaturize
all open windows, then drag the mouse cursor to the window she wants to bring
forward. With EyeExposé, the user can hit the F11 key, then bring forward a
window of interest by tapping a keyboard key. Also, Kumar has modified the
"scroll lock" key on a keyboard in an application called EyeScroll: as a
person reads, the screen slowly reveals more text. In addition, Kumar is
testing a modified version of the "page up" and "page down" keys. When a
person reads to the bottom of a page, the software automatically scrolls down
one page; in order to help a reader keep her place, the most recently looked
at part of the screen is highlighted.

The important thing about the Stanford research, says Shumin Zhai, researcher
at IBM Almaden Research Center in San Jose, CA and pioneer in the
eye-tracking field, is that Kumar "has been working on making eye tracking
practical for everyday tasks." However, Zhai says that there may still be a
barrier for the average person because she needs to go through a calibration
process in which the software measures how quickly her eyes move.

There are some signs that eye-tracking technology could find its way to the
consumer market soon. Apple's desktops and laptops are now equipped with a
built-in camera for videoconferencing. If a higher-resolution camera,
infrared LEDs, and software were added, Apple's machines would be able to
support applications from the GUIDe project, says Kumar. If eye tracking
proves appealing to the consumer, and the hardware costs drop to a reasonable
range, eye-tracking interfaces could provide an alluring and entertaining
alternative to the mouse or laptop track pad. "It's almost like magic when
it's working," says Tufts's Jacob. "The sensation you get is that the
computer's reading your mind, and that's really very powerful."