The Selective Tuning Model of Visual Attention

1
Gaze-Controlled Human-Computer Interfaces
Marc Pomplun Department of Computer
Science University of Massachusetts at
Boston E-mail marc_at_cs.umb.edu Homepage
http//www.cs.umb.edu/marc/
2
Gaze-Controlled Human-Computer Interfaces

• Overview
• Using Eye Movements as a Response Modality in
  Psychophysics
• Typing by Eye with Dynamic Recentering
• An Advanced Typing Interface Dasher
• A Gaze-Controlled Zooming Interface

3
Using Eye Movements as a Response Modality in
Psychophysics (Stampe Reingold, 1995)

• In psychophysical experiments, subjects typically
  respond to a stimulus by pressing one out of two
  or more buttons.
• The obtained response times are used as an
  indicator of how long it took the subject to
  process the stimulus.
• However, the measured duration also includes the
  time taken for initiating and executing the
  manual response.

4
Eye Movements as a Response Modality

• If subjects can indicate their response by moving
  their eyes instead of pressing a button, this
  response overhead should be reduced.
• Therefore the signal-to-noise ratio in the
  reaction data should be improved.

5
Eye Movements as a Response Modality
6
Eye Movements as a Response Modality
7
Advantages of Gaze-Controlled Interfaces

• Allow intuitive use of computer programs
• Operators can simultaneously use their hands for
  other tasks
• Enable handicapped people to control systems and
  communicate by means of eye movements (e.g.
  typing by eye)

8
Problems with Gaze-Controlled Interfaces

• The Midas-Touch ProblemSince eye movements
  are not completely under conscious control,
  sometimes functions may be triggered
  inadvertently.

Typically, researchers try to solve this problem
by setting a minimum dwell time for triggering
events.
9
Typing by Eye with Dynamic Recentering (Stampe
Reingold, 1995)

• This is a simple typing by eye application
  using a virtual keyboard.
• Keys are triggered using a dwell time threshold.
• This threshold can be varied while using the
  system.
• The authors also implemented a mechanism of
  dynamic recentering to avoid frequent
  recalibration of the eye tracker system.

10
Typing by Eye with Dynamic Recentering (Stampe
Reingold, 1995)
11
Typing by Eye with Dynamic Recentering (Stampe
Reingold, 1995)

• The dynamic centering mechanism makes the
  (reasonable) assumption that users fixate only
  the keys on the screen.
• To compensate for drift in gaze-position
  measurement (as caused by headset shift), the
  system measures the offset between the centers of
  the keys and the fixation positions.
• If a fixation shows such an offset, the following
  measurements are shifted by about 10 of the
  offset distance, but in the opposite direction.

12
Typing by Eye with Dynamic Recentering (Stampe
Reingold, 1995)

• Dynamic recentering is able to reduce the average
  fixation error and the frequency of system
  recalibration.

13
Typing by Eye with Dynamic Recentering (Stampe
Reingold, 1995)

• One disadvantage of the mechanism is that if the
  offset is larger than half the distance between
  neighboring keys, it will draw the measurement
  towards unwanted keys.
• Moreover, this system can only compensate linear
  shifts but no rotation or other distortions of
  measurement.

14
Dasher - An Advanced Typing-by-Eye Interface
(Ward MacKay, 2002)

• The previously shown interface is the most basic
  and straightforward implementation of typing by
  eye.
• It is possible to make such interfaces more
  intelligent to allow faster and more convenient
  typing.
• One such approach is the Dasher system (freely
  available on the web).

15
Dasher

• The initial display of Dasher shows all letters
  of the alphabet in a column at the right edge of
  the screen

16
Dasher

• The letters flow leftwards, each of them followed
  by a new alphabet with the most likely continuing
  letters being the biggest ones.

17
Dasher

• Use the mouse to control the typing
• Left-right control the speed of letters
• Up-down select next letter
• Video Demonstration

18
A Gaze-Controlled Zooming Interface(Pomplun,
Ivanovic, Reingold Shen, 2001)

• We created a gaze-controlled interface that
• supports a common, important task(zooming in/out
  to inspect an image),
• can be used easily and intuitively,
• and minimizes the Midas-Touch Problem.

Demonstration of the Zooming Interface
19
A Gaze-Controlled Zooming Interface

• We compared the efficiency and practice effects
  of gaze control vs. mouse control.
• Four subjects participated in six sessions, each
  session including 50 gaze and 50 mouse trials.
• We measured response time, error rate, and the
  number of magnifications per trial as functions
  of time (sessions one to six).

20
Response Time
21
Error Rate
35
30
25
20
15
10
5
0
1
2
3
4
5
6
session number
22
Number of Magnifications per Trial
23
Conclusions

• The novel zooming interface is well-suited for
  efficient gaze control.
• With this interface, mouse control is only
  slightly more efficient than gaze control.
• Using gaze control can be learned as quickly as
  using a mouse.