Gaze-controlled HCI and the Midas-Touch problem
Hendrik Koesling
Collaborative Research Centre 673 Alignment in Communication
Centre of Excellence 277 Cognitive Interaction Technology
Bielefeld University
Introduction
¾ “Gaze mouse”
Gaze-controlled HCI: Issues …
How to select an action?
How to detect the intention to select an action?
¾ Midas-Touch problem
Distinction between attention for exploration and
attention for selection
… and solutions
Volitional action-selection methods
¾ Gaze mouse + manual action selection
¾ Fixating
¾ Blinking
¾ Separate action-selection areas
… and solutions
Volitional action-selection methods
¾ Gaze mouse + manual action selection
¾ Fixating
¾ Blinking
¾ Separate action-selection areas
Research Questions
¾ Efficiency:
Is one of the fixation or blinking methods more efficient
than the other?
¾ Usability:
Do users prefer one method to the other?
¾ Acquisition:
How easy are the two methods to learn?
Experimental method
¾ Task
¾ Gaze typing
¾ Participants
¾ 20 German native speakers
¾ 10 female, 10 male; mean age 24.3 years
¾ Apparatus
¾ LC Technologies EyeGaze eye tracker @ 120 Hz
Experimental method
¾ Stimuli & Procedure
t
¾ Task area
¾ Input control area
…
¾ Interaction area
…
¾ German sentences
¾ 6-11 words, 55-65 characters
¾ Selection threshold: 300 ms
+MB
Experimental method
¾ Design
¾ 2 blocks (fixation, blinking)
¾ 5 trials per block
¾ Balanced for number of characters,
sequence of methods & gender
+ Questionnaire
¾ IV: Method, gender, seq. of methods
¾ DV: Error rate, completion time; quest.
Results – Error rate ER
¾ Fewer errors with blinking method
¾ No gender difference
¾ Without practice, blinking & fixation equal
¾ Practice benefit for blinking method
t(19) = 4.187; p < 0.001
Females: t(9) = 2.480; p = 0.038
Males: t(9) = 3.279; p < 0.001
Females vs. males: not signif.
Blink-fix: t(9) = 2.124; p = 0.063 (trend)
Fix-blink: t(9) = 4.351; p < 0.001
Results – Completion time CT
¾ No difference in CT between methods
¾ Females’ CTs longer than males’
¾ Practice benefit for fixation method
Not signif.
Females: not signif.
Males: not signif.
Females vs. males: t(9) = 2.232; p < 0.031
Results – Practice
Usability
¾ Initial improvement through practice
¾ Followed by possible fatigue effects
¾ User ratings
¾ Method easier to get used to?
¾ … less error prone?
¾ … more accurate?
¾ … less tiring?
¾ … more intuitive?
¾ Use gaze-controlled HCI?
Blink-fix: t(9) = 4.507; p < 0.001
Fix-blink: not signif.
Conclusions
¾ Both gaze-contingent selection methods present feasible approaches to solve
the Midas-touch problem.
¾ No method can be preferred over the other.
¾ Blinking method produces fewer errors and allows for more accurate selections
¾ Without practice, fixation method may not necessarily be more error-prone. Task
completion times in particular for the fixation method benefit from practice.
¾ Thus, after practice, fixation method may be more efficient than blinking method.
¾ Furthermore, fixation method is rated as less tiring and more intuitive – relevant
with regard to user-friendliness, user satisfaction and possible long-term effects
of novel interaction methods on users.
Conclusions
¾ Blinking method appears to be well suited for applications where only few
selections are needed, that does not rely on a “fluent” input stream and does
not allow much time for practice.
¾ Possible applications: ticket vending machines, filling in forms in multiple
choice style, sending action command to robots.
¾ Fixation method better suited for continuous, long-lasting and rapid
interaction with an application. When users are appropriately trained in using
the fixation method, they may consider this interaction method an intuitive,
user-friendly means of communication with their environment.
¾ Possible applications: virtual keyboards, in particular for users with motor
control deficits.
Outlook
¾ Test handicapped participants
¾ Examine practice effects more closely
¾ Evaluate alternative gaze-controlled action-selection methods
¾…
Thank you!