5. Now let's looking back the gaze native
functions
Seeing is essentially a perception, not a means
of action to make some effects in the world.
•It finds something to operate, and gets
continuous feedbacks of operation by other
means such as hand tactics etc.
6. Existing use cases of Gaze tracking in industry
• Control in shooting games
• Control of screen navigations
• Control of cursors or mouse pointer
Commanding
• Analysis of user attentions and areas of interest
Marketing research or usability test
7. Existing use cases versus gaze native
functions
Commanding
Miss-matches with
gaze native
functions, Leads to
fatigue, and
Requires learning
Marketing research or
usability test
Matches with gaze
native functions
8. Let’s get back to the gaze native functions
Eyes are basically receptor and used as an effector
ONLY WHEN…
Starting communication
with someone else
(including pet animals) by
eye-contact
Showing interest on
something to someone else
(even animals use this eye
gesture to request
something to human)
9. Applications on top of gaze tracking should
take advantage of its native functions
• As a receptor: Finding
• As an effector: Showing interest of/with
communication partner
Natural, No-fatigue, No-learning UX
[JPA2017-204737]
10. Further implications
As a receptor: Finding
• Tracking of the flow of
attention is more important
than the traditional point of
gaze.
As an effector: Showing interest
of/with communication partner
• Environment SLAM used
together with gaze tracking is
more important than before.
• Many use cases would require
only rough estimation of gaze.
12. Special hardware
Hard to adopt for volume uses
High price, limited use cases, and low usability
Traditional Gaze tracker requires Infrared light sources and expensive
imaging device (zoom camera, two cameras, and so on).
13. Use case limitation and low usability
• Body attachment is bothering
• Low usability
Infrared light source
in head mounted
type device
• Only for environment as Desktop PC
• It can capture gaze only in near distance (~1m)
• Suffers from outdoor light (eg, smart phone on the
road is under sun light)
Infrared light source
attached with
graphical monitor
14. Gaze tracking by commodity camera
• Implementation
• They may use Face/Eye-Ball models and track iris to approximate the point of
gaze.
• Advantages
• Low cost
• Unlike wearable type, users can have an open view
• Distance flexibility (a few meters)
• Free from environment with reasonable lighting condition
15. Challenge is the Accuracy
• Purple positions are also clearly captured under infrared light. The
infrared light source gives Purkinje reflections which give the
geometry of eye-balls and purples.
• On the other hand, the camera method rely on face/eye-ball models
and iris position tracking.
• It is not easy for commodity camera to track iris positions precisely.
• It is not easy for monocular camera to acquire precise Face/eye-ball 3D
models.
17. Calibration Hurdle
It is bothering. It makes mass-adoption difficult.
Traditional gaze tracker requires initial setup of personal
parameters by asking user to see several known points.
18. Background calibration
• Implementations
• They may calibrate personal eye parameters such as Eye-ball center positions,
and Visual/optical axis delta
• They may use iterative algorithm to optimize parameters
• They may take advantages of user’s operation of object selection (mouse
click, touch etc) to calibrate parameters in the background
• Advantage
• Users are not aware of calibration process.
19. Challenges of Background calibration
Smooth
UX
Face, Eye-
ball
Model
Sensor-
object
calibration
JPO PA2018-173743
20. Challenge: Smooth UX
Previous works
• Background
calibration doesn’t
bother users.
• But UX
before/during/after
calibration are
different.
Challenge
• Smooth transition
before/after
background
calibration are
required.
JPO PA2018-173743
21. Challenge: Face, Eye-ball 3D Model
Previous works
• Eye parameters
(eye ball center
depth,
optical/visual axis
delta) gives a
geometry
Challenge
• Eye ball position in
face model and iris
gives a geometry.
It needs Face
model calibration
besides that of
eyes.
22. Challenge: Sensor-object calibration
Previous work
• The attention
target object
positions relative
to camera position
are pre-calibrated
or fixed.
Challenge
• Imagine sensors in
a robot which can
move. Needs a
calibration of the
positions b/w
sensor and objects
23. Summary: Factors for Gaze tracking to be
among main-stream UIs
Natural use of
gaze
Commodity
device
Background
calibration
• Face 3D
• UX
• De-coupling
sensor-object
calibration