EyeGrip proposes a novel and yet simple technique of analysing eye movements for automatically detecting the users objects of interest in a sequence of visual stimuli mov- ing horizontally or vertically in front of the user’s view. We assess the viability of this technique in a scenario where the user looks at a sequence of images moving horizontally on the display while the user’s eye movements are tracked by an eye tracker. We conducted an experiment that shows the performance of the proposed approach. We also investigated the influence of the speed and maximum number of visible images in the screen, on the accuracy of EyeGrip. Based on the experiment results, we propose guidelines for designing EyeGrip-based interfaces. EyeGrip can be considered as an implicit gaze interaction technique with potential use in broad range of applications such as large screens, mobile devices and eyewear computers. In this paper, we demonstrate the rich capabilities of EyeGrip with two example applications: 1) a mind reading game, and 2) a picture selection system. Our study shows that by selecting an appropriate speed and maximum number of visible images in the screen the pro- posed method can be used in a fast scrolling task where the system accurately (87%) detects the moving images that are visually appealing to the user, stops the scrolling and brings the item(s) of interest back to the screen.

1. IT UNIVERSITY OF COPENHAGEN
EyeGrip: Detecting Targets in a Series of
Uni-directional Moving Objects
Using Optokinetic Nystagmus
Eye Movements
Shahram Jalaliniya - Diako Mardanbegi
IT University of Copenhagen
Pervasive Interaction Technology Lab

2. IT UNIVERSITY OF COPENHAGEN
MOTIVATION
• Information age & overwhelming users with data
• Data is getting more visual (e.g. Web, Facebook)
• Scrolling among visual data is becoming more popular

3. IT UNIVERSITY OF COPENHAGEN
• Scrolling includes:
- Scrolling
- Stopping the page
- Bringing the desired content back
(not always an easy task)

4. IT UNIVERSITY OF COPENHAGEN
EYEGRIP
EyeGrip automatically detects moving images that seem
to be interesting for a user among other scrolling images
by monitoring and analyzing user’s eye movements.

5. IT UNIVERSITY OF COPENHAGEN
EYEGRIP WORKS BASED ON OKN EYE MOVEMENTS
• OKN is an eye movement that tend to track the motion of
one element at a time in a set of unidirectional moving
stimuli
• OKN: Optokinetic nystagmus is a combination of saccadic
and smooth pursuit eye movements.

6. IT UNIVERSITY OF COPENHAGEN
HOW DOES EYEGRIP WORK?
When one of the images grabs our
attention we follow that image for
a longer time that creates a peak
in the OKN signal.
180$
200$
220$
240$
260$
280$
300$
320$
340$
360$
380$
6000$ 7500$ 9000$ 10500$12000$13500$15000$16500$18000$19500$21000$22500$24000$25500$27000$28
Original$data$
180$
200$
220$
240$
260$
280$
300$
320$
340$
360$
380$
6000$ 7500$ 9000$ 10500$ 12000$13500$ 15000$ 16500$ 18000$19500$21
Original$data$

7. IT UNIVERSITY OF COPENHAGEN
EXPERIMENT GOAL
Testing the feasibility of EyeGrip in different scrolling
conditions:
- Different speeds
- Maximum number of visible images on the screen (manipulated by
changing image width)

8. IT UNIVERSITY OF COPENHAGEN
EXPERIMENTAL DESIGN (3 × 2 )
• 20 participants
• 3 speeds
26.5, 37.5, and 49 °⁄sec
• 2 image widths
18°(
𝑊 𝑖𝑚𝑎𝑔𝑒
𝑊 𝑑𝑖𝑠𝑝𝑙𝑎𝑦
= 0.6) and 9°(
𝑊 𝑖𝑚𝑎𝑔𝑒
𝑊 𝑑𝑖𝑠𝑝𝑙𝑎𝑦
= 0.3)

9. IT UNIVERSITY OF COPENHAGEN
APPARATUS
• Head-mounted eye tracker with the Haytham open
source gaze tracker (20 Hz sampling rate)
• Laptop to display the scrolling images & collect eye data
34.5 cm
19.5cm
Small width condi ons: 1, 3,and 5
(a)
34.5 cm
19.5cm
Big width condi on: 2, 4, and 6
(b)

10. IT UNIVERSITY OF COPENHAGEN
EXPERIMENT TASK
• Visual search among faces: Participants should
press space key as soon as they see Bill Clinton’s
picture among other faces
• Participants repeated the task for all 6 conditions
• 40 random images of famous people is displayed
in each condition where 7 was Bill Clinton’s photos

11. USER ERROR
The error rate: total missing target images by
participants divided by total number of targets
Precision(NoEvent)"
n"6"
Window=10"
Window=16"
Window=20"
Window=30"
0"
1"
2"
3"
4"
5"
6"
7"
8"
9"
10"
Condi<on"
1"
Condi<on"
2"
Condi<on"
3"
Condi<on"
4"
Condi<on"
5"
Condi<on"
6"
Error$rate$(%)$
(b)$
Usererrorrate(%)
Cond. 1 2 3 4 5 6
Speed slow slow med med fast fast
Image
width
small big small big small big

12. IT UNIVERSITY OF COPENHAGEN
DATA ANALYSIS
• Cleaning data: Removing 5 participants with less than 75% data
• Normalization: finding left & right eye coordinates during the
experiment by displaying 2 red circles a the beginning of each
task. We used these coordinates to bring all the data in the same
range (min-max normalization)
• Aggregation: we aggregated the data from all 15 participants
• Labeling data: we used the space key to label the collected data

13. IT UNIVERSITY OF COPENHAGEN
EVENT DETECTION ALGORITHM
• We used the default setting for the Multilayer perceptron
algorithm in the WEKA with a single hidden layer
• Horizontal coordinate of the pupil center was the only
feature
• Sliding window is selected based on maximum
performance
- 30 frames for conditions 1,2,4
- 20 frames for conditions 3,6
- 16 frames for condition 5

14. EFFECT OF SPEED & MAX NO. OF IMAGES
• 10 cross-fold validation is used
• Repeated measure ANOVA
- Significant effect of image width
F(1,14) = 34.9, p < :0001
- No significant effect for speed
- But higher speed caused more error
(missing targets)
0"
10"
20"
30"
40"
50"
60"
70"
80"
90"
100"
Small"width" Big"width" Small"width" Big"width" Small"width" Big"width"
Slow" Medium"speed" Fast"
Accuracy'of'the'classifica1on'(%)'
(a)'
70"
75"
80"
85"
90"
95"
100"
Slow" Medium"
speed"
Fast"
Accuracy'of'the'classifica1on(%)'
(b)'
Small"image"width"
Big"image"width"
70
75
80
85
90
95
100
Accuracy'of'the'classifica1on'(%)'
0"
10"
20"
30"
40"
50"
60"
70"
80"
90"
100"
Small"width" Big"width" Small"width" Big"width" Small"width" Big"width"
Slow" Medium"speed" Fast"
Accuracy'of'the'classifica1on'(%)'
(a)'
70"
75"
80"
85"
90"
95"
100"
Slow" Medium"
speed"
Fast"
Accuracy'of'the'classifica1on(%)'
(b)'
Small"image"width"
Big"image"width"
"
"
"
"
"
"
"
"
"
"
"
Small"width" Big"width" Small"width" Big"width" Small"width" Big"width"
Slow" Medium"speed" Fast"
(a)'
70"
75"
80"
85"
90"
95"
100"
Slow" Medium"
speed"
Fast"
Accuracy'of'the'classifica1on(%)'
(b)'
Small"image"width"
Big"image"width"
70"
75"
80"
85"
90"
95"
100"
Small"image"
width"
Big"image"width"
Accuracy'of'the'classifica1on'(%)'
(c)'
Slow"
Medium"speed"
Fast"

15. IT UNIVERSITY OF COPENHAGEN
DESIGN GUIDELINES
• Images should move in a one direction at a
certain speed
• There should be a balance between moving
speed & max number of images
• The visual search task should not be very
complex otherwise all images will draw equally
high attention that increases falls positives

16. IT UNIVERSITY OF COPENHAGEN
STUDY 1: A PICTURE SELECTION SYSTEM
• 8 participants
• Speed: 37 °⁄sec
• Image width: 18 °
• Selecting Clinton pictures
0%#
10%#
20%#
30%#
40%#
50%#
60%#
70%#
80%#
90%#
100%#
Accuracy# Precision# Recall#
1"
2"
3"
4"
5"
Mental'
Demand'
Physical'
Demand'
Temporal'
Demand'
Performance' Effor t' Frustra: on'
5'lickert'scale'

17. IT UNIVERSITY OF COPENHAGEN
STUDY 2: MIND READING GAME
• 10 participants
• Select 1 of 4 characters
• Participants are asked to count
repetitions of selected person
• Accuracy:100%

18. IT UNIVERSITY OF COPENHAGEN
OTHER SUGGESTED APPLICATIONS
• Interaction with scrolling menus (e.g. scrolling cards
on Google Glass)
• EyeGrip for browsing Facebook page
• Advertisement on public displays
• Text reading assistant for small displays (slowing
down text when a user has problem with a word)
• Assistant for visual inspection in production lines
(automatically detecting unqualified products)

19. IT UNIVERSITY OF COPENHAGEN
RELATED WORK
• Pursuit [1]: is a calibration-free
technique to detect limited number
of moving objects on the screen
using smooth pursuit eye
movements.
• EyeGrip detects unlimited number
of unidirectional moving objects.
[1] M´elodie Vidal, Andreas Bulling, and Hans Gellersen. 2013. Pursuits: Spontaneous
Interaction with Displays Based on Smooth Pursuit Eye Movement and Moving Targets. In
Proceedings of UbiComp ’13. ACM, 439–448.

20. IT UNIVERSITY OF COPENHAGEN
CONCLUSIONS
• EyeGrip is a calibration-free and implicit eye
interaction technique to select an object among
other moving unidirectional objects (top-down
attention)
• EyeGrip can be used in gaze-contingent user
interfaces to detect what draws users attention
(bottom-up attention)
• Simpler algorithms (e.g. threshold-based method)
can be applied to detect the event in EyeGrip

21. IT UNIVERSITY OF COPENHAGEN