eye tracking: principles and applications 廖文宏 wen-hung liao [email protected] 12/10/2009
TRANSCRIPT
Outline
Eye Tracking Device Application areas State-of-the-art technology
Eye tracker @NCCUCS 1.0 (wearable) Eye tracker @NCCUCS 2.0 (remote) Gaze-based HCI Demo Conclusions
Eye Tracking Device
An eye tracker is a device for measuring eye positions and eye movements.
The most popular variant uses video images from which the eye position is extracted.
Input source: visible spectrum vs. infrared
Eye Movements Eye movements are typically divided into
fixations (when the eye gaze pauses in a certain position) and saccades (when it moves to another position).
The resulting series of fixations and saccades is called a scanpath.
Web Design (U. of Manchester)
Heat map Gaze plot
Usability Study:Google Search Evaluation
Gaming
Driving Behavior
[Andrew T. Duchowski]
Character Input
EyeWrite [Andrew T. Duchowski]
Media Research
The Poynter Institute : http://www.poynter.org Published first eye track study in 1991. More results published in 2000, 2004 and
2007. Eyetrack ’07:
http://www.poynter.org/content/content_view.asp?id=105035
Research Communities
ACM SIG/CHI Eye Tracking Research and Applications
(ETRA) Symposium: http://www.e-t-r-a.org/ COGAIN: Communication by Gaze
Interaction http://www.cogain.org References:
http://www.cogain.org/downloads/
State-of-the-art Eye tracking Technology
Hi-Speed (SMI iView X™ Hi-Speed) Head-Mounted System Remote tracking Integrated with LCD monitor
Mobile Eye Tracker 1.0 @ NCCUCS
eye camera
scene camera
Key Objectives
Low-cost (NTD 10,000 vs. 1,000,000) Mobility Easily customized for specific applications On-line processing Sampling rate? Accuracy?
System Architecture
Eye image
Preprocessing
Pupil detection
Gaze point projection
Scene image
Calibration
9 pairsof points
Calibration process
Pupil Detection (I):extracting feature points
Pupil Detection (II): Ellipse fitting using RANSAC
Pupil Detection (III):checking fitness measure
Ellipse model:
20,sin
coscossinsincos cy
cxba
yxp
n
bfScore
m
afScore
n
iiin
in
m
iiout
out
11
)(,
)(
,0
,)(,1
otherwise
PaThresholdapiff outbrightout
,0
,)(,1
otherwise
PbThresholdbpiff indarkin
Number of bright vs. dark points inside and out the perimeter
Results
HCI Application: Eye Scrolling
HCI Application: Eye Gaming
HCI Application: Eye Typing
Mobile Eye Tracker 2.0 @ NCCUCS
Improve the pupil detection algorithm to alleviate corneal reflection problem.
Enhance the accuracy by compensating for head movement.
Construct and test a remote eye tracker. More HCI applications using the remote eye
tracker. Use the eye tracking device to assist mobile
user interface design.
Improved Pupil Detection
Allowing Head Movement
Remote Eye Tracker
Experimental Results (Wearable)
Calibration
point
Original error Compensate for head
movement
(error/standard deviation)
1 98.17 32.91 (12.7)
2 137.54 55.89 (10.94)
3 108.20 29.69 (9.57)
4 76.30 28.79 (21.19)
5 113.30 22.49 (7.79)
6 117.86 34.83 (10.96)
7 116.53 33.31 (8.01)
8 112.56 16.03 (6.78)
9 146.73 26.49 (9.66)
1 cm = 38 Pixels
Accuracy (Wearable)
Experimental Results (Remote)Calibration point Original error Compensate for head movement
(error/standard deviation)
1 238.73 39.69 (24.6)
2 227.51 63.56 (22.82)
3 132.97 44.30 (21.49)
4 222.97 51.66 (30.94)
5 306.58 31.15 (16.71)
6 280.79 51.66 (21.83)
7 311.89 69.03 (41.57)
8 344.24 69.88 (31.71)
9 347.03 64.66 (24.41)
Accuracy (Remote Eye Tracker)
Demo: Web Browsing
Demo: Photo Viewing
Demo: Interactive Story Telling
Demo: Tic-Tac-Toe
Summary
Eye tracking device Low cost (<$10,000 NTD) Head-mounted with mobile functionality Remote eye tracking Allow slight head movements Accuracy: suitable for some HCI applications,
image viewing tasks, not high enough for reading HCI applications
http://www.cs.nccu.edu.tw/~whliao/dct/
More materials
Gaze Path Analysis
Recursive intersection Find the similarity between two scan paths Order of scanning is irrelevant Suitable for processing fixation data
Modified dynamic time warping Order of scanning is taken into account Can handle both fixation and saccade data
Recursive Intersection
Path 2
Path 1
Recursive Intersection: Example
Dynamic Time Warping (DTW)
Modified DTW (MDTW)
X-axisY-axis
time
MDTW Result