Eye movement detection project

klipitiklopwarrenΛογισμικό & κατασκευή λογ/κού

7 Νοε 2013 (πριν από 4 χρόνια και 6 μέρες)

99 εμφανίσεις

Eye movement
detection project

Final project by

Nir Arcushin and Ellen Plosker.

Introduction

What captures our attention as humans when
we look at different pictures?


Is there any part of the picture which is most
probable to capture our mind?


How can we study whose issues?

The method


The answer to those and many other questions
could be found in experiments using the
EyeLink II video
-
based eye tracking system
developed by SR Research to measure
human eye movements.


EyeLink II


As shown in the figure below, EyeLink II uses
a headset with infrared cameras mounted on
it to obtain information on a subject

s
current gaze position.



The method


Our project divided into
2
parts:



Part
1
: Installation of the system and
programming an experiment.



Part
2
: Running an experiment.

Part
1
-

Installation


Integration and configuration of the system
using supplied software from SR Research
Data Viewer.



Understanding provided templates written
in C and programming an experiment using
those templates.


Installation


General Scheme

Respond Box


Host PC


Hardware

Host PC


Hardware

EyeLinkII PCI Card

EyeLink II headband.


Host PC


Hardware

Host PC


Software

ROM
-
DOS
7.1




Introduced by Datalight
©

in
1989
as an MS
-
DOS
compatible
operating system engineered specifically
for embedded developers.



Require as little as
186
CPU



Small amount of RAM



Current version has kernel support for features
such as:
TCP/IP Stack, FAT
32

Host PC


Software

ROM
-
DOS
7.1

-

Continue

Display PC


Hardware



Standard
100
/
10
Mps Ethernet card



>
1
GHz Processor



40
-
80
GB hard drive



256
MB RAM



19


Monitor



USB license key (Data Viewer)

Display PC


Software



Windows
2000
/XP



Visual C
6.0
+



Data Viewer



ELSDK

Display PC


Software



EyeLink Software Development Kit



The tool kit is made up of two high level shared libraries



eyelink_core
-

implements all tracker related functions



eyelink_core_graphics


implements all eyelink graphics related
functions



Such as: the display of camera image, calibration, validation,
and drift correct



The eyelink_core_graphics is currently implemented using
Simple Direct Media Layer (SDL)

ELSDK

Code Scheme

main()/WinMain()

parseArgs()

readCfgFile()

run_trials()



main loop

do_picture_trial()



load & draw image



start recording

app_main()



edf creation



tracker init

bitmap_recording_trial()



record eye movment



capturing keystrokes

Alternatives For C Libs



Pylink


The Python EyeLink
module, allowing access to API



Used with VisionEgg (Python to
OpenGL)



Experiment Builder
-

End to End Solution by

SR
Research


Data Viewer


Analyzing results


End Solution by “SR Research”


Capable reading the native EDF format


Distinguish between different trails


Displaying gaze movement over image

Part
2
-

The experiment

In experiment the subject exposed to serias of
pictures, which include faces of different
people. His eye movements were recorded
and introduced to the researcher as the
original picture with stamps of gazes of the
subject.

The objectives


We assumed that when subject exposed to
picture in which there are faces he will
concentrate on them.



To approve this assumption we found
pictures of people that contain faces and
showed them to the subjects, whose eye
movements are recorded.

The results

After showing the serias of pictures to subject
Data Viewer puts stamps on the original
picture that show the gazes of subject.


We ran this experiment on
2
subjects and
those are the results :



The results

The results

The results

The results

The results

The results

The results

The results


The results

The conclusion


As you can see in the output images the
points of gazes are very close to the faces on
the pictures, even that there are other
interesting and colorful parts of image.


There is some small inaccuracy in gaze
points because of sensitivity of the headset.


So disregarding this small inaccurancy we
can see that indeed faces capture our mind
more then other part of image.