Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
9
, No
.
2
,
Febr
ua
ry
201
8
,
pp.
438
~
446
IS
S
N:
25
02
-
4752
,
DOI: 10
.11
591/
ijeecs
.
v9.i
2
.
pp
4
38
-
4
46
438
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Ga
z
e T
rack
ing
Alg
orith
m Us
ing
Nigh
t
Visio
n Camera
T.A.
Iz
z
uddin
1
*
,
M.H.
Jali
2
,
A.R.
Abdull
ah
3
,
R
. Sudir
m
an
4
1, 2, 3
Cent
er
for Robotic a
nd
Ind
ustria
l
Autom
at
i
on,
Facu
lty
of El
ec
tr
ic
a
l
Engi
ne
er
ing
Univer
sit Te
knik
al
Ma
lay
sia
Mel
aka
,
Hang Tuah
Ja
y
a
,
76100
Dur
ia
n
Tungga
l
,
Me
la
ka
,
Ma
lay
sia
4
Facul
t
y
of Elect
ric
a
l
Eng
ineeri
n
g,
Univer
si
ti T
ek
nologi
Ma
lay
sia
,
81310
UTM,
Johor Ba
hru,
Johor
Malay
s
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
un
9
, 201
6
Re
vised
N
ov
2
0
, 2
01
6
Accepte
d
Dec
11
, 201
6
Now
aday
s,
the
adva
nc
ement
of
m
edi
ca
l
te
chn
olog
y
h
as
given
birt
h
into
m
an
y
innova
t
ive
m
ac
hine
s
and
devi
c
es
to
improv
e
our
hea
lt
h
life,
espe
cia
l
l
y
to
those
who
are
disabl
ed
or
gif
t
ed.
On
som
e
peopl
e
with
seve
r
e
disabi
litie
s
such
as
quadr
ip
le
gi
a,
th
e
hum
an
e
y
e
-
ba
ll
does
not
onl
y
serv
e
as
a
vision
s
y
stem,
but
al
so
a
m
ea
ns
of
co
nve
y
ing
informa
ti
on
and
in
te
nt
i
on
to
othe
r
peopl
e
.
Thi
s
is
bec
ause
a
lt
hou
gh
quadr
ipl
eg
ic
pat
ie
n
ts
suffer
the
loss
of
m
oto
r
sensory
f
unct
ions
from
the
nec
k
and
bel
o
w,
upper
nec
k
fu
nct
ions
suc
h
as
the
vision
s
y
stem
is
nor
m
al
l
y
spare
d
.
Thi
s
e
nabl
es
the
pa
ti
e
nt
to
cont
rol
the
m
ovement
o
f
his/he
r
e
y
e
ball
s
to
conv
e
y
d
esire
d
informati
on
.
Although
m
an
y
sim
il
ar
r
e
sea
rch
es
h
ave
b
ee
n
d
one
,
th
is
pape
r
proposes
the
use
of
image
proc
essing
on
image
ca
p
t
ure
d
using
web
c
am
with
it
s
Infr
a
-
Red
(IR)
fil
ter
removed
(a.
k.
a
night
visi
on)
to
ac
hi
eve
robustness.
Thi
s
al
lows
th
e
al
gorit
hm
to
pro
per
l
y
tr
ac
k
th
e
l
oca
t
ion
of
the
ir
i
s
despit
e
of
it
s
a
nd
t
he
pupil
col
or
v
ari
a
ti
ons.
Two
image
pro
ce
ss
ing
a
lgori
th
m
s
are
the
n
use
d,
e
ac
h
with
owns
tra
deof
f
b
et
wee
n
spe
ed
a
nd
ac
cur
acy
.
Anal
y
sis
on
bot
h
al
gorit
hm
s
show
s good
tra
c
king
per
form
an
c
e
despi
te of
th
e m
ent
ione
d
tra
d
e
off
Ke
yw
or
d
s
:
Im
age p
r
ocessi
ng
Gaze
de
te
ct
ion
Pupil
d
et
ect
io
n
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
T.A. I
zz
uddi
n
,
Ce
nter fo
r
R
obotic an
d Ind
us
tria
l A
uto
m
at
io
n,
Faculty
of E
le
ct
rical
En
gi
ne
erin
g
Un
i
ver
sit
Te
kn
ikal
Ma
la
ysi
a Me
la
ka,
Hang T
ua
h
Jay
a,
76100 D
ur
ia
n
T
un
gg
al
,
Mel
aka,
Ma
la
ysi
a
Em
a
il
:
ta
r
m
iz
i
@u
te
m
.ed
u.
m
y
1.
INTROD
U
CTION
Our
fr
ee
do
m
to
m
ov
e
f
reely
is
inv
al
ua
ble
to
us.
Howe
ve
r
not
e
ver
y
on
e
sh
a
red
the
sa
m
e
ta
ste
of
fr
ee
do
m
.
For
s
om
e,
su
ch
as
a
quad
riplegic
patie
nt,
or
a
pa
raly
zed
per
s
on,
unable
t
o
m
ov
e
in
dep
e
nde
ntly
is
dem
enting.
To
ov
e
rco
m
e
this
m
atter
m
any
researc
hes
ha
ve
been
do
ne
in
the
m
edical
fiel
d
to
i
m
pr
ove
the
qu
al
it
y
of
li
fe
of
a
quad
riple
gic.
T
he
fact
t
hat
m
os
t
qu
ad
r
iplegic
retai
n
m
os
t
of
it
s
uppe
r
nec
k
f
unct
io
n,
giv
e
way
for
resea
rch
e
r
to
le
verage
this
into
assist
ing
them
to
m
ov
e
inde
pende
ntly
again
[1
]
-
[3
]
.
I
n
m
os
t
researc
hes
an
e
le
ct
ric
wh
eel
c
ha
ir
is
us
ed
to
a
chieve
this
obje
ct
ive
[
1],
[4
]
.
Bi
o
-
sig
nals
s
uc
h
as
the
m
ov
em
ent
of
the
ey
e
ball
[
5],
[4
]
,
E
EG
sign
al
s
an
d
EM
G
sig
nals
a
re
t
ran
sla
te
d
in
co
m
m
and
to
m
ove
the
wh
eel
c
ha
ir
[
4].
This
pro
j
ect
pro
poses
the
use
of
im
age
pr
ocessin
g
te
ch
ni
qu
e
to
trac
k
the
m
ov
em
ent
of
the
ey
ebal
l,
by
locat
ing
sub
j
ec
t’s
pu
pil.
T
he
l
ocati
on
ca
n
la
te
r
be
tra
ns
la
te
d
int
o
c
o
m
m
an
ds
t
o
m
ov
e
t
he
w
heelchair
to
le
ft
or
rig
ht.
This
le
ads
the
res
earc
her
s
t
o
de
velo
p
a
syst
e
m
that
can
help
the
disable
pe
rs
on
by
m
ov
in
g
th
e
ey
e
balls.
Gen
e
rall
y,
gaz
e
detect
ion
syst
e
m
is
a
syst
e
m
that
captur
e
d
im
age
data
of
ey
e
pupil,
de
te
ct
ing
a
nd
trac
king
m
ov
e
m
ents
of
t
he
us
er
’s
ey
es,
a
ppr
ox
im
at
ing
the
li
ne
-
of
-
sig
ht
vect
or
,
an
d
m
ov
e
the
el
ect
ric
wh
eel
c
hair
in d
esi
re
d
directi
on
b
ase
d
on
th
e
per
s
on’s
ey
es
and
it
is
a
syste
m
based
on
e
ye
bio
m
et
rics
[
6].
Ey
e
trackin
g
syst
e
m
/gaze
trackin
g
syst
em
us
e
the
im
a
ge
pr
oc
essing
te
ch
niq
ue
to
c
onve
rt
the
in
pu
t
dat
a
int
o
dig
it
al
form
an
d
so
m
e
m
a
them
at
ic
al
op
erati
on
s
a
re
ap
plied
to
the
data
to
create
m
or
e
i
m
age
en
ha
ncem
ent
[7
]
.
Using
im
age
processi
ng
te
chn
i
qu
e
,
the
im
age
of
the
picture
will
th
en
be
ide
ntifi
ed
on
col
or
s
,
sh
a
des,
con
t
rasts
an
d
s
hap
e
s
that
can
no
t
be
see
n
by
norm
al
ey
e.
Subseque
ntly
,
after
pe
rfor
m
i
ng
im
age
pr
oc
essin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Ga
ze
Tr
ackin
g Al
gorit
hm Usi
ng Ni
ght Vi
sio
n
C
am
er
a
(
T.
A
. I
zz
uddi
n
)
439
te
chn
iq
ue,
i
nfo
rm
ation
on
the
locat
ion
of
th
e
pupil
will
then
be
se
nt
to
a
m
ic
ro
process
or
that
will
con
t
ro
l
the
directi
onal
m
ov
em
ent o
f
a
wheel
chair
m
oto
r
[
16]
.
Ed
ge
detect
io
n
is
an
im
age
processin
g
te
c
hniqu
e
for
fin
ding
the
boun
dar
i
es
of
ob
j
ect
s
w
it
hin
im
ages
[8
]
.
T
his
can
be
achie
ved
by
detect
ing
di
sco
ntinu
it
ie
s
in
bri
ghtness
.
Ed
ge
detect
io
n
is
cast
of
f
for
i
m
age
segm
entat
ion
and
data
e
xtract
ion
in
areas
s
uc
h
as
im
age
processin
g,
c
om
pu
te
r
visio
n,
a
nd
m
achine
vision.
Com
m
on
edg
e
detect
ion
al
gorithm
s
include
Sobel,
Ca
nny,
Pr
ewit
t,
Ro
be
r
ts,
an
d
f
uzzy
log
ic
m
et
ho
ds.
But
in
this
case,
the
us
e
of
ca
nn
y
m
et
ho
d
is
esse
ntial
as
the
in
pu
t
fr
am
e
is
pr
oces
se
d
for
dig
it
al
i
m
ages
[9
]
.
Thi
s
m
et
ho
d
is
no
t
finali
zed
sinc
e
it
is
in
t
he
m
idd
le
of
pro
cess.
T
he
res
ul
t
of
ed
ge
det
ect
ion
m
et
ho
d
will
be
con
ti
nue
d wit
h som
e
m
or
phol
og
y
f
undam
ental
f
or
pupil e
xtracti
on.
This
pa
per
pro
po
s
es
two
ty
pe
of
m
et
ho
d
in
order
to
d
et
ect
the
iris.
The
fi
r
st
ste
p
is
to
detect
the
us
er
face
by
us
in
g
sk
in
colo
r
segm
entat
ion
and
fin
d
the
fac
e
of
the
per
s
on
ov
e
r
the
im
age.
This
m
et
hod
cat
egorized
the
i
m
age
into
fac
ia
l
and
non
-
fa
c
ia
l
reg
ion
si
nc
e
hu
m
an
sk
in
colo
r
is
dissim
il
ar
fr
om
the
na
t
ur
e.
Af
te
r
t
he
face
of
the
per
s
on
is
detect
ed,
the
propose
d
us
e
d
of
ci
rc
ular
H
ough
Tr
ans
f
or
m
per
m
itted
to
locat
e
the
iris
in
the
face
re
gion.
T
his
pro
j
ect
wil
l
fo
cu
s
m
or
e
on
ey
e
detect
io
n.
T
he
H
ough
trans
form
is
a
featur
e
abstracti
on
te
c
hn
i
qu
e
us
e
d
in
i
m
ag
e
analy
sis,
com
pu
te
r
vi
su
al
iz
at
ion
,
a
nd
di
gital
i
m
age
processi
ng
[
10]
.
Th
e
pur
po
se
of
thi
s
trans
f
or
m
is
to
disc
over
im
perfect
occa
sio
ns
of
ob
j
ect
in
side
a
ce
rtai
n
cl
ass
of
s
hap
e
s
by
a
vo
ti
ng
proce
dure
.
T
his
m
et
ho
d
is
c
onside
r
ed
t
o
be
s
uitable
since
the
ci
r
cular
sh
a
pes
can
detect
t
he
pu
pil
reg
i
on [11
]
,
[
12]
.
Re
gionpro
ps
is
al
so
us
ed
in
this
pro
j
ect
,
whic
h
it
can
us
ed
to
cal
culat
e
o
bj
ect
pro
pe
rtie
s
fr
om
the
bin
a
ry
i
m
age.
Ever
y
re
gion
prop
s
has
s
om
e
s
pecific
pro
per
t
ie
s
that
can
be
us
e
d
to
m
easur
e
the
bin
ary
im
a
ge.
So
m
e
of
exam
ple
pro
per
ti
es
that
can
be
use
is
boundi
ng
box
pro
per
ti
es.
This
is
re
pr
es
ented
as
a
4
-
ve
ct
or
wh
e
re
the
first
two
e
ntries
a
r
e
the
x
a
nd
y
c
oor
din
at
es
of
t
he
up
per
le
ft
c
orner
of
the
bo
unding
box,
an
d
the
two
la
st
entrie
s
are
the
wi
dth
a
nd
th
e
heig
ht
of
the
box
[13].
T
he
su
it
able
prop
e
rtie
s
that
ca
n
be
use
are
centr
oid
pro
pe
rtie
s
that
cal
cu
la
te
the
cente
r
co
ordinati
on
of
bi
nar
y
obj
e
ct
.
The
cent
ro
i
d
will
be
us
e
d
gaze
detect
ion p
urp
os
e
d.
2.
RESE
ARC
H M
ET
H
OD
This
syst
em
c
on
sist
s
of
3
pa
rts
w
hich
is
the
ha
rdwa
re
dev
el
op
m
ent
f
or
t
he
fi
rst
pa
rt.
A
fter
t
he
hard
war
e
co
nfi
gurati
on
is
read
y,
the
ne
xt
ste
p
will
be
the
al
gorithm
to
trac
k
the
pupil
re
gio
n.
The
sec
ond
part
will
sh
ows
the
process
un
der
ta
ken
to
e
xtra
ct
pupil
bo
unda
ry.
T
he
la
st
pa
rt
is
to
for
m
ulate
the
gaze
tr
ackin
g
directi
on.
Figure
1. O
veral
l gaze
detect
ion sy
ste
m
2.1. Hea
dset
Configur
at
io
n
This
pro
j
ect
c
on
sist
s
of
se
ve
ral
dev
ic
es
th
at
need
to
be
consi
der
e
d
f
or
the
accuracy
of
web
cam
placem
ent.
Ther
ef
or
e
,
a
head
set
is
design
by
us
ing
S
OLID
WO
RK
s
o
th
at
the
web
cam
can
be
placed
on
to
p
of the
hea
d
to
iden
ti
fy t
he gaz
e posit
ion.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
2
,
Fe
br
uary
201
8
:
4
3
8
–
4
4
6
440
Figure
2. Desi
gn of
gaze
dete
ct
ion
headset
2.2. Nig
ht Visi
on
C
amer
a
The
co
ntour
a
nd
c
olor
for
bo
t
h
pupil
an
d
iris
of
Asia
n
pe
op
le
is
m
or
e
to
da
r
k
brown,
un
li
ke
Euro
pean
pe
ople
w
hich
it
s eye
s intensit
y between pupil
and
iris can
be
see
n
by clea
r
visi
on
or
norm
al
c
olored
web
cam
[14].
To
overc
om
e
this
pro
blem
,
the
use
of
ni
gh
t
visio
n
ca
m
era
can
be
us
e
d
to
overc
om
e
the
pro
blem
s since it ca
n
cl
early
diff
e
re
ntiat
e
be
tween
pupil a
nd iris
of the
hu
m
an
ey
es.
The
use
or
real
Night
Visio
n
Cam
era
is
expensive
.
H
ow
e
ve
r
this
can
be
a
chieve
by
rem
ov
i
ng
t
he
IR
filt
er
of
a
nor
m
al
cheap
we
bc
a
m
su
ch
as
s
how
n
i
n
Fi
gure
3.
By
rem
ov
in
g
the
IR
filt
er,
it
al
lows
the
in
fr
a
red
li
gh
t
to
be dete
ct
ed
by t
he nig
ht v
isi
on cam
e
ra.
Norm
al
LED
te
nd
s
to
m
ake
the
patie
nt
or
us
e
r
feels
uncom
fo
rta
ble
du
e
t
o
li
gh
t
intensit
y
that
i
m
m
erse
towa
r
ds
the
ey
es.
T
he
f
unct
ion
of
the
IR
L
ED
is
to
pro
vid
e
an
add
it
io
nal
li
ght
towards
t
he
hum
an
visio
n
s
o
t
hat
it
can
be
detect
ed
by
night
vis
ion
cam
era.
The
a
dvanta
ges
of
this
cam
era
are
it
detect
th
e
iris
and pr
ov
i
de
a
ddit
ion
al
li
ght t
hat can
not b
e s
een
by no
rm
al
visio
n.
Figure
3.
W
e
bc
a
m
w
it
h
ni
ght visio
n
2.3. Alg
orit
h
m for G
az
e Trackin
g
Fo
r
both
al
gori
thm
s,
the
proc
ess
of
Im
age
Acquisi
ti
on,
E
dg
e
Detect
io
n,
Im
age
Fil
le
d
and
Creat
es
a
Stru
ct
ur
al
Ele
m
ent
are
i
den
t
ic
al
[15].
T
he
dif
fer
e
nt
c
ome
s
w
he
n
the
pupil
is
detect
e
d
a
nd
it
ne
eds
to
be
centrali
zed.
Bo
th
Al
gorithm
s
will
us
e
diff
e
r
ent
m
ediu
m
in
orde
r
to
i
den
ti
fy
the
c
oord
i
na
te
s
and
to
highli
gh
t
the b
i
nar
y
reg
i
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Ga
ze
Tr
ackin
g Al
gorit
hm Usi
ng Ni
ght Vi
sio
n
C
am
er
a
(
T.
A
. I
zz
uddi
n
)
441
Figure
4:
Dev
e
lop
m
ent o
f
g
az
e tracki
ng alg
ori
thm
2.4.
I
ma
ge Ac
quisi
tion
First
par
t
of
this
pro
j
ect
is
by
captu
rin
g
the
i
m
age
wh
i
ch
will
be
in
an
RGB
i
m
age.
The
im
age
acqu
isi
ti
on
wil
l
fr
am
e
these
im
ages
from
re
corde
d
vid
e
o
a
nd
st
or
e
d
in
th
e
MATLAB
director
y.
T
he
i
m
age
after
that
will
be
co
nv
e
rted
into
a
gr
ay
scal
e
i
m
age
since
the
pix
el
store
d
in
the
i
m
ages
is
m
os
tly
br
igh
t
a
nd
dark.
Af
te
r
tha
t,
the
gray
scal
e
i
m
age
is
conver
te
d
into
bi
na
ry
i
m
ages
tha
t
rep
r
esent
0
f
or
blac
k
an
d
w
hi
te
fo
r
1.
This
will
re
su
lt
in
black
a
nd
w
hite
im
ages.
As
we
know
that
our
pu
pi
l
is
m
os
tl
y
black,
t
hus
it
will
s
easi
ly
getti
ng
t
he regi
on as
we
c
onve
rt it
into bina
ry
.
2.5.
Ed
ge
D
e
te
ction
Af
te
r
the
bin
ar
y
i
m
age
is
ob
ta
ined
with
desired
th
reshol
d
le
vel,
the
ne
xt
process
w
ould
be
detect
in
g
the
e
dg
es
of
t
he
area
of
inte
re
st
in
the
im
age
w
hich
is
the
pupil
portio
n,
in
orde
r
t
o
rem
ov
e
t
he
portio
n
of
the
pupil
out
f
r
om
the
e
nh
a
nce
d
i
m
age.
The
c
hose
n
filt
er
wh
i
ch
is
the
can
ny
filt
er
has
bee
n
im
ple
m
ented
to
the
i
m
ages
in
orde
r
to
get
the
noti
ceable
edg
e
s.
The
ca
nn
y
e
dge
detect
or
first
sm
oo
ths
the
im
age
to
el
im
inate
and
reduce
noise
.
I
t
then
fin
ds
th
e
i
m
age
gra
dient
to
hi
gh
li
ght
reg
i
on
s
with
hi
gh
sp
at
ia
l
de
ri
vati
ves
w
hich
m
ean
the
sh
a
rp
cha
nges
in
im
age
bri
ghtness
t
o
ge
t
i
m
po
rtant
in
f
or
m
at
ion
from
the
im
age.
Ca
nn
y
filt
er
bee
n
c
ho
s
e
n
du
e
to
t
he
c
harac
te
risti
c o
f
t
he
eye
s which
is roun
d
a
nd cur
ved, th
us i
t i
s can pro
du
ce
d
t
he
d
esi
re
d res
ult.
2.6. Fi
ll
Ho
le
s
Af
te
r
t
he
fin
al
resu
lt
we
will
get
the
desir
ed
ed
ge
wh
ic
h
is
the
pu
pil
center,
it
is
an
esse
ntial
requirem
ent to det
ect
the m
ove
m
ent o
f
t
he h
um
an
ey
es. A
s
we kno
w,
pupi
l porti
on h
a
s
be
en used
for t
r
ackin
g
the
gaze
m
ov
em
ent
ei
ther
righ
t
or
le
ft.
I
n
order
to
m
ake
t
he
pupil
portion
to
be
m
ore
no
ti
ceable
,
di
la
ti
on
process
has
be
en
ap
plied
t
oward
s
t
he
im
age.
The
n
the
m
orpholo
gical
ope
rati
on
is
im
plem
ented
to
the
i
m
age.
Fil
l
ho
le
s
act
as
a
fu
ncti
on
th
at
fill
an
i
m
age
with
wh
it
e
pi
xel.
It
fill
s
al
l
the
ho
le
s
that
pr
ese
nt
in
the
im
age
thu
s
m
ake th
e im
age full
in w
hite pixels
.
2.7. Cre
at
in
g Struc
tu
ri
n
g
E
le
ments
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
2
,
Fe
br
uary
201
8
:
4
3
8
–
4
4
6
442
This
process
usi
ng
str
uctu
ral
el
em
ent
in
order
to
c
reates
a
la
rg
e
s
hap
e
s
of
bin
a
ry
im
a
ge
from
the
pupil
re
gion.
S
ince
the
syst
e
m
us
ing
a
ni
ght
vision
c
am
era
,
the
re
g
i
on
t
ha
t
sh
ows
t
he
da
r
kest
col
or
is
on
ly
the
pupil
reg
i
on.
Thu
s
the
us
e
of
structu
rin
g
el
em
ent
tog
et
her
with
dilat
ion
c
an
help
to
crea
te
s
i
m
age
with
m
or
e
prom
inent an
d
la
rg
er
. S
truct
uri
ng
elem
ent co
m
es w
it
h
m
any
p
aram
et
ers
su
ch
as d
ia
m
on
d, rectang
le
s, oct
a
gon
and
dis
k
str
uct
ur
i
ng
el
em
ents
.
All
of
t
he
pa
ram
et
ers
com
e
with
s
pecific
set
ti
ng
f
or
the
structu
ral
el
em
ents.
Fo
r
this
projec
t,
the
s
uitable
par
am
et
er
is
disk
str
uctu
rin
g
el
e
m
ent.
Disk
structu
rin
g
el
e
m
ents
will
create
s
a
disk
-
sha
pe
d
of
bin
a
ry
i
m
age
with
de
sire
d
ra
diu
s
.
But
the
de
sired
rad
i
us
m
us
t
sat
isfy
the
non
-
ne
gative
i
ntege
r
.
The
la
r
ge
r
the
value o
f
ra
dius
, th
e
struct
ur
i
ng elem
ents w
il
l bec
om
e larger.
2.8. Pupil
Loc
aliz
at
ion
usin
g Reg
i
on
Pro
p
s a
n
d
Circul
ar
Houg
h
Wh
e
n
the
pupi
l
portio
n
has
be
en
detect
ed,
t
he
ne
xt
ste
p
is
to
tr
ack
this
portio
n
from
the
rest
of
the
i
m
age.
It
ca
n
be
ob
ta
ine
d
by
detect
in
g
the
com
par
ison
of
the
black
an
d
w
hite
of
the
im
age.
T
o
ide
nt
ify
the
portio
n
of
wh
i
te
pix
el
s,
t
he
s
yst
e
m
fo
r
Algorit
hm
1
use
r
egio
npr
op
s
f
unct
ion
to
create
s
pro
per
ti
es
t
ha
t
can
m
easur
es
an
d
highli
gh
t
the
i
m
age
reg
io
n
with
boun
ding
box,
wh
il
e
al
gorithm
2
will
us
e
Ci
rcu
la
r
Hou
gh
Transf
or
m
to
detect
the
rou
nd
im
age
and
centrali
zed
the
i
m
age.
The
Im
age
will
be
highli
gh
te
d
w
i
th
ci
rcle
ou
tl
ine t
hat act
ually
the pr
op
e
rtie
s of Circ
ular
H
ough
Tra
nsfo
rm
.
2.9. Gaz
e Tra
cking
Directi
on
The
in
pu
t
im
a
ge
give
n
to
th
e
MATLAB
f
or
processin
g
will
pr
od
uce
a
n
outp
ut
i
m
age
with
the
coor
din
at
es o
f
the
pu
pil.
T
he
ou
t
pu
t
im
age w
il
l
be
a
fi
xed s
iz
e
an
d
the
siz
e
will
be
di
vide
d
int
o
si
x
bloc
ks
a
nd
the
pu
pil
coor
din
at
es
in
the
center
will
indi
cat
e
it
as
strai
gh
t.
Wh
ere
the
blo
c
k
in
the
s
econd
row
a
nd
third
colum
n
will
ind
ic
at
e
the
po
si
ti
on
of
the
pu
pil
as
le
ft
and
si
m
i
la
rly
the
blo
c
k
in
the
s
econd
r
ow
an
d
first
colum
n
will
ind
ic
at
e
the
pupi
l
po
sit
ion
a
s
Ri
gh
t.
[
1]
By
set
ti
ng
the
co
ord
inate
s
of
eac
h
blo
c
k.
T
he
out
pu
t
i
m
ages w
il
l be
consi
der
e
d
a
s a
r
ect
an
gle
with
certai
n
le
ng
t
h and o
f br
ea
dth.
But
a
m
od
ific
at
ion
of
this
m
et
ho
d
would
be
pro
posed
i
ns
te
ad
of
us
in
g
the
c
oord
i
na
te
s
fo
r
im
age
siz
e,
the
pro
posed
m
et
ho
d
w
ou
l
d
be
ide
ntif
yi
ng
the
di
recti
on
of
the
gazing
by
cal
culat
ing
t
he
an
gle
of
sight
.
The
distance
f
r
om
us
er
ey
e
can
be
c
al
culat
ed
f
ro
m
the
we
bcam
and
the
distance
of
t
he
us
er
ey
e
ca
n
a
lso
be
ca
lc
ulate
d.
W
it
h
these
two
va
lues,
the
val
ue
of
θ
can
be
determ
ined.
This
can
be
un
de
rstood
by
lookin
g
at
the
T
heorem
Pyt
hagor
as
co
ncep
t.
But,
th
e
we
bcam
it
se
lf
has
bee
n
pr
ov
i
ded
with
th
e
viewi
ng
an
gl
e.
Th
e
viewin
g
an
gle
ta
ke
place
with
60
de
gr
ee
to
ta
l
as
we
ta
ke
an
exam
ple
of
320*240
ou
t
put
i
m
ages,
the
320
is
the
x
-
axis
of
t
he
f
ram
es
and
the
viewi
ng
a
ng
le
of
the
we
bcam
will
be
synch
ronize
w
it
h
it
.
The
co
ndit
io
n
would be
li
ke
t
his:
-
a.
Bl
ock
(2,
1)
&
(2,
2)
:
T
he
ou
tpu
t
c
oor
din
at
e
s
of
the
i
ris
sat
isfyi
ng
belo
w
conditi
on
will
detect
the
pupil
po
sit
io
n
as
Ri
ght.C
on
diti
on
: i
f
(
(Cent
ro
i
d
>=
Bl
ock
(2,
2) &
(2,
1)))
b.
Bl
ock
(2,
3)
&
(2,
4)
:
T
he
ou
tpu
t
c
oor
din
at
e
s
of
the
i
ris
sat
isfyi
ng
belo
w
conditi
on
will
detect
the
pupil
po
sit
io
n
as
Str
ai
gh
t.C
onditi
on:
if
((
Ce
ntr
oi
d
>
Bl
ock
(2,
3)
an
d
cent
ro
i
d
<
Bl
ock
(2,
2)
&&
ce
ntr
oid
>
Bl
ock
(2,
4) an
d
ce
ntr
oid
<
Bl
ock (
2, 5))
c.
Bl
ock
(2,
5)
&
(2,
6)
:
T
he
ou
tpu
t
c
oor
din
at
e
s
of
the
i
ris
sat
isfyi
ng
belo
w
conditi
on
will
detect
the
pupil
po
sit
io
n
as
Lef
t. Co
nd
it
io
n: if
(
(ce
ntr
oid
s
>=
Bl
ock
(2,
5) &
(2,
6)))
Figure
5. Direc
ti
on
detect
ion
m
et
ho
d
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Ga
ze
Tr
ackin
g Al
gorit
hm Usi
ng Ni
ght Vi
sio
n
C
am
er
a
(
T.
A
. I
zz
uddi
n
)
443
3.
RES
ULT A
ND D
I
SCUS
S
ION
The
fi
rst
ste
p
of
the
local
iz
at
ion
of
t
he
pu
pil
is
by
captu
re
the
im
age
of
th
e
hu
m
an
ey
e
in
RGB
c
olor
for
the
im
age
acqu
isi
ti
on
process.
T
he
im
age
form
at
will
be
JP
G
f
orm
a
t
that
can
be
read
by
M
ATL
AB
com
m
and
.
T
he
i
m
age
al
so
m
us
t
be
store
d
i
n
MAT
LAB
di
rector
y
s
o
th
at
the
com
m
and
window
ca
n
re
ad
th
e
i
m
age
.
Figure
6:
Nor
m
al
RGB i
m
age in
to
night
vis
ion
im
age
Figure
6
s
how
s
the
im
age
of
an
ey
e
in
ori
gi
nal
RGB
c
olor
captu
red
f
ro
m
night
visi
on
w
ebcam
.
The
night
visio
n
c
a
m
era
al
lows
t
he
in
fr
a
-
re
d
li
gh
t
t
o
im
m
erse
to
it
.
T
her
e
f
or
e
,
the
pupil
reg
i
on
ca
n
be
cl
early
identifie
d.
Figure
7.
Proce
ss of
pupil l
oca
li
zat
ion
Figure
8. Im
age is d
et
ect
ed
3.1. Alg
orit
h
m Perf
orm
ance
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
2
,
Fe
br
uary
201
8
:
4
3
8
–
4
4
6
444
Ba
sed
on
the
Figure
9,
it
can
be
s
how
n
th
at
the
blu
e
li
ne
process
th
e
im
age
faster
c
om
par
ed
to
th
e
red pl
otted li
ne
. T
his is
du
e
to
the
functi
on
use
d by the
b
l
ue
li
ne
is t
aki
ng less tim
e to process.
Figure
9.
Per
f
orm
ance o
f
t
he f
un
ct
io
n use
d
i
n M
at
la
b
Table
1.
Tim
e Perfo
rm
ance f
or A
l
gorithm
1
Line Nu
m
b
e
r
Co
d
e
Calls
Total Ti
m
e
% Ti
m
e
12
d
ata=
g
etsn
ap
sh
o
t (
v
id
);
299
1
3
.96
4
s
4
5
.05
%
23
i
m
sh
o
w (
d
ata);
299
9
.41
9
s
3
0
.3%
1
v
id
=
v
id
eo
i
n
p
u
t(‘
win
v
id
..
1
1
.94
1
s
6
.3%
18
e=
ed
g
e(BW
6
,
’can
n
y
’)
;
299
1
.33
1
s
4
.3%
20
BW=
i
m
f
ill
(e,
’ho
les’);
299
0
.74
3
s
2
.4%
All oth
er
Lines
3
.64
5
s
1
1
.7%
Totals
3
1
.04
3
s
100%
Table
2.
Tim
e Perfo
rm
ance f
or A
l
gorithm
2
Line Nu
m
b
e
r
Co
d
e
Calls
Total Ti
m
e
% Ti
m
e
29
[
cent
ersBrig
h
t,
rad
iiBrig
h
…
299
1
1
.81
9
s
3
0
.3%
11
d
ata=
g
etsn
ap
sh
o
t (
v
id
);
299
9
.71
3
s
2
4
.9%
24
i
m
sh
o
w (
d
ata);
299
9
.35
4
s
2
4
.0%
1
v
id
=
v
id
eo
i
n
p
u
t (
‘winv
id
..
1
1
.86
7
s
4
.8%
17
e=
ed
g
e(BW
5
,
’can
n
y
’)
;
299
1
.20
0
s
3
.1%
All oth
er
Lines
5
.00
8
s
1
2
.9%
Totals
3
8
.95
3
s
100%
3.2. Accur
ac
y of G
az
e D
etec
tion
The
la
st
is
the
te
st
fo
r
accu
ra
cy
,
Table
3,
Table
4
an
d
Tab
le
5
sh
ows
the
accuracy
of
ga
ze
detect
ion
by
m
ov
in
g
the
ey
eball
to
the
le
ft
and
rig
ht.
Both
te
chn
i
que
pr
od
uce
acc
ur
at
e
gaze
a
nd
the
detect
ion
can
be
cl
early
seen.
But
the
pro
blem
occu
rs
f
or
bo
t
h
te
ch
nique
s
w
hen
the
ey
e
gazes
t
o
the
m
os
t
righ
t
w
hich
i
t
ind
ic
at
es the
a
ng
le
of
36 d
e
gree. T
his is
du
e
to the su
rro
undi
ng
li
ght t
hat im
m
erse tow
ar
ds
t
he
ey
es.
Fo
r
this p
r
oj
ect
, th
e requirem
e
nt is f
or
f
ast
com
pu
ta
ti
on
al
w
it
h
high accur
ac
y. Fo
r
both alg
or
it
hm
s,
it
us
es
ca
nn
y
e
dge
detect
io
n
wh
ic
h
detect
s
the
ed
ge
bo
un
dar
y
of
pupil
reg
i
on.
But
f
or
tim
e
co
m
pu
ta
ti
on
al
,
Algorithm
1
proces
s
faster
c
om
par
ed
to
A
lgorit
hm
2
with
ti
m
e
diff
ere
nt
of
7.9
1
sec
onds
.
The
refo
r
e
for
rob
otics app
li
c
at
ion
that requi
res
fast com
pu
ta
ti
on
, A
l
gorith
m
1
is
m
or
e su
it
able and
can be i
m
ple
m
ented
into
a r
obotics w
he
el
chair syste
m
.
Table
3.
Acc
uracy
o
f
P
up
il
fo
r
Ce
nte
r Gaze
0
5
10
15
20
25
0
10
20
30
40
T
i
m
e
(
s)
T
i
m
e
P
e
r
f
o
r
m
a
n
ce
d
u
e
t
o
F
u
n
ctio
n
Use
d
N
u
m
b
e
r
o
f
F
u
n
ctio
n
Ca
l
l
s
A
l
g
o
r
i
th
m
1
A
l
g
o
r
i
th
m
2
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Ga
ze
Tr
ackin
g Al
gorit
hm Usi
ng Ni
ght Vi
sio
n
C
am
er
a
(
T.
A
. I
zz
uddi
n
)
445
Para
m
eters
Can
n
y
with Reg
io
n
Pr
o
p
s
Ho
u
g
h
Circle
Cen
troid
at
x
-
ax
is
83
88
Cen
ter=12
deg
ree
Table
4.
Acc
uracy
o
f
P
up
il
fo
r
Le
ft G
aze
Para
m
eters
Can
n
y
with Reg
io
n
Pr
o
p
s
Ho
u
g
h
Circle
Cen
troid
at
x
-
ax
is
109
109
Left
=
-
2
4
deg
ree
Cen
troid
at
x
-
ax
is
136
117
Left
=
-
36
d
eg
ree
Table
5: A
cc
ur
acy
o
f
P
up
il
fo
r
Ri
ght Gaze
Para
m
eters
Can
n
y
with
Reg
io
n
Pr
o
p
s
Ho
u
g
h
Circle
Cen
troid
at
x
-
ax
is
44
58
Rig
h
t =
2
4
deg
ree
Cen
troid
at
x
-
ax
is
136
117
Rig
h
t= 36
d
eg
ree
4.
CON
CLUSION
This
pap
e
r
has
presente
d
t
he
de
velo
pm
ent
and
analy
sis
of
a
n
al
gorit
hm
that
ca
n
detect
the
pupil
reg
i
on
by
us
in
g
re
gionpro
ps
an
d
Ci
rcu
la
r
Hou
gh.
T
he
pupil
re
gion
is
then
form
ulated
to
giv
e
a
re
f
eren
ce
ang
le
that
can
be
us
e
d
by
r
obotic
wheel
chair.
Detect
ion
of
pu
pil
reg
i
on
is
do
ne
by
us
in
g
a
night
visio
n
ca
m
era
that
is
m
od
ifie
d
fr
om
a
no
rm
al
web
ca
m
.
The
us
e
of
ni
gh
t
visio
n
ca
m
er
a
can
help
to
m
axi
m
i
ze
the
edg
e
boun
dar
y
reg
i
on
since
i
t
can
dif
fere
nti
at
e
betwee
n
iri
s
an
d
pu
pil.
T
he
we
bcam
is
at
ta
ched
with
4
IR
LED
t
hat
can
giv
e
e
xtra
il
lu
m
inati
on
to
wa
rd
s
t
he
ey
es
s
o
that
the
pupil
will
be
m
or
e
prom
inent.
T
his
is
an
adv
a
ntage
f
or
t
he
ga
ze
detect
i
on
beca
us
e
t
he
pupil
can
sti
ll
be
de
te
ct
ed
ev
en
in
a
da
rk
r
oom
.
More
im
p
or
ta
nt,
The
IR
L
ED
does
not
m
ake
the
patie
nt
or
use
r
feels
un
c
om
fo
rtable
sinc
e
the
IR
L
ED
c
an
only
be
dete
ct
ed
by
night
visi
on
ca
m
era.
For
fu
t
ure
im
pr
ov
em
ent,
the
web
cam
c
an
be
re
place
d
with
a
cam
era
that
can
zo
om
ou
t
X:
83
X:
109
X:
136
X:
44
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
2
,
Fe
br
uary
201
8
:
4
3
8
–
4
4
6
446
from
wide
distance.
T
he
cu
r
ren
t
cam
era
is
placed
in
fro
nt
of
t
he
use
r
and
t
his
m
igh
t
m
ake
the
us
e
r
feel
un
c
om
fo
rtable
.
ACKN
OWLE
DGE
MENTS
The
a
uthor
s
w
ou
l
d
li
ke
t
o
e
xpress
since
re t
hanks
to
Ce
ntr
e of Ro
boti
c
s, Inst
ru
m
entat
ion
a
nd
Au
t
om
ation
(CERIA
)
a
nd Uni
ver
sit
i Te
kn
i
ka
l M
al
ay
sia
Melak
a
(U
TeM
) for the
f
ina
ncial
su
pp
or
t
f
or
t
his
pro
j
ect
.
REFERE
NCE
S
[1]
Gauta
m
,
G.
,
Sum
ant
h,
G.,
Kart
hikey
an
,
K.
C
.
,
Sundar,
S.,
&
Venka
ta
r
aman,
D.
(2014).
E
y
e
m
ovement
base
d
el
e
ct
r
oni
c
whee
l
cha
ir
for
phy
sic
al
l
y
ch
allen
ged
per
sons
.
Int
ernati
onal
Jour
nal
of
Sci
entific
&
Technol
ogy
Re
search
,
3
(2)
,
206
-
212.
[2]
L.
Masek
,
“
Recogniti
on
of
Hu
m
an
Iris
Patt
ern
s
for
Biom
et
ric
Ide
nti
f
ic
a
ti
on
,
”
B.
S.
the
sis
,
Dept.
Com
pute
r
Scienc
e
and
Software
En
gi
nee
r
ing, Weste
rn
Aus
tra
lia
Uni
ver
sit
y
,
Crawl
y
,
Aus
tra
li
a
,
200
[3]
J.
Cann
y
,
“
A
Com
puta
ti
onal
App
roa
ch
to
Edg
e
Detect
ion
,
”
IE
EE
Tr
ans.
Pat
te
rn
Anal
.
Ma
ch.
Int
el
l
.
,
vol
.
PA
MI
-
8,
no.
6
,
pp
.
679
–
6
98,
Nov.
1986.
[4]
3.
D
M.
A.
Z
ia,
U.
Ans
ari
,
M.
Ja
m
il
,
O.
Gil
la
ni
,
and
Y.
A
y
az,
“
F
ace
and
e
ye
d
etec
t
ion
in
images
using
skin
col
o
r
segment
ati
on
an
d
ci
rcular
hough
transfor
m
,
”
2
014
Int.
Conf.
Robot.
Emerg.
Alli
ed
Technol
.
Eng.
,
pp.
211
–
2
13,
Apr.
2014.
[5]
L.
Masek
,
“
Recogniti
on
of
Hu
m
an
Iris
Patt
ern
s
for
Biom
et
ric
Ide
nti
f
ic
a
ti
on
,
”
B.
S.
the
sis
,
Dept.
Com
pute
r
Scienc
e
and
Software
En
gine
er
ing, Weste
rn
Aus
tra
lia
Uni
ver
sit
y
,
Crawl
y
,
Aus
tra
li
a
,
2003
.
[6]
H.
M.
Peixot
o,
A.
M.
G.
Guerre
iro,
and
A.
D.
D.
Neto,
“Im
age
Proce
ss
ing
for
Ey
e
Det
ec
t
ion
and
Classifi
cat
io
n
of
the
Gaz
e
Dir
ec
t
i
on,
”
Inte
rn
ation
al
Join
t
Confe
re
nce
on
Neura
l
N
et
works
,
pp
.
247
5
–
2480,
2009
.
[7]
As
te
ria
dis,
S
.
,
N
ikol
ai
d
is,
N.
,
Ha
jdu,
A.
,
&
Pit
as,
I.
(2006,
M
arc
h
).
An
ey
e
detec
t
i
on
algorit
hm
using
pix
e
l
to
edge
inf
orm
ati
on
.
In
I
nt.
S
y
m
p
.
on
Co
ntrol
,
Com
m
un.
and
Sig
n.
Proc
.
[8]
J.
C
ann
y
,
“
A
Com
puta
ti
onal
App
roa
ch
to
Edg
e
Detect
ion
,
”
IE
EE
Tr
ans.
Pat
te
rn
Anal
.
Ma
ch.
Int
el
l
.
,
vol
.
PA
MI
-
8,
no.
6
,
pp
.
679
–
6
98,
Nov.
1986.
[9]
A.
Ektesabi
an
d
A.
Kapoor
,
"Ex
ac
t
pupi
l
an
d
iris
boundary
detec
t
ion,"
The
2nd
Int
ern
atio
nal
Conf
ere
nc
e
on
Control
,
Inst
rum
ent
a
ti
on
and
Aut
om
at
ion
,
Shir
az,
2011,
pp.
1217
-
1221.
[10]
D.
Dec
ker
and
J
.
A.
Piepmei
er
,
“G
aze
Tr
ack
ing
Inte
rfac
e
for
R
oboti
c
Con
trol,
”
2008
40th
Sout
hea
st.
S
y
m
p
.
S
yst.
The
or
y
,
pp.
274
–
278,
Mar
.
2008
.
[11]
Mohd
Saad,
Norhashim
ah
and
Abdulla
h,
Abdu
l
Rahi
m
(2011)
Brai
n
le
sion
seg
m
ent
at
ion
from
diffusion
weighted
MRI
base
d
on
ada
pti
v
e
thre
shol
ding
and
gra
y
level
co
-
oc
cur
ren
c
e
m
at
rix.
Journa
l
of
Tele
commun
ic
ati
on
,
Elec
tronic
and
Computer
E
ngine
ering
(
JTEC)
.
pp.
1
-
13.
ISS
N 2180
-
1843
[12]
a.
E
l
-
Z
aa
rt
and
H.
Mathkour,
“
A
new
approach
for
pupil
de
tection
in
iris
rec
ogn
it
ion
syst
em,
”
2010
2nd
Int.
Co
nf
.
Com
put.
Eng
.
T
ec
hnol
.
,
pp.
V4
–
415
–
V4
–
419,
20
10.
[13]
B.
Suhaimi,
“
Com
pute
r
Cursor
by
Hum
an
Optica
l
Inte
r
ac
t
ion,”
B
.
S.
the
sis,
Dept.
El
e
ct
ri
ca
l
Engi
n
ee
ring
.
,
UT
eM.
,
Mela
ka
,
Ma
lay
si
a,
2012
[14]
Y.
Jung,
D.
Kim
,
and
J.
Kim
,
“Eye
detec
ti
on
for
ey
egl
ass
w
eare
rs
in
iris
rec
ognit
ion
,
”
18th
IEE
E
Int
.
S
y
m
p.
Consum
.
El
ectro
n.
(ISCE
2014),
no.
1
,
pp
.
1
–
2
,
J
un.
2014
.
[15]
N.
Mohd
Saad,
S.
A.
R.
Abu
-
Baka
r,
S.
Muda
,
M.
M.
Mokji
a
nd
L.
Sa
la
hudd
i
n,
"Brai
n
le
sion
segm
ent
at
ion
o
f
Diffusion
-
weigh
te
d
MRI
using
g
ra
y
le
v
el
co
-
oc
c
urre
nce
m
at
rix
,
"
2011
IE
EE
In
ter
nati
onal
Conf
e
renc
e
on
Imagin
g
Syste
ms
and
Te
c
hnique
s
,
Pen
ang, 2011, pp. 284
-
2
89.
[16]
Jali
,
M.
H.,
et
al.
"A
N
FIS
-
PI
D
C
ontrol
ler
for
Arm
Reha
bil
itati
on
Devic
e.
"
I
nt
ernati
onal
Journal
of
Engi
ne
ering
and
Technol
ogy
(
IJET
)
7.
5
(2015):
1
589
-
1597.
Evaluation Warning : The document was created with Spire.PDF for Python.