In
te
r
n
ation
a
l Jou
rn
al
o
f Po
we
r
Elec
tron
ic
s an
d
D
r
ive S
y
stem
(IJ
PED
S
)
V
o
l.
11, N
o.
1, Mar
ch 20
20,
p
p.
417~
4
2
4
IS
S
N
: 2088-
86
94,
D
O
I
:
10.11
59
1
/ij
ped
s
.
v11
.
i
1.pp
4
17-
42
4
417
Jou
rn
a
l
h
o
me
pa
ge
:
ht
tp:
//i
j
p
eds.i
a
esco
re
.com
IoT base
d facial r
ecognitio
n door access control home
security system using rasp
berry pi
A. R.
S
yafeeza
1
,
M. K.
Mohd
Fitri A
lif
2
,
Y. Nursyifaa
A
t
h
irah
3
,
A
.
S
.
J
aaf
ar
4
,
A
.
H
. N
or
ih
an
5
,
M
.
S
.
S
ale
h
a
6
1,
3
,
4,
5
Fakul
ti Kejurut
e
raan E
lekt
r
o
nic
an
d K
e
ju
ru
teraan
Komp
u
ter,
Un
i
v
ers
iti
T
e
kni
k
al
M
alays
i
a,
M
alaysia
6
Fakul
ti Teknologi Ke
juruteraan, U
n
i
v
e
rsi
t
i
Tek
n
i
k
al M
alaysia Me
l
a
ka, M
al
aysia
2
S
c
h
o
o
l
o
f El
ectrical
En
g
i
n
eering,
F
aculty
o
f E
n
g
i
neeri
n
g
,
Un
i
v
e
rsiti Tekno
l
ogi
Mal
a
ysia,
Mal
a
y
s
ia
Art
i
cl
e In
fo
ABSTRACT
A
r
tic
le hist
o
r
y
:
R
e
c
e
i
v
e
d
Au
g
1
9
,
2
019
Re
vise
d N
ov
2
7
,
201
9
A
c
c
e
pte
d
D
ec 11,
2
0
1
9
Th
e
ho
m
e
s
ecur
i
ty
s
yst
e
m
has
becom
e
v
it
al
f
or
e
very
hou
se.
Prev
iously,
mo
st
d
oo
rs
c
a
n
b
e
o
p
e
n
b
y
u
s
in
g
tra
d
it
io
na
l
w
a
ys,
su
c
h
a
s
ke
y
s
,
s
ecu
rity
cards
,
pas
swo
rd
o
r
p
a
tt
e
r
n
.
H
o
w
ev
er,
inci
den
t
s
such
a
s
a
k
e
y
l
o
s
s
h
a
s
l
e
d
t
o
much
w
orrying
cases
s
uch
as
r
obb
ery
and
identity
fraud.
T
h
i
s
h
a
s
b
ecom
e
a
si
gnif
i
c
a
n
t
i
ss
ue.
To
o
v
e
rc
o
m
e
th
is
p
ro
bl
e
m
,
f
a
ce
re
cog
n
i
t
i
on
u
s
i
ng
d
eep
learn
i
n
g
t
echni
q
u
e
w
a
s
i
n
t
r
odu
ced a
n
d
Intern
e
t o
f
T
h
i
ng (IoT
)
a
lso
b
een used
to
p
erf
o
rm
e
ffi
c
ien
t
d
o
o
r
access
con
t
ro
l
s
y
stem.
Ras
pberry
P
i
i
s
a
pro
g
ram
m
a
bl
e
small
com
p
u
t
er
b
o
a
rd
a
n
d
u
sed
as
t
h
e
m
ai
n
co
nt
ro
l
l
er
f
o
r
f
ace
recog
n
i
tio
n
,
y
o
u
t
h
s
yst
e
m
and
l
o
ck
ing
s
y
stem.
Th
e
cam
era
i
s
u
se
d
t
o
captu
re
im
ag
es
o
f
th
e
p
e
rso
n
i
n
f
r
o
n
t
of
t
he
d
o
o
r.
I
oT
s
ys
te
m
enabl
e
s
t
he
u
ser
t
o
con
t
ro
l t
h
e d
o
o
r
acc
ess.
K
eyw
ord
s
:
Deep
l
earn
in
g
Fa
cial r
ecog
n
i
t
i
o
n
H
o
m
e
se
c
urity
s
ys
t
e
m
Interne
t
of
th
in
gs (
IoT)
Ra
s
pberry
p
i
Th
is
is a
n
o
p
en acces
s a
r
ti
cle u
n
d
e
r t
h
e
CC
B
Y
-S
A
li
cens
e
.
Corres
pon
d
i
n
g
Au
th
or:
Sya
f
ee
za
Ahm
a
d Ra
dz
i
,
F
a
kult
i
K
e
j
uru
t
e
r
aa
n Elek
tro
n
i
k
a
nd K
e
juru
te
r
aan
K
om
put
e
r
,
U
n
i
v
ersi
ti
T
e
k
ni
ka
l Ma
la
ysia
,
Jala
n
Hang Tu
a
h
Jaya
,
7610
0
Durian
T
ung
g
a
l
,
Mela
k
a,
M
alay
sia.
Em
ail:
s
y
af
eeza@ut
em.
e
du.
my
1.
I
N
TR
OD
U
C
TI
O
N
N
o
w
a
days,
ho
me
s
ecur
i
t
y
s
y
s
tem
is
a
c
ruc
i
al
i
ss
ue.
Indee
d
,
t
h
is
s
y
s
te
m
is
t
o
e
n
s
u
r
e
p
r
o
p
e
r
t
i
e
s
a
n
d
lo
ves
o
n
e
s
i
s
a
l
w
a
ys
s
afe
an
d
pro
t
ecte
d
.
F
o
r
t
h
e
pa
st
f
e
w
y
e
a
rs
,
it
is
i
m
port
a
nt t
o
ha
ve
a
s
ol
id
s
ec
urit
y
syste
m
f
o
r
h
o
m
e,
w
hic
h
can
s
ec
u
r
e
i
n
t
h
e
m
o
s
t
i
d
ea
l
and
safe
w
a
y
[
1
]
.
Ma
n
y
c
ou
n
t
ries
a
re
s
t
e
p
by
s
te
p
de
pl
o
y
e
d
home
se
c
u
ri
ty
s
yste
m [2].
Th
e
imp
or
t
a
nt par
t
of a
n
y
hom
e se
curi
ty
s
yste
m is t
he
p
erso
n
i
d
en
t
i
fica
t
i
o
n
t
o e
n
ter
and
e
x
i
t
t
he
h
o
u
se
.
P
r
e
v
i
o
us
ly,
pe
op
l
e
u
se
t
he
t
radi
t
i
o
n
a
l
m
e
t
h
od
for
their
hom
e
s
ecur
i
ty
s
ys
tem
.
T
h
e
t
r
a
d
i
t
i
o
n
a
l
s
e
c
u
r
i
t
y
s
y
s
t
e
m
r
e
l
i
e
s
o
n
t
h
e
u
s
e
o
f
e
x
t
e
r
n
a
l
t
h
i
n
g
s
s
u
c
h
a
s
key,
p
as
sw
ord
an
d
ID
c
a
r
d
t
o
g
a
i
n
a
c
c
e
s
s
[
3
]
.
H
o
w
e
v
e
r
,
d
u
e
t
o
s
o
m
e
l
i
m
i
t
a
t
i
o
n
,
b
i
o
m
e
t
r
i
c
t
a
k
e
s
p
l
ace
t
o
de
li
ver
suc
h
a
p
romis
i
ng
s
ecur
ity
sy
st
em.
Th
e
bio
m
et
ri
c i
s
a
uni
qu
e
and
q
u
a
nti
f
i
a
b
l
e p
a
ramet
e
r f
o
r in
di
v
i
du
a
l
re
c
o
g
n
i
tio
n [
4
]. B
i
o
m
e
tric
s
y
s
tem
requ
ire
d
t
he
u
s
e
d
of
s
pec
i
a
l
iz
e
d
h
a
r
dw
ar
e
su
ch
a
s
fi
nge
rpri
nt
s
c
a
n
n
er
,
pa
l
m
p
rin
t
s
c
a
n
n
e
r,
D
NA
analyz
er
a
nd
e
t
c
.
F
u
r
t
h
e
r
more,
t
h
i
s
s
p
e
ci
fi
c
ma
ch
i
n
e
requ
i
r
ed
t
h
e
t
a
r
g
e
t
t
o
to
uc
h
the
hardw
a
re
t
o
a
c
q
u
i
re
d
ata
of
h
um
an
u
n
i
q
u
e
f
e
a
t
u
r
e
s
.
B
i
o
m
e
t
r
i
c
t
e
c
h
n
o
l
o
g
y
i
s
v
i
e
w
e
d
a
s
a
s
t
a
n
d
o
u
t
a
mon
g
t
h
e
m
ost
secur
e
v
erifica
tio
n
sy
ste
m
ac
cessi
ble,
b
y
gi
v
i
n
g
a
m
ore
e
l
e
v
a
t
ed
a
mo
un
t
o
f
s
ec
uri
t
y
t
h
an
c
on
ve
nt
i
ona
l
me
t
h
o
d
[
1].
F
a
ce
re
cog
n
i
t
i
o
n
is
t
he mos
t fa
mous me
t
ho
d
in b
iom
e
tric
tec
hno
lo
g
y
be
s
ide
s
fi
n
ge
rpr
i
nt character
i
s
tics [2]. This
is due to m
o
r
e
st
a
b
il
it
y
as
f
ac
e
co
nta
i
ns
m
or
e
fea
t
ure
s
[
3].
Be
s
i
de
s,
i
t
i
s
c
o
ns
ide
r
ed
h
ig
h
l
y
secur
e
a
s
fa
c
e
c
an
no
t
be
s
t
o
l
e
n,
borrow
e
d
or fo
r
ge
i
n order t
o
e
nt
e
r
the
h
ouse
.
F
ace
r
ecogn
i
t
i
o
n
is
l
i
k
e
ly
t
h
e
m
ost
na
tur
a
l
a
ppr
oac
h
t
o
pe
rform
bi
ome
t
r
i
c
ve
rifica
ti
on
be
tw
ee
n
ind
i
v
i
dua
ls
[
2
]
.
F
a
ce
de
tec
t
io
n
is
t
he
f
irst
s
te
p
of
t
he
f
ac
e
r
e
c
o
gn
i
t
i
on
s
y
s
t
em
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN: 2088-
8694
I
nt
J
P
ow
Elec
& Dr
i
S
y
st V
ol.
11,
N
o.
1
, Ma
r
202
0
:
417
–
42
4
41
8
Fa
ce
pic
t
ures
can
b
e
cau
g
h
t
a
t
a
d
ista
nc
e
wi
t
h
t
he
u
se
o
f
a
we
b
c
am
era
.
T
he
i
n
d
i
v
idua
l
ca
n
be
r
e
c
o
g
n
i
ze
d
witho
u
t
ph
y
s
i
c
a
l
c
on
tac
t
o
n any
sp
ecia
l
h
ard
w
a
r
e
t
o
p
erc
e
ive
th
e pe
rs
on
's
i
de
n
tit
y.
F
a
c
e
r
ec
ogn
i
t
i
on
us
in
g
de
ep
l
ear
nin
g
t
ec
h
n
i
q
ue
i
s
use
d
.
D
eep
l
e
a
r
n
i
n
g
i
s
a
pi
e
ce
of
t
he
m
or
e
exte
ns
i
v
e
gr
ou
p
o
f
m
ach
i
n
e
lea
r
n
i
n
g
m
e
t
h
ods
b
ase
d
o
n
l
ear
ning
d
a
t
a
repre
s
en
tat
i
o
n
s,
a
s
o
ppose
d
t
o
tas
k
-
spec
ific
a
l
g
orithm
s
.
Le
arni
n
g
can
b
e
ma
naged,
s
em
i
-
d
i
r
ecte
d
o
r
unsu
p
e
r
v
i
se
d.
W
i
t
h
t
h
e
dee
p
l
e
a
rni
ng,
the
sys
t
em
i
s
im
pro
v
ed
f
r
o
m
t
i
m
e
t
o
t
i
me
.
S
o
m
e
i
m
a
ges
of
a
u
t
hor
i
z
in
g
use
r
a
re
u
se
d
a
s
t
he
d
a
t
a
b
ase
of sy
s
tem
an
d
t
h
e
s
y
s
t
e
m
w
il
l
trai
n
the
face
r
ecog
n
i
t
i
on
a
u
t
oma
ti
c
a
l
l
y.
T
h
u
s,
t
he
a
c
c
u
ra
cy i
s incre
a
se
d.
H
om
e
secur
ity
i
s
an
e
xam
p
le
o
f
a
n
I
nte
r
ne
t
of
T
h
i
n
g
s
(
I
oT)
app
l
i
c
a
t
i
o
n
s.
I
o
T
r
ef
e
r
s
t
o
t
h
e
n
et
wo
rk
o
f
a
s
so
ci
a
t
ed
p
h
y
s
i
cal
obj
e
c
t
s
t
h
a
t
can
i
nt
erac
t
and
t
r
a
d
e
i
n
fo
rmat
i
o
n
amo
n
g
t
h
em
sel
v
es
w
i
t
h
o
u
t
t
h
e
n
eed
o
f
an
y
hum
a
n
i
n
t
e
rv
eni
n
g
[5
].
I
o
T
i
s
a
fu
t
u
ri
sti
c
t
e
c
h
nol
og
y
wh
e
r
e
d
e
vi
ce
s
a
nd
i
nt
erne
t
i
s
i
nt
e
r
c
o
nne
c
t
ed
.
It
i
s
di
ff
eren
t
f
r
o
m
t
h
e
i
nt
e
r
n
e
t
du
e
t
o
i
nt
ern
e
t
e
x
c
e
e
d
co
nn
e
c
t
i
v
it
y
by
a
l
l
o
w
i
ng
a
n
y
em
bed
d
e
d
c
ir
cui
t
t
o
com
m
u
ni
c
a
t
e
w
ith
e
a
c
h
o
the
r
u
si
ng
t
he
c
urre
nt
i
n
t
e
r
ne
t
infrastruc
t
ur
e.
N
o
d
ou
bt
I
oT
h
e
l
ps
u
sers
t
o
co
nt
rol
o
n
e
o
r
mo
re
dev
i
ce
s an
d
ca
pab
i
l
i
t
i
es
t
o
a
u
toma
t
e
w
i
t
h
m
a
ny
d
a
i
ly
c
h
o
res.
B
y
u
s
i
ng
I
o
T
,
it
ca
n he
lp
i
n
c
o
n
t
ro
lli
n
g
t
he
d
o
o
r
a
c
c
e
ss
a
n
d
al
so
s
e
n
t
not
i
f
i
cat
io
n
t
h
roug
hout
t
h
e
i
n
t
ern
e
t
.
I
n
t
his
s
y
st
e
m
,
Bly
n
k
a
pps
a
re
u
sed.
B
l
ynk
a
p
p
s
i
s
a
n
a
p
p
t
h
a
t
e
n
a
b
l
e
s
u
s
t
o
c
o
n
t
r
o
l
t
h
e
d
o
o
r
a
c
c
e
s
s
b
y
d
e
s
i
g
n
i
ng
the
grap
hic
a
l
i
n
t
erface
i
n
the
a
pps
a
cc
or
d
i
n
g
to
t
he
s
pe
c
i
fic
fu
nc
t
i
on
t
o
p
erform
.
I
t
a
lso
a
b
l
e
t
o
sen
d
n
o
tif
ic
a
tio
n
t
o
c
om
p
u
t
e
r,
s
ma
rtp
h
o
n
e
a
n
d
ot
he
r
smart
d
e
v
i
ce
s.
2.
LI
T
E
R
A
TU
R
E
R
EV
I
E
W
2.1.
F
a
c
e
rec
ogniti
o
n
techn
o
log
y
C
u
rre
ntly,
t
h
e
num
ber
o
f
t
he
fts
a
n
d
i
d
e
n
t
i
t
y
fr
au
d
ha
ve
fre
que
n
t
l
y
bee
n
r
e
porte
d
an
d
has
be
c
o
m
e
sign
ifica
n
t issu
e
s
. Tr
aditio
na
l
w
a
ys for perso
nal
i
d
e
n
t
i
fica
t
i
o
n req
u
ire
s
ex
t
er
nal e
l
e
m
e
n
t
,
suc
h
as key, secur
ity
passw
ord,
R
F
I
D
car
d,
a
nd
ID
c
a
r
d
to
h
a
v
e
acc
ess
int
o
a
p
riva
te
a
s
se
t
or
e
n
t
e
r
in
g
p
u
b
l
ic
s
pa
ce
[
1
]
.
Many
proce
s
se
s
s
u
c
h
a
s
dra
w
ing
o
u
t
mo
ney
fr
o
m
b
anks
r
eq
ui
re
s
pa
ssw
o
r
d.
O
ther
s
uc
h
pa
rki
ng
i
n
p
riva
t
e
s
pac
e
w
o
u
l
d
a
l
so
n
e
e
d
p
ark
i
n
g
t
i
c
ke
t.
F
or
s
om
e
ho
uses,
the
ho
use
ke
y
i
s
v
er
y
im
por
tan
t
.
H
o
w
e
ver,
a
ll
th
is
m
e
t
h
o
d
als
o
h
as
s
e
v
er
al
d
isa
d
va
nta
g
e
s
s
uc
h
a
s
l
os
i
n
g
ke
y
an
d
for
g
e
tti
n
g
pa
ssw
or
d
[3]
.
W
he
n
th
is
h
ap
pe
ns,
it
c
a
n
b
e
hass
le
t
o
re
cov
e
r
bac
k
.
This
m
etho
d
i
s
s
lo
w
l
y
ta
ke
n
o
v
er
by
b
i
o
me
t
r
i
c
m
etho
ds
a
s
i
t
i
s
the
poss
i
b
l
e
w
a
y
t
o
so
l
v
e
those
pr
ob
l
e
ms.
Th
is
t
e
c
hn
i
que
r
e
qui
re
d
t
o
u
se
t
h
e
s
pec
i
a
l
ha
rdw
a
r
e
s
uch
a
s
f
ing
e
rprin
t
s
ca
nne
r
,
p
a
l
m
p
r
in
t
sca
n
n
e
r,
DNA
an
al
y
z
er
t
o
g
a
t
h
er
i
nfo
r
mat
i
on
f
o
r
t
h
e
v
ast
m
a
jori
t
y
o
f
the
b
i
om
etric
app
l
icat
i
o
n
s
a
nd
t
he
targe
t
o
b
j
e
c
ts
h
ave
t
o
t
o
u
ch
w
ith
t
he
h
ar
d
w
are
to
a
cq
uir
e
i
nfor
m
a
t
i
o
n
[
6].
A
s
b
i
o
m
e
t
r
ic
i
s
a
te
ch
n
i
que
t
hat
d
i
s
tin
gui
shin
g
p
h
y
s
i
cal
h
igh
lig
ht
s
o
f
p
e
opl
e
ac
co
rdi
ngl
y
it
h
a
s
a
n
e
x
ten
s
i
v
e
var
i
e
t
y
o
f
u
til
iza
t
i
o
n
in
s
e
c
u
ri
ty
fra
m
e
w
orks
a
nd
it
is
v
iew
e
d
a
s
one
o
f
t
h
e
most
s
e
c
ure
m
e
th
o
d
s
[
1].
Basical
l
y
,
bi
ome
t
r
i
c
s
c
an
b
e
clas
s
i
fie
d
i
n
tw
o
ca
te
gor
i
e
s
w
h
i
c
h
are
ph
y
s
ica
l
a
nd
b
e
ha
vi
oral.
Rec
e
n
t
l
y
,
t
h
e
f
a
c
e
r
e
c
o
g
n
i
t
i
o
n
t
e
c
h
n
o
l
o
g
y
h
a
s
e
n
g
a
g
e
d
a
n
overw
h
e
lmi
ng
num
ber
of
r
e
s
e
a
rc
hers
a
nd it i
s
g
ra
dua
l
l
y
s
u
p
p
la
n
t
i
n
g
ot
her
bi
om
etric secu
rit
y
f
ram
e
w
o
rk
s
[7]
.
F
a
ce
r
ecogn
i
tion
i
s
a
ls
o
kn
o
w
n
a
s
i
m
a
ge
m
atchi
n
g.
I
t
is
a
r
a
p
id
ly
g
rowi
ng
f
i
e
ld
w
h
e
re
i
t
is
h
ea
din
g
in
a
direc
t
i
o
n
s
u
c
h
t
hat
i
t
w
il
l
r
e
pl
a
c
e
t
h
e
tr
a
d
i
tio
na
l
me
th
o
d
.
Fa
c
e
re
cog
n
i
tio
n
is
m
ore
sta
b
le
a
m
o
ng
o
thers
bi
ome
t
r
i
c
i
d
e
n
tif
i
c
at
i
o
n
me
th
od
as
i
t
i
s
u
sing
t
he
h
uma
n
f
a
c
e
t
ha
t
re
su
lt
s
in
h
i
g
h
acc
urac
y,
l
ow
e
s
t
false
rec
o
g
n
it
io
n
ra
te
a
nd
it
doe
s
n
o
t
c
h
a
nge
i
n
p
e
op
le
’s
l
i
f
e
[3]
.
T
h
u
s
,
t
h
i
s
m
e
t
h
o
d
i
s
m
u
c
h
p
r
a
c
t
i
c
a
l
f
o
r
a
l
o
t
o
f
usa
g
e,
incl
u
di
n
g
fa
ce
rec
ogn
it
io
n for
the
un
loc
k
ing
h
ouse
d
oor.
2.2.
Meth
o
d
use
d
for
f
ace
recogn
i
t
i
o
n
I
n
t
h
i
s
new
er
a,
f
ace
r
ec
og
nit
i
on
pl
a
y
s
a
n
i
m
por
tan
t
r
ole
in
s
e
curi
ty
a
nd
o
b
serva
t
i
o
n.
C
o
n
se
que
n
t
l
y
,
t
h
e
r
e i
s
a
r
equ
i
reme
n
t
f
o
r
a
p
ro
fi
ci
ent
an
d
cost
-e
f
f
e
c
ti
ve
sy
s
t
em
.
F
a
ce
r
e
c
ogn
i
t
i
o
n is a te
c
h
n
i
que t
hat
i
s
a
ble
to
i
d
ent
i
f
y
and
v
e
ri
fy
p
e
o
pl
es
[
8].
Ac
co
rd
in
g
t
o
[
9
]
,
f
a
c
e
r
ec
o
g
n
i
t
i
o
n
,
d
efin
e
as
s
t
e
ps
t
o
id
en
ti
fy
,
d
i
stin
gui
s
h
a
nd
proce
s
se
d
fac
e
is
c
o
m
par
e
d
w
ith
t
he
i
ma
ge
s
t
h
a
t
s
t
o
red
i
n
t
he
d
a
t
a
b
ase
t
o
v
eri
f
y
w
ho
the
pers
on
i
s
.
Thi
s
f
ac
e
rec
o
g
n
it
io
n
ha
s
bec
o
m
e
a
s
ig
ni
fican
t
te
c
h
n
i
que
f
or
u
s
e
r
id
e
n
ti
f
ic
a
t
i
o
n
[1
0].
There
a
r
e
ma
ny
tech
n
i
qu
es
t
ha
t
ca
n
be
u
se
d
fo
r
fa
ce
r
e
c
ogn
i
tio
n
bu
t
t
h
e
P
r
inc
i
p
l
e
C
o
mp
o
n
en
t
A
n
a
l
ysis
(
P
C
A
)
i
s
on
e
of
t
he
m
os
t
po
pu
l
a
r
t
e
c
hni
qu
e
s
u
sed
fo
r
f
a
c
e
r
e
c
o
g
n
it
i
o
n
.
T
hi
s
me
t
h
od
i
nvo
lve
s
a
m
a
them
at
ic
al
p
r
o
ce
d
u
re
t
o
t
r
an
sform
a
num
b
e
r
of
p
o
ssib
l
y
c
o
r
r
e
late
d
va
ri
ab
l
e
s
int
o
a
n
um
b
e
r
of
u
nc
orre
lated
var
i
a
b
l
e
s
know
n
a
s
p
r
i
nc
i
p
le
c
om
pone
n
t
[
10]
.
G
e
ne
rall
y, the
P
CA
tec
hn
iq
ue
for fa
ce re
co
g
n
it
io
n
w
i
l
l
u
ti
li
ze
t
h
e use of
E
ige
n
face
s
[6]
.
It is the
ef
f
ec
ti
v
e
an
d
efficie
n
t
w
a
ys
t
o
repre
s
e
n
t
pi
cture
s
i
n
t
o
E
i
g
e
nfa
c
e
s
c
om
po
nen
t
a
s
i
t
c
a
n
reduc
e t
h
e
s
i
ze
o
f the data
ba
se
o
f
t
h
e
tes
t
i
m
a
ge.
N
u
m
e
r
ous
m
et
h
od
is
d
e
v
e
l
op
ed
a
n
d
d
e
p
l
o
y
e
d
i
n
o
r
d
er
t
o
i
mp
ro
v
e
t
h
e
p
erf
o
rman
c
e
of
f
a
c
e
re
co
gn
i
t
i
o
n
t
e
c
h
nol
ogy
.
2.3.
Deep learnin
g
D
e
ep
l
ea
rn
in
g
has
be
ne
fi
ted
t
h
e
h
u
m
a
n
k
i
n
d
for
ye
ars
now
.
In
t
h
e
mode
r
n
s
oc
iet
y
,
a
dee
p
l
ear
n
i
ng
tech
n
i
q
u
e,
e
sp
ec
i
a
l
l
y
c
o
n
v
o
lut
i
ona
l
ne
ura
l
n
e
t
w
o
rk
u
se
d
i
n
m
an
y
a
pp
l
i
ca
ti
o
n
s
s
u
ch
a
s
li
ce
nse
pl
at
e
rec
o
g
n
it
io
n
,
fin
g
er-ve
i
n
i
d
e
n
t
i
fica
t
i
o
n
[
6-
11],
ge
nde
r
re
c
ogn
i
t
i
on
[6]
,
f
ac
e
rec
o
gn
iti
on
[
8
-1
2]
,
e
m
o
tio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
Int J
P
o
w
E
l
e
c
&
D
ri S
yst
IS
S
N
:
2088-
86
94
IoT
based f
a
c
i
al re
c
o
g
n
iti
on
do
or ac
ce
ss
co
nt
ro
l
hom
e se
cur
i
ty
sys
t
em
usi
n
g
ras
p
be
rry
pi (A
. R. Sy
af
e
e
za)
41
9
rec
o
g
n
it
io
n
[1
3]
,
a
nd
o
t
her
app
l
i
c
a
t
i
ons.
Ba
sed
on
[9],
d
ee
p
l
e
a
rni
n
g
tec
h
n
i
que
i
s
h
i
gh
ly
u
sed
f
o
r
co
m
puter
vi
si
o
n
a
p
p
l
i
ca
tio
n.
B
y
us
ing
C
o
n
v
o
l
u
ti
o
n
al
N
eur
a
l
N
e
tw
ork
(CN
N
)
,
i
t
resu
l
t
s
i
n
b
e
tt
e
r
p
erf
o
rma
n
ce
f
o
r
f
a
ce
detec
t
i
on
a
nd
fac
e
r
ec
ogn
it
i
on
[11]
.
There
a
r
e
ma
ny
a
dva
nta
g
es
by
us
in
g
CN
N
as
it
c
a
n
per
cei
ve
p
a
tter
n
s
wi
t
h
h
i
g
h
v
a
ria
b
ili
ty
a
nd
r
ob
u
s
tn
e
ss
to
d
is
to
rt
ion
s
a
n
d
si
mpl
e
ge
om
et
ric
trans
f
orm
a
tio
ns
l
i
k
e
tra
n
s
l
at
i
o
n
,
scal
in
g,
r
ot
a
t
io
n,
s
q
u
e
e
z
i
n
g
,
st
r
o
ke
w
id
t
h
a
nd
a
n
o
ise
[1
2].
Be
side
s,
O
rie
n
te
d
F
A
S
T
a
nd
Ro
ta
te
d
B
R
I
E
F
(O
RB)
a
l
so
a
s
one
o
f
t
h
e
te
ch
ni
que
s
use
d
f
o
r
f
a
c
e
re
cogn
it
ion.
I
t
i
s
use
d
f
or
f
eat
ure
e
x
t
r
acti
o
n,
w
hich
u
ti
liz
es
a
f
a
st
b
i
n
a
r
y
de
scri
pt
o
r
d
e
p
en
d
e
nt
o
n
BRIEF
an
d
i
s
r
ot
a
t
i
o
n
a
l
l
y
i
nvaria
n
t.
T
y
p
i
ca
ll
y,
D
e
e
p
Lea
r
ni
n
g
i
s
base
d
on
s
u
p
er
v
i
se
d
l
e
a
r
ni
ng.
T
he
a
i
m
o
f
D
e
e
p
L
e
a
r
ni
ng
i
s
t
o
m
a
k
e
a
m
ac
h
i
ne
c
a
p
ab
le
t
o
c
o
rrec
tly
c
lass
i
f
y
i
m
age
s
.
Th
us,
d
u
rin
g
t
he
s
upe
r
v
ise
d
l
ea
rn
i
n
g
Ra
spb
e
rry
p
i
a
r
e
sh
o
w
n
an
ima
g
e
t
o
p
r
o
d
u
c
e
a
n
ou
tpu
t
i
n
t
h
e
form
o
f
vec
t
or
s
score
s
,
w
i
t
h
e
ac
h
ca
t
e
gor
y one
.
2.3.
F
a
c
e
rec
ogn
iti
o
n
in
r
asp
b
err
y p
i
The
first
r
e
sea
r
ch
o
n
fa
ce
re
cog
n
iti
on
g
o
es
w
ay
b
a
c
k
i
n
1
9
5
0
i
n
t
h
e
f
i
e
ld
o
f
p
s
y
c
ho
l
ogy
.
Th
e
a
c
t
u
al
wo
rk
o
f
au
to
mat
i
c
mac
h
i
n
e
re
co
gn
i
tio
n
o
f
f
ac
es
r
ea
l
l
y
s
t
a
rt
ed
i
n
19
7
0
[
1
4
].
F
r
o
m
all
the
resea
r
ch
don
e,
t
he
re
tw
o
typ
e
s
of
f
ac
e
re
c
o
g
n
iti
o
n
m
e
t
hod
w
hic
h
a
re
t
he
i
m
a
ge-
b
a
s
ed
fa
ce
r
ec
og
n
i
ti
on
a
n
d
vi
de
o-
base
d
re
co
gn
i
t
i
o
n
.
V
id
e
o
-
b
a
s
e
d
f
ac
e
re
co
gni
tio
n
i
s
t
h
e
p
ro
ce
ss
o
f
f
in
din
g
3
D
im
ages
from
i
t
s
2
D
w
hi
l
e
t
he
i
m
a
ge
-
base
d
r
eco
g
n
i
t
i
on
m
et
ho
d,
i
s
t
h
e
proc
ess
b
y
w
h
i
c
h
h
u
ma
n
tr
ain
t
he
m
ach
i
n
e
us
in
g
a
c
a
m
era
by
sh
ow
i
n
g
the
ca
me
ra
s
e
t
s
of
s
til
l
i
m
age
s
.
A
F
a
c
e
R
ec
og
ni
t
i
on
S
y
s
t
em
i
s
a
fra
me
w
o
rk
w
hic
h
c
ons
eq
uen
t
ly
r
ec
og
ni
z
e
s
an
d
add
i
tio
na
l
l
y c
h
ecks
the
ide
n
ti
ty
o
f
a pe
rso
n
fr
o
m
di
g
i
ta
l im
age
s
or a
vide
o
o
u
tli
ne
f
rom
a
vi
de
o so
urc
e
[
15].
M
a
n
y
re
se
arch
ers
c
hoo
se
t
o
u
s
e
emb
e
dd
ed
d
evi
c
e
ca
l
l
e
d
as
R
as
pb
e
rry
P
i
for
training
a
nd
ide
n
tif
i
c
at
i
on
pur
pose
.
T
he
f
un
da
me
nta
l
r
ea
so
ns
w
h
y
t
he
y
ha
ve
p
i
c
k
e
d
t
h
i
s
p
arti
c
u
l
a
r
co
mp
on
en
t
b
e
c
a
u
se
i
t
has
hi
g
h
h
a
n
d
l
i
ng
l
i
m
i
t
,
l
ow
c
os
t,
a
nd
i
t
s
c
apa
c
i
t
y
a
d
j
u
s
t
s
in
var
i
o
u
s
pro
g
ra
m
m
ing
m
odes
[1].
B
y
us
i
n
g
Ra
s
pberry P
i,
i
t he
l
p
s t
o
reso
l
ve t
he
l
imi
t
a
tio
n o
f
P
C suc
h
a
s
it
s w
e
i
g
h
t
, si
ze and
h
i
g
h
po
w
e
r
consump
t
i
on [3]
.
R
a
s
pberry
P
i
i
s
a
d
evic
e
tha
t
can
d
ivi
d
e
the
so
ftw
a
r
e
p
a
r
t
in
t
o
t
hree
p
ar
t
s
w
h
i
c
h
a
re
r
ec
ording
im
a
g
e
s
,
trai
ni
n
g
a
nd
fa
ce
r
ecog
n
i
t
i
on
[
1].
A
ccor
d
in
g
to
[
1]
a
nd
[
3]
a
s
the
y
d
e
p
l
o
ye
d
t
h
e
used
o
f
Ra
spbe
rry
P
i
fo
r
ima
g
e
cap
t
u
rin
g
s
ys
tem
,
t
he
s
yste
m
bec
o
m
e
s
li
ttler
,
l
i
g
ht
er
a
nd
h
as
l
ow
er
pow
er
u
ti
liz
at
ion.
S
o
it
is
m
ore
con
v
e
n
ie
n
t
c
o
m
p
a
red
to P
C-ba
se
d face
re
c
og
n
i
t
i
on
syste
m
.
2.3.
IoT in
fa
c
e r
e
c
ogn
iti
o
n
I
o
T
has
bee
n
a
ppl
ied
i
n
f
a
c
e
r
e
c
og
ni
tio
n
in
m
a
ny
a
p
p
lic
at
ions
s
uc
h
a
s
u
n
m
anned
a
r
ial
v
e
hic
l
e
[16]
,
sma
r
t
c
l
assro
o
m
[
17]
,
home
se
curi
ty
s
ys
te
m
[2,
18],
smart
h
ouse
[
19],
sm
art
survei
l
l
a
n
ce
a
n
d
m
an
y
m
o
r
e
app
l
ica
t
i
o
ns.
The
prev
i
o
u
s
i
m
p
lem
e
n
t
at
ion
of
I
oT
i
n
fac
e
r
ec
ogn
i
t
io
n
a
r
e
usi
n
g
c
onv
e
n
ti
ona
l
me
t
hod
su
c
h
loca
l
b
i
nary
p
a
tter
n
[
2
0
],
n
e
u
r
a
l
ne
tw
ork
[21
,
22],
su
pp
ort
vec
t
o
r
m
a
c
h
i
n
e
[23],
a
n
d
k
n
e
are
s
t
ne
ig
hb
o
r
[
2
4
]
.
H
o
w
e
ve
r,
i
n th
i
s
r
ese
a
rc
h dee
p
lea
rn
in
g w
a
s bei
n
g use
d
.
3.
METHODOLOG
Y
3.1.
Ove
r
vie
w
Th
is
p
ro
j
e
c
t
w
ill
des
i
gn
fa
ce
re
cog
n
iti
on
for
rea
l
-ti
m
e
use.
I
t
i
s
integrated
w
i
t
h
IoT
t
o
p
er
form
s
m
a
rt
home
se
cur
i
t
y
sys
t
e
m.
A
d
ee
p
le
arn
i
ng tec
h
niq
u
e
is use
d i
n
t
hi
s
p
rojec
t
.
In
o
rder
t
o
ens
u
r
e
the ex
p
ec
te
d
r
e
s
u
lt
are
obta
i
ned,
s
eve
r
al
m
ajor
s
t
e
ps
n
ee
d
to
b
e
c
onduc
t
e
d
suc
h
a
s
da
ta
c
o
l
le
cti
o
ns,
im
pl
e
m
e
n
ti
ng,
t
es
t
i
n
g
,
and
tro
u
b
l
e
s
hoo
ti
n
g
.
These
s
t
ep
s
a
r
e
used
t
o
a
n
a
l
yze
the
da
ta
a
nd
o
u
t
p
u
t
.
W
i
t
h
t
h
e
s
e
s
t
e
p
s
,
t
h
i
s
p
r
o
j
e
c
t
a
r
e
a
b
l
e
t
o
be
e
val
u
a
t
e
d
.
3.2.
Face
rec
ogn
iti
o
n
Th
e
p
r
ot
o
t
y
p
e
i
s
b
u
il
t
by
c
o
m
b
i
ni
ng
t
h
e
p
a
r
t
o
f
f
a
c
e
r
e
c
ogni
tio
n
an
d
IoT
t
oge
t
h
er.
F
a
c
e
r
e
c
og
n
i
t
i
on
is
o
pera
t
e
d
a
t
f
i
r
st
p
lac
e
.
T
h
ere
ar
e
five
s
te
ps
i
n
fac
e
r
ecog
n
it
i
o
n,
w
hi
ch
a
re
c
ol
lec
t
i
n
g
ima
g
es,
c
r
ea
tin
g
data
ba
se
,
pre-
proce
s
si
n
g
im
a
g
e
s,
traini
ng
im
ages
a
n
d
te
s
t
i
n
g
im
ages.
F
i
rst
l
y,
i
m
a
ges
ar
e
col
l
ec
te
d
.
T
hese
i
ma
ge
s
are
ob
ta
i
n
e
d
b
y
ca
p
tur
i
ng
u
sing
c
a
m
e
r
a
a
nd
use
d
t
he
exi
s
ti
ng
i
m
age
s
.
Thi
s
i
m
a
g
e
i
s
u
s
e
d
f
or
t
ra
ini
n
g
p
u
r
p
o
s
e
f
o
r
th
e
s
y
s
t
e
m
t
o
be
m
ore
ac
cura
t
e
w
he
n
d
e
a
lin
g
w
ith
n
e
w
i
ma
ge
s.
A
t
ota
l
o
f
fi
ve
p
er
sons,
eac
h
w
i
t
h
f
i
v
e
p
i
ct
ur
es
i
s
t
a
k
e
n
fro
m
d
i
f
f
eren
t
p
o
siti
on
s.
E
ac
h
pic
t
ure
i
s
a
ppr
oxim
a
te
l
y
2
6
8
x
3
5
0
p
i
x
e
l
s
of
h
e
i
gh
t
an
d
w
i
d
t
h.
Ima
g
es
t
h
a
t
a
r
e
c
o
l
l
ect
ed
a
re
s
t
o
red
in
t
h
e
data
ba
se
a
s sh
ow
n i
n
F
ig
ur
e
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N: 2
0
8
8
-
86
94
I
nt
J
P
o
w
E
l
e
c
&
D
r
i
S
yst
V
o
l.
11,
N
o.
1
,
Mar
202
0
:
417
–
42
4
42
0
F
i
gur
e
1.
N
umber
e
d
a
n
d
labe
le
d
im
ages
S
i
nce
fac
e
r
e
c
o
g
n
iti
on
f
r
a
me
w
o
r
k
a
n
e
e
d
l
ar
ge
num
ber
o
f
i
m
a
ge
s,
e
x
i
sti
n
g
i
m
a
g
es
h
av
e
b
een
a
ugme
n
t
e
d.
T
hi
s
i
s
d
one
b
y
usi
ng
a
n
a
l
g
o
r
it
hm.
Eac
h
p
i
c
t
u
r
e
p
er
p
er
s
on
w
i
ll
a
ugm
ent
in
t
o
100
p
i
ctur
e
s
,
r
e
sulti
n
g
2
5
00
im
ages
s
t
o
r
e
d
i
n
d
a
t
a
b
ase
.
T
h
e
i
ma
ges
var
y
i
n
br
i
ght
ne
ss,
c
o
l
our
,
in
t
e
nsit
y,
a
n
d
a
n
g
l
e
.
T
hi
s
i
s
to
e
nsur
e
t
h
a
t
f
a
ce
r
e
cogn
i
t
i
o
n
s
y
s
t
e
m
c
a
n
de
tec
t
e
ven
i
n
d
if
fe
r
e
n
t
c
on
d
i
t
i
on
s.
F
igu
r
e
2
(
a
)
show
s
the
tr
a
n
sf
or
m
a
ti
on
f
r
o
m
t
h
e
or
i
g
i
n
a
l
i
ma
g
e
i
n
the
data
base
i
n
t
o
the
p
r
o
ce
sse
d
p
h
o
t
o.
T
he
r
e
s
u
l
t
of
e
ac
h
p
e
r
s
on
c
a
teg
o
r
i
z
e
d
i
nt
o
e
a
c
h
f
o
l
der
.
N
ext,
t
he
c
r
o
p
p
i
n
g
pr
oc
ess
t
a
kes
p
l
ac
e.
T
h
i
s
pr
ocess
wil
l
crop
t
he
e
xac
t
f
ac
e
f
r
o
m
t
h
e
ima
g
es.
Th
is
p
r
o
c
e
s
s
i
s
car
r
i
ed
out
b
y
us
i
n
g
a
n
a
lgor
i
t
h
m
.
T
h
e
p
ix
e
l
o
f
e
a
c
h
p
i
c
t
u
r
e
i
s
r
e
d
u
c
e
d
t
o
4
8
x
4
8
p
i
x
e
l
s
o
f
h
ei
g
h
t
a
nd
w
i
d
t
h
.
F
i
g
ur
e
2(
b)
s
how
s
extr
ac
t
fea
t
ur
e
s
by
sepa
r
a
ti
n
g
t
he
f
a
ce
fr
o
m
t
he
ba
ck
gr
o
u
nd.
(a)
(b
)
F
i
gur
e
2.
(
a
)
D
a
t
a
a
ugme
n
t
a
t
i
on
o
f
I
ma
ges
(
b
)
Exa
c
t
F
ace
O
bta
i
n
ed
3.
3.
D
e
e
p
l
earn
in
g
Ex
ist
i
ng a
r
ch
i
t
ecture
wa
s use
d
i
n t
h
e t
r
a
i
n
i
n
g
proce
s
s
.
I
m
ages
were train
u
sing
d
eep
l
ea
rn
ing
me
t
hod
us
in
g
Co
n
v
o
l
u
t
i
o
nal
N
e
ur
al
N
etw
o
r
k
(
C
N
N
)
t
e
c
hn
iq
ue.
The
cur
r
en
t
a
r
c
hi
tec
t
ur
e
use
d
i
s
A
l
exN
e
t
w
h
ic
h
co
ns
i
s
t
of
e
i
g
h
t
l
ay
ers.
T
h
i
s
arc
h
it
e
c
t
u
r
e
b
u
i
l
d
s
with
s
e
v
era
l
laye
r
and
a
c
t
iva
t
ion
fu
nct
i
on
suc
h
a
s
C
o
n
v
o
lu
ti
o
n
,
Ma
xp
o
o
l
i
n
g,
F
la
tte
n,
D
ense,
A
c
ti
vat
i
on
a
n
d
D
r
opou
t
.
The
ent
i
r
e
n
e
u
r
a
l
n
e
tw
or
k
a
ppr
oa
c
h
w
as
im
pl
e
m
e
n
te
d
i
n
P
yt
h
on
la
ng
ua
ge
a
n
d
K
er
a
s
l
i
b
r
a
r
y
[
25]
.
The
t
r
ai
n
i
ng
i
nvo
lv
es
1
00
ep
o
c
h
s
a
t
f
i
r
s
t
a
nd
r
e
pea
t
e
d
w
i
t
h
2
0
e
poc
hs
a
f
t
e
r
t
he
t
est
i
n
g
p
h
ase
.
F
ig
ur
e
3
w
i
l
l
ill
us
t
r
ate
t
h
e
t
r
ai
ni
ng
p
ro
c
e
s
s
o
f
t
h
e
d
a
t
a
s
et
.
A
f
t
e
r
tr
a
i
n
i
ng
p
r
o
ce
ss
w
a
s
done
i
m
a
ge
t
es
tin
g
is
r
eq
uir
e
d
t
o
d
e
ter
m
ine
t
h
e
a
c
c
u
r
acy
a
c
h
ie
ve
d
b
y
t
he
s
ys
tem
.
I
n
t
h
i
s
sta
g
e,
i
m
a
ge
t
ha
t
ar
e
no
t
in
t
he
d
a
t
a
b
a
s
e
ar
e
use
d
a
s
t
es
t
i
m
age
s
.
Ther
e
are
ten
ima
g
es
t
e
s
t
e
d
for
ea
ch
Evaluation Warning : The document was created with Spire.PDF for Python.
Int J
P
o
w
E
l
e
c
&
D
ri S
yst
IS
S
N
:
2088-
86
94
IoT
based f
a
c
i
al re
c
o
g
n
iti
on
do
or ac
ce
ss
co
nt
ro
l
hom
e se
cur
i
ty
sys
t
em
usi
n
g
ras
p
be
rry
pi (A
. R. Sy
af
e
e
za)
42
1
l
a
b
e
l
e
d
whi
c
h
are
rec
o
g
n
i
z
ed
a
n
d
u
n
re
c
ogniz
e
d
p
e
rso
n
.
E
a
c
h
i
m
ag
e
tes
t
e
d
w
ill
labe
le
d
t
h
e
im
age
w
i
t
h
n
am
e
or a
s a
n
unk
no
w
n
.
F
i
gure
3. Tr
a
in
ing
t
h
e da
ta
set
3.4.
I
n
ter
n
et
of things (
IoT)
B
l
y
n
k
i
s
a
fa
m
ous
a
p
p
s
si
n
c
e
it
has
been
dow
nl
oa
ded
m
o
re
t
ha
n
1
00
t
h
o
u
san
d
u
sers.
Bl
yn
k
i
s
a
pla
t
form
f
or
i
O
S
a
nd
A
ndro
i
d
a
pps
t
hat
m
a
na
ge
d
to
c
o
n
t
rol
Ra
spb
e
rry
p
i
a
n
d
m
a
ny
o
t
he
r
mic
r
oco
n
tr
o
llers.
It
is
a
d
ig
ita
l
da
s
h
b
o
ar
d
t
h
a
t
d
e
s
igne
d
for
the
user
t
o
cre
a
te
t
he
i
r
o
w
n
g
r
a
p
h
i
c
i
n
t
e
r
f
a
c
e
f
o
r
t
h
e
p
r
o
j
e
c
t
.
I
t
i
s
e
a
s
y
and
s
i
mple
t
o
use
a
s
t
he
u
se
r
ca
n
sim
p
ly
d
r
a
g
a
n
d
dro
p
t
h
e
w
i
d
g
et
s
th
at
t
h
e
y
n
e
ed
a
c
c
ord
i
n
g
t
o
t
h
e
i
r
pro
j
e
c
t
ty
pe. Th
i
s a
p
p is
u
se
d i
n
I
oT par
t
.
B
ly
nk s
t
a
r
t o
n
l
i
ne a
s the
R
asp
b
e
rry
Pi
co
nn
e
c
t
e
d to
t
he
i
nt
e
r
n
e
t
o
v
e
r W
i
-Fi
.
B
e
s
i
de
s,
i
t
is
a
lso
w
ill
g
e
t
o
n
line
b
y
lin
k
t
o
t
he
i
n
t
erne
t
thr
ou
g
h
t
he
E
t
h
e
r
net
or
t
he
n
e
w
E
S
P
826
6
c
h
i
p
.
F
o
r
con
d
i
t
i
on
w
h
e
r
e
fac
e
cann
o
t
be
r
ec
og
ni
z
e
d
,
that
p
ers
on
c
a
n
pre
s
s
t
h
e
d
o
o
r
bel
l
a
nd
n
o
tif
i
c
at
i
on
a
r
e
sent
t
o
sma
r
tph
one
o
f
ho
use
ow
ne
r.
H
ence
,
li
ve
s
tr
ea
ming
v
i
de
o
w
i
l
l
a
p
p
ea
r
t
o
i
de
n
t
i
f
y
t
h
e
p
e
rson
tryi
n
g
t
o
un
l
o
ck
t
h
e
doo
r.
4.
RESULT
AND
A
NALYSI
S
F
a
c
e
r
ecogn
it
ion
i
s
t
e
s
t
e
d
o
n
tw
o
t
ypes
w
h
ic
h
ar
e
b
y
t
e
s
ti
ng
im
age
a
n
d
r
eal-tim
e
t
o
d
e
t
e
r
m
i
ne
t
he
s
y
s
t
e
m
a
c
c
u
r
a
c
y
.
F
o
r
t
e
s
t
i
n
g
i
m
a
g
e
,
t
h
e
r
e
a
r
e
t
e
n
i
m
a
g
e
s
t
h
a
t
a
re
not
i
n
t
h
e
da
t
a
base
a
re
t
es
ted
for
ea
ch
l
abe
l
w
h
ic
h
are
au
t
hor
i
z
ed
a
n
d
u
nk
n
o
w
n
p
e
r
so
n.
T
he
t
este
d
i
m
a
g
e
w
ill
ha
v
e
l
a
b
e
l
e
d
t
he
i
m
a
ge
w
it
h
na
m
e
s
for
aut
h
orize
d
p
e
r
son
w
h
i
l
e
u
n
k
now
n
for
u
n
au
th
orize
d
p
er
son.
F
igure
4
(
a),
(b),
(
c
)
a
nd
(d)
shows
the
tested
i
m
ag
e
wi
th
p
o
s
i
tiv
e
a
n
d
ne
g
a
tiv
e
re
sult
s
f
o
r
a
u
t
h
o
r
i
zed
a
nd
u
n
k
n
o
w
n.
R
ea
l
-
t
i
m
e
f
a
ce
re
co
gn
i
tio
n
i
s
perform
ed
u
si
ng
w
e
b
c
a
m
e
ra
.
A
n
a
ut
ho
r
i
z
e
d
pers
on
ca
n
be
r
ec
o
g
n
i
z
e
d
t
hr
ou
g
h
t
he
s
y
s
tem
an
d
v
i
ce
versa
.
The
na
me
o
f
t
h
e
u
s
er
w
i
l
l
be
s
h
o
w
n
b
e
l
ow
t
he
ir
f
ac
e
a
s
s
how
n
i
n
F
i
g
ure
5
(a)
w
h
ile
u
nau
t
ho
r
i
z
e
d
pe
r
s
on
i
s
show
n in F
igur
e 5
(b).
(a)
(b)
(c)
(d)
F
i
gure
4.
(
a) P
osit
i
v
e r
e
sul
t
f
or
a
ut
horize
d
(
b)
P
osit
i
v
e
re
sul
t
fo
r u
n
kno
wn
(c
) N
e
gati
ve
r
esu
l
t
for
aut
hori
z
e
d
(
d)
N
ega
t
i
v
e
re
sul
t
for un
k
n
ow
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N: 2
0
8
8
-
86
94
I
nt
J
P
o
w
E
l
e
c
&
D
r
i
S
yst
V
o
l.
11,
N
o.
1
,
Mar
202
0
:
417
–
42
4
42
2
(a)
(
b
)
F
i
gur
e
5.
(
a)
A
ut
h
o
r
i
ze
d
per
s
on
la
be
led
w
i
th
n
am
e
(
b
)
una
ut
hor
i
z
e
d
p
er
son
la
be
led
w
i
t
h
t
he
u
n
k
n
o
w
n
(a)
(
b
)
F
i
gur
e
6.
(
a
)
N
oti
f
ic
a
t
i
o
n
sen
d
t
hr
ou
g
h
B
l
y
n
k
(
b)
V
ideo
s
tr
e
a
m
i
ng
i
n
B
ly
nk
F
a
ce
r
ecogn
it
i
o
n
a
n
d
I
o
T
ar
e
inte
gr
ate
a
n
d
bu
i
l
d
in
p
r
o
to
t
ype.
Wh
en
p
e
r
so
n
fa
ce
c
a
n
b
e
re
co
gn
i
zed
by
t
h
e
sys
t
em
,
the
do
or
w
il
l
ope
n
au
t
o
ma
tic
al
ly
a
s
sh
ow
n
i
n
F
ig
ur
e
4.
9.
I
f
fa
ce
c
an
no
t
r
e
c
o
gn
i
z
e
by
t
h
e
sys
t
em
,
do
or
w
ill
re
ma
i
n
c
lose
d
a
s
i
l
l
us
tra
t
e
d
i
n
Fi
gure
7
(a
)
an
d
(b).
D
oor
a
c
cess
ca
n
a
l
so
b
e
c
o
ntr
o
l
l
e
d
t
h
r
o
ugh
I
o
T
u
si
ng
Bl
y
nk
a
pp
.
(
a
)
(
b
)
F
i
gur
e
7.
(
a
)
Door
i
s
u
n
loc
k
in
g
(
b
)
D
oor
i
s
locke
d
5.
CONCLUSION
A
s
a
c
onc
lus
i
o
n
,
secur
ity
s
yst
e
m
by
us
ing
fa
ce
r
e
cogn
i
tio
n
com
b
i
n
e
d
w
i
t
h
I
o
T
i
s
s
u
c
c
e
s
s
f
u
l
l
y
d
o
n
e
.
The
fa
ce
rec
o
gni
t
i
o
n
i
s
ab
le
t
o
re
co
g
n
ize
t
h
e
fa
ce
a
n
d
a
ble
t
o
send
n
oti
f
i
c
a
tio
n
t
o
a
u
ser
wh
en
a
n
unk
no
wn
be
in
g
has
be
e
n
d
e
t
ec
te
d
thr
o
u
g
h
I
oT.
O
n
t
he
o
t
h
er
h
a
n
d
,
t
h
i
s
pr
ojec
t
i
s
t
his
pr
ojec
t
sti
l
l
h
as
a
b
ig
r
oom
o
f
i
m
p
r
ov
e
m
e
n
t
to
b
e
do
n
e
,
e
s
pe
ci
al
ly
i
n
t
h
e
ef
fi
ci
en
cy
o
f
the
i
m
a
ge
p
r
o
c
e
ssi
ng
par
t
.
D
u
e
to
t
he
m
od
u
l
e
used
w
h
ic
h
is
Ras
p
b
er
r
y
P
i
3,
t
he
p
r
o
cess
i
n
g
t
i
m
e
of
t
he
c
o
d
i
ng
too
k
a
l
o
n
g
t
i
m
e
s
o
p
r
o
c
e
s
s
t
h
e
i
m
a
g
e
t
a
k
e
n
a
n
d
take
a
c
tio
n.
B
y
usi
ng
an
o
t
her
be
tter
m
odu
l
e
,
t
h
is
p
r
o
j
e
c
t
c
an
b
e
im
pr
ove
d
g
r
eatl
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Int J
P
o
w
E
l
e
c
&
D
ri S
yst
IS
S
N
:
2088-
86
94
IoT
based f
a
c
i
al re
c
o
g
n
iti
on
do
or ac
ce
ss
co
nt
ro
l
hom
e se
cur
i
ty
sys
t
em
usi
n
g
ras
p
be
rry
pi (A
. R. Sy
af
e
e
za)
42
3
ACKNOW
LEDG
E
MEN
T
S
The
au
t
hors
w
o
u
l
d
l
i
ke
t
o
th
ank
U
n
i
v
ersi
ti
T
ekn
i
kal
Mal
a
ys
ia
M
e
l
a
ka
(
U
T
eM)
and
Mi
nis
t
ry
o
f
Ed
uca
t
i
on f
o
r sup
p
o
rt
in
g
t
h
i
s
r
esea
rch
under
P
J
P
/
201
8
/
F
T
K
(
9D
)/S
0
1
603
.
REFE
RENCES
[1]
Y.
J
an
uzaj
,
A.
L
um
a,
Y
.
Jan
u
zaj
,
V
.
R
am
aj.,
"Real
t
i
m
e
acce
ss
c
o
n
t
r
o
l
b
as
ed
on
f
ace
reco
gni
ti
on
,"
i
n
International
Con
f
eren
ce on
Netwo
r
k s
ecur
i
ty &
Co
mp
ut
er
S
c
ien
ce (
I
CNS
C
S-15)
,
pp.
7
-1
2,
2015.
[2]
M
.
S
ah
ani,
C
.
N
a
nd
a,
A
.
K
.
S
ah
u,
B
.
P
a
tt
nai
k
.,
"Hom
e
securi
ty
s
y
s
t
e
m
bas
e
d
on
face
r
ecog
n
i
tio
n,
""
2
015
In
t.
Conf
.
Cir
c
uits,
Po
wer Co
mpu
t
.
T
echnol.
[
I
CCP
CT
-2
01
5
]
, p
p. 1–
6
, 2
01
5.
[3]
G.
S
ent
h
ilk
um
a
r
,
K.
G
o
p
a
l
akrish
nan,
V
.
S.
K
um
ar.
,
"Em
b
ed
ded
i
m
a
ge
capt
u
rin
g
s
y
s
t
e
m
u
s
ing
ras
pberry
p
i
syst
e
m
,
”
v
o
l
.
3
,
No
.
2
,
p
p
. 2
13
-2
1
5
, 2
01
4.
[4
]
M. R
. M
ul
l
a
.
, "F
acial
i
mage b
ased
secu
rit
y
s
y
s
te
m u
s
i
n
g
PCA
,
"
pp.
5
48
-55
3
,
2015.
[
5
]
M
.
H
.
Ju
so
h
&
F.
B
i
n
Jam
al
i
,
'
'H
ome
security sys
t
e
m u
s
ing
i
nt
ern
e
t
o
f
t
h
i
ng
s,
'
'
2017.
[6]
S.
S
.
Liew,
M.
K
halil-Hani
,
S.
A
hmad
R
adzi,
R
.
B
a
k
ht
eri
.,
'
'G
end
e
r
clas
sificat
io
n
:
A
c
on
vo
lutional
n
e
ural
n
et
work
app
r
oach,
'
'
T
u
r
k
is
h J.
El
ect
r.
E
n
g.
Co
mp
ut
.
S
c
i.
,
v
o
l.
24
, No.
3
, p
p.
12
4
8
-
1
2
6
4
, 2
01
6.
[7]
M.
S
a
j
jad
et a
l
.
,
''Ras
p
be
rry
p
i
as
si
s
t
ed
f
ace
recognit
i
on
f
ramewor
k
f
or
e
nhanc
ed
l
aw
-enf
orcem
ent
s
e
rvices
i
n
sm
art
cities,'
'
F
u
t
u
r
.
G
e
ner.
C
o
mpu
t
.
Syst.
, 2
01
7
.
[8]
A.
R
.
S
y
a
f
eeza,
S
.
S
.
L
iew
,
R
.
Bakht
eri.,
'
'
Co
nv
olu
t
i
o
n
a
l
n
eural
net
w
o
r
ks
w
ith
f
used
l
ay
ers
ap
pl
ied
to
f
ac
e
recognit
ion
,
''
In
t
.
J.
C
o
mpu
t
. Intel
l
.
A
ppl.
,
v
o
l
. 14
,
N
o. 3
,
2
0
1
5
.
[9]
A.
R
.
S
y
af
eeza,
M
.
K
h
alil
-Hani
,
S
.
S.
L
iew
,
R
.
Bak
h
teri
.,
'
'
C
on
volu
t
io
nal
n
e
ural
n
etwo
rk
f
or
f
ace
reco
gn
iti
on
wit
h
po
se
a
n
d
i
ll
umi
n
ati
o
n
vari
ati
on,
''
In
t. J.
En
g.
Tech
no
l.
,
Vol.
6
,
N
o
.
1
,
pp.
4
4
-
57
,
2014
.
[10
]
K
.
S
y
azana-It
qa
n
,
A
.
R.
S
y
a
f
e
eza,
N
.
M
.
S
aad,
N.
A
.
Ha
m
i
d,
W
.
H
.
B
in
M
o
h
d
S
aad.,
''A
r
ev
iew
of
f
ing
e
r-vein
bi
om
e
t
ri
cs
i
dentificati
on
app
r
oac
h
es,
'
'
In
d
i
an
J. Sci.
T
e
c
h
no
l.
, v
o
l
.
9
, No
. 3
2,
20
1
6
.
[11
]
S
.
A
h
m
a
d
Radzi
,
M
.
K
h
alil-H
a
ni,
R.
B
akhteri.
,
'
'
F
ing
e
r-vei
n
bio
m
etri
c
id
entifi
catio
n
u
s
in
g
conv
ol
utio
nal
neu
r
al
net
w
ork
,
'
'
Turk
i
s
h J.
E
l
ectr. E
ng.
Comput.
Sci.
,
v
o
l
.
2
4
,
N
o.
3
,
p
p
.
18
63
-1
87
8,
2016.
[1
2]
S
.
Ah
m
a
d
R
a
dz
i.,
''A
M
A
T
L
A
B-ba
se
d
c
o
n
v
o
l
utio
na
l
n
e
ura
l
n
e
t
work
a
pproach
f
o
r
f
ace
recognit
i
on
s
y
s
tem,”
J.
B
i
oi
nf
o
r
m
a
.
P
r
ot
e
o
m
i
c
s
R
e
v
,
vol.
2(1
)
,
pp
.
1
-
5,
2
0
16.
[13
]
M
.
K
.
M
.
F.
A
lif,
A.
R
.
S
y
a
f
eeza,
P
.
M
a
rz
u
k
i
,
A
.
N
.
A
lisa.
,
'
'
F
u
sed
co
nv
ol
ution
a
l
n
e
ural
n
etwork
f
or
f
aci
al
exp
r
essi
on
r
ecog
n
i
t
i
on,
''
i
n
S
y
mp
osi
u
m
on
Electr
i
ca
l,
M
echatr
onics
an
d A
p
p
lied
S
c
ience
201
8 (
S
EM
A’1
8
)
,
v
o
l.
2
0
1
8
, n
o.
No
v
e
mber, p
p.
73
-
74
, 20
1
8
.
[1
4]
T
.
Tat
,
M
.
S
t
u
d
e
n
t, L
.
C
.
W
in
g,
P
. M.
Nu
mb
er., '
'Imag
e
-Bas
ed
F
ace Det
ecti
on S
y
stem
,'
'
.
[15
]
P
.
Kam
e
ncay
,
M
.
B
enco
,
T
.
M
izdo
s,
R
.
Radil,
''A
n
ew
m
et
h
o
d
f
or
f
ace
recognition
u
s
ing
convol
utional
neural
net
w
ork
f
ace rec
o
g
n
i
t
i
on sy
s
t
e
m
- st
a
te of
th
e art,
'
'
pp
.
6
63
-6
72
,
2
0
1
7.
[1
6]
N
.
H
.
M
o
t
l
a
g
h
,
M.
B
ag
aa,
T
.
Taleb.,
''UA
V
-b
ased
i
o
t
p
l
a
t
f
o
rm:
a
crowd
s
u
rvei
ll
ance
u
s
e
c
a
s
e
,'
'
IE
EE
Com
m
un.
Mag
,
v
o
l
.
55
,
N
o. 2
, p
p.
12
8
-
13
4, 20
1
7
.
[17
]
C
.
H.
C
hang
.,
'
'
Sm
art
classroo
m
ro
ll
cal
ler
s
y
stem
w
ith
I
O
T
arch
itecture,
''
i
n
Pr
oceedi
n
g
s
- 2
011
2
n
d
In
ter
natio
nal
Con
f
eren
ce on
Inn
o
va
ti
ons i
n
B
i
o
-Insp
ir
ed Computi
ng an
d Appli
c
at
io
n
s
,
IBICA
2011
, p
p.
35
6
-
3
6
0
,
20
11
.
[18
]
J
.
S
ee
&
S
.
W
.
L
ee,
'
'
A
n
integ
r
ated
v
i
s
i
on-b
a
se
d
arch
itect
ure
f
o
r
h
o
m
e
s
ecuri
ty
s
ys
tem,'
'
IEEE T
r
ans.
Con
s
u
m
.
El
ectro
n
,
v
o
l
.
5
3
, No
. 2
, p
p
. 48
9
-
4
98
, 20
0
7
.
[1
9]
L
. Y
. M
ano
et
a
l
.
,
''Ex
p
l
o
iti
ng
Io
T
tech
nol
ogies
f
or
e
n
h
an
cin
g
h
e
a
lth
smart
h
o
m
e
s
thro
ug
h p
a
ti
ent
identi
f
i
cation
and
e
m
o
tio
n re
c
o
gn
it
io
n,'
'
Co
mput.
Com
m
un
,
vo
l. 8
9-
9
0
, pp
.
17
8
-1
90
,
20
16
.
[2
0]
Y
.
P.
C
he
n
,
Q
.
H.
C
h
e
n,
K
.
Y.
C
h
o
u
,
R
.
H.
W
u
,
'
'Low-c
os
t
f
ace
reco
gni
ti
on
s
ys
tem
b
a
sed
on
e
x
t
en
ded
lo
cal
b
in
a
r
y
pat
t
ern,
”
20
16
In
t
.
Auto
m.
Control Co
n
f
. CA
CS
2
0
16
, pp
.
1
3
-
18
, 2
01
7.
[21
]
N
.
A.
O
thm
a
n
&
I.
A
y
d
i
n
,
'
'
A
face
re
cog
n
i
tio
n
m
e
th
od
i
n
t
h
e
In
t
e
rnet
o
f
Thi
ngs
f
or
s
e
c
u
r
it
y
app
l
i
cati
ons
i
n
sm
art
ho
m
e
s
an
d
ci
ti
e
s
,
'
'
in
Pr
oceed
in
gs -
20
18
6
t
h In
ter
n
a
t
i
o
n
a
l Ist
anbul
S
m
art Grids
a
nd Cit
i
es Congres
s and
F
a
ir
,
ICSG
20
18
,
pp
. 2
0-2
4
,
2
01
8.
[22
]
S
.
H.
O
h,
G
.
W.
K
im,
K
.
S
.
Lim
.
,
''
Com
p
ac
t
deep
l
earn
e
d
f
e
atu
r
e-bas
e
d
f
ace
r
eco
gn
it
ion
f
o
r
V
i
su
a
l
I
nt
e
r
n
e
t
of
Thing
s
,
'
'
J
.
Su
pe
rc
omp
u
t.
, 2
01
8
.
[2
3]
N
.
Fu
na
biki
,
D.
P
ra
ma
diha
n
t
o
,
R
.
Arr
i
d
h
a
,
S
.
S
uk
a
r
id
h
o
to.
,
''C
lassi
f
i
cat
i
on extens
ion
base
d o
n
IoT-b
ig
dat
a anal
y
t
ic
f
o
r
smart
env
i
ro
nm
ent
mo
n
ito
rin
g
a
nd
a
nal
y
tic
i
n
real-tim
e
sy
st
em
,
'
'
In
t
.
J. Sp
ace-B
a
sed
S
i
tuated
Comp
u
t
,
vo
l.
7
,
N
o
,
2
,
pp.
8
2
,
2
017.
[24
]
U
.
S
.
S
h
a
n
t
ham
a
ll
u
,
A
.
S
p
an
ias,
C
.
Teped
e
len
liogl
u,
M
.
S
t
a
nl
ey
.,
'
'A
b
ri
ef
s
u
r
vey
o
f
m
achi
n
e
l
earni
ng
m
e
th
od
s
an
d
th
eir
s
e
ns
or
a
n
d
I
oT
a
pp
li
catio
ns,'
'
201
7
8t
h In
t.
Con
f
.
In
f
o
rm
a
tio
n,
Int
e
ll.
Syst.
Ap
p
l
.
IISA 2
0
1
7
,
vol.
201
8-J
a
nu
ary,
p
p
. 1-8
,
20
18
.
[25
]
F
.
C
holl
e
t.
,
'
'
K
e
ra
s
:
T
he
P
yth
o
n
Deep
L
earn
i
ng
l
ib
rary
,
'
'
K
e
ra
s.I
o
,
20
15.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N: 2
0
8
8
-
86
94
I
nt
J
P
o
w
E
l
e
c
&
D
r
i
S
yst
V
o
l.
11,
N
o.
1
,
Mar
202
0
:
417
–
42
4
42
4
BIOGRAPHI
E
S
OF
AUT
HORS
M
o
ham
m
ad
F
itri
Alif
M
oha
m
m
ad
K
asai
c
urren
tly
i
s
a
P
h
D
candi
dat
e
i
n
Elect
rical
E
ng
in
eeri
ng
f
r
om
U
n
i
versiti
Tek
nol
ogi
M
ala
y
si
a.
H
e
receiv
e
d
B.En
g
d
e
gree
in
M
ech
atro
nics
E
n
g
in
eerin
g
in
2
0
1
1
f
rom
Un
iversi
ti
S
elan
go
r
and
h
i
s
M.En
g
deg
r
ee
in
M
echat
ron
ics
and
Automat
i
cs
C
on
t
r
ol
i
n
2
0
1
3
f
ro
m
Universi
ti
T
ek
no
lo
g
i
M
alay
si
a.
H
is
P
hD
r
esearch
i
s
F
a
ci
al
E
xp
ress
io
n
Reco
gn
it
ion
an
d D
eep L
earnin
g
.
S
y
afeeza
A
h
m
a
d
Rad
z
i
receiv
e
d
h
e
r
B.
Eng
deg
r
ee
in
E
lect
rical-E
l
ect
ro
n
i
c
En
g
i
neerin
g
i
n
2
003
an
d
her
M
.
E
n
g
degree
i
n
E
lect
ric
a
l
-
Elect
ron
i
c
&
Tel
e
co
m
m
unica
ti
on
E
ngineeri
n
g
in
2
00
5
fro
m
U
n
iv
ersi
ti
T
ekn
o
lo
gi
M
al
aysia.
S
he
a
lso
recei
ved
h
e
r
P
h
D
d
e
gree
i
n
Electri
cal
E
n
g
in
eerin
g
f
r
om
t
h
e
sam
e
u
n
i
versity
i
n
20
14.
S
he
i
s
curren
tly
a
S
eni
o
r
Lect
urer
a
t
the
F
acu
lty
of
E
lect
roni
c
E
ngi
neeri
n
g
an
d
Com
p
u
t
er
E
n
g
ineeri
n
g
,
U
n
i
versiti
Tekn
ik
a
l
M
a
l
ay
s
i
a
M
e
l
a
k
a
(
U
T
e
M
)
.
S
h
e
h
a
s
b
een
a
n
a
cad
e
m
i
c
ian
in
U
Te
M
s
i
nce
20
06
.
S
h
e
ded
i
cat
e
he
rs
elf
t
o
uni
versit
y
teachi
n
g
and
co
nd
ucti
ng
r
esearch
.
Her
res
e
a
r
ch
i
nt
erests
i
n
c
lu
de
e
m
b
edd
e
d
s
y
s
t
e
m
,
p
a
tt
e
r
n
reco
gni
ti
on,
m
achi
n
e
learni
n
g
, deep
learn
i
n
g
,
im
age
proces
si
n
g
and
bi
o
m
e
tri
c
.
Evaluation Warning : The document was created with Spire.PDF for Python.