Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
3
,
J
un
e
201
9
, pp.
1805
~
18
13
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
9
i
3
.
pp
1805
-
18
13
1805
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Develop
ment of
portable
automat
ic num
ber pl
ate re
cogniti
on
(ANPR)
system
on
Ras
pb
erry Pi
S.
F
ak
h
ar
A
.
G.
,
M. S
aad H
.
,
A.
Fau
z
an
K.,
R. A
ff
e
ndi
H.
,
M.
Aidil
A
.
Facul
t
y
of
El
e
ct
r
ic
a
l
&
E
le
c
tronic
Eng
ine
e
ring T
ec
hnolog
y
,
Univ
ersit
i
Te
kn
ika
l
Malay
s
ia Mel
ak
a
(UTe
M)
,
Ma
lay
sia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
ug
11
, 201
8
Re
vised N
ov 20, 2
01
8
Accepte
d Dec
11, 201
8
AN
PR
sy
stem
i
s
used
in
aut
o
m
at
ing
acce
ss
c
ontrol
and
se
cu
rity
such
as
ide
nti
f
y
i
ng
stolen
ca
rs
in
re
al
tim
e
b
y
installi
ng
it
to
police
p
at
r
ol
ca
rs,
an
d
det
e
ct
ing
v
ehicle
s
that
ar
e
ove
rspee
ding
on
highwa
y
s.
How
eve
r,
thi
s
technolog
y
is
stil
l
r
el
a
ti
v
ely
expe
ns
ive;
in
Novem
ber
2014,
the
Ro
y
a
l
Malay
s
ia
n
Poli
ce
(PD
RM)
pu
rch
ase
d
and
in
stal
le
d
20
uni
ts
of
AN
PR
s
y
stems
in
the
ir
pat
rol
vehicl
es
costi
ng
nea
rl
y
R
M
30
m
il
li
on.
In
thi
s
pape
r
a
che
ap
er
alter
n
at
i
ve
of
a
porta
b
le
AN
PR
sy
st
em
running
on
a
Raspber
r
y
P
i
with
Op
enCV
li
bra
r
y
is
pre
sen
te
d.
Once
th
e
c
amera
ca
p
ture
s
an
image,
image
desa
turati
on,
fil
t
eri
ng
,
segm
ent
at
ion
and
cha
ra
cter
rec
ogn
it
ion
is
all
done
on
the
Ras
pber
r
y
Pi
b
efo
re
the
ext
r
acte
d
n
um
ber
pla
t
e
is
d
isplay
ed
on
the
LCD
and
s
ave
d
to
a
data
base
.
Th
e
m
ai
n
cha
l
le
nges
in
a
porta
b
le
appl
i
ca
t
ion
in
clude
cru
c
ial
ne
ed
of
an
eff
i
ci
en
t
cod
e
an
d
red
uced
computat
ion
al
complexi
t
y
while
offe
ring
improve
d
fle
xibilit
y
.
The
per
form
ance
ti
m
e
is
al
so
pre
sente
d
,
where
t
he
whole
proc
es
s
is
run
with
a
not
ic
e
able
3
se
conds
dela
y
in
g
et
ti
ng
th
e
f
ina
l
o
utput
.
Ke
yw
or
d
s
:
Im
age p
r
ocessi
ng
Nu
m
ber
recog
niti
on
Op
e
nCV
Po
rta
ble
ANP
R
Ra
sp
be
rr
y
Pi
Copyright
©
201
9
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
:
Sh
am
su
l
Fakha
r
A
bd
Gan
i
,
Faculty
of
Ele
c
tric
al
&
Ele
ct
r
on
ic
En
gin
ee
ri
ng Tec
hnology
,
Un
i
ver
sit
i Te
knikal M
al
ay
sia
Mel
aka (UTe
M),
Hang T
ua
h
Jay
a, 76
100 D
ur
ia
n
T
unggal
, Me
la
ka,
Mal
ay
sia
.
Em
a
il
: sha
m
su
lfak
har@utem
.ed
u.m
y
1.
INTROD
U
CTION
Au
t
om
atic
Num
ber
Plat
e
Re
cogniti
on
(
A
N
PR)
te
c
hnolog
y
has
play
ed
a
substanti
al
ro
l
e
in
a
uto
m
at
ed
traff
ic
la
w
en
f
or
cem
ent.
An
ANPR
syst
e
m
is
essenti
al
ly
a
m
eans
of
ide
ntifyi
ng
a
ve
hi
cl
e
nu
m
ber
plate
by
extracti
ng
the
inf
or
m
at
ion
from
an
i
m
age
file
us
in
g
i
m
a
ge
proces
sin
g
te
chn
iq
ues
.
Th
e
process
ty
pic
al
ly
consi
sts
of
im
a
ge
ac
qu
isi
ti
on,
i
m
age
pr
e
proc
essing,
dete
rm
i
nation
an
d
e
xtr
act
ion
of
reg
i
on
of
inte
rest
(
ROI),
and
inte
rpreti
ng
the
pix
el
s
into
nu
m
erical
ly
read
able
char
act
ers
usi
n
g
op
ti
cal
char
act
e
r
recogn
it
i
on
(O
CR
)
[
1]
-
[
3]
.
Ther
e
are
m
any
te
chn
iq
ues
use
d
t
o
im
pr
ov
e
the
acc
ur
acy
of
t
he
syst
em
;
so
m
e
com
m
on
ones
a
re
R
GB
,
Ycb
C
r,
im
age
filt
er,
f
uzzy
a
lgorit
hm
,
i
m
ag
e
bin
a
rizat
ion,
support
vecto
r
m
achines,
ge
netic
al
gorith
m
and
arti
fici
al
neu
ra
l
network
[
4]
,
[5]
.
This
pa
pe
r
aim
s
to
dev
el
op
an
A
N
PR
syst
e
m
runn
i
ng
s
olely
on
the
Ra
sp
be
rr
y
Pi u
sing O
pe
nCV
.
Ra
sp
be
rr
y
Pi
is
a
card
-
siz
e
d
cheap
m
ini
-
co
m
pu
te
r
aim
ed
to
m
ake
com
p
uting
acce
ssi
ble
to
the
public.
Wh
il
e
the
or
i
gi
nal
intenti
on
of
the
Ra
s
pb
e
rry
Pi
is
to
pro
vid
e
a
ba
se
f
or
ki
ds
to
le
ar
n
pro
gr
am
m
ing
,
it
i
s
al
so
gaining
po
pu
l
arit
y
a
m
on
g
te
ch
-
e
nthusiast
s
as
it
ca
n
be
us
e
d
t
o
do
dif
fer
e
nt
ty
pes
of
c
ommerci
al
pro
gr
am
m
ing
.
It
serv
es
as
an
eff
ic
ie
nt
ba
se
du
e
to
it
s
low
c
os
t
an
d
the
num
ber
of
interfaces
a
va
il
able.
The
Ra
spber
ry
Pi
can
be
us
e
d
instea
d
of
a
per
s
onal
com
pu
te
r,
but
with
so
m
e
lim
it
ation
s
due
to
it
s
lim
it
e
d
processi
ng po
wer
[
6]
,
[
7]
.
Ta
ble 1 f
ur
t
her il
lustrate
s t
he
s
pe
ci
ficat
ion
detai
ls of the
Pi 3
used i
n
this
pro
je
ct
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
201
9
:
1805
-
1813
1806
Tab
le
1.
Ra
s
pb
err
y Pi
3 Speci
ficat
ion
Detai
ls
Sp
ecif
icatio
n
Details
So
C
Bro
ad
co
m
BCM2
8
3
7
CPU
4
× ARM
Co
rtex
-
A
5
3
,
1
.2GHz
GPU
Bro
ad
co
m
Video
Co
re
IV
RAM
1
GB LPDDR2
(90
0
M
Hz)
Netwo
rkin
g
1
0
/1
0
0
E
th
ernet, 2.
4
GHz 80
2
.11
n
wir
eless
Blu
eto
o
th
Blu
eto
o
th
4.1
Clas
sic, Blu
eto
o
th
L
o
w E
n
ergy
Sto
rage
Extern
al
m
icroSD
GPIO
40
-
p
in
head
er,
po
p
u
lated
Po
rts
HDM
I,
3.5
m
m
an
a
lo
g
u
e aud
io
-
v
id
e
o
jack, 4×
USB 2
.0, E
th
ernet,
Ca
m
e
ra
S
erial
I
n
t
erface
(CSI)
,
Disp
la
y
Se
r
ial I
n
terface
(
DS
I)
An
8
m
egap
i
xe
ls
(MP)
Ra
s
pber
ry
Pi
cam
era
can
be
at
ta
c
hed
to
t
he
on
-
bo
a
r
d
Ra
spbe
r
ry
Pi
cam
er
a
connecto
r,
a
nd
this
create
s
a
n
i
m
age
captur
e
syst
e
m
with
em
bed
ded
com
pu
ti
ng
that
can
extract
inform
at
ion
from
i
m
ages
without
the
ne
ed
f
or
an
e
xte
rn
al
processi
ng
unit
.
T
he
m
ulti
ple
GPIOs
avail
able
c
an
i
nterf
ace
with
e
xter
nal
de
vices
an
d
can
be
us
e
d
to
m
ake
res
ults
avail
able
to
ot
her
de
vi
ces.
Co
ns
ide
r
ing
t
he
requirem
ents
of
i
m
age
proces
sing
c
om
par
ed
to
the
Ra
spber
r
y
Pi
’s
processi
ng
m
odule
and
it
s
per
iph
e
rals
,
it
is
decide
d
that
t
he
syst
em
is
capab
le
o
n
e
xe
cuting
the
ta
sk
s
s
pecifie
d.
Ex
per
im
ental
resu
lt
s
s
how
that
th
e
desig
ne
d
syst
e
m
is
decen
t
e
noug
h
to
ru
n
the
i
m
age cap
t
ur
i
ng
an
d
im
age
r
ecognit
ion al
gorithm
[8]
.
2.
PROP
OSE
D DESIG
N
The
ove
rall
syst
e
m
design
c
an
be
cat
e
gorized
into
2
pa
r
ts,
the
ha
rdwa
re
desi
gn
a
nd
the
softwar
e
desig
n.
T
he
ha
rdwar
e
desi
gn
as
de
picte
d
in
Figure
1
will
be
entirel
y
run
on
a
Ra
s
pber
r
y
Pi
3
with
Ra
sp
bia
n
Jessie
OS
insta
ll
ed,
with
ad
dit
ion
al
pe
rip
her
a
ls
of
a
n
8
MP
c
a
m
era
to
captu
re
i
m
ages,
an
d
a
3.5”
TFT
LC
D
to
disp
la
y
the
re
su
lt
s.
T
he
rec
ognized
num
ber
plate
s
will
al
so
be
l
ogge
d
to
a
cl
ou
d
database
i
ns
ide
Pi
.
The
syst
em
is i
nten
ded to
be
por
ta
ble so it
is
powe
red by c
onnecti
ng it
t
o
a
m
ob
il
e p
ower
bank.
Figure
1. Ha
rdwar
e
r
e
qu
i
rem
ents
In
a
ddit
ion
t
o
t
he
Pyt
ho
n
pro
gr
am
m
ing
la
nguag
e
that
is
na
ti
vely
us
ed
i
n
Ra
sp
be
rr
y
Pi
e
nv
i
ronm
ent,
this
syst
e
m
wi
ll
al
so
us
e
O
pe
nCV,
a
com
pu
te
r
visi
on
si
m
ula
ti
on
from
In
te
l
[
9]
to
he
lp
with
im
age
pr
e
-
processi
ng
t
ha
t
is
desig
ned
with
crit
erias
of
res
ource
optim
iz
at
ion
,
lo
w
power
co
nsum
ption
an
d
i
m
pr
ov
e
d
sp
ee
d.
The
pro
po
s
ed
po
rtable
AN
PR
syst
em
so
ftwa
re
i
m
ple
m
entat
ion
can
be
cat
eg
or
iz
ed
into
7
pr
oc
edural
ste
ps
as
il
lustr
at
ed
in
Fi
gure
2.
Eac
h
proces
s
is
cru
ci
al
be
cause
the
res
ult
of
on
e
pa
rtic
ular
process
w
il
l
be
delivere
d o
n
to
the
nex
t
proce
ss.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Develo
pm
e
nt
of
p
ort
able
auto
ma
ti
c
num
ber plate rec
ogniti
on (AN
PR)
…
(
Shamsul
Fak
har A
bd
Gani,
)
1807
Figure
2. Im
ple
m
entat
ion
f
l
owcha
rt
2
.
1.
Ima
ge a
cquisi
tion
An
8
MP
Pi
No
IR
cam
era
ca
pab
le
of
ta
king
inf
rar
e
d
phot
os
up
to
3280
×
2464
pix
el
s
is
us
ed
.
Th
is
ca
m
era
is
al
so
capab
le
of
capt
ur
i
ng
vi
deo
at
1080p3
0,
72
0p60
a
nd
64
0
×
480p90
res
olu
ti
ons
w
hich
is
a
high
qu
al
it
y
vid
e
o
[
10
]
.
T
o
re
duce
the
com
pu
ta
ti
on
al
loa
d
on
t
he
Ra
spber
ry
Pi,
this
cam
era
is
set
up
to
ca
pture
i
m
ages
with
only
640×
480
pi
xels.
Be
f
or
e
te
sti
ng
the
syst
e
m
with
a
li
ve
im
age,
a
te
st
i
m
age
is
prel
oa
de
d
to
te
st t
he
al
gorit
hm
as sh
ow
n
i
n
Fi
gur
e
3.
Figure
3. Loa
de
d
te
st i
m
age
2
.
2
.
Ima
ge
d
es
atur
at
i
on
The
nex
t
ste
p
is
conver
ti
ng
t
he
col
our
im
a
ge
into
gray
scal
e
by
app
ly
in
g
im
age
desatur
at
io
n
[
11]
.
This
is
do
ne
to
re
duce
the
c
om
plexity
of
processi
ng
c
o
l
our
im
age.
A
gr
ay
scal
e
di
gital
i
m
age
is
a
sing
le
sam
ple, car
ryi
ng only
intensi
ty
inf
orm
ation
,
and this ca
n fa
ci
li
ta
te
the p
rocessi
ng b
et
te
r
[
12
]
. O
pen
C
V h
as the
cvCvtCol
or
f
unct
ion
to
co
nv
ert
col
or
im
ag
es
to
gr
ay
scal
e
an
d
the
ef
fect
of
this
f
un
ct
io
n
is
dem
on
stra
te
d
in
Figure
4.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
201
9
:
1805
-
1813
1808
Figure
4. Im
age cover
te
d
int
o g
raysca
le
2
.
3
.
Ima
ge
t
hresh
ol
ding
Thr
e
sholdi
ng
i
s
the
proces
s
of
c
onve
rtin
g
the
gray
scal
e
im
age
into
a
bi
-
le
vel
im
age.
It
does
not
identify
m
ai
n
obj
ect
s,
but
rat
her
se
pa
rate
th
e
m
fr
om
the
ba
ckgr
ound
in
order
t
o
ob
ta
in
the
inf
orm
ation
that
we
ha
ve
t
o
de
al
with.
Durin
g
th
res
ho
l
ding,
it
is
cru
ci
al
to
set
the
c
orre
ct
t
hr
es
ho
l
d
va
lue
w
hich
wi
ll
best
determ
ine
a
pix
el
as
an
ob
j
ec
t
or
a
bac
kgr
ound.
W
e
us
e
O
pe
nCV
'
s
cvT
hr
e
sh
ol
d
to
ac
hiev
e
this,
an
d
the
resu
lt
is as il
lustrate
d i
n
Fi
gure
5.
Figure
5. Im
age af
te
r
t
hr
es
hol
ding
2
.
4
.
Ga
us
sian
f
il
te
r
Gau
s
sia
n
filt
er
is
blu
rr
i
ng
a
n
i
m
age
by
appl
yi
ng
a
Gau
s
sia
n
f
un
ct
io
n.
It
is
us
ed
to
re
duce
im
age
no
ise
a
nd
re
du
ce
unwa
nted
de
ta
il
.
The
vis
ua
l
eff
ect
of
t
his
blurri
ng
te
ch
nique
is
a
sm
oo
th
blur
re
sem
blin
g
that
of
viewin
g
the
im
age
throu
gh
a
tra
ns
l
ucen
t
sc
reen
.
Gau
s
sia
n
fi
lt
er
is
ty
pical
l
y
us
ed
as
a
pre
-
pro
cessi
ng
sta
ge
in
com
pu
te
r
visio
n
al
gorithm
s
in
or
de
r
to
enh
a
nce
im
age
structu
re
s
at
diff
eren
t
s
cal
es
[13]
.
Th
e
eff
ect
of ap
plyi
ng thi
s fun
ct
io
n
is
d
e
m
on
strat
ed
in
Figure
6.
Figure
6. Im
age af
te
r Ga
us
sia
n fil
te
r
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Develo
pm
e
nt
of
p
ort
able
auto
ma
ti
c
num
ber plate rec
ogniti
on (AN
PR)
…
(
Shamsul
Fak
har A
bd
Gani,
)
1809
2
.
5
.
Mo
r
ph
olo
gica
l t
ra
nsform
at
i
on
Mor
phologica
l
trans
form
at
io
n
ins
pects
geom
et
rical
structur
e
within
a
n
i
m
age
by
pro
bi
ng
it
wit
h
sm
a
ll
patte
rn
s
cal
le
d
structu
ri
ng
el
em
ents.
The
res
ult
is
a
nonlinear
im
a
ge
operat
or
th
at
is
well
-
su
it
ed
f
or
exp
l
or
i
ng
ge
om
et
rical
and
to
po
l
og
ic
al
str
uc
ture.
The
ope
ra
tor
is
a
ppli
ed
t
o
a
n
im
age
in
order
to
m
ake
certai
n
featur
e
s
ap
parent,
an
d
disti
nguis
h
m
eaningfu
l
in
form
at
io
n
f
ro
m
irrelev
ant
disto
rtions
by
reducin
g
i
t
to
a
sk
el
et
on
[
14
]
.
This
proces
s
ha
s
4
kinds
of
op
e
rati
ons:
ex
pansi
on,
co
rro
sion,
openi
ng
and
cl
os
in
g
op
erati
on.
Fig
ure
7
s
hows
the e
f
fect
of th
e i
m
age af
te
r
m
or
phologica
l
trans
form
ation
.
Figure
7. Im
age af
te
r
m
orpho
log
ic
al
tra
ns
f
or
m
at
ion
2
.
6
.
Segme
ntati
on
Im
age seg
m
ent
at
ion
is the
pro
cess of
div
idi
ng a
n
im
age in
to m
ulti
ple p
arts
,
ty
pical
ly
u
se
d
to i
den
ti
fy
obj
ect
s
or
oth
e
r
relev
ant
in
for
m
at
ion
in
di
gital
i
m
ages
[15]
.
Af
te
r
m
orphol
og
ic
al
tr
a
ns
f
or
m
at
ion
,
2
se
gm
ents
wer
e
ide
ntifie
d
as
po
te
ntial
cand
i
dates
as
de
picte
d
in
Fig
ur
e
8.
T
he
seg
m
ents
will
then
go
thr
ough
a
no
t
he
r
process
wh
e
re
the
syst
e
m
wil
l
determ
ine
wh
ic
h
one
has
th
e
gr
eat
est
possi
ble
char
a
ct
er.
T
he
se
gm
ent
will
be
isolat
e
d
in
a
s
qu
are
im
age
,
an
d
it
is
te
ste
d
so
that
the
lo
ng
e
st
li
st
of
pote
nt
ia
l
char
act
er
w
il
l
be
determ
in
ed
as
the
act
ual
nu
m
ber
plate
.
T
he
la
st
i
m
age
with
the
m
os
t
po
te
ntial
char
act
er
s
in
it
will
then
be
ch
os
e
n
to
go
f
or
char
act
e
r
rec
og
niti
on
i
n
t
he ne
xt pr
ocess
.
Figure
8. Ca
nd
idate
s for rec
ogniti
on
highli
ghte
d
2
.
7
.
Charac
ter r
ec
ognitio
n
To
rec
ognize
al
ph
a
nu
m
eric
char
act
e
rs
f
rom
the
segm
ented
i
m
age,
the
po
te
ntial
num
ber
plate
is
segm
ented
f
ur
t
her into
in
div
i
du
al
c
ha
racters
b
e
fore
OCR t
est
ing
as
s
how
n i
n
Fig
ure
9.
Figure
9. I
nd
i
vi
du
al
c
har
act
e
r
seg
m
entat
ion
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
201
9
:
1805
-
1813
1810
This
syst
e
m
us
e
s
the
K
-
N
ear
est
N
ei
ghbour
(KNN)
m
et
ho
d
w
h
ic
h
re
quir
es
runn
i
ng
a
s
et
of
te
st
on
the
char
act
e
r
li
br
ary
agai
ns
t
the
pix
el
s
of
the
separ
at
e
d
ind
ivi
du
al
ch
aracte
rs
f
ro
m
the
i
m
age
[16
]
.
A
si
m
il
arity
fu
nc
ti
on
(
)
can
be
de
rive
d
f
r
om
a
distance
f
un
ct
i
on
(
)
.
Wh
e
n
(
1
,
2
)
[
0
,
1
]
w
e
de
fine
t
he
si
m
il
arity as
(
1
,
2
)
=
1
−
(
1
,
2
)
ℎ
0
≤
(
1
,
2
)
≤
1
1
2
T
he
syst
em
will
deter
m
ine
the
ou
tp
ut
of
the
al
ph
an
um
eric
nu
m
ber
plate
by
pr
int
in
g
it
on
the
i
m
age
.
KNN
al
gorith
m
op
erate
s
by
giv
in
g
a
set
of
trai
ni
ng
dat
a
to
gen
e
rate
the
data
struct
ur
e
.
For
this
syst
e
m
,
m
ul
ti
ple
data
structu
re
has
been
ge
ne
rated
to
get
the
highest
per
ce
nt
age
of
ac
cu
r
ac
y
of
the
ch
aracte
r
recog
niti
on
[
17]
.
3.
E
X
PERI
MEN
TAL
RES
UL
TS
A
ND
D
IS
CUSSIO
N
The
syst
em
is
then
set
to
cap
ture
im
ages
fr
om
the
cal
ibra
te
d
[18]
Pi
NoIR
ca
m
era
and
sa
m
ples
of
100
im
ages
w
ere
ca
ptured
a
nd
te
ste
d.
It
is
obser
ve
d
t
hat
the
syst
em
m
anag
e
s
to
deli
ver
go
od
res
ul
ts
w
hen
the
sub
j
ect
is
within
2
m
et
e
rs
f
ro
m
the
ca
m
era.
Sam
pl
es
of
the
s
ucc
essfu
l
recog
niti
on
s
im
age
w
ere
a
s
il
lustrate
d
in
Fi
gure
10
–
13.
Figure
10. Suc
cessf
ul r
ec
ogni
ti
on
Figure
11. Suc
cessf
ul r
ec
ogni
ti
on
Figure
12. Suc
cessf
ul r
ec
ogni
ti
on
Figure
13. Suc
cessf
ul r
ec
ogni
ti
on
So
m
et
i
m
es
the
syst
e
m
fail
s
t
o
recog
nize
ch
aracte
rs
co
rr
ec
tl
y,
wh
ere
le
tt
e
rs
are
filt
ered
ou
t
causi
ng
the
cha
racter
no
t
to
be
reg
is
te
red
in
the
da
ta
structur
e
as
dep
ic
te
d
i
n
Figure
14.
I
n
som
e
cases,
the
syst
e
m
conf
us
es
sim
ilar
le
tt
ers
s
uc
h
as
D
f
r
om
O
or
6
f
r
om
8
and
s
o
on.
T
hi
s
is
il
lustrate
d
in
Fig
ure
15
a
nd
ultim
at
ely in Tab
le
2.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Develo
pm
e
nt
of
p
ort
able
auto
ma
ti
c
num
ber plate rec
ogniti
on (AN
PR)
…
(
Shamsul
Fak
har A
bd
Gani,
)
1811
Figure
14. Eli
m
inate
d
cha
rac
te
r
er
ror
Figure
15. C
ha
racter m
is
m
a
tch
e
rror
Table
2.
C
omm
on
Re
cogniti
on Err
or
s
Actu
al
Reco
g
n
ized
Actu
al
Reco
g
n
ized
B
8
,
6
R
P
D
0
,
O
U
V
G
6
V
U
I
1
W
V
K
X
0
D,
Q
M
N
1
I
N
M
3
B
O
0
6
G
P
R
8
B
Q
0
Ba
sed
on
10
0
s
a
m
ples
ta
ken,
i
t
is
fou
nd
that
t
he
processi
ng
on
the
Ra
s
pber
ry
Pi
3
to
ok
2
to
3
sec
onds
to
process
the im
age a
nd co
m
e o
ut
with t
he re
cognized
num
ber
plate
s,
wit
h a succes
s r
ec
ogniti
on
rate o
f 85%.
Figure
16. T
he
co
m
plete
syst
e
m
4.
CONCL
US
I
O
N
Th
e
de
velo
p
m
ent
of
a
porta
bl
e
AN
PR
syst
em
us
ing
Ra
spb
err
y
Pi
3
is
dem
on
strat
ed
in
this
arti
cl
e,
i
m
ple
m
enting
Op
e
nCV
as
th
e
cor
e
im
age
proces
sin
g
soft
war
e
.
The
syst
e
m
per
f
or
m
s
m
agn
ific
ie
ntly
,
bu
t
it
is
hope
d
that
t
he
Ra
spber
ry
Pi
fou
nd
at
io
n
ca
n
com
e
ou
t
wit
h
a
new
e
r
m
od
el
with
bette
r
processi
ng
po
wer
t
o
el
i
m
inate
the
3
seconds
delay
need
e
d
to
pro
du
ce
t
he
res
ults.
In
t
he
m
eantim
e,
fu
ture
w
ork
m
ay
be
do
ne
to
i
m
pr
ove
the
ac
cur
acy
i
n
the
r
ecognit
ion
pr
oc
ess
as
wel
l
as
sp
ee
d
up
the
t
i
m
e
ta
ken
to
pro
duce
the
outpu
t
on
the alg
or
it
hm
side.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
201
9
:
1805
-
1813
1812
ACKN
OWLE
DGE
MENTS
Th
e
aut
hors
w
ou
l
d
li
ke
to
th
ank
Un
i
ver
sit
i
Tekn
i
kal
Ma
la
ysi
a
Me
la
ka
(U
TeM
)
f
or
pr
ov
i
ding
the
su
pp
or
t
nee
de
d
to
com
plete
the
w
ork
he
rein
via
gra
nt
nu
m
ber
:
RAGS
/
1/2
015/ICT
01/FTK/
03
/B
00
115
a
nd
PJP/2
016/P
KA/
FTK
-
C
ACT/S
01512
.
REFERE
NCE
S
[1]
H.
Raj
put
,
T
.
S
om
,
and
S.
Kar
,
"
An
Autom
at
ed
Vehic
l
e
L
icen
se
Plat
e
Rec
ogn
it
ion
S
y
s
te
m
,
"
Computer
(
Long
.
Be
ach
.
Ca
li
f)
.
,
v
ol.
48
(
8
)
,
pp
.
56
–
61,
2015
.
[2]
S.
R.
Aher
and
N.
D.
Kapa
le
,
"
Autom
at
ic
Num
ber
Plat
e
Recogniti
on
S
y
stem
for
Vehic
le
Ide
nti
ficat
ion
Us
ing
Optic
a
l
Char
acte
r
Rec
ogn
it
ion
,
"
I
nt.
Re
s.
J
.
Eng.
Technol
.
,
pp
.
23
95
–
56,
2017
.
[3]
S.
Du,
M.
Ibr
ahim
,
M.
Shehata,
and
W
.
Bad
aw
y
,
"
Autom
at
ic
Li
c
ense
Plate
Re
co
gnit
ion
(ALPR):
A
Stat
e
-
of
-
the
-
Art
Review,
"
I
E
EE
Tr
ansacti
ons
on
Circu
it
s and
Syste
ms
for Vide
o
Technol
og
y
,
v
ol.
23
(
2
),
pp
.
31
1
–
325,
2013
.
[4]
Z.
X.
Chen
,
C.
Y.
Li
u,
F. L.
Ch
ang,
and
G.
Y
.
W
ang,
"
Autom
at
ic
Lice
nse
-
Pl
at
e Loc
a
ti
on
and
R
e
cogni
ti
on
b
ase
d
on
Feat
ure
Salienc
e
,
"
IEEE
Tr
ans.
V
eh.
Te
chnol.
,
vol
.
58
(
7
)
,
pp
.
3781
–
3785,
2009
.
[5]
A.
Mutholi
b,
T
.
S.
Gunawan,
J
.
Chebi
l
,
and
M.
Kart
iwi,
"
Deve
lopment
of
Portabl
e
Autom
at
i
c
Num
ber
Pl
ate
Rec
ognition
S
y
s
te
m
on
Android
Mobile
Phone
,
"
in
IOP
Con
fe
re
nce
S
erie
s:
Mat
erial
s
Scienc
e
a
nd
Engi
ne
ering
,
vol.
53
(
1
)
,
2013
.
[6]
D.
P.
R.
D.
Ms
.
Seja
l
V.
Gawan
de,
"
Raspber
r
y
Pi
Te
chnol
og
y
,
"
Int.
J.
Adv.
Re
s.
Comput.
Sci
.
So
ft
w.
Eng
.
,
vol
.
5,
no.
4
,
pp
.
37
–
40
,
2015.
[7]
The
MagPi
Mag
az
in
e,
"
Raspber
r
y
Pi
3
is out now!
Specs,
Benc
hm
ark
s &
Mor
e,
"
MagPi
Mag
.
,
pp.
1
–
14,
2016.
[8]
Senthi
lkumar
G,
Gopala
krishnan
K,
and
Sati
sh
Kum
ar,
"
Embedde
d
Im
age
Capt
uring
Sy
st
em
U
sing
Raspber
r
y
Pi
S
y
stem,
"
In
t. J
.
Eme
rg.
Tr
ends
Technol.
Comput
.
Sci.
,
vol
.
3
(
2
)
,
p
p.
213
–
215
,
201
4.
[9]
M.
Te
en
a
Raval
i
and
R.
Sai
Ko
m
ara
giri
,
"
Im
ag
e
Proce
ss
ing
Plat
form
On
Raspber
r
y
Pi
For
Fa
ce
Rec
ogn
it
ion
,
"
Glob.
J
.
Adv. En
g.
Techno
l. ISSN
,
vol
.
3
,
pp
.
2277
–
6370,
2014
.
[10]
W
.
F.
Aba
y
a
,
J.
Basa,
M.
S
y
,
A.
C.
Abad
and
E
.
P.
Dadios,
"
Low
Cost
Sm
art
Secur
ity
Camera
wi
th
Night
Vision
Capa
bilit
y
Us
in
g
Raspber
r
y
P
i
and
OpenCV,
"
Inte
rnationa
l
Confe
renc
e
o
n
Hum
anoid,
Nanote
chnol
og
y,
Information
Tec
hnology
,
Comm
unic
ati
on
and
C
ontrol,
Environm
ent
and
Manage
ment
(
HNICEM)
,
Pala
wan,
pp
.
1
-
6,
2014
.
[11]
G.
Xie
and
W
.
Lu,
"
Im
age
E
dge
Detect
ion
Based
O
n
Openc
v,
"
Int
.
J.
Elec
tron.
El
e
ct
r.
Eng.
,
vo
l.
1
(
2
)
,
pp.
104
–
106
,
20
13.
[12]
H.
Sugano
and
R.
Mi
y
amoto,
"
Highl
y
Optimiz
ed
Im
ple
m
entation
of
OpenCV
for
the
Ce
ll
Bro
adba
nd
Engi
ne
,
"
in
Computer
Vi
s
ion
and
Imag
e
U
nderstanding
,
vo
l.
114
(
11
)
,
pp
.
1
273
–
1281
,
2010
.
[13]
Z.
Zi
vkov
ic
,
"
I
m
prove
d
Adapti
ve
Gauss
ia
n
Mixture
Model
for
Bac
kground
Subtraction,
"
in
Pr
oce
ed
ings
of
the
17th
Int
ernati
on
al
Conf
ere
nce o
n
Pattern
R
ec
og
nit
ion, 2004
.
IC
PR
2004
.
,
v
ol
.
2
, p
p.
28
–
31
,
2004
.
[14]
H.
İ.
Çelik,
L
.
C.
Dülg
er,
an
d
M.
Top
al
b
ek
irogl
u,
"
Fa
bric
Defe
ct
Det
ec
t
io
n
Us
ing
Li
n
ea
r
Filt
e
ring
and
Morphologic
a
l O
per
at
ions,
"
Ind
ian
J
.
Fi
bre
Text.
R
es.
,
vol
.
39
(
3
)
,
2014
.
[15]
N.
Singh
and
M.
J.
Delwic
he
,
"
Mac
hin
e
Vision
Methods
for
Defe
c
t
Sorting
Stonefr
uit
,
"
vol
.
37
(
6
)
,
pp.
1989
–
1997
,
Nov 1994.
[16]
G.
Am
at
o
and
F
.
Falc
hi
,
"
kNN
base
d
Im
age
Cla
s
sific
a
ti
on
Rely
i
n
g
on
Loc
al
Fea
t
ure
Sim
il
ari
t
y
,
"
Proc.
Thir
d
Int.
Conf.
SImi
larit
y
Searc
h
App
l.
-
S
ISAP
’10
,
pp
.
10
1,
2010
.
[17]
R.
A.
Ham
za
h,
M.
S.
Ham
id,
A.
F.
Kadm
in,
and
S.
F.
A.
Ghani,
"
Im
prove
m
ent
of
Stere
o
Cor
respo
nding
Algorit
hm
base
d
on
Sum
of
Abs
olut
e
Diffe
ren
ce
s
and
Edg
e
Preserving
Filt
er
,
"
in
Proceedi
ng
s
of
the
2017
IEE
E
Inte
rnat
iona
l
Confe
renc
e
on
S
ignal
and
Image Proce
ss
ing
App
l
ic
ati
ons
,
ICSIP
A
2017
,
pp.
222
–
2
25,
2017
.
[18]
R.
A.
Ham
za
h,
S.
F.
Abd
Ghani,
A.
F
.
Kadm
in
and
K.
A.
A.
Aziz,
"
A
Prac
tical
Method
for
Camera
Cal
ibr
at
io
n
in
Stere
o
Vision
Mobile
Robo
t
Naviga
t
ion,
"
I
E
EE
Stud
ent
Conf
ere
nce
on
R
ese
a
rch
and
Dev
el
op
ment
(
SCOReD)
,
Pulau
Pinang
,
pp
.
104
-
108
,
2012
.
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Sh
ams
ul
Fakh
ar
bin
Ab
d
Gani
gra
dua
te
d
from
Univer
siti
Malay
sia
Perli
s
(Uni
MA
P)
in
Bac
helor
of
Engi
ne
eri
ng
(
Com
pute
r
Engi
n
ee
ring)
with
hon
ours
in
2006
and
la
t
er
r
ec
e
ive
d
h
i
s
Master
'
s
d
egr
e
e
in
Inte
rne
t
&
W
eb
Com
puti
n
g
in
2015
fro
m
Roy
al
Melbour
ne
Instit
ute
of
Te
chno
log
y
(R
MIT)
Aus
tra
li
a
.
He
start
ed
his
ca
r
ee
r
as
an
R&D
el
ectroni
c
engi
n
ee
r
spec
ializing
in
software
design
f
o
r
m
et
er
cl
uster
d
eve
lopment
in
Siemens
VDO
Autom
oti
ve
Penang
(la
t
er
known
as
Conti
nent
al
Autom
oti
ve
Mal
a
y
si
a).
Befor
e
l
e
avi
ng
Cont
ine
n
t
al
,
Sham
sul
plays
an
a
ct
iv
e
ro
le
i
n
Mitsubishi
Fus
o
proje
c
ts
as
a
sof
twar
e
develope
r
and
al
so
softw
are
project
m
an
age
r.
Sham
sul
is
now
a
le
ct
ure
r
in
el
e
ct
roni
c and c
om
pute
r
engi
n
eering
t
ec
hnolog
y
depa
rtment
of
F
TK
EE
UT
eM.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Develo
pm
e
nt
of
p
ort
able
auto
ma
ti
c
num
ber plate rec
ogniti
on (AN
PR)
…
(
Shamsul
Fak
har A
bd
Gani,
)
1813
Mohd
Saad
b
orn
in
1981,
gra
du
at
ed
from
Multi
m
edi
a
Univer
sit
y
(MM
U)
where
he
recei
ved
his
B.
Eng
m
aj
oring
in
Com
pute
r
in
2003.
H
e
the
n
continues
working
with
Telek
om
Malay
si
a
Berha
d
an
d
la
t
er
joi
n
ed
Sie
m
ens
VDO
Auto
m
oti
ve
Penang
(
la
t
er
known a
s Conti
nental
Auto
m
oti
ve
Malay
si
a
)
as
a
softwar
e
eng
i
nee
r.
He
devel
oped
firmw
are
in
m
et
er
c
luste
r
developm
ent
f
or
both
local
a
nd
int
ern
at
ion
al
cus
tomer
in
the
area
of
Control
ler
Area
Network
a
nd
ECU
dia
gnostic
during
his
t
e
nure
in
the
compan
y
.
In
2014
he
rec
ei
ved
his
M.
En
g
m
aj
ori
ng
in
Com
pute
r
and
Com
m
unic
at
ion
from
Univer
siti
Keba
ngsaa
n
Malay
s
ia
(UK
M)
.
Curre
n
tly
h
e
is
a
le
ct
ur
er
in
Univer
siti
Te
knik
al
Malay
s
ia
Mela
ka
teac
h
ing
progra
m
m
ing,
e
m
bedde
d
s
y
st
em a
nd
m
i
cro
cont
r
oll
er
subjects.
Ah
mad
Fau
z
a
n
bin
Ka
dmin
C
Eng.
P.T
e
ch.
c
u
r
ren
tly
atta
che
d
with
UTe
M
as
a
rese
arc
her
,
Ahm
ad
Fauza
n
bin
Kad
m
in
CEng.
P.T
ec
h.
h
as
over
1
4
y
e
ars
of
experie
nc
e
in
e
le
c
tr
onic
&
comput
er
engi
ne
eri
ng
fi
eld
with
te
chnic
al
expe
r
t
ise
in
R&D
engi
neering,
computer
vision
&
m
edi
cal
el
e
ct
roni
cs.
He
g
rad
uated
wi
th
a
Bac
h
el
or
Degr
ee
i
n
E
lectr
oni
c
Engi
neering
fro
m
Univer
siti
Sai
n
s
Malay
s
ia
(US
M)
and
Master
D
egr
ee
in
Com
pute
r
&
Com
m
unic
ation
Engi
n
ee
r
i
ng
from
Univer
siti
Keba
ngsaa
n
Ma
l
a
y
si
a
(UK
M)
.
Pr
evi
ousl
y
,
he
wor
ked
with
M
ega
st
ee
l
Sdn.
Bhd.
,
S
ams
ung
SD
I
(M
)
Sdn
.
Bhd.
and
Agensi
Angkasa
Nega
ra
.
He
publ
ished
seve
r
al
te
c
hnic
a
l
and
engi
n
ee
ring
pape
rwor
k
s
in
image
proc
essing
and
m
edi
c
al
el
e
ct
roni
cs.
Ros
tam
Affen
di
Ham
z
ah
gra
du
at
ed
from
Univer
siti
Te
knolog
i
Malay
s
ia
(UTM
)
where
he
rec
e
i
ved
his
B.
Eng
m
aj
o
ring
in
Elec
tronic
Engi
n
ee
ring
in
2000.
In
2010
he
recei
v
ed
his
M.
Sc.
m
aj
oring
in
El
e
ct
roni
c
S
y
s
tem
Design
Engi
n
ee
ring
from
Univer
siti
Sa
ins
Ma
lay
s
ia
(US
M)
.
In
2017,
h
e
re
cei
ved
PhD
m
aj
oring
in
El
ectroni
c
Im
agi
ng
from
Uni
ver
siti
Sains
Malay
si
a
(US
M)
.
Curre
ntly
he
is
a
le
c
ture
r
in
Uni
ver
siti
Te
knik
al
Malay
sia
Mel
a
ka
teac
h
ing
dig
it
al
elec
troni
cs
and
digital
ima
ge
proc
essing.
Muhammad
Aidil
bin
Az
har
gra
duated
from
Univer
siti
Te
kn
i
kal
Ma
lay
sia
M
el
ak
a
(UT
eM)
i
n
Bac
he
lor
of
Com
pute
r
Engi
neering
Te
chn
o
log
y
(Com
pute
r
Sy
s
te
m
)
with
H
onours
in
2018.
He
is
cur
ren
t
l
y
worki
ng
as
a
t
est
and
val
id
at
ion
enginee
r
in
a
firmw
are
dep
art
m
ent
in
W
este
rn
Digital
(M
al
a
y
s
ia
).
Evaluation Warning : The document was created with Spire.PDF for Python.