Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
10
,
No.
3
,
June
201
8
,
pp.
12
21
~
1226
IS
S
N:
25
02
-
4752
, DO
I: 10
.11
591/
ijeecs
.
v
10
.i
3
.pp
12
21
-
12
26
1221
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Emb
edded Aut
om
ate
d
Vi
sion for
Double P
ar
kin
g
Identific
atio
n
System
No
r
asyiki
n F
adi
lah
,
See
Yoo
n Soon,
Hadz
fiz
ah
R
adi
Fa
cul
t
y
of Electr
ic
a
l
and
E
le
c
tron
ic
s E
ng
ine
er
ing,
Univer
siti
Malays
ia
Pahang
,
2660
0
Pekan, Paha
ng
,
Mal
a
y
s
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Dec
26
, 201
7
Re
vised
Ma
r
2
,
201
8
Accepte
d
Ma
r
2
0
, 201
8
The
a
im
of
thi
s
work
is
to
assist
the
c
ity
admini
strat
ion
issue
which
invo
lve
the
tr
aff
i
c
flow
disrupti
on
in
an
urba
n
ar
ea
.
One
of
the
ca
uses
of
tra
ffi
c
flow
disrupti
on
is
dou
ble
p
ark
ing;
thus
,
in
thi
s
work,
an
aut
om
ated
doub
le
p
ark
ing
ide
nti
f
icati
on
an
d
al
ert
s
y
s
te
m
was
deve
lope
d
usi
ng
embedde
d
vision
sy
st
em
and
interne
t
of
thi
ngs.
A
ca
m
er
a
was
uti
l
iz
ed
t
o
ac
quir
e
the
i
m
age
of
a
par
king
ar
ea,
a
nd
the
imag
e
was
proc
essed
using
Bea
gl
e
bone
Black
proc
essor.
A
co
m
pute
r
vision
algorithm
was
deve
lope
d
to
p
roc
es
s
the
image
using
bac
kgrou
nd
subtrac
t
ion,
reg
ion
of
int
er
e
st
ide
nt
ifi
c
at
ion
,
and
co
lor
ana
l
y
sis.
W
hen
a
double
par
k
ed
v
ehi
c
le
is
de
tected,
the
d
ata
was
sent
int
o
the
cl
oud
aut
om
atic
al
l
y
to
a
le
rt
the
ci
t
y
administrator
for
furthe
r
ac
t
ion.
Th
e
deve
lop
ed
s
y
s
tem
ac
hie
v
ed
91%
accurac
y
in
de
t
ec
t
ing
th
e
tra
ffi
c
violati
on
of
double
p
ark
ing
Ke
yw
or
d
s
:
Be
agleb
on
e
Com
pu
te
r
visi
on
Em
bed
ded syst
e
m
ap
plica
ti
on
Sm
art ci
t
y
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
:
Norasyi
kin Fa
dilah
,
Faculty
of Elec
tric
al
an
d El
ect
ronics E
nginee
rin
g,
Un
i
ver
sit
i M
al
ay
sia
Pah
a
ng, 266
00 Pe
kan,
Paha
ng, Mal
ay
sia
1.
INTROD
U
CTION
In
t
his
m
od
er
n
era,
t
he
af
for
da
bili
ty
of
purc
hasin
g
veh
ic
le
s
increase
s
no
wad
ay
s
.
It
dire
ct
ly
resu
lt
s
in
increasin
g
nu
m
ber
of
ve
hicle
s
an
d
t
hu
s
pr
oduce
s
high
vo
l
um
e
of
tra
ff
ic
i
n
the
ur
ban
are
a.
A
n
iss
ue
of
il
le
gal
parkin
g
has
be
com
e
m
or
e
ap
par
e
nt
prob
le
m
f
aced
by
t
he
ci
ty
adm
inistr
at
ion
as
it
is
on
e
of
t
he
reas
on
s
that
le
ad
to
bo
tt
le
ne
ck
on
r
oad
a
nd
soo
n
c
onge
sti
on
.
Be
fore
e
nfor
ci
ng
the
la
w
to
s
ol
ve
thi
s
pro
blem
,
detect
ing
su
c
h
vio
la
ti
on
has
bee
n
a
cha
ll
eng
e
by
m
ult
iple
par
ti
es,
as
the
ta
sk
of
de
te
ct
ion
so
l
el
y
dep
e
ndent
on
hu
m
an
op
e
rato
r for s
urveil
la
nce
[
1].
To
ai
d
with
th
e
detect
ion
of
i
ll
egal
parkin
g,
m
any
te
chn
iq
ue
s
ha
ve
been
use
d
wh
ic
h
util
i
zed
var
io
us
sens
or
s
.
I
n
se
nsor
-
ba
sed
syst
em
,
veh
ic
le
s
are
detect
ed
us
in
g
diff
e
re
nt
ki
nd
of
se
nsor
s
su
c
h
as
in
duct
ive
loop
,
m
agn
et
ic
sens
or,
ultras
onic
sens
or
a
nd
i
nfra
red
se
nsor
.
I
nductive
lo
op
[
2]
and
m
agn
et
ic
sens
or
[3
]
rely
on
t
he
change
of
m
a
gn
et
ic
value
due
to
par
ts
of
the
veh
ic
le
to
ob
ta
in
the
si
gnal
thu
s
it
is
pr
evail
in
g
on
trackin
g
m
ov
ing
ve
hicle
s.
Alth
ough
t
hese
sens
or
s
pr
ovi
de
high
accuracy
in
detect
ing
ve
hi
cl
es,
the
dura
bili
ty
,
instal
la
ti
on
an
d
m
ai
ntenan
ce
effor
t
pose
a
s
m
ajo
r
dra
w
backs
as
they
require
pa
ve
m
ent
cutti
ng
[
4]
,
[
5].
Ultraso
nic
an
d
infr
ar
ed
se
nsors
are
ca
pab
l
e
to
determ
ine
wh
et
he
r
a
ta
rg
et
ed
spo
t
is
de
te
ct
ed
with
ve
hicle
,
howe
ver
due
to
the
natu
re
of
us
in
g
wav
el
et
s
to
se
ns
e
ob
j
e
ct
s,
these
se
nsors
nee
d
direct
li
ne
-
of
-
sig
ht
on
the
ta
rg
et
ed
s
po
t.
Th
us
,
they
ar
e
ver
y
hard
to
protect
against
dust
or
ac
ci
den
ta
l
dam
a
ge.
Mo
reover
,
since
ind
ivi
du
al
se
nsor
has
to
be
pl
aced
on
eac
h
ta
rg
et
e
d
pa
rk
i
ng
spot,
im
ple
m
entat
ion
of
s
uc
h
sen
sors
on
a
la
rg
e
parkin
g
are
a
a
nd
i
rr
e
gu
la
r
s
urface
will
be
costly
and
i
m
pr
act
ic
al
du
e
to
m
any
har
dware
i
ns
ta
ll
at
i
on
a
nd
pav
em
ent cu
tt
ing p
r
ocedur
e
s.
Com
pu
te
r
vision
m
et
ho
d
is
ge
ner
a
ll
y
m
or
e
rob
us
t
an
d
has
le
ss
dep
e
nd
e
nc
y
on
the
c
harac
te
risti
cs
of
the
r
oa
d
s
urfac
e
an
d
s
pecifici
ty
on
in
div
id
ual
ve
hicle
detect
ion. It
pro
vid
e
s i
m
age
visu
al
iz
at
ion
o
f
a w
ide
area
[6]
-
[7
]
eve
n
on
irregular
s
urfa
ce
and
path
wa
y.
Im
age
pr
oce
ssing
te
c
hn
i
ques
ha
ve
t
o
be
c
on
si
der
e
d
to
prov
i
de
the
eff
ic
ie
ncy
and
e
ff
ect
i
veness
of
ve
hicle
detect
ion.
Be
f
ore
the
im
age
is
processe
d
f
or
the
detect
io
n
pur
pose,
a
pr
e
-
processin
g
te
ch
nique
is
need
e
d
t
o
e
nh
a
nce
a
nd
im
pr
ove
the
a
cq
uire
d
im
age
fo
r
be
tt
er
i
m
age
proc
essi
ng
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
.
10
, N
o.
3
,
June
201
8
:
12
21
–
12
26
1222
eff
ect
ive
ness.
In
[8
]
,
S
oo
pr
opos
e
d
that
in
order
t
o
detect
a
veh
ic
le
in
a
par
ti
cular
pla
ce
m
ent,
the
de
te
ct
ion
reg
i
on
m
us
t
be
locat
ed
at
t
he
re
gion
of
interest
(R
OI).
It
m
eans
that
the
RO
I
s
hould
be
placed
on
th
e
pro
hib
it
ed
are
a
fo
r
de
te
ct
ion
of
double
-
pa
rk
e
d
ca
r.
By
def
inin
g
RO
I
,
the
process
of
detect
in
g
ve
hicle
avail
abili
ty
ca
n
be
sim
plifi
ed
an
d
us
e
l
ess
proces
sin
g
powe
r.
Ba
c
kgr
ound
s
ub
t
ra
ct
ion
al
gorith
m
will
e
m
ph
asi
ze
out
the
obj
ect
on
the
fo
re
gro
un
d
w
hile
rem
o
ving
the
sta
ti
c
backgro
und
im
age
wh
ic
h
i
s
the
back
gro
und
m
od
el
.
T
his
m
ea
ns
t
hat
in
pract
ic
al
,
m
ov
ing
obj
ect
s
li
ke
ve
hicle
s
an
d
hum
a
ns
will
be
plac
ed
out
and
sho
wed
w
hile
the
backgroun
d
m
od
el
is
rep
la
ce
d
wit
h
a
sing
le
c
olor.
As
propose
d
i
n
[9
]
,
t
his
ba
ck
gro
und
su
bt
racti
on
al
gorithm
is
the
sim
plest
al
go
rith
m
with
op
tim
a
l
perform
ance
to
be
us
ed
in
te
rm
of
obj
ect
detect
ion.
In
[
10
]
,
li
ve
vide
os
wer
e
capt
ur
e
d
an
d
analy
sed
by
com
par
in
g
each
f
ram
e
with
res
pect
to
any
backg
rou
nd
sc
enes.
As
lo
ng
as
the
captu
rin
g
de
vice
is
sta
ti
c,
the
obj
ect
detect
ion
c
ou
l
d
be
util
i
zed
in
f
ull
perform
ance.
Ba
sed
on
the
adv
a
ntage
s
of
us
in
g
vi
deo
a
cqu
isi
ti
on
an
d
com
pu
te
r
visi
on
al
gorithm
i
n
detect
in
g
veh
ic
le
s
f
or
e
nfo
rcem
ent
app
li
cat
ion
s
as
dis
cusse
d
in
[
11]
,
we
aim
to
dev
el
op
a
com
pute
r
visio
n
al
gorithm
that wil
l aut
oma
ti
cal
ly
d
et
ect
t
he violat
io
n of
double
-
parked
veh
ic
le
s.
2.
RESEA
R
CH MET
HO
D
2.1. H
ardwar
e a
n
d Sof
twar
e Set
up
The
aut
om
at
ed
d
ouble
parkin
g
identific
at
io
n and
alert
syst
em
h
ardware
c
onsist
ed
of a U
S
B Log
it
ec
h
C310
cam
era
at
ta
ched
on
B
eagleb
on
e
Bl
ack
(BBB
)
m
icr
oc
ontrolle
r
.
A
m
od
el
of
a
pa
rk
i
ng
sp
ace
a
rea
was
const
ru
ct
e
d
as
il
lustrate
d
in
Figure
1
a
nd
the ca
m
era
and
B
BB
wer
e
place
d
at
a
heigh
t
th
at
gen
erate
d
el
evated
bir
d
-
ey
e v
ie
w of
t
he
pa
rk
i
ng
sp
ace area
. A
r
ow
be
hind the par
king slots
w
as d
efi
ned
a
s the pr
oh
i
bited parki
ng
a
rea, w
hich
if
a v
e
hicle
p
a
rks
in
the
area
, it
would be
consi
der
e
d
as
a
double
p
a
r
ked v
e
hi
cl
e.
Figure
1. Mo
de
l of a
parkin
g space a
rea
The
com
pu
te
r visi
on algorit
hm
o
f
the syst
em
w
as d
evel
oped
inside BBB
by u
sin
g Op
e
nC
V
li
br
aries
.
To
e
xecu
te
a
nd
com
p
il
e
the
cod
i
ng
withi
n
BB
B,
Op
e
nC
V
wa
s
first
i
nst
al
le
d
inside
th
e
hard
war
e
.
Fi
gure
2
il
lustrate
s the
pro
posed
alg
or
it
hm
f
or
double
parkin
g
i
den
ti
f
ic
at
ion
syst
em
.
Figure
2. Pro
pose
d
al
gorithm
for
double
parkin
g
ide
ntific
at
ion
syst
e
m
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
Em
bedde
d
A
uto
m
ated
Visi
on
for Do
ub
le
Par
ki
ng
I
den
ti
fi
cation Syste
m
(
No
ra
syi
ki
n Fa
dila
h
)
1223
2.2.
B
ackgr
ound
an
d
For
eg
round
Det
ec
tion
In
this
sta
ge,
the
ca
m
era
star
te
d
to
acq
uir
e
the
cur
re
nt
im
age
into
the
BB
B
storag
e.
Then,
tw
o
functi
ons
we
re
execu
te
d
in
pa
rall
el
.
The
fir
st
functi
on
c
onve
rted
th
e
captu
red
im
age
fr
om
RGB
into
H
S
V
colo
r
sp
ac
e,
wh
il
e
the
sec
ond
f
un
ct
i
on
e
xecu
te
d
the
ba
ckgr
ound
sub
tract
ion
al
go
rithm
us
ing
m
ixtur
e
of
Gau
s
sia
n (MO
G)
m
et
hod
[
12
]
. F
inall
y, the
r
egio
n of i
ntere
st (RO
I)
of the
i
m
age w
as
def
i
ned.
Be
fore
ge
ne
rati
ng
t
he
f
or
e
gro
und
m
od
el
,
the
bac
kgrou
nd
m
od
el
of
the
parkin
g
area
was
init
ia
li
zed.
The
backgro
und
m
od
el
consi
ste
d
of
the
parkin
g
area
with
ou
t
a
ny
ve
hicle
,
w
hich
was
gen
e
rated
duri
ng
t
he
init
ia
li
zation
(s
ee
Figu
r
e
3(
a
)).
Then,
the
ba
ckgr
ound
subtr
act
ion
us
i
ng
G
aussion
Mi
xtur
e
Mod
el
[13
]
,
[14]
was
im
ple
m
ented
on
t
he
RG
B
i
m
age
(see
Figure
3
(b))
t
o
gen
e
rate
dyna
m
ic
fo
regrou
nd
m
od
el
sh
owing
the
veh
ic
le
s as
s
hown in Fi
gure
3(c).
(a)
(b)
(c)
Figure
3. Sam
ple i
m
age o
f
(
a
)
Ba
ckgrou
nd m
od
el
,
(b)O
rigina
l im
age,
(c
)Fo
regrou
nd m
od
el
Mult
iple
ROIs
wer
e
se
gm
ente
d
on
the
proh
i
bited
parki
ng
a
rea
as
sh
ow
n
in
Fig
ur
e
4.
Each
ROI
wa
s
separ
at
e
d
an
d
processe
d
in
div
id
ually
by
the
i
m
age
pr
ocess
ing
al
gorithm
discusse
d
in
th
e
fo
ll
owin
g
se
ct
ion
s
.
The
se
par
at
io
n
of
R
OI
s
was
im
ple
m
ented
in
to
the
or
igi
nal
RGB
i
m
age,
f
oreg
rou
nd
m
od
e
l
and
HSV
im
a
ge
t
o
ease t
he
e
xtrac
ti
on
of p
a
ram
eter
s
need
e
d.
Figure
4. Mult
iple R
O
Is
se
gme
nted o
n
the
proh
i
bited
parkin
g
a
rea
2.3. Par
ame
te
rs Initi
aliz
at
ion
Four
pa
ram
et
e
rs
wer
e
init
ia
li
zed
be
fore
det
ect
ing
any
ve
hi
cl
es
on
the
for
egro
und
m
od
el
.
They
wer
e
set
to 0
durin
g t
he
fir
st f
ram
e
of the im
age.
T
he param
et
ers
are li
ste
d
as
b
e
low:
Count
:
ind
ic
at
es
the
num
ber
of
ti
m
es
the
R
OI
detect
ed
th
ere
is
a
parkin
g
vi
olati
on
.
If
Count
=2
,
it
m
e
ans
that a ve
hicle
i
s im
m
ob
il
e fo
r 2
pro
gr
am
r
uns
.
Mea
n(R,G,B)
:
ind
ic
at
e the
m
e
an
values
for r
ed,
green
and
bl
ue
s
paces
respec
ti
vely
.
Mea
n(H,S,V)
: in
dicat
e t
he
m
e
an values
for h
ue,
sat
ur
at
io
n a
nd v
al
ue
s
pace
s r
es
pecti
vely
.
Area
:
area
of
the
f
or
e
gro
und
obj
ect
(
blob)
by
cal
culat
ing
the
non
-
zer
o
pix
el
in
the
f
oreg
rou
nd
m
odel
i
m
age.
2.4. V
ehic
le
D
etectio
n
The
A
rea
of
th
e
blob
was
cal
culat
ed
by
c
ounting
t
he
no
n
-
zero
pix
el
s
in
t
he
RO
I
of
the
foregr
ound
m
od
el
.
To
def
i
ne
the
bl
ob
as
a
ve
hicle
,
the
rati
o
of
deter
m
inati
on
,
Ca
was
int
rod
uce
d.
T
his
rati
o
was
pr
e
-
determ
ined
by
placi
ng
a
ve
hi
cl
e
to
t
he
sp
eci
fic
ROI,
wit
h
m
ult
iple
at
tem
pts
of
cal
ibr
at
ion
an
d
de
fining
on
how
s
ensiti
ve
the
detect
io
n
s
hould
be
.
F
or
exam
ple,
whe
n
Ca
=0
.5,
the
cal
culat
ed
are
a
has
to
be
m
ore
than
half
of
the
total
area
of
the
ROI
to
de
fine
th
at
there
is
a
ve
hicle
on
the
ROI.
I
f
the
area
is
le
ss
than
half
of
the
total
area
of
ROI,
it
m
eans
that
there
is
no
veh
ic
le
det
ect
ed,
th
us
the
Count
=0.
E
quat
ion
(1)
sho
ws
the
conditi
on whe
n
the
re is a
v
e
hi
cl
e d
et
ect
ed o
n
the
RO
I,
w
he
re ROI.
Ar
ea
in
dicat
es the t
otal area
of
t
he
R
OI
.
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
.
10
, N
o.
3
,
June
201
8
:
12
21
–
12
26
1224
Area >
=
C
a*R
OI
.
Are
(1)
2.5. Viol
at
i
on
Ident
ific
at
i
on
To
ide
ntify
w
he
ther
the
re
is
a
ny
pa
rk
i
ng
viol
at
ion
,
the
par
a
m
et
ers
on
t
he
c
urren
t
fr
am
e
and
previ
ou
s
fr
am
e
wer
e
co
m
par
ed
by
cal
culat
ing
the
c
ha
ng
e
s
in
Mea
n(
R,G
,B)
and
M
ean
(
H,S,V)
values.
∆P
aramet
er
was
c
al
culat
ed
a
nd
com
par
ed
wit
h
a
n
Id
e
ntific
a
ti
on
Pa
ram
et
er
Ra
nge
(
IPR
)
t
o
s
how
t
he
flu
ct
uation
val
ue
fro
m
pr
e
vious
f
ram
e
to
def
i
ne
w
hether
the
detect
ed
ve
hicle
was
the
sam
e
fr
om
current
f
ram
e.
We
co
ns
ide
re
d
tw
o
cases:
Ca
se
1:
∆Par
amet
er
>
IPR
:
the
fl
uctuati
on
is
hig
h,
the
refo
re
the
ve
hicle
s
from
cur
ren
t
a
nd
pr
e
vi
ou
s
fr
am
es are d
i
fferent.
Co
un
t
w
il
l be set to
1,
wh
ic
h
m
eans a
v
e
hicle
is im
mo
bile
for on
e
ti
m
e.
Ca
se
2:
∆Parameter
<
=
IPR
: t
he
fluctuati
on is
low,
th
us
ve
hicle
s
detect
ed
fr
om
cur
re
nt
and
p
re
vious
fr
am
e
s
are
the
sam
e.
Cou
nt
w
il
l
be
increa
sed
by
on
e
to
si
gn
ify
the
sam
e
veh
ic
le
is
im
m
ob
il
e
fo
r
an
ad
diti
on
a
l
Count
.
IPR
w
as
determ
ined
ex
per
im
ental
l
y
by
ta
bu
la
ti
ng
the
pa
ram
eter
s
c
ollec
te
d
f
ro
m
222
set
of
data.
Wh
e
n
t
he
C
ount
num
ber
r
ea
c
hed 6, the
v
e
hi
cl
e w
as
i
den
ti
fi
ed
as
v
i
olati
ng
the do
ub
le
pa
r
king
ru
le
s.
3.
RESU
LT
S
A
ND AN
ALYSIS
In
this
sect
io
n,
it
is
exp
la
ined
the
res
ul
ts
of
resea
rch
and
at
the
sam
e
tim
e
is
giv
en
th
e
com
pr
ehe
ns
ive
discussi
on.
Re
su
lt
s
can
be
presented
i
n
fi
gures,
gr
a
phs,
ta
bl
es
and
ot
her
s
t
hat
m
ake
the
r
eade
r
unde
rstan
d
eas
il
y [2
]
, [5]. T
he
d
isc
us
sio
n
ca
n be m
ade in
s
ever
al
s
ub
-
c
ha
pters.
3.1.
D
ata Co
ll
ection
Fo
r
eve
ry
pr
ogram
execu
ti
on,
pr
e
vious
im
age
file
s
wer
e
rep
la
ce
d
with
curre
nt
file
s
to
c
on
se
r
ve
m
e
m
or
y
in
BB
B.
Th
ree
ty
pes
of
im
ages
w
er
e
store
d
f
or
ea
ch
RO
I
w
hic
h
consi
ste
d
of
th
e
RGB
im
age,
HSV
i
m
age
and
the
foregr
ound
i
m
age.
A
data
lo
gg
e
r
(see
Fi
gure
5)
was
produced
t
o
save
t
he
pa
ram
et
ers
neede
d
wh
ic
h
incl
ude
Count
,
A
rea
,
Mea
n
(
R,
G,
B)
,
Mea
n(H,S,V)
and
∆P
ar
amet
er
.
Figure
5. Data
logger
3.2.
Vehicl
e
D
etectio
n
Since
the
syst
e
m
ob
j
ect
i
ve
is
to
detect
a
ve
hi
cl
e
that
vio
la
te
the
double
pa
rk
i
ng
ru
le
s
,
it
sh
oul
d
be
able
to
detect
wh
et
her
the
ve
hicle
on
the
s
pe
ci
fic
ROI
is
the
sam
e
or
dif
fer
e
nt
ve
hicle
.
Thu
s
,
tw
o
de
ci
ding
factors
we
re
o
btained
i
n
determ
ining
the
s
uc
cess
rate.
222
consecuti
ve
fra
m
es
wer
e
ob
t
ai
ned
to
te
st
f
or
t
he
sam
e
or
diff
e
r
ent
ve
hicle
det
ect
ion
.
Ta
ble
1
shows
t
he
detect
ion
acc
ur
ac
y
fo
r
each
dif
f
eren
t
par
am
et
e
r.
T
he
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
Em
bedde
d
A
uto
m
ated
Visi
on
for Do
ub
le
Par
ki
ng
I
den
ti
fi
cation Syste
m
(
No
ra
syi
ki
n Fa
dila
h
)
1225
changes
of
pa
r
a
m
et
ers
wer
e
set
as
∆R
<
=
5,
∆G
<
=
5,
∆B
<
=
5,
∆H
<
=
15, ∆S
<
=
1.3,
∆V <
=
2.1.
All
para
m
et
ers
wer
e
a
ble
to
de
te
ct
the
sa
m
e
veh
ic
le
correct
ly
with
100%
a
ccur
acy
.
Howe
ver,
us
i
ng
eac
h
par
am
et
er
al
on
e
the
fail
ur
e
rate
of
detect
in
g
different
ve
hicle
can
be
as
high
as
60.
81%.
Th
us
,
t
he
c
ombinati
ons
of
m
ulti
ple
par
am
et
ers
with
AND
l
og
ic
wer
e
te
ste
d,
w
hich
both
dete
ct
ion
s
pro
du
ce
d
100%
detect
ion
rate
for
∆R
&
G
&
B <
=
5 & ∆
H
<
=
15 &
∆
S <
=
1.3 as
the
best
pa
ram
et
ers
com
b
inati
on
.
Table
1.
Detect
ion
Acc
ur
acy
with
Diff
e
re
nt
∆Par
am
et
er
Com
bin
at
ion
Detectio
n
Par
a
m
et
ers
Sa
m
e
Vehicle
Dif
f
erent Vehicle
Tr
u
e I
d
en
tif
y
False Iden
tif
y
Tr
u
e I
d
en
tif
y
False Iden
tif
y
∆R <
=
5
1
0
0
.00
%
0
.00
%
7
4
.32
%
2
5
.68
%
∆G
<
=
5
1
0
0
.00
%
0
.00
%
3
9
.19
%
6
0
.81
%
∆B <
=
5
1
0
0
.00
%
0
.00
%
7
6
.13
%
2
3
.87
%
∆H
<
=
15
1
0
0
.00
%
0
.00
%
6
3
.51
%
3
6
.49
%
∆S
<
=
1
.3
1
0
0
.00
%
0
.00
%
7
7
.48
%
2
2
.52
%
∆V
<
=
2
.1
1
0
0
.00
%
0
.00
%
4
9
.55
%
5
1
.80
%
∆H
<
=
1
5
&
∆V
<
=
2
.1
1
0
0
.00
%
0
.00
%
7
0
.27
%
2
9
.73
%
∆S
<
=
1
.3
&
∆V
<
=
2
.1
1
0
0
.00
%
0
.00
%
8
6
.94
%
1
3
.06
%
∆H
<
=
1
5
&
∆S
<
=
1
.3
1
0
0
.00
%
0
.00
%
9
1
.89
%
8
.11
%
∆H
<
=
1
5
&
∆S
<
=
1
.3 &
∆V
<
=
2
.1
1
0
0
.00
%
0
.00
%
9
1
.89
%
8
.11
%
∆R
&
G
&
B <
=
5
1
0
0
.00
%
0
.00
%
9
6
.85
%
3
.15
%
∆R
&
G
&
B <
=
5
&
∆H
<
=
15
1
0
0
.00
%
0
.00
%
9
9
.55
%
0
.45
%
∆R
&
G
&
B <
=
5
&
∆H
<
=
1
5
&
∆S
<
=
1
.3
1
0
0
.00
%
0
.00
%
1
0
0
.00
%
0
.00
%
3.3.
Viol
at
i
on
Ident
ific
at
i
on
In
validat
in
g
the
par
am
et
ers
as
m
entione
d
in
Sect
ion
3.
2,
a
n
ex
per
im
ent
to
detect
t
he
veh
ic
le
vio
la
ti
on
wa
s
c
onduct
ed.
T
he
resu
lt
s
s
howe
d
that
out
of
15
0
data
colle
ct
e
d,
t
her
e
we
re
t
hr
ee
cases
i
dent
ifie
d.
In
t
he
first
cas
e,
S1
(see
Fi
gu
re
6(a)
),
t
he
vi
olati
on
was
c
orrectl
y
identifi
ed.
T
he
sec
ond
case
,
S
2
(see
Figure
6(b)
in
dicat
es
that
the
syst
e
m
cou
l
d
not
ident
ify
there
was
a
veh
ic
le
vio
la
te
the
ru
le
or
m
isc
ounted
the
Count
wh
e
ne
ver
di
ff
e
ren
t
car
pas
sed
by.
In
the
fina
l
case,
S3
there
was
an
undete
ct
ed
veh
ic
le
or
m
isdete
ct
ed
e
m
pt
y
ROI
with
a
ve
hicle
.
This
c
ou
l
d
be
due
to
inc
on
sist
e
nt
li
gh
ti
ng
c
ondi
ti
on
s,
w
hic
h
cause
d
inacc
uracy
i
n
backg
rou
nd subtracti
on a
nd for
e
gro
und dete
ct
ion
.
(a)
(b)
(c)
Figure
6.(a)
Case S1,
(b)
Ca
se
S2 and
(c) C
as
e S3
A
total
of
143
d
at
a
we
re
co
rrec
tl
y
identifie
d
as
case
S1
;
w
her
eas 6
a
nd
1
d
at
a
we
re
ide
ntifie
d
as
cas
e
S2
a
nd
S
3
res
pe
ct
ively
.
This
r
esulte
d
with
95.
3%
detect
ion
rate
with
tr
ue
identific
at
ion.
In
te
rm
s
of
viol
at
ion
identific
at
ion,
75
vio
la
ti
on
oc
cur
a
nces
were
te
ste
d
and
t
he
syst
em
coul
d
ide
ntify
91
%
of
the
vio
l
at
ion
s
accuratel
y.
4.
CONCL
US
I
O
N
In
t
his
pa
per,
we
prese
nt
a
dev
el
op
m
ent
of
al
gorithm
for
do
ub
le
park
veh
ic
le
detect
ion
,
wh
ic
h
i
m
ple
m
ents
co
m
pu
te
r
visio
n
te
ch
niques.
We
first
ge
ne
rated
t
he
bac
kgr
ound
s
ubtra
ct
ion
t
o
gen
e
r
at
e
the
foregr
ou
nd
m
od
el
,
w
hich
det
ect
s
the
ve
hicle
on
R
OI
s
w
hi
ch
are
sel
ect
e
d
on
the
pro
hi
bited
pa
r
king
area.
The
n,
t
he
c
ol
or
sp
ac
es
in
f
orm
ation
of
R,
G,
B,
H,
S
and
V
are
ob
t
ai
ned
.
We
us
e
these
pa
ram
e
te
rs
to
determ
ine
wh
e
ther
the
ve
hicle
is
i
mm
ob
il
e
on
the
RO
I.
Fi
nally
,
the
veh
i
cl
e
wh
ic
h
is
im
m
ob
il
e
fo
r
m
or
e
tha
n
6
co
unts
is
ide
ntifie
d
as
vio
la
ti
ng
the
doubl
e
park
r
ules.
T
he
res
ult
sho
w
s
that
out
al
gorithm
achieves
91%
accuracy.
H
ow
ever,
it
is
ob
se
rv
e
d
t
hat
the
pe
rfor
m
ance
is
poor
wh
e
n
t
he
li
gh
ti
ng
is
to
o
br
i
gh
t
or
to
o
d
ark.
I
n
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
.
10
, N
o.
3
,
June
201
8
:
12
21
–
12
26
1226
the
f
uture,
we
will
inv
est
igat
e
i
m
age
en
hance
m
ent
te
chn
iq
ues,
i
ntell
igent
te
chn
i
qu
es
an
d
m
otion
detec
ti
on
t
o
address
su
c
h
is
su
e t
o
im
pr
ov
e
the
perform
ance.
ACKN
OWLE
DGE
MENTS
This
wor
k was
su
pp
or
te
d by
Un
i
ver
sit
i M
al
ay
sia
Pah
a
ng (R
DU
1703
227).
REFERE
NCE
S
[1]
J.
T. Lee
,
et
a
l
,
“
Rea
ltim
e
de
tection of
i
ll
eg
al
l
y
pa
rke
d
veh
ic
l
es
usi
ng
1D t
ran
sform
at
ion
,
”
2007
IE
E
E
Conf
ere
nce on
Adv
anc
ed Vi
deo
and
Signal B
ase
d
Surve
i
ll
anc
e
,
London,
2007
,
p
p.
254
-
259
.
[2]
S.
Sheik
Moham
m
ed
Ali,
B.
George
,
L
.
Van
aj
akshi
and
J.
Venka
tra
m
an
,
"A
Multi
ple
Ind
uct
iv
e
Loop
Ve
hic
l
e
Dete
c
ti
on
S
y
st
e
m
for
Hete
roge
neous
and
La
n
e
-
Le
ss
Tra
ff
ic,"
in
IEE
E
Tr
ansacti
ons
on
Instrum
ent
ati
on
an
d
Me
asur
eme
nt
,
v
ol.
61
,
no
.
5
,
pp
.
1353
-
1360,
Ma
y
2012.
[3]
J.
W
olff,
T.
H
e
uer
,
Haib
in
Gao
,
M.
W
ei
nm
ann
,
S.
Voi
t
and
U.
Hart
m
ann,
"P
a
rking
m
onit
or
sy
stem
base
d
on
m
agne
ti
c
fi
el
d
s
enso,
"
2006
IEEE
Inte
l
li
gen
t
Tr
anspor
tat
ion
Syste
ms
Confe
renc
e
,
Toront
o
,
Ont.
,
2006,
pp.
1275
-
1279.
[4]
R.
L.
Anderson,
"Ele
c
tromagneti
c
loop
vehicle
d
et
e
ct
ors,"
in
IE
E
E
Tr
ansacti
ons
on
Ve
hic
u
lar
Technol
ogy
,
vol.
1
9,
no.
1
,
pp
.
23
-
30
,
Feb
1970
.
[5]
J.
H.
Kell
and
I.
J.
Full
ert
on
,
Tra
ffi
c
de
tecto
r
handbook,
2n
d.
ed
.
,
W
ashin
gton,
D.C
.
:
U.
S.
Depa
rtment
of
Tra
nsporta
ti
on,
Feder
al Highwa
y
Adm
ini
str
at
ion
,
1990
.
[6]
R.
O'
Mal
ley
,
E.
Jones
and
M.
Gl
avi
n,
"Re
ar
-
La
m
p
Vehic
le
Det
ection
and
Tracki
n
g
in
Low
-
Exposure
Color
Video
for
Night
Condit
ions,"
in
IE
EE
T
rans
act
ions
on
Inte
lligent
Tr
anspor
tat
ion
Syst
ems
,
vol.
11
,
no
.
2,
pp.
453
-
462,
June
2010.
[7]
Y.
K.
Mac
,
H.
H
abe
,
et
a
l,
“
Deve
lopment
of
par
ki
ng
lot
surveill
an
ce
s
y
st
em
with
a
n
outdoor
ca
m
er
a
using
m
ac
hine
le
arn
ing”
,
The
In
stit
ute of
Elec
tronic
s,
In
formatio
n
and
Comm
uni
cat
ion
Engi
n
ee
rs
t
,
111(441)
,
pp
.
1
65
-
169,
2012
.
[8]
S.
Soo,
“
Object detecti
on
using
Haa
r
-
ca
s
ca
de
Cl
assifie
r,”
Univ
er
sit
y
of Tar
tu, pp. 1
–
12,
2014.
[9]
D.
Hall
et
al
.,
"Com
par
ison
of
ta
rge
t
detec
t
ion
al
gorit
hm
s
using
ada
pti
ve
bac
kground
m
odel
s,"
2005
IEE
E
Inte
rnational
W
orkshop
on
Vi
su
al
Surve
il
lan
ce
and
Pe
rform
anc
e
E
val
uat
ion
o
f
Tr
ack
ing
and
Surve
il
lan
ce
,
2005
,
pp.
113
-
120
.
[10]
V.
As
ari
,
W
ide
Area
Survei
llan
c
e,
Volum
e
6.
Ne
w York
Dordrecht L
ondon
:
53
Springer
He
idelbe
rg,
2013
.
[11]
M.
Bom
m
es,
A.
Faze
kas
,
T.
Vol
kenhof
f,
M.
Oes
er,
“
Video
bas
e
d
Inte
l
li
gen
t
Tr
a
nsportat
ion
S
y
s
t
ems
–
stat
e
of
t
he
art
and
fu
ture development”
,
Tr
anspor
tat
ion
R
ese
arch
Proce
d
ia
,
2
016,
pp
.
4
495
-
4
504
Vol.
14.
[12]
Z.
Z
ivkovi
c
,
"I
m
prove
d
ada
ptive
Gauss
ia
n
m
ixt
ure
m
odel
for
bac
kground
subtraction,
"
Pro
c
ee
dings
of
th
e
1
7th
Inte
rnational
Co
nfe
renc
e
on
Patt
ern
Recogni
t
ion, 2004. ICP
R
200
4
,
2004
,
pp
.
28
-
31
Vol.
2
.
[13]
Thi
err
y
Bouwm
ans,
Fida
El
B
af,
Ber
tra
nd
V
ac
hon.
B
ac
kgro
und
Modeli
ng
using
Mixture
of
Gauss
ia
ns
for
Foreground
Det
ec
t
ion
-
A
Survey
.
R
ecent
Pat
ents
on
Com
pute
r
Scie
nc
e,
B
ent
ha
m
Sci
ence
Publi
shers
,
2008,
1
(3),
pp.
219
-
237.
[14]
Z.
Z
ivkovi
c
,
"I
m
prove
d
ada
ptive
Gauss
ia
n
m
ixt
ure
m
odel
for
bac
kground
subtraction,
"
Pro
c
ee
dings
of
th
e
1
7th
Inte
rnational
Co
nfe
renc
e
on
Patt
ern
Recogni
t
ion, 2004. ICP
R
200
4
.
,
2004,
pp.
28
-
31
Vol.
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.