TELKOM
NIKA
, Vol.14, No
.1, March 2
0
1
6
, pp. 56~6
3
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i1.2798
56
Re
cei
v
ed Se
ptem
ber 30, 2015; Revi
se
d Jan
uary 3, 2016; Accept
ed Ja
nua
ry 1
8
, 2016
Adaptive Traffic Controller Based On Pre-Timed Syst
em
Fredd
y
Kurnia
w
a
n*
1
, Har
uno Sajati
2
, Okto
Dinary
anto
3
1
Departme
n
t of Electrical En
gi
neer
ing,
Sek
o
l
ah T
i
nggi T
e
knolo
g
i Adis
utji
pto
2
Departme
n
t of Informatic Eng
i
ne
erin
g, Sekol
ah T
i
nggi T
e
knolo
g
i Adis
utji
pto
3
Departme
n
t of Mechan
ical E
ngi
neer
in
g, Sekola
h T
i
nggi T
e
kno
l
og
i Adis
utjipto
Jln. Janti, Blok
R, Kompleks L
anu
d Adis
utjipt
o
, Yog
y
akart
a
,
Ph. +
62-274-
4
512
62, F
a
x. +
6
2-27
4-45
12
65
*Corres
p
o
ndi
n
g
author, e-mi
a
l
: fredd
ykurn
i
a
w
a
n
@stta.ac.i
d
A
b
st
r
a
ct
Adaptiv
e traffic controll
er systems b
a
sed
on i
m
a
ge proc
essin
g
have
b
een d
e
vel
o
p
e
d
w
i
dely.
N
e
ve
rthe
l
e
ss, in
a
d
e
v
e
l
op
in
g
co
un
tr
y, the s
ystems
often c
oul
d n
o
t be
e
a
s
ily a
ppl
ie
d b
e
c
ause
al
l types
of
vehicle use the sam
e
road. Theref
ore, to
overcome the
problem
,
the
new conc
ept of the systems is
proposed. The system
s were develope
d from
a pre-timed traffic controller system
that
based on AV
R
m
i
cr
ocontr
o
ller
. By default, the system
s
use the signal-
t
im
i
ng plans t
o
control t
he
vehicle flow. To
acco
mmo
date
the traffic va
riatio
ns, a n
e
w
metho
d
of vehicl
e detecti
on
h
a
s bee
n built.
T
h
e
met
h
o
d
calcul
ated
a
n
intens
ity histo
g
ra
m sta
ndar
d
dev
iati
o
n
of the
i
m
age
re
p
r
esenti
ng a d
e
tection
ar
ea to
deter
mi
ne traffic dens
ity of each inters
ectio
n
lan
e
. The systems
mo
difi
ed the gr
een-t
i
me of each
l
a
n
e
base
d
on th
e traffic dens
ity. The
meth
od co
u
l
d det
ect all
typ
e
s of vehic
l
es
and w
o
rk pro
p
e
rly in
a day a
n
d
a
night tim
e
.
Ke
y
w
ords
:
deter
mi
ne traff
i
c d
ensity, sta
ndar
d d
e
vi
atio
n, inte
nsity
his
t
ogra
m
, si
gn
al-
t
imi
n
g
pl
an, A
V
R
micr
ocontr
o
ll
er
.
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
1.1.
Bac
k
grou
nd of the Proble
m
A traffic jam i
s
a
daily ph
e
nomen
on i
n
every big
city in the
wo
rld.
It commo
nly
happ
en
s
at an interse
c
tion controll
ed by a conv
entional tra
ffi
c co
ntrolle
r. One of
the weakne
sses of
th
e
conve
n
tional
traffic co
ntroll
er is
una
ble to accom
m
od
ate the traffic volume varie
t
ies be
cau
s
e
it
use
s
a fixed-timed system.
It means tha
t
the
systems have a same
green time a
ll day. This can
cau
s
e
a heav
y traffic jam i
n
pea
k ho
urs
becau
se of
th
e long vehi
cle
s
qu
eue a
nd t
here i
s
wa
stin
g
time in off p
eak ho
urs b
e
c
au
se
only a
few ve
hicle
s
pa
ss the i
n
tersectio
n
. Th
at phe
nome
n
o
n
become
s
the major fa
ctor o
f
causi
ng a
traffic
jam in s
o
me inters
ec
tions
.
One m
e
thod
to overcome
this p
r
oble
m
i
s
to
u
s
e
ada
ptive traffic controlle
r
syst
ems. Th
e
system
s cont
rol the vehi
cl
es flow
ba
se
d on the real
time traffic de
nsity. In developed
cou
n
tri
e
s,
the system
s have been d
e
velope
d wid
e
ly. Usually,
the system
s use
severa
l sensors, su
ch
as
indu
ctive loo
p
dete
c
tor, i
n
frared, o
r
camera
to det
ect the vehi
cle and th
en
cal
c
ulate t
r
af
fic
den
sity. By d
o
ing so, it can redu
ce the
air pollu
tant
by minimizin
g
the amount
of accel
e
rati
on
and braki
ng o
f
the vehicle [1].
One of the vehicl
e dete
c
tion metho
d
s t
hat
has
been
develope
d rapidly is u
s
in
g a rea
l
time video a
nd imag
e proce
s
sing. Th
e dete
c
tion u
s
e
s
on
e or
several
cam
e
ras to d
e
tect
the
vehicle
s
such as
ca
r, tru
ck, b
u
s, o
r
tracto
r on
every interse
c
tion lane. Th
e
system i
s
n
o
w
comm
only used in the dev
elope
d co
untries that the
ro
ad is inten
d
e
d
for motori
ze
d vehicle
s
onl
y.
Unfortu
nately
,
most road
s in developin
g
c
ount
ry like in Indone
si
a are intend
e
d
for all
kind
s of moto
rize
d and n
o
n
-moto
r
ized vehicl
es.
The
vehicle dete
c
tion applied i
n
the develop
ed
cou
n
trie
s
can
not d
e
tect t
he n
on-motorized
vehicl
es su
ch
a
s
bi
cycles,
padi
ca
bs
or
ri
cksha
w
s.
The ad
aptive traffic co
ntroll
er ba
se
d on t
he vehicl
e de
tection like in
the develop
e
d
cou
n
try co
u
l
d
not be
prope
rly applied
in t
he d
e
velopin
g
country.
Up
to no
w, m
o
st intersectio
n
s
in
the
co
un
try
use the fixed
-
timed traffic
controlle
r systems, t
hus th
e traffic jam
become
s
on
e
of the seri
o
u
s
probl
em
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Adaptive T
r
af
fic Cont
rolle
r Based
On Pr
e-Tim
ed System
(Fredd
y K
u
rnia
wa
n)
57
1.2
Related Res
earch Studie
s
Many re
sea
r
ch stu
d
ie
s ha
ve been
con
ducte
d
to improve the a
d
aptive traffic controlle
r
based on an
image pro
c
e
ssi
ng. The common ima
g
e
pro
c
e
ssi
ng
method to detect vehicl
e
s
is
usin
g ed
ge
detectio
n
. Ho
ngjin Z
hu
prese
n
ted
a m
o
ving vehi
cle
dete
c
tion th
at com
p
ri
se
d
a
hori
z
ontal e
d
ge detectio
n
method an
d auto co
rrel
a
ti
on. It is possible to detect
each individ
ual
vehicle eve
n
if the vehicle
s
are overl
a
p
p
ing [2]. Fazli
obtained ve
hicle
s
cl
assifi
cation b
a
sed
on
neural networks for a
n
ada
ptive
traffic controlle
r syst
em [3]. Khan and Askerza
de impleme
n
ted
an image p
r
o
c
e
ssi
ng an
d fuzzy logi
c control, and th
en se
nt the result to a mi
cro
c
o
n
troll
e
r
to
drive th
e traf
fic si
gnal
[4
]
-[5
]
. After that, Sutjiadi [
6
]
tried to
e
x
tract the
ba
ckgro
und
u
s
i
ng
Gau
ssi
an Mixture Mod
e
ls
Algorithm in o
r
de
r to detect
the vehicle
s
.
The
advanta
ge m
e
thod
of
vehicl
e d
e
te
ction
i
s
usi
n
g
Ca
scad
e Ha
ar. Chirag an
d
Ri
pal
impleme
n
ted
the method b
y
training the
classifie
r
for
15 stag
es, 1
7
kind
s of Ha
a
r
feature
s
, an
d
a size imag
e
of 35 × 20
pi
xels. The det
ection
re
su
lts can b
e
improved by traini
ng the cl
assif
i
er
on the l
a
rg
est set of
samp
les [7]. Th
e traffic d
e
n
s
ity can
be
cal
c
ul
ated by
com
parin
g the
re
a
l
time frame of
live video by refere
nce im
age an
d se
archin
g vehicl
e
s
-o
nly on the
road a
r
ea [8
].
The vehicl
e d
e
tection a
nd trackin
g
ca
n b
e
done by extractin
g
the video fram
e se
quen
ce [9].
Some
research studie
s
h
a
ve
bee
n co
ndu
cted
to
meet the traffic
c
h
arac
teris
t
ics
in
Indone
sia. Jatmiko pre
s
e
n
ted
the architecture
of
decentrali
ze
d
self-org
ani
zi
ng traffic
co
ntrol
system i
n
re
al situatio
n
even
on
non
-structu
re i
n
tersectio
n
like
in Ja
ka
rta [
10]. Kurnia
wan
develop
ed a
pre
-
timed a
n
d
co
ordi
nated
traffic co
nt
rol
l
er sy
st
em
s b
a
se
d
on AVR Microcontroll
er.
The sy
stems
manag
e the vehicl
e flow
ba
sed o
n
sig
nal
-timing pla
n
s
[11].
Almost none
of the vehicle dete
c
tion
me
thods b
a
sed on im
ag
e pro
c
e
ssi
ng
can be
prop
erly ap
pli
ed in the dev
elopin
g
co
unt
ry. They
can
not detect all
types of vehi
cle
s
. Accordi
ng
to Haa
r
Ca
scade meth
od, the variou
s types of
vehi
cles can be
cl
assi
fied, but they need h
u
ge
sampl
e
cla
s
sifiers and
in
crea
se co
mpu
t
ational
lo
ad signifi
cantly. An
ada
ptive traffic controll
er
system
ba
se
d imag
e p
r
ocessing li
ke
ap
plied in
devel
oped
co
untrie
s
is not
reliab
l
e appli
ed in
th
e
developin
g
country. To overcome the p
r
oble
m
, a
ne
w co
ncept of the adaptive
traffic cont
ro
ller
system i
s
pro
posed. The
system is b
a
sed on p
r
e
-
ti
m
ed sy
stem in
the previo
us
rese
arch [11].
To
accomm
odat
e the t
r
affic v
a
riation
s
, the
syste
m
s det
ermin
e
the
traffic de
nsity f
r
om
an i
n
ten
s
ity
histog
ram
sta
ndard d
e
viation of a
n
ima
ge repr
esenti
ng the l
ane
a
nd mo
di
fy the green
-time
of
each interse
c
tion lane.
2. Resear
ch
Method
2.1. The Dev
e
lopment o
f
the Pre
-time
d Sy
stems
The
system i
s
a
develo
p
m
ent of the
pre
-
tim
ed traffic
controlle
r sy
stem. As me
ntioned i
n
the [11], the system u
s
e
s
ATmeg
a12
8A microc
ont
rolle
r as th
e
heart of
cont
rolling traffic
as
s
h
ow
n
in
F
i
gu
r
e
1
.
By def
ault, the syst
ems
cont
rol t
he v
ehi
cle flo
w
ba
se
d on t
he si
gnal
-timing
plan
s. In the
plan, a
day i
s
se
gme
n
ted i
n
to ten
tim
e
slots. The
r
e are
thre
e
pl
a
n
s availabl
e
t
hat
can b
e
allo
cated to seve
ral types of
day:
wee
k
da
ys, Saturday
s, and Sun
d
a
ys. The pla
n
s
compri
sing green, yellow,
and red-clearance time of all lanes
are stored in the
EEPROM of the
microcontroll
er. The op
era
t
or can m
odif
y
the pl
an from the Traffic
Manag
eme
n
t Center p
r
o
g
ram
to meet the chara
c
te
risti
c
s of the tr
affic
by reffering the s
t
atis
tic
result
.
Figure 1. The
system blo
ck diagra
m
The sy
stems
determi
ne th
e traffic den
si
ty by
detecting all obje
c
ts i
n
a vehicle d
e
tection
area
of eve
r
y lane
with f
our
wirele
ss
IP came
ra
s.
After that, the sy
stems m
odify the curren
t
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 56 – 63
58
gree
n-time
of a lane. As th
e previo
us
re
sea
r
ch
, the case
study is
on Gon
dom
a
nan Interse
c
ti
on,
one
of the b
u
s
y interse
c
tio
n
s i
n
Yogya
k
arta
City
.
Figure 2 sho
w
s a
map
of
the vehicle dete
c
tion
area
and
ca
mera p
o
sitio
n
of each la
ne
. The dete
c
tion area is l
o
cated a
bout
40 meters fro
m
a
traffic light.
With a vehi
cl
e sp
eed
at a
gree
n-ti
me
about 3
0
km/
h
, the vehicl
es move f
r
o
m
the
area to the traffic lights in
about 5 seco
nds.
Figure 2. Map of the case study
Each
cam
e
ra
captu
r
e
s
the
entire ve
hicl
e waitin
g a
r
e
a
, but the p
r
o
g
ram
only de
tects th
e
vehicle i
n
the
vehicle
dete
c
tion a
r
ea.
O
n
lane
2 an
d
4,
the area i
s
slightly shifted to the
cente
r
of
the lane be
cause there is a turn-left priority.
The ca
mera
send
s
a real-tim
e video su
rveilla
nce
with its re
sol
u
tion of 3
20
×
240 to
a
com
puter via
a
wi
-fi ro
uter
with
a fram
e rate
of 7 fra
m
e
s
p
e
r
se
con
d
. The detectio
n
is p
e
rform
ed at the last fram
e.
2.2. Dete
rmine the Tr
affic Densi
t
y
us
ing Histogr
a
m
The syste
m
s
detect all obj
ects in the ve
hicle
dete
c
tio
n
area a
s
sh
own in Figu
re 2. The
area
is divide
d into th
ree
region
s of i
n
tere
st (ROI
s).
The
dete
c
tio
n
is
only p
e
rf
orme
d in
RO
Is.
For saving th
e comp
utatio
n, the color i
m
age
is tran
sformed into a
gray imag
e.
It can be co
nclu
ded fro
m
the previou
s
resear
ch th
at the existe
nce of vehi
cl
e woul
d
make
on
obvi
ous
differe
nce between
th
e
vehicl
e a
n
d
the ba
ckgrou
nd colo
r [12
]
[
13]. This would
make th
e sta
ndard deviati
on of the inte
nsity histog
ra
m highe
r. A lane with
no v
ehicl
e will result a
histog
ram
wit
h
its lo
w sta
ndard
deviati
on. Otherwi
se, a lane
co
ntaining a l
o
t of vehicle
s
will
result a histo
g
ram
with its high stand
ard deviati
on. The dete
r
min
i
ng traffic de
nsity is done
by
cal
c
ulatin
g a
n
inten
s
ity hi
stogram of
al
l ROI
s
. After that, the sy
stem
s calcul
ate the
stand
ard
deviation of the histo
g
ra
m. If the value is less
than a
low thre
sh
old, the system
s co
ncl
ude th
at
the traffic
dens
ity is
low.
In this
case, the
system
s sho
r
ten
the gree
n-time. Otherwise,
if
the
value is hi
gh
er than
a hig
h
thre
shol
d, the sy
stem
s
concl
ude that
the traffic de
nsity is hig
h
, and
then the syst
ems extend t
he gre
e
n
-
time. Determi
n
in
g traffic den
si
ty is only performe
d
for a lane.
Figure 3. Det
e
ction time a
nd co
untdo
wn numbe
r
2.3. Modif
y
ing the Co
untdo
w
n
Num
b
er
The traffic de
tection is onl
y performe
d
for a lane th
at gets a gre
en-time a
s
shown in
Figure 5. When lane
gets the green-ti
me, the systems can
mo
dify the green-ti
me for the lane.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Adaptive T
r
af
fic Cont
rolle
r Based
On Pr
e-Tim
ed System
(Fredd
y K
u
rnia
wa
n)
59
This i
s
d
one
by modifying
the co
untdo
wn
n
u
mbe
r
o
f
gree
n-time
of the lane
(
). If the traffic
den
sity is low, the green
-time can b
e
ended shor
tl
y. This can b
e
done when
the countdo
wn
numbe
r is g
r
eater than fi
ve. Otherwi
se, if the tr
affic den
sity is high, the gre
en-time
can
be
extended. Thi
s
can b
e
do
n
e
wh
en the
countdo
wn
nu
mber i
s
b
e
tween five an
d
eight. Wh
en t
h
e
system
s mo
d
i
fy the count
down num
be
r of a lan
e
that gets
a g
r
een
-time, th
e syste
m
s
also
modify the countdo
wn nu
mber of t
he other lan
e
s t
hat get red-ti
me (
). At
the time,
and
whe
r
e
, are not displaye
d.
Whe
n
the co
untdo
wn num
ber of gre
en-time at lane
is
greater then five,
the s
y
s
t
ems
execute al
gorithm 1 for mo
difying
and
.
Algorithm 1: Modifying the
green
-time o
f
lane
and re
d-time of lane
(
)
1.
2.
.
3.
∑
4. If
<
_
then
a.
∆
=
- 5
b.
= 5
c.
=
-
5, for
d. return
5. If
(
< 8 and
_
<
_
and
>
_
) then
a.
++
b.
_
++
c.
++
, for
d. Return
The first ste
p
of the algorit
hm is to
cal
c
ulate the inte
nsity histog
ra
m.
denotes t
h
e
intensity histogra
m
of the
pixels in all
ROIs at the lane
,
denotes the gray level (0, 1, 2, ...
255), an
d
denote
s
the numbe
r of gra
y
level
. Afte
r that, in line 2, systems
cal
c
ulate the
mean of the histog
ram (
). In the equation,
denotes
the numbe
r o
f
pixels in all
ROIs. An
d
then line 3 ca
lculate
s
the st
anda
rd deviat
i
on of the hist
ogra
m
.
In line 4, if th
e value is less than the lo
w thre
shol
d (
_
), the system
s co
ncl
ude th
at
the traffic de
nsity is lo
w.
T
herefore,
system
s shorte
n the gr
een
time by doin
g
the follo
wi
ng
step
s: (a
) cal
c
ulate
a diffe
ren
c
e b
e
twe
e
n
cu
rrent cou
n
tdown num
b
e
r an
d five (
∆
), (b)
set the
c
o
un
td
ow
n
nu
mb
er
(
) to five, and (c)
subtra
ct the countdo
wn
nu
mber of re
d-ti
me of the other
lane
s (
r
j
, for
) by
∆
.
Otherwise, in
line
5, the
sy
st
em
s extend
the g
r
ee
n-ti
me. Thi
s
ca
se can
be
don
e if some
para
m
eters
meet the req
u
irem
ent: (a)
is less than
eight, (b) the
addition of
that has b
e
e
n
done (
_
) is less than the maximum nu
mber of gree
n-time additi
on (
_
), and (c) the
stand
ard
devi
a
tion is
great
er tha
n
the hi
gh threshold
(
_
). The
system
will extend th
e green
-
time by incre
a
sin
g
the three paramet
ers
:
(
a
)
c
u
rr
en
t c
o
un
td
ow
n
nu
mb
er
(
), (b)
the gree
n-tim
e
addition valu
e (
_
), and (c) the cu
rre
nt co
untdo
wn num
ber
of red
-
tim
e
of the other lanes (
,
for
). In this ca
se, all co
u
n
tdown num
bers
do not
appe
ar to ch
ange b
e
cau
s
e they also
decrea
s
e eve
r
y se
cond.
3. Result an
d Discus
s
io
n
3.1. The Sta
ndard Dev
i
ation of the
Histogr
am
Several expe
riment
s have
been cond
u
c
ted to dete
r
mine the lo
w and high th
resh
old.
Figure 4(a
)
and (b
) sh
o
w
the imag
e rep
r
e
s
enti
ng vehicle
arrival
s
at the ea
st lane of
Gond
oman
an
interse
c
tion
at a day time. When the
r
e is no vehicl
e at all three ROIs as
sho
w
n
in
Figure 4(a
)
, the histo
g
ra
m of the image become
s
nea
rly the same
as sho
w
n in Figure 4(c). T
h
e
intensity of
a
l
most
pixels
are
in
a
ran
g
e
bet
wee
n
1
20 a
nd
160.
The m
ean
a
nd the
stand
ard
deviation of the inten
s
ity histogram are 145 an
d 17.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 56 – 63
60
(a) th
ree ROI
s
with no vehi
cle
(b) th
ree ROI
s
with a lot of vehicle
s
(c) the histo
g
ram of the three ROI
s
Figure 4. The
three ROIs o
f
the detectio
n
area o
n
the
lane 2
Figure 4
(
b
)
shows th
e a
r
e
a
containin
g
a lot of
vehi
cl
es. Th
e inte
n
s
ity of the pix
e
ls
are
in
a ra
nge
bet
ween
0 a
nd
25
5 a
s
sho
w
n
i
n
Figu
re
4(c). The
mea
n
a
nd the
sta
n
d
a
rd
deviation
of
the inten
s
ity histog
ram
are
127 a
nd 51.
The me
an
of
the histo
g
ra
m
may cha
nge
with a vari
ation
of the sun li
ght intensity
over the ho
ri
zon
a
nd
wea
t
her, but the
standa
rd d
e
v
iation doe
s
not
signifi
cantly chang
e. The i
n
tensity histo
g
ram va
rie
s
with the varia
t
ion of the vehicle. Th
e hig
h
e
r
the traffic de
nsity of the detection a
r
ea
and the
high
er the perce
n
t
age the roa
d
are occu
pied
b
y
the vehicle a
nd the high
er
the int
ensity histog
ram
sta
ndard deviati
on.
3.2. The Mod
i
fied Green
-time
Some comm
on ca
se
s ca
n occur in th
e lane. In a
norm
a
l traffic,
decrea
s
e
s
every
se
con
d
as sh
own in case 1 of Figure 5. At this
case, the default gre
en-time is 1
5
se
con
d
s. In the
pea
k hou
r, generally the traffic volume
is highe
r
than
usual. Ope
r
a
t
or sho
u
ld m
odify the green-
time default t
o
a
c
commo
d
a
te the t
r
affic va
riation.
Ho
wever, i
n
some
cases,
so
metime
s,
the
traffic volume
can be hig
h
e
r than p
r
edi
cted. Ca
se 2
in Figure 5
sho
w
s the co
untdo
wn num
ber
whe
n
the traffic density is high. At
this ca
se,
doe
s not chan
ge
at
t
= 8 until 15 seco
nd
s.
Mean
while,
case
3 in Fig
u
r
e 5
sho
w
s a
con
d
ition wh
en the g
r
ee
n
-
time is
given
to a lane, the
t
r
af
f
i
c den
sit
y
is low.
A
t
t
h
is ca
se,
sy
st
e
m
s set
= 5.
A
f
t
e
r t
hat
,
syst
em
s cou
n
t
d
own
until
‘zero’ and g
r
e
en-time run
s
out.
0
500
1
000
1
500
2
000
2
500
0
5
0
100
150
2
0
0
2
5
0
N
u
m
b
e
r
of
pi
xe
l
s
G
r
ay
sc
al
e of
i
n
t
ensi
t
y
spac
e
N
o
v
ehi
cl
e
A
l
o
t
of
ve
hi
cl
es
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Adaptive T
r
af
fic Cont
rolle
r Based
On Pr
e-Tim
ed System
(Fredd
y K
u
rnia
wa
n)
61
Figure 5. The
countin
g-d
o
w
n process
The G
ond
om
anan
interse
c
tion is control
l
ed by
fixed
-
time traffic co
ntrolle
r
syste
m
s that
its default green-tim
e
(
) for lane 1, 2, 3
,
and 4 are resp
ec
tively 30, 36, 43, an
d 30 second
s.
Those are the gree
n-time
need
ed at the
peak h
o
u
r
. Mean
while, the
yellow-time
(
) and the red
-
clea
ran
c
e (
) of all lanes
are 3 an
d 5
second
s. The pre
s
et for green, yello
w, and re
d-
clea
ren
c
e tim
e
at day time are sho
w
n in
Table 1.
Table 1. Gree
n and re
d time of all lanes
at the night time
i
(sec)
(sec)
(sec)
_
_
(sec)
′
(sec)
′
(sec)
1 30
3
5
30
80
136
28
101
2 36
3
5
25
75
130
18
95
3 43
3
5
20
70
123
23
88
4 30
3
5
25
70
136
22
101
Whe
n
the g
r
e
en-time i
s
giv
en to a la
ne
and the
co
unt
down num
ber
decrea
s
e
s
, vehicl
es
start to mov
e
and leave
the queue.
In this ca
se,
the traffic density of the lane gen
era
lly
decrea
s
e
s
, therefo
r
e, the standar
d devia
tion decrea
s
e
s
. Figure 6 s
hows the inte
nsity histog
ra
m
stand
ard
dev
iation
whe
n
a g
r
een
-time
is
given to
each la
ne
at the d
a
y an
d
night tim
e
.
The
hori
z
ontal
ax
is of
the fig
u
re
is the
countdo
wn
nu
mber at th
e
pre-time
d
systems an
d
the
prop
osed sy
stems at the d
a
y and night t
i
me. The
cou
n
tdown numb
e
rs
at the pre
-
timed sy
ste
m
s
are al
way
s
shown. Mean
while, at the
prop
osed
s
ystems, they are sh
own whe
n
the value
o
f
the
green-time from five to z
e
ro.
The hi
gh thresh
old of the
stand
ard de
viation for e
a
c
h la
ne
(
_
)
is defined high
er
than the maxi
mum value
of the stan
dard
deviation
tha
t
sometime
s
occurre
d
, be
cause the p
r
e
s
et
gree
n-time
h
a
s b
een
set to the green
-time nee
ded
i
n
the pea
k h
ours. The
r
efo
r
e, the sy
ste
m
s
hardly eve
r
extend the g
r
een
-time. M
ean
while, the
low thre
sh
ol
d (
_
) is defined slightly
highe
r than t
he stan
da
rd
deviation ave
r
age
wh
en th
ere i
s
no vehi
cle in the
RO
Is. The high
and
low thre
sh
old
of the standa
rd deviation f
o
r
all lane
s can also be sh
own in Ta
ble
1.
(a) L
ane 1: n
o
rth
(b) L
ane 2: e
a
st
(c) Lan
e 3: so
uth
(d) L
ane 4: west
Figure 6. The
intensity hist
ogra
m
stan
da
rd dev
iation a
t
a green
-time at day and night time
1
H
mi
n
_
1
H
max
_
1
H
2
H
max
_
2
H
mi
n
_
2
H
3
H
mi
n
_
3
H
max
_
3
H
4
H
ma
x
_
4
H
mi
n
_
4
H
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 56 – 63
62
Figure 6
sh
o
w
s fluctu
atio
n of the i
n
ten
s
ity hist
o
g
ra
m stan
da
rd d
e
viation of all
pixels i
n
the all
ROIs
at the imag
e
rep
r
e
s
entin
g
the vehi
cle
arrival
of ea
ch lane
wh
en
a green
-time
is
given to the lane. In gene
ral, in day time, the gr
ee
n-t
i
me allows all
vehicle
s
que
ue to enter a
n
d
pass th
rou
g
h
the interse
c
t
i
on. Neve
rthe
le
ss at the
e
nd of the
gre
en-time
(
=
0), there a
r
e
some
arrival vehicle
s
in la
ne 1, 3, and
4. In th
is ca
se, the standa
rd deviation i
s
still high
er t
han
the lo
w threshold
as sh
o
w
n i
n
Fig
u
re
6(a
)
, 6
(
c), a
nd 6
(
d
)
, so t
hat the
co
un
ting do
wn
go
es
norm
a
lly for lane 1, 3, and
4.
Mean
while,
a
t
lane
2 i
n
Fi
gure
6
(
b
)
, th
e g
r
ee
n-time
also all
o
ws
all vehi
cle
s
q
ueue
to
enter an
d pa
ss throug
h the intersectio
n
at t
he day
time. When th
e countd
o
wn numbe
r of green-
time is t
en, th
ere i
s
no ve
hi
cle i
n
all
ROIs. Sy
stem
s shorten
the green-tim
e
by
setting the
cu
rrent
cou
n
tdo
w
n n
u
mbe
r
to five
and t
hen th
e
gre
e
n
-
time i
s
e
nde
d in fi
ve se
co
nd
s. At this case,
the
system
s
sav
e
the g
r
e
en-ti
me for five seco
nd
s. Figu
re 7
(
a
)
sho
w
s the
video
surveillan
c
e
when
the cou
n
tdo
w
n numbe
r at lane 2 go
es from five to zero.
Whe
n
a night
time, at lane
1, all vehicles
queue ente
r
and pa
ss thro
ugh the intersection
whe
n
the
co
untdo
wn n
u
m
ber
of the
green-tim
e
is
seven. As
ca
n
be
sho
w
n
in
Figu
re 6
(
a
)
,
the
system
s
sho
r
ten the g
r
ee
n-time a
nd
save the g
r
ee
n-t
i
me f
o
r t
w
o se
co
nd
s.
This
ca
se
al
so
happ
en
s in the other la
ne
s. It can be sh
own in Fi
gur
e
6(b), (c), a
n
d
(d) that the
saving time for
lane
2, 3, a
n
d
4
at the
nig
h
t time a
r
e
re
spe
c
ti
vely
18,
20,
a
nd 8 se
con
d
s. The shorten
ed gre
en-
time of all lanes at the nigh
t time (
′
) ca
n also be
see
n
in Table 1. Figure 7
(
b
)
sho
w
s the video
surveill
an
ce a
t
the night time wh
en the g
r
een-tim
e
is gi
ven to lane 2.
At the case, the countd
o
wn
numbe
r of green an
d red
-
t
i
me is not sh
own b
e
cau
s
e
the numbe
r is gre
a
ter tha
n
five.
(a) at a day time
(b) at a nig
h
t time
Figure 7. Rea
l
Time Monito
r wind
ow
sho
w
s the vide
o surveill
an
ce o
f
all lanes
Whe
n
the g
r
een-tim
e
of a lane i
s
de
crea
sed, the
red-time of th
e other l
ane
s will be
decre
sed to
o
.
The re
d-tim
e
of the pre-timed syste
m
(
) a
nd the
red-time
of the pro
p
o
s
ed
sy
st
em (
′
) in
a night time can be see
n
in Table 1. In
this ca
se, the red
-
time of all lanes ca
n be
redu
ce
d a
bou
t 20%. It ca
n
also
be
con
c
l
uded
that the
system
s
not
only save the
gre
e
n
-
time,
but
also
shorte
n
the
red
-
time
of all l
ane
s.
In ot
he
r
wo
rds,
the
syst
ems can
red
u
ce
the ve
hi
cle
waiting
-
time
at all lanes.
4. Conclusi
on
An ad
aptive t
r
affic
co
ntroll
er
ba
sed
on
p
r
e-time
d syst
ems
can be
i
m
pleme
n
ted based on
an ATme
ga1
28A micro
c
o
n
trolle
r. The
system
use
s
the
sig
nal-timing pla
n
s as
a ba
si
s for
controlling
th
e vehicl
e flo
w
. To
accom
m
odate th
e
i
n
stanta
neo
us traffic vari
ation, the
syste
m
s
modify the
current g
r
ee
n-time of a
lan
e
a
c
cordi
ng t
o
the t
r
affic
d
ensity of th
e
lane. Th
e traffic
den
sity can be determi
ne
d by calcul
ating the
inten
s
ity histogra
m
standa
rd
deviation of the
image
rep
r
e
s
enting the ve
hicle d
e
tectio
n are
a
of
the
lane. The m
e
thod
s work
prop
erly in d
a
y
and
night tim
e
an
d n
o
t o
n
ly savin
g
th
e waste
-
time
at the
gree
n-time, b
u
t a
l
so
red
u
ci
ng
the
waiting
-
time
at the red-tim
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Adaptive T
r
af
fic Cont
rolle
r Based
On Pr
e-Tim
ed System
(Fredd
y K
u
rnia
wa
n)
63
Ackn
o
w
l
e
dg
ements
The research
is fully su
pp
orted by T
h
e
Mi
nistry of
Re
sea
r
ch, T
e
ch
nolo
g
y an
d Hig
her
Educatio
n of the Rep
ubli
c
of Indone
sia
unde
r grant “
Hiba
h Bersai
ng Di
kti
”. The
authors woul
d
lik
e to thank
to “
Kopertis Wilayah
V
” and
“
STT Adisutjipto
” for thei
r sup
port.
Referen
ces
[1]
Dobr
e C. Usin
g Intelli
gent T
r
affic signals t
o
Red
u
ce Ve
hicle Emiss
i
on
s.
Internationa
l
Journal of
Innovativ
e Co
mp
utin
g, In
formati
on a
nd C
o
ntrol
. 201
2; 8(9
)
: 6283-6
3
0
2
.
[2]
Z
hu H, F
an H,
Guo S. Movin
g
Vehicl
e D
e
tecti
on
and T
r
acki
ng i
n
T
r
affic Images
base
d
o
n
Horizo
ntal
Edges.
T
E
LKOMNIKA Indone
sian Jo
urna
l of Electrical E
ngi
neer
ing
. 2
013;
11(1
1
): 647
7-6
483.
[3]
F
a
zli S, Moh
a
mmadia
S, Ra
hman
i M. Neu
r
al N
e
t
w
ork b
a
sed V
e
h
i
cle
Classific
a
tio
n
for Ada
p
ti
v
e
T
r
affic Control.
Internation
a
l J
ourn
a
l of Softw
are Eng
i
n
eeri
n
g & Applic
atio
n
s
. 2012; 3(3): 1
7
-22.
[4]
Khan
BA, L
a
i
NS. An
Adv
ance
d
F
u
zz
y
Log
ic b
a
se
d
T
r
affic Contro
ll
er.
Internati
o
n
a
l J
ourn
a
l
o
f
Innovati
on i
n
the Digit
al Econ
o
m
y
.
20
14; 5(1):
31-40.
[5]
Askerzad
e IN,
Mahmo
od M.
Contro
l the
E
x
t
ensi
on T
i
me
of T
r
affic Light i
n
Sin
g
le
Ju
nctio
n
b
y
Usi
n
g
Fuzz
y
Logic.
Internati
o
n
a
l Jo
urna
l of Electri
c
al & Co
mput
e
r
Sciences IJECS – IJENS
. 2010; 10(
2):
48-5
5
[6]
Sutjia
di R, Se
t
y
a
d
i E, Lim R. Adaptive B
a
ckgro
und E
x
t
r
action for Vid
eo Base
d T
r
a
ffic Counte
r
Appl
icatio
n U
s
ing Ga
ussia
n
Mixtur
e Mod
e
ls Alg
o
rithm.
T
E
LKOMNIKA Indon
esia
n
Journ
a
l of
Electrical E
ngi
neer
ing
. 2
015;
13(3): 10
06-
10
13
[7]
Chira
g
IP, Rip
al P. Co
unti
n
g
Cars i
n
T
r
affic Usin
g Casc
ad
e Ha
ar
w
i
th
K
L
P.
Internati
o
n
a
l Jo
urna
l o
f
Co
mp
uter and
Electrical E
ngi
neer
ing
. 2
013;
5(4): 435-
43
7.
[8]
Abbas N, T
a
yya
b
M, Qadri MT
. Real T
i
me
T
r
affic
Densit
y Co
unt
using Image
Processin
g
.
Internatio
na
l Journ
a
l of
Co
mputer App
lic
ation
. 201
3; 83(9)
: 16-19.
[9]
Cha
p
le M, P
a
ygud
e SS. Veh
i
cle D
e
tection
a
nd T
r
acking fr
om Vid
eo F
r
a
m
e Seq
u
e
n
ce.
Internatio
na
l
Journ
a
l of Scie
nce & Engi
ne
e
r
ing R
e
searc
h
. 201
3; 4(3): 1-7
.
[10]
Jatmiko W
,
Azurat A, W
i
bo
wo
AH, Marihot
H, Wicaksana
M,
T
a
k
aga
w
a
I
,
Seki
yama
K, F
u
kuda
T
.
Self-Organizing Ur
ban T
r
affic
Control Arc
h
it
ecture
w
i
th S
w
a
rm-Self Organizing M
a
p in
J
a
karta: Signal
Contro
l S
y
ste
m
and Sim
u
lat
o
r.
Internatio
na
l Journ
a
l o
n
S
m
ar
t Se
nsin
g a
nd Ad
aptive Sy
stems
. 20
10
;
3(3): 443-
46
5.
[11]
Kurni
a
w
a
n F
,
Derma
w
a
n D,
Dinar
ya
nto O, Ira
w
at
i M. Pre-T
i
med and Co
ordi
nated T
r
affic Control
l
e
r
S
y
stems Bas
e
d on AVR Mic
r
ocontro
ller.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mputi
ng Electro
n
ics
and C
ontrol
. 2
014; 12(
4): 787
-794.
[12]
Dun
ne P, Matu
sze
w
ski BJ. Hi
stogram-b
ased
Detect
ion of
Moving Object
s for
T
r
acker Initialization i
n
Surveil
l
a
n
ce Vi
deo.
Intern
atio
nal Jo
urna
l of Gr
id and D
i
stri
buted C
o
mp
uti
ng.
201
1; 4(3): 71-7
8
.
[13]
W
ang J, Sun
X, Guo
J. A Regi
on T
r
acking
-Based Ve
hic
l
e Detect
io
n Al
gorithm i
n
Nig
ht time T
r
affic
Scenes.
Sensors
. 2013; 13: 1
647
4-16
49
3.
Evaluation Warning : The document was created with Spire.PDF for Python.