TELKOM
NIKA
, Vol. 13, No. 4, Dece
mb
er 201
5, pp. 1446
~1
455
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i4.2236
1446
Re
cei
v
ed
Jun
e
22, 2015; Revi
sed
No
ve
m
ber 4, 2015
; Accepte
d
Novem
b
e
r
22, 2015
Double Difference Motion Detection and Its Application
for Madura Batik Virtual Fitting Room
Rima Tri
w
a
h
y
u
n
i
ngrum*,
Indah Agu
s
ti
en Siradjudd
in, Yonathan
Fer
y
Hendra
w
a
n
,
Arik Kurniaw
a
ti, Ari Kusumaningsih
Informatics De
partment, Univ
ersit
y
of T
r
unoj
o
y
o
Mad
u
ra, T
e
la
ngra
y
a
Stre
et, Kamal, Mad
u
ra
*Corres
p
o
ndi
n
g
author, em
ail
:
rima_t
w
@
yah
oo.co.id
A
b
st
r
a
ct
Madur
a Batik
Virtual F
i
ttin
g
Roo
m
usin
g do
ubl
e diffe
rence
alg
o
rith
ms
moti
on d
e
t
ection i
s
prop
osed
in th
is researc
h
. T
h
is virtua
l fittin
g
roo
m
co
nsis
ts of three mai
n
st
ages, i.e.
moti
on d
e
tecti
on,
deter
mi
natio
n
of the
reg
i
on
o
f
interest
of th
e d
e
tect
ed
mo
tion, su
peri
m
p
o
sed
the
virtu
a
l c
l
othes
i
n
to
th
e
regi
on
of int
e
r
e
st. T
he d
o
u
b
l
e
d
i
fference
a
l
gorith
m
is
use
d
for th
e
moti
on
detecti
on st
age, s
i
nce
i
n
t
h
is
alg
o
rith
m, the
empty fra
m
e
a
s
the r
e
ferenc
e
frame is
n
o
t r
equ
ired.
T
h
e
dou
ble
d
i
fferen
c
e al
gor
ith
m
u
s
es
the prev
io
us a
nd n
e
xt frame t
o
detect th
e
motion
in
th
e cur
r
ent fra
m
e. Per
c
eptio
n T
e
st Ima
ges S
e
q
uen
ces
Dataset ar
e us
ed as
the
data
of the ex
per
i
m
e
n
t to
me
as
ure the
perfor
m
a
n
ce
accur
a
cy of this a
l
g
o
rith
m
before
the
al
go
rithm is
use
d
fo
r the M
adur
a
b
a
tik virtu
a
l fittin
g
ro
o
m
. T
he
ac
curacy is
5
7
.31
%
, 99.7
1
%,
an
d
78.52
% for th
e
sensitiv
ity,
sp
ecificity, an
d
b
a
la
nced
accur
a
cy, resp
ective
ly. T
he b
u
il
d M
adur
a b
a
tik virt
ual
fitting ro
o
m
in t
h
is r
e
searc
h
c
a
n b
e
used
as
the
ad
ded
featu
r
e of th
e M
a
d
u
r
a b
a
tik
onl
ine
stores, he
nce
th
e
consu
m
er is
a
b
le to
se
e w
het
her th
e cl
othes
is fitted to
the
m
or n
o
t, and
this virtu
a
l fittin
g
roo
m
is
also
can
be use
d
as the
promotio
n of Madur
a batik b
r
oad
ly.
Ke
y
w
ords
: Motion Detection,
Double
Difference, Augm
ented Re
a
lity, Virtual F
i
tting Ro
o
m
Copy
right
©
2015 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Motion dete
c
t
i
on is
an imp
o
rtant
stage i
n
many
ap
pli
c
ation
s
, e
s
pe
cially obje
c
t t
r
ackin
g
appli
c
ation. T
he mo
re
accu
rate of m
o
tio
n
dete
c
ti
on, t
hen the
a
ccu
racy of the
obj
ect tra
c
king
will
increa
se. Th
e motion d
e
tection
and o
b
ject tra
c
kin
g
ca
n be u
s
ed in ma
ny appli
c
ation
s
, fo
r
instance, nav
igation, gait recogniti
on, surveillance
sy
stem, etc. Th
erefore, many
research focus
on the motion
detection al
g
o
rithm [1-5].
This resea
r
ch
propo
se
s mo
tion detectio
n
algorithm fo
r Madura Batik Virtual fitting room.
The Buil
ding
of Mad
u
ra
B
a
tik virtu
a
l fitting
room
is p
r
opo
se
d in
th
is
re
sea
r
ch
si
nce
currently, a
home i
ndu
stry of Madu
ra
batik i
s
g
r
o
w
t
h
in
cre
a
si
ngl
y. The g
r
owth
is
cau
s
e
d
by
the existe
nce of
Surama
du Bridge. In term of Madura b
a
t
ik, the
Suramadu Brid
ge gives two mai
n
benefits to the
Madu
ra peo
p
l
e. First, because of the bridg
e
, t
he touri
s
m in Ma
dura i
s
incre
a
sin
g
, and m
any
tourist
s
b
oug
ht Madu
ra
ba
tik a
s
thei
r
so
uvenir.
Se
co
nd, the
brid
g
e
ma
ke
s the
delivery p
r
o
c
ess
of Madura b
a
tik from Ma
dura to the
region o
u
ts
id
e
Madura is e
a
sie
r
. He
nce, these two m
a
in
benefits ma
ke the sale of
Madu
ra bati
k
is increa
sin
g
nowaday
s.
Some of
hom
e ind
u
stry
of
Madu
ra
batik notic
es the
o
pportu
nity to i
n
crea
se t
he
Madu
ra
batik; therefo
r
e they use the info
rm
atio
n techn
o
logy
as the p
r
om
o
t
ion of Madu
ra batiks, su
ch as
online
sto
r
e,
e-comme
rce,
etc. T
h
e
promotion
of
Madu
ra batik
usi
ng onlin
e
sto
r
e also
g
i
ves
benefit to the consume
r
, i.e. t
he consu
m
er wo
uld n
o
t need to travel a long wa
y to buy Mad
u
ra
batik. Th
e
co
nsum
er is abl
e to
cho
o
se the Ma
du
ra b
a
tik fro
m
the
web,
and
the
tran
sa
ction
will
be do
ne in
a
n
onlin
e tra
n
s
a
c
tion. However, unli
k
e t
he phy
sical store, the o
n
li
ne sto
r
e
doe
s not
provide th
e fitting ro
om. He
nce, the
co
nsumer i
s
n
o
t a
b
le to try the
clothe
s, whet
her th
e cl
othe
s
is fitted to them or not. The
existence of virtual
fitting room is requi
red in the onli
ne sto
r
e a
s
a
n
adde
d feature. This featu
r
e makes th
e
con
s
um
er
see whethe
r the color
or t
he pattern of
the
clothe
s is fitted to them or not.
The Virtu
a
l fitting ro
om i
s
prop
osed in
this
re
sea
r
ch
usin
g the
dou
ble-diffe
ren
c
e
motion
detectio
n
alg
o
rithm. Thi
s
algorith
m
is
use
d
, si
n
c
e i
n
the previo
us research
usin
g the fra
m
e
differen
c
e mo
tion detectio
n
,
the first fra
m
e of the dat
a video shoul
d be an e
m
pt
y frame (the
re is
no foregroun
d in the f
r
am
e) [6]. Follo
wing is t
he re
mainde
r of t
he pa
pe
r, the se
co
nd
se
ction
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 144
6 – 1455
1447
explain
s
the
motion d
e
tect
ion alg
o
rithm,
the third
section explain
s
the buildi
ng
of virtual fitting
room, the fou
r
th se
ction de
scribe
s the e
x
perime
n
t
an
d the last se
ction is the co
nclu
sio
n
se
cti
o
n
2. Double Differen
ce Mo
tion De
tec
t
ion
Motion d
e
tection alg
o
rith
m is re
quire
d in thi
s
re
search fo
r
bui
lding th
e virt
ual fitting
room. Th
e ob
jective of moti
on dete
c
tion i
n
this res
earch is to find th
e location of the con
s
ume
r
in
each frame o
f
the data vid
eo. The simp
lest al
go
rithm
for motion detection is frame differen
c
e
algorith
m
. Thi
s
al
gorith
m
compa
r
e
s
e
a
ch pixel i
n
the
su
ccessive f
r
ame o
r
co
mp
are
s
e
a
ch pix
e
l
in the
refe
ren
c
e
and
curren
t frame. If the
pixel in
the
same lo
catio
n
from the
com
pare
d
frame
s
is
cha
nge
d, the
n
the pixel i
s
recogni
ze
d a
s
the m
o
ving
pixel. To com
pare t
he pixel
,
the differen
c
e
betwe
en the comp
ared pixels is calculat
ed as sho
w
n
in (1). The pi
xel is detecte
d as the motion
pixel or th
e f
o
reg
r
o
und, if
the re
sult of t
he differen
c
e
in (1
) i
s
bi
gg
er tha
n
the
set thre
shol
d. On
the contrary, i
f
the result of
the
differen
c
e le
ss
tha
n
th
e set threshol
d, the
pixel i
s
re
co
gni
zed
a
s
the backg
rou
nd pixel.
background
as
set
then
)
,
,
(
foreground
as
set
then
)
,
,
(
)
,
(
)
,
,
(
)
,
,
(
0
Th
t
y
x
D
if
Th
t
y
x
D
if
y
x
F
t
y
x
F
t
y
x
D
(1)
whe
r
e
F
(
x
,
y
,
t
) is
a pixel
val
ue o
n
(
x
,
y
) i
n
frame
t
, a
nd
F
0
(
x
,
y
) i
s
th
e
pixel on
(
x
,
y
) in
the refe
ren
c
e
frame.
Gene
rally, the referen
c
e frame i
s
the first
frame of the data vide
o, where the frame is
empty fram
e
or the f
r
am
e con
s
ist
s
of
only the
ba
ckgro
und
im
age
and
the
foreg
r
ou
nd i
s
not
available
in
referen
c
e
fra
m
e a
s
depi
ct
ed in
Fig
u
re
1
.
As
see
n
in
F
i
gure
1.
The
referen
c
e
fra
m
e i
s
empty frame, hen
ce the foregro
und o
r
th
e moving
pixel in the curre
n
t frame is de
tected ea
sily.
Figure 1. Left: the reference frame, Middle: t
he c
u
rrent
frame, Right:
the detec
ted
motion [5]
In the
real
system, the
re
feren
c
e
fram
e that
con
s
i
s
ts of
an
emp
t
y frame i
s
di
fficult to
achi
eve. Usu
a
lly, the foreg
r
oun
d imag
e is unp
re
di
cta
b
l
e, i.e. the foregro
und im
ag
e can
be in th
e
first frame, o
r
it appea
rs in
the se
con
d
frame,
o
r
prob
ably the fore
grou
nd ima
g
e
appe
ars in
any
frame of the
data video. Hence, the
ref
e
ren
c
e th
at consi
s
ts of a
n
empty frame i
s
ra
re in the
real
sy
st
em.
Ho
wever, if the differen
c
e is cal
c
ul
at
ed between
the su
cce
ssive frame
s
,
or the
differen
c
e is
cal
c
ulate
d
be
tween the cu
rre
nt
and previous fram
e
then the ghosting effect
is
appe
ar i
n
th
e re
sult of t
he differen
c
e
.
Figure
2
shows the
gh
ost effe
ct of
frame diffe
re
nce
method. The
Figure 2 sho
w
s that
the previous fram
ei
s used as th
e
refere
nce fra
m
e, and the the
obje
c
tapp
ears in the p
r
ev
ious
and the
curre
n
t fram
e. Therefore,
the re
sult of
the differen
c
e
between the successive fr
ames present
s the ghostin
g effects, and this effect will decrease the
motion dete
c
tion accu
ra
cy.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Dou
b
le Difference Motion
Dete
ction an
d Its App
licati
on for Mad
u
ra Batik Virtua
l … (Rim
a T)
1448
Figure 2. Left: the previous Fr
ame, Middle: the c
u
rrent
fram
e, Right:
the detec
ted
motion
Dou
b
le diffe
rence ima
ge
g
eneration
alg
o
rithm fo
r m
o
tion dete
c
tion
is
use
d
to
overcome
the probl
em. This alg
o
rith
m calculate
s
two kind
s
of differen
c
e
s
o
f
the successive frames, i.e,
first, the
current frame
an
d the p
r
eviou
s
fra
m
e,
an
d
se
con
d
, the d
i
fference b
e
twee
n the
cu
rrent
frame an
d the next frame. The re
sult of the diffe
rences is
combi
n
ed with the AND o
peration
[6].
The diag
ram
of the algorith
m
can be
see
n
in Figure 3.
Figure 3. Dou
b
le Differe
nce Motion Det
e
ction
Diag
ra
m [5]
The u
s
e of th
e two
kind
s o
f
differences
cal
c
ulatio
n a
nd the result is combin
ed
with AND
operation ma
ke
s the gho
st
ing effect is di
sap
pea
red.
The empty
f
r
ame a
s
the
referen
c
e frame
is not required in the double diff
eren
ce
algorith
m
. He
nce thi
s
algo
rithm is more
suited to impl
ement in the real system.
3. Rese
arch
Metho
d
The M
adu
ra
batik virtu
a
l fitting ro
om i
s
p
r
opo
se
d in thi
s
resea
r
ch.
T
he virtual
fitting room
is u
s
ed fo
r th
e co
nsume
r
to se
e wh
eth
e
r the
colo
r
and the p
a
ttern of the
clot
hes
are fitted
to
them or not.
The build o
f
virtual fittin
g
room
in th
is re
sea
r
ch
use
d
the au
gmented
real
ity
techn
o
logy, i.
e. the te
chno
logy that cre
a
tes th
e
com
b
ination
of th
e re
al o
b
je
cts a
nd the
virt
ual
obje
c
ts such that the boun
d
a
ry betwe
en the obje
c
ts
ca
n’t be disting
u
ish
ed [8, 9].
In this re
se
arch the M
adu
ra batik virtu
a
l
clothe
s play
s a
s
the virtu
a
l obje
c
t. Me
anwhile
the bo
dy of th
e con
s
ume
r
p
l
ays a
s
th
e
re
al obj
ect. T
h
e
pu
rpo
s
e
of th
e Ma
dura b
a
tik virtu
a
l fitting
room i
s
to
co
mbine the
s
e
obje
c
ts. To
combine th
e
vi
rtual
clothe
s i
n
to the
bo
dy
of the co
nsu
m
er
Fra
m
e
t
-
1
Fra
m
e
t
Fra
m
e
t
+
1
diffe
r
e
nc
e
d
iffe
r
e
nc
e
D
i
ffe
r
e
nc
e
I
m
age
i
-1
D
i
ffe
r
e
nc
e
I
m
age
i
b
inarization
b
inarization
AND Ope
r
ation
Double Differ
e
nce M
o
tion Detection
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 144
6 – 1455
1449
in the data v
i
deo, the a
u
g
mented
re
al
ity need the
locatio
n
of th
e pla
c
eme
n
t. Two
kin
d
s
of
placement
of
virtual
obje
c
t
s
into
the
re
a
l
obje
c
ts, th
e
y
are
ma
rker
and
marke
r
le
ss. A
ma
rker
is
requi
re
d in th
e marke
r
a
u
g
m
ented
realit
y. This ma
rke
r
is
used
as t
he lo
cation
of
the pla
c
em
e
n
t
of the virtual
obje
c
ts into
the real o
b
je
cts.
The exa
m
ple of ma
rker and th
e ma
rker au
gment
ed
reality are
sh
own in Fig
u
re
4.
This research
use
s
marke
r
l
e
ss augm
ent
ed rea
lity. To combi
ne the
virtual obje
c
t and the
real obj
ect, th
is re
se
arch u
s
e
s
the re
gio
n
of in
tere
st of the body o
f
the con
s
um
er in
stead of
the
marker ima
g
e
. He
nce, the
first
stag
e i
n
this re
se
arch
is motion
det
ection.
The
motion
dete
c
tion
is used in the
next stage, i.e. the deter
mi
nation of the region of interest.
Figure 4. Left: Mark
er, Right: Augmented Reality with Marker
The dia
g
ra
m
of the Mad
u
ra b
a
tik virt
ual fitting ro
om ca
n be
seen in Fig
u
re 5. The
Madu
ra
batik virtual fitting
room
con
s
ist
s
of
thre
e m
a
in sta
g
e
s
. Fi
rst, motion
det
ection
to
dete
c
t
the motion in
the frame of
data.Seco
nd,
the deter
mi
n
a
tion of the region of inte
rest ba
sed
on
the
detecte
d moti
on. The regio
n
of intere
st is dete
r
mine
d
from the enti
r
e pixel that
are
recogni
zed
as th
e foreg
r
ound
imag
e
or the
movin
g
pixel. Th
e example of
the
regio
n
of intere
st
from
the
detecte
d motion ca
n be seen in Figu
re
6. The fi
nal stage i
s
sup
e
rimp
osi
ng the Madu
ra b
a
tik
virtual
clothe
s o
n
the
b
o
d
y
of the
co
nsumer a
c
co
rdi
ng to th
e
regi
on of th
e i
n
te
rest. T
h
e
virtual
clothes are al
so resi
zed based on
the region of interest. Hence,
the virtual clothes will be fitted
to the body of the con
s
ume
r
Figure 5. The
virtual fitting
room di
agra
m
mot
i
on
detection
using
double
difference
deter
m
ination
of
reg
i
on
of
inter
e
st
superimposing
madura
batik
virtual
cl
ot
hes
into
the
body
of
the
consumer
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Dou
b
le Difference Motion
Dete
ction an
d Its App
licati
on for Mad
u
ra Batik Virtua
l … (Rim
a T)
1450
Figure 6. Reg
i
on of Interest
of the detected motion
4. Results a
nd Analy
s
is
Two
main
experim
ents are condu
cted
i
n
this
re
se
arch. Th
e first
experim
ent is done
to
measure the accuracy pe
rforma
nce of the doubl
e differen
c
e alg
o
rithm before the algo
rithm i
s
impleme
n
ted
in th
e virtual
fitting ro
om
appli
c
ation. I
n
this expe
ri
ment, we u
s
e the
Perce
p
t
ion
Test Image
Sequen
ce d
a
t
aset. The ground trut
h of
the detectio
n
motion is
provide
d
in the
databa
se. T
h
e first
experi
m
ent is com
p
leted
wi
th
motion d
e
tection usi
ng
co
nventional f
r
ame
differenc
e (a
firs
t frame for refe
ren
c
e
fra
m
e)
as the
co
mpari
s
o
n
for
the do
uble
dif
f
eren
ce
motio
n
detectio
n
method.
The
se
con
d
e
x
perime
n
t is the impl
ement
ation of th
e d
ouble
differe
n
c
e m
o
tion
det
ection
for the Madu
ra batik virtual
fitting room.
There are five enviro
n
men
t
s of the data
s
et in
the first
experim
ent. Two d
a
ta a
r
e
use
d
in
the experi
m
e
n
t in ea
ch e
n
v
ironme
n
t.We used b
a
la
n
c
ed
accu
ra
cy to mea
s
ure the pe
rform
a
n
c
e
of the motion detectio
n
alg
o
rithm a
s
sh
o
w
n in (2
)
;
F
N
T
P
TP
y
Sensitivit
;
F
P
T
N
TN
y
Specificit
2
y
Specificit
y
Sensitivit
curacy
BalancedAc
(2
)
TP (True Po
sitive) is the
numbe
r of t
r
ue d
e
te
cte
d
motion pixel
s
, TN
(True
Neg
a
tive) is
the
numbe
r of tru
e
dete
c
ted
b
a
ckgroun
d pi
xels, FP (F
al
se Po
sitive) i
s
the
numb
e
r of false
dete
c
ted
motion pixels,
and FN (Fal
se Neg
a
tive) is the numbe
r of false dete
c
ted backg
rou
nd pixels.
The
se
nsitivity is
used to
mea
s
u
r
e
th
e a
c
curacy
o
f
detecte
d m
o
tion pixel
s
and th
e
spe
c
ificity i
s
use
d
to
mea
s
ure th
e a
ccura
cy of
det
ected
ba
ckg
r
ound
pixel
s
.
Mean
while
th
e
balan
ce
d accura
cy is the
averag
e a
c
cura
cy.The
a
c
curacy of th
e first expe
ri
ment usi
ng t
h
e
doubl
e differe
nce motio
n
d
e
tection i
s
sh
own in Ta
ble
1.
As seen
in
T
able
1, the
a
v
erage
sen
s
itivit
y accu
ra
cy for the
fram
e
differen
c
e
m
e
thod i
s
61.63%, me
anwhile fo
r t
he do
uble
di
fference m
e
thod i
s
5
7
.31
3
%. The av
erag
e
spe
c
ifi
c
ity
accuracy fo
r frame diffe
ren
c
e a
nd
doubl
e difference metho
d
is 9
7
.48
%
and 99.7
1
%,
respe
c
tively.
The avera
ge result sho
w
s thate
fram
e d
i
fference i
s
b
e
tter
sin
c
e
se
nsitivity rate i
s
highe
r
comp
are to
the
d
ouble
differe
nce
metho
d
. Ho
weve
r, the fra
m
e
differen
c
e
(with
out
aempty frame
as the refe
re
nce frame
)
method yield g
host effect in
the motion de
tection p
r
o
c
e
ss
n as d
epi
cted
in Figure 7. The sen
s
itivity rate
for the
frame differe
nce i
s
hig
her,
since when t
h
e
detecte
d moti
on is supe
ri
mposed to th
e provided
groundtruth, th
e num
ber of
TP pixels
(white
pixels) i
s
more compa
r
e
to the number of TP pixels from
double differen
c
e meth
od.
Unfortu
nately
,
the frame
di
fference
will g
i
ve gho
st
effect, and thi
s
ef
fect will give f
a
lsely d
e
tecte
d
motion. Th
erefore,
spe
c
ifi
c
ity is
also
requir
ed i
n
th
e mea
s
u
r
em
ent of a
c
cura
cy. Table
1
s
hows
the spe
c
ifity accu
ra
cy fo
r the
do
uble
differe
nce
method
is hi
gher
comp
are to the
fra
m
e
differen
c
e me
thod.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 144
6 – 1455
1451
Table 1. The
Accu
ra
cy of the Fra
m
e Diff
eren
ce a
n
d Double
Differe
nce Algo
rithm
for motion
detec
tion, SN is
s
e
ns
it
ivity
accuracy and SP is
specifi
c
ity accuracy
Curre
nt Frame
Grou
nd T
r
uth
Detected Motion
Performance Acc
u
rac
y
(%)
Frame
Difference
Double
Difference
Frame Differenc
e
Double
Differenc
e
SN SP
SN SP
62.64
99.64
55.59
99.98
65.42
95.79
50.89
98.73
62.25
97.07
61.44
100
72.07
94.05
70.56
99.95
72.99
98.32
67.02
99.84
55.65
96.88
52.57
99.3
46.26
99.8
46.26
99.98
64.71
99.02
64.43
100
53.16
96.29
44.57
99.47
61.14
97.92
57.31
99.71
Average accuracy
61.629
97.478
57.313
99.707
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
9
30
Dou
b
le Difference Motion
Dete
ction an
d Its App
licati
on for Mad
u
ra Batik Virtua
l … (Rim
a T)
1452
(a)
(b)
(c
)
(d)
(e)
Figure 7. (a)
Previou
s
Fra
m
e; (b)
Curre
n
t Frame
(c) grou
ndtruth; (d) dete
c
ted m
o
tion with fra
m
e
differen
c
e; (e
) detect
ed mo
tion with dou
b
le differen
c
e
The ave
r
ag
e
of balan
ce
d
accu
ra
cy for doubl
e
diffe
ren
c
e m
e
tho
d
is
78.51%
with the
sen
s
itivity accuracy is o
n
l
y
57.31%, and sp
ecif
i
c
ity achieve
d
9
9
.71%. This
mean
s that the
doubl
e difference alg
o
rith
m gives l
e
ss
perfo
rman
ce
for dete
cting
the motion
pi
xels; mea
n
while
the algo
rithm
gives th
e be
st pe
rform
a
n
c
e fo
r det
e
c
ti
ng the
ba
ckg
r
oun
d pixel
s
.
As seen i
n
T
able
1, ho
wever
even the
se
nsitivity accu
racy
is lo
w,
the result
of
dete
c
ted
m
o
tion i
s
sufficient
enou
gh to provide the regi
on of intere
st
in the virtual fitting room proves.
The improve
m
ent of doub
le differen
c
e
algorit
h
m
for motion dete
c
tion com
pare to frame
differen
c
e
algorithm i
s
th
e refe
ren
c
e f
r
ame i
s
n
o
t required to
b
e
the em
pty frame
(the ta
rget
obje
c
t is do
e
s
not exi
s
t in
the frame
)
. Unfortu
nat
ely
,
to detect th
e motion of t
he current fra
m
e,
the next frame is re
quire
d in orde
r to
eliminat
e the gho
sting ef
fect cau
s
e
d
by the difference
cal
c
ulatio
n b
e
twee
n fram
es. Th
e exa
m
ple of the
d
i
fference cal
c
ulation
to det
e
ct
the motio
n
in
the data
s
et can be
see
n
in
Figure 8. Fig
u
re 8
sh
ows t
h
at com
b
inati
on wi
th
A
ND operator of
the
both difference cal
c
ulation
(current - previous fram
e and current
–
next frame)
will eliminate the
gho
sting effe
ct that cau
s
e
d
by di
fference cal
c
ulatio
n pro
c
e
ss.
(a)
(b)
(c
)
(d)
(e)
(f)
(g)
Figure 8. (a)
previou
s
fram
e; (b) current
frame;
(c) ne
xt frame; (d) differen
c
e bet
wee
n
cu
rrent
and previou
s
frame; (e
) differen
c
e b
e
twe
en cu
rrent
an
d next frame; (f) the dete
c
ted motion ; (g
)
grou
nd truth i
m
age
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 144
6 – 1455
1453
The
se
con
d
e
x
perime
n
t in i
m
pleme
n
tatio
n
of the
do
ubl
e
differe
nce
motion al
go
rithm, i.e.
building M
a
d
u
ra bati
k
virtual fitting room. To build
the virtual fitting room u
s
ing a
ugme
n
t
ed
reality, then v
i
rtual o
b
je
cts
that are
com
b
ined i
n
to
the
real
obj
ect
s
are
create
d
first. Th
e Ma
d
u
ra
batik
virtual clothes plays as
the
virtu
a
l obje
c
ts
in the
resea
r
ch. Th
erefo
r
e
we h
a
ve created t
he
virtual cl
othe
s u
s
in
g bati
k
origi
nally
co
me from
Ma
dura,
so
me o
f
cre
a
ted
Ma
dura
bati
k
virtual
clothe
s can b
e
see
n
in Fig
u
re 9.
These virtual
clothe
s are e
m
bedd
ed to the bod
y of the con
s
um
er
based on the
location
of the
regi
o
n
of i
n
tere
st
of the
dete
c
ted
moti
on.
The
expe
ri
ment i
s
d
o
n
e
in
the
different
environ
ment of
a
virtu
a
l
fitting room. Thi
s
scena
ri
o
proves th
at the
motion
dete
ction al
gorith
m
is
work corre
c
tly in a different environme
n
t. The exper
i
m
ent of the virtual fitting ro
om can b
e
se
en
in Figure 10 a
nd Figu
re 11.
Figure 9. Virtual cloth
e
s from Madu
ra b
a
tik
Figure 10. Top left: the firs
t frame; Top righ
t: the third frame; Bottom: the s
i
xth frame
Figure 1
0
sh
ows that
the
locatio
n
of t
h
e
Ma
dura b
a
t
ik virtual
cl
othes is move
d alo
n
g
with the loca
tion of the body of the consume
r
,
sin
c
e the virtual
clothe
s is e
m
bedd
e
d to
the
regio
n
of inte
rest
of dete
c
ted motio
n
. Fi
gure
10
also
sho
w
s that th
e refe
ren
c
e
frame i
s
not th
e
empty frame,
sin
c
e fo
reg
r
ound
obje
c
t i
s
foun
d in th
e first frame.
This
re
sea
r
ch overcom
e
s our
limitation in the previou
s
rese
arch [4]. In the
previo
u
s
re
se
arch, the refe
ren
c
e
frame
shoul
d
be
the empty frame; hence the motion in the curre
n
t fra
m
e is dete
c
te
d easily.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Dou
b
le Difference Motion
Dete
ction an
d Its App
licati
on for Mad
u
ra Batik Virtua
l … (Rim
a T)
1454
Figure 11. Virtual fitting room in the different environ
m
ent
The va
riety of
environme
n
t
in the
virtual f
i
tting ro
om
ca
n be
seen
in
Figure 1
1
. Fi
gure
1
1
sho
w
s
that, the virtual fitting room
works
well e
v
en in the different envi
r
on
ment.
Unfortu
nately
,
this re
sea
r
ch ha
s a d
r
awback, i.
e. the virtual fitting room do
es
not
work i
n
real time. Since in the dou
ble differen
c
e
motion detection algorithm
, to detect th
e motion in th
e
curre
n
t fram
e, we ne
ed t
o
cal
c
ulate t
he differe
nce
betwe
en the
curre
n
t fram
e and p
r
evio
us
frame, a
nd
al
so
we
ne
ed t
o
calculate t
he diffe
ren
c
e
between
the
cu
rrent fram
e an
d the
ne
xt
frame. He
nce
,
this method
is one fra
m
e
delayed fro
m
the real time.
4. Conclusio
n
The Mad
u
ra
batik virtual fitting room is
built in this re
sea
r
ch. Motion dete
c
tion algorith
m
is re
quired to
obtain the lo
cation of the
regio
n
of
inte
rest, he
nce th
e Madu
ra b
a
tik virtual cl
oth
e
s
is embe
dde
d
to the regio
n
. The doubl
e differen
c
e
algorith
m
is
use
d
for the motion dete
c
tion
stage
and
the
advantag
e of
this al
gorith
m
is the
refe
rence fram
e d
oes
not h
a
ve
to be the
em
pty
frame. T
he
di
sadva
n
tage
o
f
this
algo
rith
m is th
e
one
frame
delaye
d
from
the
re
al sy
stem.Th
e
re
are two exp
e
r
iment
s are
condu
cted in this re
se
ar
ch.
The first experim
ent is to measure the
perfo
rman
ce
accuracy o
f
double diff
eren
ce
m
o
tion dete
c
tion
algorithm,
and the
se
cond
experim
ent is to implement
the algorith
m
for the
virtual fitting roo
m
. The first e
x
perime
n
ts show
that even
the
algorith
m
giv
e
s l
o
w p
e
rfo
r
mance i.
e.
78
.51% a
c
cura
cy. However this
pe
rform
a
n
c
e
is suffici
ent enoug
h for the
determinatio
n of region in
terest in the
virtual fitting room stag
e. The
impleme
n
tation of this alg
o
rithm for the
Madura
batik virtual fitting room is d
o
ne in the se
cond
experim
ent. The expe
rim
ent sho
w
s that the moti
on
detection al
g
o
rithm give
s a good
result for
the emb
eddin
g
of virtual
cl
othes i
n
the
real obj
ec
t
s
in
the video.
Unfortunately, t
he cha
r
a
c
teri
stic
of the alg
o
rit
h
m ma
ke
s th
e virtual fittin
g
ro
om i
s
n
o
t
compl
e
tely
works in
re
al
time, sin
c
e
the
algorith
m
req
u
ire
s
next fra
m
e to det
ec
t
motion in the c
u
rrent frame.
Referen
ces
[1]
ME Martin and
APd Pobi
l.
Rob
u
st Motion Det
e
ction i
n
Re
al-
L
ife Scen
ari
o
s
. Spring
er. 201
2.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 144
6 – 1455
1455
[2]
T
Schlogl, C Belez
nai, M W
i
nter and H Bisc
hof.
"Performa
nce Eval
uatio
n
Metrics for Mo
tion Detecti
o
n
and T
r
acking".
IEEE
. 2004.
[3]
Curtler
R an
d
Davis
LS. "R
obust re
al-tim
e per
io
d
i
c
mot
i
on detecti
on, ana
l
y
sis,
a
nd app
licati
ons".
Pattern Analys
i
s
and Mach
in
e Intelli
genc
e, IEEE Transactio
n
s on
. 20
02: 7
81-7
96.
[4]
A Rahm
an
N
Y
and
N S
a
tya
W
i
do
do. "
C
olor
ed B
a
l
l
Positio
n
T
r
acking M
e
tho
d
fo
r Goalke
ep
er
Huma
noi
d Ro
b
o
t Soccer".
T
e
l
k
omnik
a
. 201
5; 11(1): 11-1
6
.
[5]
RT
W
a
h
y
un
in
grum, YF
He
n
d
ra
w
a
n, IA Sir
adj
udd
in, A K
u
rni
a
w
a
ti
an
d
A Kusuma
ni
ng
sih. "F
rame
Differenc
ing M
o
tion D
e
tectio
n for Madur
a Batik Virtual F
i
tting Ro
om". i
n
Reg
i
o
nal C
o
nferenc
e o
n
Co
mp
uter and
Information En
gin
eeri
n
g
, Yog
y
ak
arta. 201
4: 116-
119.
[6]
IA Sirad
j
ud
di
n, MR W
i
d
y
a
n
to
an
d T
Basaru
ddi
n. "Partic
l
e
F
ilter
w
i
th
Gau
ssian
W
e
ig
htin
g for H
u
ma
n
T
r
acking".
T
e
lkomnik
a
. 201
2; 10(6): 80
1-8
0
6
.
[7]
Y Kameda
an
d M Minoh. "A
human m
o
tio
n
estimati
o
n
m
e
thod
usin
g 3-
successiv
e
vid
eo frames". in
Internatio
na
l C
onfere
n
ce o
n
Virtual S
y
stems
and Mu
ltimed
ia
(VSMM), Gifu, Japan. 1
996:
135-
140.
[8]
RT
Azuma. "A
surve
y
of aug
mented re
alit
y"
.
T
e
leoper
ators
and Virtua
l En
vironme
n
ts. 19
97; 6(4).
[9]
P Milgr
a
m, H
T
a
kemura,
A Utsumi a
nd F
Kishi
no. “
Augm
en
te
d re
al
i
t
y: A cl
ass of
displays on
t
h
e
reality-virtu
a
lity
contin
uu
m
”. i
n
SPIE Proce
edi
ngs Vo
l. 2
351:
T
e
leman
i
pul
ator an
d T
e
le
prese
n
c
e
T
e
chnolog
ies, Boston.
Evaluation Warning : The document was created with Spire.PDF for Python.