TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 11, Novembe
r
2014, pp. 77
9
8
~ 780
7
DOI: 10.115
9
1
/telkomni
ka.
v
12i11.63
71
7798
Re
cei
v
ed
Jun
e
5, 2014; Re
vised Septem
ber
14, 20
14;
Accept
ed O
c
tober 2, 201
4
Mosquito Tracking by Image S
e
gmentation of Optical
Flow Field
Jahan
gir Ala
m
SM, Hu Guoqing*
Schoo
l of Mechan
ical & Auto
motive Eng
i
ne
er
in
g, South C
h
in
a Univ
ersit
y
of
T
e
chnolo
g
y
,
Room 3
14, Bui
l
din
g
No 2
9
, 51
064
1,
T
i
anhe District, Guang
zhou, Gua
ngd
ong, Ch
in
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: gqhu
@scut.e
du.cn
A
b
st
r
a
ct
High s
pee
dy
Mosqu
i
to tracking a
nd a ti
me effici
ent techn
i
qu
e hav
e bee
n pres
e
n
ted b
y
consi
deri
ng o
n
ima
ge se
g
m
e
n
tation of
the
optica
l
flow
w
h
ich has b
e
e
n
compute
d
by i
m
a
ge succ
essi
v
e
frames to trac
k the Mosq
uito
of a
specific r
egi
on of i
n
tere
st on the re
gio
n
of fiel
d w
i
th seg
m
e
n
ted flo
w
regi
ons. T
he o
p
tical fl
ow
has
bee
n estab
lis
hed
by the su
ccessive tw
o frames to co
nsi
der ac
quir
i
ng t
h
e
imag
e for c
o
mputin
g. A fu
zz
y
anta
g
o
n
is
m i
n
dex
has
be
en
indic
a
ted
as
th
e d
egre
e
of th
e co
nsistency
of
flying M
o
sq
uit
o
. T
he i
m
ag
e
frames
are
us
ed to s
e
g
m
ent
the o
p
tica
l flo
w
field. T
he
i
m
a
ges
hav
e b
e
e
n
seg
m
e
n
ted i
n
flow
field w
i
th in
the different consiste
ncy
of regi
on of inter
e
st. T
he s
pecific regio
n
of inter
e
st
can be d
e
tecte
d
in the differe
nt regio
n
of interest s
paces.
T
herefore, the
Mosqu
i
to can b
e
tracked fro
m
tw
o
subse
q
u
ent i
m
ages. H
o
w
e
ver
,
the detect
ed
specific re
gi
on
of interest is a
sub-re
gio
n
of r
egi
on of i
n
tere
st.
T
he specific re
gio
n
of interest
is smal
ler i
n
the
i
m
a
ge fra
m
es w
h
ich can
be red
u
ce
d the time to co
mp
ute
the sp
ecific re
gio
n
of i
n
terest
. It is
the facil
i
tating r
e
a
l
-time
process
of Mo
squito tr
ackin
g
. In the
prop
os
e
d
techni
qu
e, it h
a
s be
en
de
mo
nstrated i
m
ag
e
sequ
enc
es
of
movin
g
Mos
q
uito for
detecti
on a
nd
pos
itio
n
tacking.
Ke
y
w
ords
: mosquito, tim
e
efficient, tra
cking,
segm
entation, optical flow
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Tra
cki
ng an
d
locali
zation
of flying Mosquito det
e
c
tio
n
is the m
o
st
intere
sting a
nd funny
but ch
allen
g
i
ng research
work to
save
huma
n
lif
e from Mala
ria.
This
re
sea
r
ch wo
rk ha
s o
t
her
appli
c
ation
s
i
n
defen
se vi
sion
analy
s
is. Some
co
n
v
entional me
thods
are
prese
n
ted in t
he
literature [1-3
] for tra
c
king
of moving
fro
m
a m
o
tion
si
ght. The
r
e
are some
techn
i
que
s a
r
e
use
d
with morphol
ogical metho
d
s. But the inherent pr
obl
em of these tech
niqu
es h
a
s
bee
n u
s
ed
for
the morp
holo
g
ical o
peratio
ns.
The imag
e flow [4-7] from
a 2D pe
rsp
e
ctive
sp
ecifi
e
s the am
ou
nt of the pixels of an
image move
s between t
w
o adja
c
e
n
t time-ord
ere
d
frame
s
[8]. The flow field has be
en
con
s
id
ere
d
to refer a
s
the
optical flo
w
field [6]. The optical flow i
s
su
ch an effici
ent and effect
ive
techni
que
tha
t
su
spe
c
t o
b
j
e
ct m
o
tion
wi
th image
inte
nsity variatio
n
s
whi
c
h i
s
co
mputed
between
two co
nsecutive image fra
m
es [9, 10]
a
nd the optical
flow score is
the techni
que
in the literatu
r
e
[11-15]. T
he
optical
flow approxim
ation
is
capability to find i
n
de
tecting Mosquit
o
’s pattern.
It
is
deriving
in th
e 3D motion
and
stru
cture
of the
Mo
sq
uito in a
n
im
age frame. T
he opti
c
al flo
w
techni
que
s in
clud
e re
al-tim
e multiple Mo
squito tr
ackin
g
. The dete
c
tion of Mosqui
to is motion
of
neon
atal sei
z
ure
s
[16]. It can be segme
n
ted of 3D m
o
tion [17].
The o
p
tical
flow a
s
se
ssm
ent metho
d
rep
r
e
s
ent
s th
e chang
es i
n
Mo
squito’
s
image
brightn
e
ss. The partial de
ri
vative constraint equat
ion
can be u
s
e
d
for time com
p
utation. The two
simila
r eq
uati
ons the o
p
tical flow
co
nst
r
aint
e
quatio
n (O
FCE
)
an
d the
contin
u
i
ty equation
of
fluid-dyna
mics
can
be
u
s
e
d
[7]. It assu
mes that
the
Mosq
uito’s i
m
age i
n
ten
s
ity is
stationa
ry due
to time refe
re
nce. T
here a
r
e two comp
utation of
opt
ical flo
w
such as re
gula
r
i
z
ation
and
m
u
lti-
con
s
trai
nt-b
a
s
ed a
pproa
ches. Regula
r
ization
ap
proached mo
d
e
l of the optical flow field
estimated
a
s
an ill-po
sed
p
r
oble
m
[18].
The mi
nimize
d an
d regul
arized
ha
s b
e
e
n
con
s
ide
r
ed
b
y
an ap
propri
a
tely weighte
d
sm
oothn
ess
con
s
trai
nt. In this
anal
ysis, the
velocity ha
s b
e
e
n
evaluated
at
every poi
nt i
n
the
Mo
squi
to im
age.
Re
gulari
z
atio
n-b
a
se
d a
pproa
che
s
usi
ng t
he
OFCE
ca
n b
e
found
in [1
9-21]. It has the
uniq
uen
ess.
The
reg
u
lari
zation p
r
o
c
e
s
ses
also h
e
lp t
o
determi
ne of
the Mosquit
o
’s
shap
e [2
0]. Thus it
can be h
e
lpful
to determin
e
the Mo
squi
to’s
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Mosq
uito Tra
cki
ng b
y
Im
a
ge Segm
enta
t
ion of Optical
Flow Field (Jahan
gir Alam
SM)
7799
pattern. Th
e
conto
u
rs of the opti
c
al flo
w
can
b
e
asse
ssed to d
e
tect the flying Mosquito.
The
image b
r
ight
ness statio
na
ry con
d
ition i
s
the mu
lti-constraint-ba
s
ed app
roa
c
h
e
s flying-inva
riant
function
like
contrast,
ave
r
age,
varia
n
ce, entropy
, curvature, mo
ments,
gra
d
i
ent ma
gnitud
e
of
colo
r spe
c
tru
m
, local inte
n
s
ity, images
obtaine
d.
Th
ese m
e
thod
s are inte
nde
d
at formative
the
inversi
on o
r
p
s
eu
do-i
n
versi
on of the coe
fficient
matrix of these functions
. The e
q
u
ivalent optical
flow me
cha
n
i
s
m
s
are u
s
e
d
to set the
con
s
trai
nt fo
r image [21,
2
2
]. To brig
ht of intensity, it is
use
d
the second-order
pa
rtial derivativ
es to co
mput
e the flow m
e
ch
ani
sm [23
,
24]. The main
theme of these techniq
u
e
s
is in the image pi
xel
neigh
borhoo
d
s
simila
rity of velocity. T
he
neigh
bori
ng
p
i
xels a
r
e
in
a
smooth
o
p
tical flow fiel
d
[7]. This multi
point m
e
thod
[25] app
ro
ach
is
usually involving as,
a)
Sequen
ce of i
m
age is p
r
e
-
filtered to re
gul
arize the initial data [21], [24], [26];
b)
Colle
cting a l
a
rge n
u
mb
er
of con
s
traint
s for large n
e
ig
hborhoo
d [27
]
;
a)
The estim
a
te
d optical flo
w
fields
are use
d
for po
st-filtering [24];
All are use
d
to smooth out
the optical flow
me
cha
n
ism obtained in
the multi-con
s
traint
-
based a
p
p
r
o
a
ch
es. It is e
x
tended S
c
h
unck’
s meth
o
d
[28,
29]
by
Ne
si et al. [3
0] to optical f
l
ow
control (EOF
C) eq
uation.
It uses the di
vergen
ce
of the flow field of the image brightn
e
ss. T
he
EOFC ha
s b
een u
s
ed to
estimate the
optical flo
w
mech
ani
sm i
n
a motional
flying position
.
The
local
optical f
l
ow e
s
timatio
n
metho
d
s
a
r
e
relati
vely faster but le
ss
stable
and
larg
er th
an t
he
global e
s
tima
tion method
s (GEM).
There a
r
e two optical flow co
m
pon
ents su
ch
a
s
Co
h
e
rent Optical Flow (COFF
)
regi
on
s
and in
coh
e
rent Optical
Flow
(IOFF
)
regio
n
s.
Th
e co
herent
can
be a
r
isen out of a
c
tual
displ
a
cement
s of
flying M
o
squito
an
d th
e in
coh
e
re
nt region
s
ca
n b
e
a
r
isen
out o
f
cha
nge
s in t
h
e
intensity level. These ch
a
nge
s can b
e
occu
rred du
e to in surfa
c
e refle
c
tan
c
e and ambie
n
t
illumination
condition
s. But the se
con
d
movement
s a
r
e the no
n-ref
l
ective in the
motion or flyi
ng
scene. T
here
are
seve
ral
regio
n
s
matching of o
p
tical flow te
chni
que
s have
b
een p
r
op
ose
d
to
get better the
preci
s
io
n of the optical flow by
eliminati
ng differentiat
i
on of image i
n
tensitie
s [7].
In this
re
sea
r
ch, the
mo
sq
uito can b
e
d
e
tection
from
the opti
c
al flo
w
field
or flying imag
e
in the motion
al flying scen
e to follow th
e con
s
trai
nts of
spatial coh
e
ren
c
e over very
small
ti
me
[31]. It can b
e
exploited in
the present treatm
ent to focu
s on a smaller regio
n
of the motional
flying sce
ne o
f
the Mosquit
o
for the
com
putation of the optical field.
The flow fiel
d can be
co
mputed the
pre
s
ent an
d
next image frames for
effective
segm
entation
[7] of the
re
gion
s of i
n
terest
(ROI).
A
specific region of
interest is called sub-
regio
n
of interest (S
ROI).
The SROI ca
n be det
e
c
ted
base
d
on its
coh
e
re
nce co
nstrai
nt, density
of the opti
c
al
flow,
sub
s
e
quent flo
w
fi
eld of
the
ROI’s n
e
ighb
orhood
s of th
e
dete
c
ted S
R
OI.
Re
sults of th
ese p
r
o
p
o
s
ed
method ha
s
been p
r
e
s
ent
ed on the tra
cki
ng of imag
e seq
uen
ce
s
of
flying Mosq
ui
to. The timing req
u
ire
m
en
ts of the pro
posed meth
o
d
are the m
a
in fact to detect
Mosq
uito. He
re, the co
nventional Horn
and Sch
u
n
ck’
s method i
s
a
l
so re
po
rted.
2. Mathem
atica
l
O
v
er
v
i
e
w
The math
em
atical ove
r
vie
w
of the
opti
c
al flo
w
field,
fuzzy
neu
ral
con
c
e
p
ts
an
d logi
c
index of th
e
Mosq
uito’s I
m
age
pixel
n
e
ighb
orh
ood
has be
en
ske
t
ched.
The
r
e
are
limitation
s
of
the inhe
rent
optical flo
w
field metho
d
d
ue to Ho
rn a
nd Sch
u
n
ck [
6
] has b
een
remove
d. Th
e
prop
osed im
a
ge segm
entat
ion p
r
o
c
e
ss
h
a
s b
een
im
proved the time
efficien
cy an
d effectivene
ss
of the optical
flow to com
p
ute by fuzzy theory
con
c
ep
ts [32]. The smoothne
ss constraints
ha
ve
been
in
corpo
r
ated
in th
e
optical
flow
computat
ion
method
s to
regula
r
ize the
way
by m
e
a
n
s
of
smooth
n
e
s
s i
n
tegratio
n of
data. The
sm
oothne
ss
co
h
e
ren
c
e
con
s
traints
su
ppo
se that the flyi
ng
Mosq
uito in
t
he m
o
tional
i
m
age
are
structurally
inte
gral
and
smo
o
th. The
opti
c
al flo
w
ba
se
d
flying or m
o
ving pattern
of the Mo
sq
uito is
a seq
uen
ce of tim
e
in o
r
de
red
image
s
cau
s
e
s
seq
uential v
a
riation
s
a
r
e
exclu
s
ively due to
ima
g
e
motion
s. T
hen the
se
q
uen
ce of im
age
s
indicates for t
he e
s
timation
of 2
D
di
screte imag
e di
spl
a
cem
ents o
r
velocitie
s
. It can b
e
referre
d
to as the optical field or M
o
sq
uito imag
e velocity
filed. The intensi
t
ies of the Mosquito ima
ge and
its de
rivatives a
r
e
used t
o
explain
th
e motion
al
fl
ying pa
ram
e
ters of the
M
o
sq
uito’s ima
ge
scene
s. Th
e
optical flo
w
e
x
plains t
w
o
mech
ani
sm o
f
the motiona
l flying Mosq
uito of a
regi
on
feature
such as probo
scis,
head, throat
,
abdome
n
or
wing in the im
age [31, 32].
Suppo
se that
,
is the image
intensity [5],
then
,
,
(
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 77
98 – 780
7
7800
Whe
r
e
is the displaceme
n
t
of the local image regi
on at
,
after time
. Taylor expansio
n
of the left hand side
can ex
pre
s
sed a
s
,
,
,
(
2
)
Whe
r
e
,
and
are respe
c
tively, the order pa
rtial de
rivatives of the Mosquito
image inte
nsity function
,
with respe
c
t to the sp
a
c
e fo
r Mo
sq
uito and tim
e
and
rep
r
e
s
ent
s th
e highe
r o
r
d
e
r term
s
whi
c
h
can
b
e
el
iminated a
n
d
the above
Equation (2)
can
rew
r
ite a
s
,
0
(
3
)
Or,
0
i.e.
0
(
4
)
Whe
r
e,
,
is the velocity of the Mosquit
o
image. The above equ
ation is the OFC
equatio
n [5]. The velo
city dire
ction
of
the Mo
squito
image i
s
th
e local
slop
e
of the inten
s
ity
function
ca
n
be
comp
uted
. It can b
e
re
ferre
d a
s
the
ape
rture
p
r
o
b
lem. Th
e e
s
timation of th
e
motion at Mo
squito i
m
age
locatio
n
is
en
ough i
n
ten
s
ity. The velocit
y
of the Mosquito imag
e
can
be determin
ed by the Equation (4).
To sati
sfy,
it need
s so
m
e
crite
r
ion
such a
s
unifo
rm
illumination,
Lambe
rtian surface refle
c
t
ance,
and transl
a
tion mo
tion parall
e
l to the Mosq
u
i
to
image pla
ne.
It can be
used a
regul
ari
z
ation te
rm t
o
app
roximat
e
the motion
s of the n
e
ig
hbori
n
g
regio
n
s with
Mosq
uito’s m
o
tion. It has
been i
n
tro
d
u
c
ed
a gl
obal
smooth
n
e
s
s
con
s
trai
nt [6] to
formali
z
e the
image flo
w
con
s
traint [7]. It has b
een g
ene
rat
ed an i
n
com
p
lete correlat
ion
betwe
en the
Mosq
uito’s
m
o
tional d
o
mai
n
and th
e im
age inte
nsity
domain. T
he
error te
rm of
the
con
s
trai
nt ha
s bee
n define
d
[5] by,
(
5
)
Whe
r
e,
,
is the Gau
s
s-Seid
el equation in
the domain
D,
is the term of weight, and
is the er
ror te
rm.
The o
p
tical
flow
based t
e
ch
niqu
es
h
a
s
anoth
e
r
probl
em. Thi
s
te
chni
que
is hi
ghly
sen
s
itive toward
s to the
smaller i
n
ten
s
i
t
y variat
ions i
n
the con
s
e
c
utive Mosq
uito image
s fra
m
e.
In the in
coh
e
r
ent o
p
tical
flow fiel
d regi
ons, th
e Mo
squito a
c
tually
ca
nnot
be d
e
tected
in th
e
movement
situation. Th
ese
uneven
ve
ct
ors are contri
buted to by m
i
nor va
riation
s
a
s
chan
ge
s in
illumination, li
ttle movements in t
he background, and use
the differential eq. in
the optical field
for computati
on. Thu
s
, the
OFF
comp
rises a
n
allo
cati
on of in
coh
e
rent flow
regio
n
s in th
e sen
s
e
of the intensity variations and co
here
n
t flow r
egio
n
s are actu
a
lly significant
travels of the
Mosq
uito in
motional pi
ct
ure. Thi
s
can
be di
st
ribute
d
as
a map
of motion ve
ctors. The
m
ap
indicated as an
inte
nsity structu
r
e,
it ca
n
be
rep
r
ese
n
ted a
s
a
bin
a
ry ima
g
e
scene
whi
c
h
ca
n be
comp
ri
sed a
s
the coh
e
ren
t
and incohe
rent flow fiel
ds. The i
n
te
nsity ca
n be
dark level
s
and
whi
c
h is directly proportion
al to the degree of mo
vement pictured b
y
the motion
vectors. Thu
s
in
the co
herent
situation fo
r t
he probl
e
m
o
f
tracking
of the moving
or flying Mosq
u
i
to in a motio
n
scene eli
m
ina
t
e probl
em of Mosq
uito ima
ge se
gm
e
n
tation. The cohe
rent an
d inco
here
n
t regi
on
s
are the re
pre
s
entatio
n of the motion ve
ctors fo
r the segm
entation
of the intensities [7]. The
segm
ented
m
o
tion
scene
can b
e
u
s
ed
for extractin
g
the cohe
rent
regio
n
s. It
ca
n be
depi
cted
in
the Mosq
uito’
s
motional pi
cture.
In [33], the
membe
r
ship
function ha
s
been defin
ed
as
,
1,2
,
3,
…
,
whe
r
e the
fuz
z
y
s
e
t,
,
,
,…
,
.
lie
s in
0
,1
. This function refle
c
ts the degre
e
of control of
the rudim
ent
s within the
fuzzy set. The high
er
value of a compon
ent, the greate
r
is the
contai
nment
of the compo
nent withi
n
th
e fuzzy set a
nd lo
wer valu
e indi
cate
s a
wea
k
e
r
cont
rol
[32]. Accordi
ng to Zade
h’
s notation [3
3], a fuzzy set
M
, then the modified
equatio
n ca
n
be
expre
s
sed a
s
,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Mosq
uito Tra
cki
ng b
y
Im
a
ge Segm
enta
t
ion of Optical
Flow Field (Jahan
gir Alam
SM)
7801
∑
,
1
,
2
,
3
,…,
(
6
)
Whe
r
e,
n
is the numbe
r of the com
pone
nts in t
he fuzzy set
and
∑
e
xpr
e
sses
a set of
comp
one
nts,
is the error o
f
the field.
(a)
(b)
(
c
)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
(k
)
(l)
(m)
(n)
(o)
(
p)
Figure 1. Pixel Intensity Information of
OFF
The O
FF
ca
n be vi
suali
z
ed a
s
a
2D fuzzy 0 o
r
1
intensity plot.
The
den
se
r
regio
n
i
s
darke
r in the
intensity plot.
But, in the IOFF regi
on
reflects
brig
hter an
d sparser inten
s
ity. The
degree
of CO
FF an
d IOFF
in the
co
mp
uted OF
F i
s
kno
w
n
by the
un
certai
nty in the i
n
tensit
y in
the neig
hbo
rhood
s of a fl
ow regio
n
. T
he di
stri
butio
n and
den
sity of the darker pixel
s
in t
h
e
comp
uted
OF
F de
pict th
e
attendan
ce
in
the n
e
igh
b
o
r
hood
of a
pa
rticular opti
c
al
flow
re
gion
[7].
A pixel neig
hborhoo
d ca
n be me
asured of the a
m
ount of ho
mogen
eity/heteroge
neity o
f
the
neighborhood of the
candi
date pixel
which i
s
i
n
the fuzzy hostility [32, 34]. T
he pixel
s
in t
h
e
neigh
borhoo
d
are
mo
re
h
o
moge
neo
us then the
fe
wer pixel
s
a
r
e in
the a
n
t
agoni
sm to
its
neigh
bors. F
r
om the correlation to the intensit
y structu
r
e of OFF that implies a de
nser
neigh
borhoo
d
that indicati
ng more CO
FF neig
hbo
rh
ood regio
n
. Ho
wever, a
pixel in an I
O
FF
neigh
borhoo
d
region i
s
more ant
ago
nism to its neigh
bors du
e to the gre
a
ter deg
ree
of
hetero
gen
eity. Suppose th
at, seco
nd-order nei
ghb
orhood g
eomet
ry, the antago
n
ism ind
e
x (
) of
the se
con
d
-o
rder nei
ghb
orhood
can b
e
defined a
s
[3
4],
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 77
98 – 780
7
7802
∑
(
7
)
Whe
r
e
is th
e ca
ndid
a
te
pixel and
,
1
,2
,
3
,
…
,8
is
the fuzzy member
ship value of
neigh
bors in the se
con
d
order neig
hbo
rhood fu
zzy subset and c is the fuzzy errors. The
li
e
s
in [0,1], maximum index is
1
, minimum index is
0
, the higher value of
then the higher
index in the candid
a
te pixel in the neigh
borh
ood.
Th
e
fuzzy index
of the pixel in a neighb
orh
ood
flow i
s
me
ntioned
of COF
F
or IOFF i
n
the compute
d
flow
regi
on
s
in a m
o
tion i
m
age
whi
c
h
can
be used bet
ween the COF
F
and IOFF.
In Figure 1 shows the di
st
ribution
of
the pixel intensity levels in 2
nd
order
neig
hborhoo
d
OFF re
gion, Figure 1a, an
d Figure 1b a
r
e not pe
rfect
l
y homogen
e
ous b
u
t Figure 1c, Figu
re
1d,
and Fig
u
re
1
g
are p
e
rfe
c
tl
y homoge
ne
ous
whi
c
h h
a
s
lea
s
t index
0
, Figure 1
e
, Figure 1f,
and Figu
re 1h are perfe
ct
ly heterogen
eou
s and the index is
1
, Figure 1i
-1p shows the
threshold
[35
]
regio
n
of h
o
moge
neity/heterog
eneity
at
0
.
5
. Thu
s
the
thre
shol
ding
at and
above
0
.
5
homo
gene
ou
s optical flow re
gio
n
s. Ho
wever,
OFF regi
on
s comp
ri
se onl
y darke
r
regio
n
s i
ndi
cating the h
o
m
ogen
eou
s
region
s that a
r
e the
esse
n
t
ial logical re
gion
s for
act
ual
motions [7] b
e
twee
n the COFF and IOF
F
.
3.
Proposed Algorithm
The Mo
sq
uito
detectio
n
an
d tra
cki
ng by
usin
g fuzzy th
eory h
a
s
bee
n se
gmente
d
of OFF
to accompli
sh
ed in Figu
re 2
that has presented in the followin
g
way
s
,
a)
Time-o
rd
ere
d
image fra
m
e
s
can be
extrac
ted from a
video se
que
n
c
e u
s
ing
stan
dard
libra
ry routine
s
and the
n
the extracted
M
o
sq
uito for the comp
utatio
n of OFF.
b)
The OFF b
e
twee
n the initial two imag
e frame
s
whi
c
h
can be
com
puted for inte
nsity
betwe
en Mo
squito image frames. Th
e O
FF is
co
mput
ed by usin
g Equation (4)
along
with Horn
an
d Schu
nck
re
gulari
z
atio
n term
whi
c
h h
a
s
given i
n
Eq
uation (5). Th
e flow
vectors are
measured at
each pixel lo
cation al
ong
with the X an
d Y dire
ction
s
put in
to the Mo
squ
i
to image vel
o
citie
s
. It can
be summa
ri
zed
as
(i)
de
nse
r
flow re
g
i
ons
corre
s
p
ondin
g
to the m
o
ving or flying Mo
squito
and
(ii)
sp
arser flo
w
region
s
equivalent to
the ambient
luminan
ce c
hang
es b
e
tween fram
es
and any noi
ses. It
may move stealthily in duri
ng the attain
ment of the video sequ
en
ce.
c)
It can be
det
ermin
ed the
optical flo
w
o
n
ROI u
s
in
g
pixel antago
n
i
sm ind
e
x wh
ich i
s
useful
seg
m
e
n
tation of the
OFF into CO
FF and IOF
F
regio
n
s i
s
th
e most
signifi
can
t
pha
se fo
r th
e re
moval of
the IOFF
re
gion
s an
d
su
bse
que
nt re
moval of the
COF
F
regio
n
s. It ca
n be
co
nsid
ered the
com
p
u
t
ed OFF to
b
e
a fu
zzy inte
nsity map
of IOFF
or COFF regi
ons, the fu
zzy antagoni
sm
index
of ea
ch flow
regi
o
n
neigh
bo
rho
o
d
pixel is
comp
uted by u
s
in
g Equation
(7) an
d then t
he OFF i
s
th
reshold
ed [3
5] at a
antago
nism i
ndex whi
c
h value of
0
.
5
. It c
o
rrespon
ds to the outer line betwe
en
the IOFF and COFF di
stribution
s
. All of fuzzy indexes a
r
e
0
.
5
that can be
comp
ri
sed
th
e filtere
d
whi
c
h i
s
out f
r
o
m
the flo
w
field [7], [35]. I
t
can
b
e
trea
ted a
s
rand
om
distu
r
ban
ce
s of in
coh
e
re
nt regi
ons
in th
e
O
FF which a
r
e
ra
rely
clu
s
te
red
jointly. Where
a
s, a pixel
wh
ich i
s
anta
g
o
n
istic
suffici
e
n
t in its nei
gh
borh
ood. It can be
unspe
cified to be a part of the incoh
e
rent re
gion
s in the OFF which i
s
re
moval
method
outco
me in the
co
here
n
t opti
c
al
flow
regio
n
s
only. It can b
e
mad
e
cohe
rent
regio
n
s fo
rm the ROI
s
for furthe
r cal
c
ul
a
t
ion of the OFF of the video succe
s
sion [
36].
d)
The S
R
OI on
ROI by d
e
tect
ing hig
h
e
s
t d
ens
ity opti
c
al
flow regio
n
s
by a pixel
position
(x, y)
of inte
rest on th
e ex
tracted
co
he
rent ROI
s
for
able to Mo
sq
uito dete
c
tion
and
its position tracking. The p
o
sition can b
e
la
id on the moving or flying ROI. At fir
s
t, the
ROI mu
st be minimum
size whi
c
h
should b
e
sp
atial coh
e
re
n
c
e con
s
traint
and
se
con
d
ly the pixel position
on the ROI m
u
st be the ma
ximum neigh
borh
ood atte
ntion
of optical
flow. The
sh
arp noi
se in
th
e OFF
must
be in th
e h
o
stility index
based
filtering pro
c
ess. The SROI on ROI of Mosquito
image whi
c
h have maximum
absolute [7, 3
1
] OFF. It me
ans that the t
r
ac
king th
e fa
stest m
o
ving
or flying
regi
o
n
. In
the first
one
can a
c
hieve
d
i
n
neig
hbo
rh
o
od OF
F atten
t
ion of ea
ch
p
i
xel is m
easu
r
ed.
Ho
wev
e
r,
m
n
neighbo
rho
od a
v
eragin
g
of the OFF is do
ne for minim
u
m size
(
3
3
) in cohe
rentl
y
for flying Mosq
uito. Thu
s
, the summati
on of the ab
solute value of
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Mosq
uito Tra
cki
ng b
y
Im
a
ge Segm
enta
t
ion of Optical
Flow Field (Jahan
gir Alam
SM)
7803
OFF v
e
ct
or
s in it
s
33
in Figure 1
c
to Figure 1l or
2
2
in
Figure 1m to Figure 1p
neigh
borhoo
d
in the X an
d
Y dire
ction
s
. At last, the pi
xel with the
maximum val
ue of
comp
uted
su
mmation is ta
ken a
s
SROI on a ROI. It can be e
s
timat
ed in a SROI
on a
desi
r
ed
ROI
even in tho
s
e motion ima
ges. It ca
n b
e
com
p
ri
sed
multiple movi
ng or
flying Mosquit
o
with
differe
nt velocitie
s
o
f
the Mo
squit
o
be
ca
use
of
the faste
r
spe
edy
Mosq
uito turn
out stron
ger
OFF whi
c
h i
s
applied fo
r the high
est op
tical flow st
re
ngth
to pertain the
high-sp
eedy
Mosq
uito.
Figure 2. Mosquito Dete
ctio
n and po
sitio
n
Tra
cki
ng Algorithm
e)
Optical flo
w
can b
e
com
p
uted in the
neighb
orh
ood
of SROI in subsequ
ent image
frame
s
conta
i
n the
movin
g
region
s th
at ca
n
con
s
ti
tute the
ba
ckgrou
nd. A
small
positio
n o
r
S
R
OI
can
be
tracked
on
a
m
o
vi
ng regio
n
whi
c
h i
s
equi
valent to tracking
the wh
ole
re
gion
whi
c
h
can redu
ce th
e time
compl
e
xity of the tracking te
ch
ni
que.
Then
it can
b
e
obtai
ned
th
e flying o
r
m
o
ving Mo
squit
o
. It is
noted
that time-orde
r
ed
frame
s
ca
n b
e
pro
c
e
s
sed
to track whi
c
h can b
e
esti
mated the sp
atial coh
e
re
n
c
e
con
s
trai
nt an
d it can be
ensure
d
the
int
egrity of the moving o
r
flying Mosq
uito
SROI. The firs
t two time ordered frames
OFF an
d a SROI on the moving or flying
Mosq
uito for dete
c
ting a
n
d
po
sition tra
cki
ng. If the
flying or m
o
ving Mo
sq
uito
is
detecte
d an
d
the po
sition i
s
tra
c
ked i
n
the nei
ghb
orh
ood p
o
int tha
t
con
s
titutes t
he
Removal of time-
o
rdere
d
Mosquito image frames
Measure the op
tical flow
bet
w
e
en
consecutive tw
o image frames
Estimate the OF
F regions of R
O
I
b
y
using pixel an
tagonism index
Detect and track
position of SROI
on ROI
in th
e hig
hest density of OFF re
gions
Determine the
op
tical flow
in the n
e
ighbor
hood
of S
R
OI in the subse
quent Mosquito
image frames
Start
Live Video Fram
es
Furthe
r processing for Ro
botic
actions (such as Destro
y
)
End
Ye
s
No
Ye
s
No
Mosquito
Detected?
Position
Tracked?
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 77
98 – 780
7
7804
ROI for
ope
ration in the
sub
s
e
que
nt frame
s
to b
e
pro
c
e
s
sed.
Tra
cki
ng of t
he
moving o
r
flying Mo
sq
uito
then syn
ony
mous to
shifting the
ROI in
agreeme
n
t
with
the OFF ca
n be dete
c
ted
SROI.
5.
Resul
t
s and
Discus
s
ions
The propo
se
d algorithm
has be
en ap
plied for det
ecting an
d p
o
sition tra
c
ki
ng of the
Mosq
uito. Th
e expe
riment
s have
be
en
con
d
u
c
ted by
the time
seri
es of th
e ima
ge fram
es [7],
[36] in the li
ve video
seq
uen
ce
s by G
i
gE 490
0C camera in
Fig
u
re
3 a
nd Fi
gure
4
whi
c
h
corre
s
p
ondin
g
the Mo
squit
o
image fram
es al
ong
with
the tracke
d S
R
OI. The al
g
o
rithm h
a
s b
e
e
n
impleme
n
ted
on Keyen
c
e
cont
rolle
r a
r
ound th
e d
e
tected S
R
OI.
In assum
p
tio
n
, the targ
et
ed
point of the
Mosquito
can
shoot by LASER
that cause
t
he M
o
squito
should
be destroyed
su
ccessfully. The mentio
ned alg
o
rith
mic techni
q
u
e
has l
e
ss
complexity. Table 1
sho
w
s the
comp
utationa
l times in se
cond
s req
u
ire
d
by
the pro
posed techni
que. The tabl
e sho
w
s evid
ent
for the
pro
p
o
s
ed t
e
ch
niqu
e. In the p
r
o
posed te
chni
que
whi
c
h
b
e
com
e
s mo
re famou
s
as the
numbe
rs, less time con
s
u
m
ing, and di
mensi
o
n
s
of
flying sequ
en
ces can ri
se. In finally, it can be
con
c
lu
ded th
at the pro
p
o
s
ed te
chni
qu
e out pe
rf
orms the o
p
tical flow comp
utation tech
n
i
que
whi
c
h takes i
n
to reflectio
n
the entire ima
ge frame
s
.
(a)
(b)
(c
)
(d)
Figure 3. Orig
inal Image
s F
r
ame
s
of the Mosq
uito
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Mosq
uito Tra
cki
ng b
y
Im
a
ge Segm
enta
t
ion of Optical
Flow Field (Jahan
gir Alam
SM)
7805
(a)
(b)
(c
)
(d)
(e)
(f)
(g)
(h)
Figure 4. Det
e
ction a
nd Po
siti
on Trackin
g
of the Mosq
uito
Table 1. Posit
i
ons a
nd their Execution Ti
me
Position (x,
y
)
Time elapsed 1
Time elapsed 2
Time elapsed 3
Avg. Time (ms)
1
(187,133.
5)
0.0345326
0.0340094
0.0343725
0.034305
2
(180,230.
5)
0.034277
0.0342373
0.0336188
0.034044
3
(169,368.
5)
0.0348935
0.0347315
0.0335772
0.034401
4
(180,246
)
0.0327388
0.0308947
0.0314708
0.031701
6.
Conclu
sion
Mosq
uito det
ection
and p
o
sition tracki
ng algo
ri
thm throug
h
the computation o
f
Optical
Flow
Field
h
a
s
bee
n expl
ained
in thi
s
resea
r
ch p
a
p
e
r. Th
e m
a
in
aim
of this
pape
r h
a
s be
en
prop
osed to
segm
ent the
optical flow
field re
gion
s comp
uted be
tween
succe
ssive
Mo
squi
to
image f
r
ame
s
of live vide
o
seq
uen
ce
s i
n
to co
herent
o
p
tical flo
w
fiel
d an
d in
co
herent opti
c
al flo
w
field by u
s
ing
fuzzy in
dex of
the flow
re
gi
ons. Ti
me
red
u
cin
g
to attai
n
by the
sele
ction of
a regi
on
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 77
98 – 780
7
7806
of intere
st an
d a spe
c
ific
region
of interest
am
ong th
e co
herent o
p
tical flo
w
fie
l
d and
sp
ecifi
c
regio
n
of inte
rest
com
putat
ion in o
p
tical
flow
field follo
wed
by se
gm
ented o
p
tical
flow field ove
r
sele
cted
nei
g
hborhoo
d of
spe
c
ific regio
n
of int
e
re
st. This
techniqu
e
is
efficie
n
t and effective of
fast detectio
n
and tra
ckin
g
pro
c
e
ss.
Ackn
o
w
l
e
dg
ements
Wa
rm exp
r
e
ssi
on
and
sin
c
ere tha
n
k
s to F
u
zh
ou
Jufen
g
Elect Ltd.
Co., 26
8
Gushan
zh
ou,
zho
u
tou, T
ong
kou
cu
n, Mingho
ujin
g
x
izhen
g, Fu
zhou, China
esp
e
ci
ally to Li
Zhang
don
g for hi
s supp
ort to a
c
co
mplish th
e e
x
perime
n
t an
d sp
eci
a
l th
anks to Xia
m
en
University, China 86
3 pro
j
ect (2
014AA
Q
002
83), Pr
oject Numb
e
r
2,Chi
na for supp
ortin
g
the
proje
c
t.
Referen
ces
[1]
Cox
IJ, Hingorani SL. An
effi
cient imp
l
em
en
tation of re
ids
multipl
e
h
y
p
o
th
esis trackin
g
a
l
gorithm a
n
d
its evalu
a
tio
n
for the purp
o
se
of visual tracki
ng.
IEEE trans. Pattern Anal
. 199
6; 18:13
8–
150.
[2]
Etoh M, Shirai
Y. Segmentation
a
nd 2D m
o
tion estim
a
tio
n
b
y
re
gio
n
fragments. In: Procee
din
g
s o
f
Internatio
na
l C
onfere
n
ce o
n
Co
mp
uter Visi
on
. 199
4; 192-
199.
[3]
W
ang JYA, A
d
elso
n EH.
Repr
esenti
ng m
o
vi
n
g
ima
ges
w
i
th
l
a
yers.
IEEE Trans. Image Process
. 19
94
;
3(5): 625-
63
8.
[4]
Dunc
an JH, C
hou T
.
On
the detection
of mo
tion a
nd th
e computati
o
n
of optical flo
w
.
IEEE Trans.
Pattern Anal.
1
992; 14(
3): 346
-352.
[5]
Beauc
hemi
n
S
S
, Barron JL. T
he comp
utatio
n of optical.
ACM Comput. Sur
v
.
1995; 27(
3): 433-
467.
[6]
Horn BKP, Schunk BG.
Determining optic
al flo
w
.
Artif
. 1981
; 17: 185-2
04.
[7]
Sing
h A. A
n
estimati
on-th
eoretic
fr
ame
w
o
r
k for
ima
ge-flo
w
com
p
utation.
I
n
: P
r
ocee
din
g
s
of
Internatio
na
l C
onfere
n
ce o
n
Co
mp
uter Visi
on.
199
0; 168-
177.
[8]
Sidd
harth
a Bh
attachar
yya, U
j
j
w
a
l
Ma
ulik, P
a
rama
rth
a
Dut
t
a, High
–Sp
e
ed targ
et tracking
b
y
fuz
z
y
hostilit
y-i
n
d
u
ce
d segm
entati
o
n Comp
utin
g o
f
Optical flo
w
fi
eld.
Ap
pli
ed S
o
ft Comp
utin
g
. 200
9; 9: 12
6-
134.
[9]
Giachetti A, Campa
n
i M, T
o
rre V.
T
he
use
of optical flo
w
for road nav
i
gatio
n.
IEEE Trans. Robot.
Atom
. 19
98; 14
(1): 34-48.
[10]
Barron J
L
, F
l
eet DJ, Bea
u
c
hemi
n
SS. Systems
a
nd
e
x
p
e
rime
nts: p
e
rformanc
e of
optica
l
flo
w
techni
qu
es.
Int.
J. Com
p
ut. Vision
. 19
94; 12(
1
)
: 43-77.
[11]
Uras S, Girosi F, Verri A,
Torre V.
A Co
mputatio
nal
ap
proac
h to moti
on p
e
rcepti
on.
Biol. Cybern
.
198
8; 60: 79-9
7
.
[12]
F
l
eet DJ, Jeps
on AD. C
o
mp
u
t
at
ion of c
o
mp
one
nt ima
ge v
e
locit
y
from
loc
a
l p
hase
infor
m
ation.
Int. J.
Com
p
ut. Vision
. 1990; 5: 77-1
04.
[13]
Szeliski R. Ba
ysian Mo
del
in
g of
Uncertai
nt
y
i
n
Lo
w
-
leve
l Vis
i
on.
Kluw
er, No
rw
ell, MA
. 1989.
[14]
Chin T
M
, Karl W
C
, W
illsky
AS. Seque
ntia
l f
ilterin
g for multi-frame vis
ual rec
onstruc
tion.
Si
g
nal
Process
. 199
2;
28: 311-3
33.
[15]
Martinez
JC, V
a
ldiv
ia
JF
. A n
e
w
appr
oac
h t
o
moti
on
patte
rn rec
ogn
itio
n
and
its a
p
p
lica
t
ion to
o
p
tica
l
flo
w
estimation.
IEEE
Trans. Sys. Man Cybern. C: Appl. Rev
. 2007; 37(
1): 39-5
1
.
[16]
Kara
yi
ann
is N
B
, varughes
e B,
T
ao G, F
r
o
s
t JD
,
W
i
se MS, Mizrahi EM. Quantif
y
i
ng m
o
tion i
n
vide
o
recordi
ngs
of
neo
nata
l
seiz
u
r
es b
y
r
egu
lari
zed o
p
tica
l flo
w
m
e
tho
d
s.
IEEE Trans. Image Proc
ess
.
200
5; 14(7): 89
0-90
3.
[17]
Mitiche A, Sek
k
ati H. Optical
flo
w
3D s
egme
n
ta
tion
and
int
e
rpretati
on: a v
a
riati
on meth
o
d
w
i
th activ
e
curve evo
l
utio
n
and lev
e
l sets.
IEEE Trans. P
a
ttern Anal
. 2
0
06; 28(1
1
): 181
8-12
9.
[18]
Had
a
mard
J. L
e
ctures
on t
he
Cauc
h
y
Pro
b
le
m in
Li
near
Par
t
ial D
i
fferenti
a
l
Eqs.
Ya
le
Un
i
v
e
r
si
ty Pre
ss,
New
Haven, C
T
. 1923.
[19]
Nag
e
l H
H
, E
n
kelm
ann
W
.
An i
n
vestig
a
t
ion
of sm
oot
hness c
onstra
i
nts for the
estimatio
n
of
displ
a
cem
ent vector fields fro
m
seque
nces. IEEE
T
r
ans.
Pattern Anal
. 19
8
6
; 8: 565-5
93.
[20]
Nesi
P. Vari
at
ion
a
l
ap
proac
h for
optic
al
f
l
o
w
estimati
on
man
agi
ng
di
scontin
uities.
Im
ag
e Vi
si
on
Co
mp
ut
. 1993;
11(7): 419-
43
9.
[21]
W
oodh
am RI.
Multipl
e
lig
ht s
ource
optic
al
fl
o
w
.
In: Proc
eedings of
3rd IEEE International Confer
enc
e
on Co
mputer V
i
sion (ICCV
’
9
0)
.
1990; 42-
46.
[22]
Mitiche A, W
a
ng YF
, Agg
a
r
w
a
l
JK. E
x
p
e
ri
ments
in c
o
m
putin
g o
p
tical f
l
o
w
w
i
t
h
the
g
r
adi
ent-bas
ed
multi-comstrai
n
t
method.
Pattern Recognit
. 19
87; 20(2): 1
73-
179.
[23]
Verri A, Poggi
o
T
.
Motion field
and o
p
tical flo
w
:
q
ual
itative p
r
operti
es.
IEEE Trans. Patern Anal
. 19
89
;
11(5): 49
0-4
9
8
.
[24]
Verri, Girosi F,
T
o
rre V.
Diffe
rentia
l techn
i
qu
es for optical fl
o
w
.
J. Opt. Soc. Am
.
1990; 7: 912-
922.
[25]
Nesi P, De
lBim
bo A, Ben-T
zvi D. A r
obust al
gorithm for o
p
ti
cal flo
w
estima
tion.
Comput. Vision Graph.
Imag
e Process
:
Image U
nder
st.
1995.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Mosq
uito Tra
cki
ng b
y
Im
a
ge Segm
enta
t
ion of Optical
Flow Field (Jahan
gir Alam
SM)
7807
[26]
Indah
Ag
ustie
n
Sira
dju
d
d
i
n, R
a
mat W
i
d
y
a
n
to
M, Ba
sar
u
d
d
in
T
.
Particle F
ilt
er
w
i
t
h
Ga
ussi
an W
e
ighti
n
g
for Human T
r
acking.
T
E
LKOMNIKA Indon
e
s
ian Jo
urn
a
l of
Electrical E
ngi
neer
ing.
2
0
1
2
; 10(6): 14
53
-
145
7.
[27] Campani M, Verri A.
Computi
ng optic
al flow
from an ov
er
constrai
ned syst
em of li
near a
l
gebr
aic eqs
.
Proceedings of
3rd IEEE Internat
ional Confer
ence on Com
p
uter
Vision (ICCV’90). 1990; 22-26.
[28] Schunck
BG.
T
he
motio
n
c
onstrai
nt eq. f
o
r optic
al fl
ow
.
Procee
din
g
s
of 7th IEEE
Internati
ona
l
Confer
ence
on
Pattern Reco
g
n
itio
n. 198
4; 20-22.
[29] Schunck
BG.
Ima
ge flow
: funda
mentals
an
d future rese
a
r
ch
. Proceedings of IEEE Conferenc
e
on
Comp
uter Visi
on Pattern R
e
c
ogn
ition (CVP
R’85). 1
985; 5
60-5
71.
[30]
Nesi P, DelBi
m
bo A, Sanz JLC.
Multi-con
s
traints-bas
ed
optica
l
flow
estimati
on a
nd s
e
g
m
e
n
tatio
n
.
Procee
din
g
s
o
f
Internation
a
l W
o
rkshop on Comp
utation
a
l
Architecture
Machi
ne Perc
eptio
n
.
19
91
;
419-
426.
[31]
Jaha
ngir
Alam
SM, Hu Gu
oq
ing, C
h
e
ng
Ch
en. C
o
mpar
ati
v
e Stud
y, P
a
ttern A
nal
ys
is a
nd D
e
tectio
n
Algorit
hm of Mosqu
i
to.
T
E
LKOMNIKA Indon
esia
n Journ
a
l o
f
Electrical Eng
i
ne
erin
g.
201
4; Accepted.
[32]
Ai
yu W
a
n
g
, H
ong
xia P
an, H
u
ili
ng
Li
u. Stud
y
on F
a
ult F
e
a
t
ure E
x
tractio
n
of Hi
gh-Sp
ee
d Autom
a
tion
.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2013; 1
1
(10):
586
7-58
75.
[33] Z
adeh
LA.
Fu
z
z
y Set. Inform
.
Control
. 196
5; 8: 338-3
53.
[34]
Bhattachar
yya
S, Dutta P, Maulik
U. Bi
nar
y o
b
ject e
x
tracti
on
using
bi-d
irecti
ona
l self-or
gan
izing
ne
ura
l
net
w
o
rk (B
DS
ONN) arch
itec
ture
w
i
th f
u
zz
y c
onte
x
t se
n
s
itive thres
h
o
l
din
g
.
Pattern Analys
is
a
nd
Appl
icatio
ns.
2
007; 34
5-3
60.
[35]
Li Jin
hu, Ma ji
an ting. Improv
ed Alg
o
rithm o
f
the L
T
E Random Access P
r
eambl
e Detec
t
ion un
der th
e
High
Sp
ee
d C
ond
ition.
T
E
L
K
OMNIKA Ind
ones
ian
Jo
urn
a
l
of Electric
al
Eng
i
ne
eri
ng.
201
3;
1
1
(8):
464
6-46
50.
[36]
Qigui Z
h
a
ng, B
o
LI, Ni Z
hang.
Researc
h
on
Auto
matic T
a
rget
T
r
acking Ba
sed on Pl
atfor
m-lens Vi
deo
Sy
s
t
e
m
.
T
E
LK
OMNIKA Indon
esia
n Journ
a
l o
f
Electrical Eng
i
ne
erin
g.
201
2; 10(7): 158
2-1
587.
Evaluation Warning : The document was created with Spire.PDF for Python.