Int
ern
at
i
onal
Journ
al of
El
e
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
11
,
No.
1
,
Febr
uar
y
2021
, pp. 21
7
~223
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v11
i
1
.
pp217
-
223
217
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om
Detectio
n and tr
ac
kin
g o
f moving
object u
sing mod
ified
b
ac
kg
ro
un
d sub
t
raction a
nd
Ka
lm
an filt
er
Jeev
i
th
S
.
H
.
1
,
Lak
s
hmik
ant
h S
.
2
1
Resea
rch
Schol
ar,
Visv
esva
ra
y
a
Tec
hnologica
l U
nive
rsit
y
,
Indi
a
2
Depa
rtment of
ECE
,
Acha
r
y
a
I
nstit
ute of Te
chn
olog
y
,
Ind
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ja
n 1
7,
2020
Re
vised
Jun
22
,
20
20
Accepte
d
J
ul
12, 2
020
Moving
obje
c
t
d
et
e
ct
ion
and
tr
acking
(MO
DT)
is
the
m
aj
or
ch
alle
nging
issue
in
Com
pute
r
Vision,
which
play
s
a
vital
rol
e
in
m
an
y
app
li
c
at
ions
li
k
e
roboti
cs,
surv
e
il
la
n
ce,
nav
igation
s
y
st
ems
,
m
il
it
ar
ie
s,
e
nvironmenta
l
m
onit
oring
et
c
.
The
re
a
re
seve
r
a
l
exi
sting
t
ec
hni
ques,
which
has
bee
n
used
to
det
e
ct
and
tr
ac
k
the
m
oving
obje
ct
in
Survei
ll
a
nce
s
y
st
em.
Therefore
it
is
nec
essar
y
to
dev
el
op
new
al
gor
ithm
or
m
odifi
ed
al
gorit
hm
whi
ch
is
robust
to
work
in
both
d
a
y
and
nigh
t
ti
m
e.
In
thi
s
p
ape
r
,
m
odifi
ed
BGS
te
chn
ique
is
proposed.
The
vi
deo
i
s
first
conve
rte
d
to
num
ber
of
fra
m
es,
the
n
the
se
fra
m
e
are
appl
i
ed
to
m
odifi
ed
bac
kgr
ound
subtrac
ti
o
n
te
chni
que
wi
th
ada
pti
v
e
thre
shold
which
give
s
det
ec
t
ed
obje
ct.
Kalman
fil
te
r
t
ec
hn
ique
is
used
for
tra
ck
ing
the
detec
t
ed
obje
c
t.
Th
e
expe
rim
ental
r
esult
s
show
s
thi
s
proposed
m
et
hod
ca
n
eff
i
c
ie
ntly
and
cor
re
c
tly
d
etec
t
and
track
the
m
oving
obje
c
ts
with
le
ss
proc
essing
t
ime
which
is c
o
m
par
ed
with
exis
ti
ng
t
ec
hniqu
es
.
Ke
yw
or
d
s
:
Ba
ckgrou
nd s
ubtract
io
n
(B
G)
Com
pu
te
r
visi
on (
C
V)
Fr
am
e d
iffer
e
nc
ing
(FD)
Kalm
an
filt
er (KF)
Mov
i
ng ob
j
ect
detect
ion an
d
trackin
g
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Jeevit
h S
.
H
.
,
Re
search
Sc
hola
r,
Visv
es
va
raya
Tech
no
l
og
ic
al
Un
i
ver
sit
y
B
el
agav
i,
India
.
Em
a
il
:
j
eevit
hshbp
@g
m
ai
l.com
1.
INTROD
U
CTION
Vide
o
sur
veill
ance
is
the
m
os
t
sign
ific
a
nt
r
esearch
a
rea
in
CV
w
hich
is
us
e
d
to
ide
ntif
y
and
trac
k
m
ov
ing
obj
ect
s
in
vi
deo.
MO
DT
has
a
la
r
ge
var
ie
ty
of
ap
pl
ic
at
ion
in
c
ompu
te
r
visi
on
s
uc
h
as
c
om
pr
ess
ion
of
vid
e
o,
visi
on
base
d
con
t
ro
l,
visu
al
su
r
veill
ance,
hum
an
com
pu
te
r
inter
faces,
r
obotics
a
nd
m
edical
i
m
age
processi
ng
[
1
-
4
].
MOD
T
in
vid
e
o
is
an
i
m
portant
chall
en
ging
ta
sk
withi
n
the
area
in
C
V.
It
is
the
pro
cess
of
analy
zi
ng
vid
e
o
f
ram
es and
e
stim
ation
of po
sit
ion
of the object
, f
i
nally
f
in
d
the
obj
ect
s tr
ajecto
ry. There
are
3
basic
ap
proac
hes
in
obj
ect
detect
ion
,
t
he
y
are
fr
am
e
d
iffe
re
ncin
g
(
FD
)
,
O
ptica
l
flo
w
an
d
bac
kgr
ound
su
bt
racti
o
n
(B
GS
)
[2
-
4
]
.
FD
is
the
sim
ples
t
te
chn
i
qu
e
to
cal
culat
e
the
di
ff
ere
nce
betw
een
tw
o
co
ns
e
cutiv
e
fr
am
es.
It
can
hav
e
str
ong
fle
xib
il
it
y
fo
r
sud
den
e
nv
ir
onm
e
nt
change,
but
diff
ic
ulty
to
reco
gniz
e
the
ob
j
ect
’s
sh
a
pe
and
obj
e
ct
in
sta
ti
on
ary.
This
m
e
tho
d
i
s
no
t
accu
rate
[5
]
.
Op
ti
cal
flo
w
is
the
te
chn
iqu
e
wh
ic
h
is
us
ed
t
o
analy
ze
the
im
age
vecto
r
fiel
d
of
m
ov
ing
obj
ect
a
nd
poss
ess
cl
us
te
ri
ng
base
d
on
opti
cal
flo
w
distrib
utio
n
char
act
e
risti
cs
of
im
age.
Us
in
g
this
te
ch
niqu
e,
the
com
plete
inf
or
m
at
ion
of
a
n
obj
ect
m
ov
em
ent
an
d
obj
ec
t
detect
ion
from
backgro
und
c
an
be
obta
ined
bette
rly
.
But
it
has
m
or
e
co
m
plex
cal
culation
,
poor
in
a
nt
i
-
no
ise
perform
ance.
As
res
ult,
it
is
no
t
su
it
able
f
or
real
tim
e
app
licat
ion
s
[
6
-
7
].
BGS
m
e
tho
d
i
s
the
si
m
ple
tech
ni
que
to ide
ntify o
bje
ct
s in
m
otion
. I
t i
s v
ery se
ns
it
ive to
the ill
um
i
nation cha
nge
and also
has
p
oor
a
nti
-
inte
rf
e
r
ence
abili
ty
[
8
-
10
].
Param
et
ers
aff
ect
ed
on
E
xisti
ng
detect
ion
te
chn
i
qu
e
s
su
c
h
as
su
dde
n
change
in
obj
ec
t
m
otion,
scen
e
il
lu
m
inati
on
,
obj
ect
s
s
had
ow,
ca
m
era
cl
arit
y
in
m
otion
[
11
-
13]
.
Scene
il
lum
inati
on
cha
ng
e
is
a
chall
eng
e
for
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
2
1
7
-
2
2
3
218
BGS,
su
c
h
as
su
dde
n
c
ha
ng
e
in
cl
im
at
e,
nig
ht,
et
c
[14
-
18]
.
From
nu
m
ber
of
ye
ars,
va
ri
ou
s
BG
S
te
c
hniq
ues
hav
e
bee
n
pro
po
s
ed
by
m
an
y
r
esearche
rs
t
hat
m
a
in
fo
c
us
is
to
bu
il
d
m
or
e
eff
ect
ive
backgro
und
m
od
el
to
handle
it
[
20
-
25]
.
In
order
t
o
ov
erco
m
e
this
prob
le
m
,
m
od
ifie
d
BGS
is
use
d
in
this
paper
to
ac
hieve
higher
le
vel
com
pu
te
r
visio
n
-
base
d
syst
em
us
ing
kalm
a
n
f
il
te
r
(KF)
[
19]
.
Fir
st,
it
m
u
s
t
be
robu
st
agai
nst
sc
en
e
illu
m
i
nation
change.
Se
cond
,
it
should
avoi
d
detectin
g
m
o
ving
bac
kgroun
d
obj
ects
suc
h
as
m
o
ving
le
av
es
,
rain,
sn
ow,
and
sha
do
ws
ca
st
b
y
m
oving
obj
ects
[
17
-
18]
.
Th
e
fo
ll
owin
g
sect
ion
s
are
as
f
ollows:
Sect
ion
2
desc
rib
es
p
r
op,
s
ect
ion
3
desc
r
ibes
re
su
lt
s a
nd a
naly
sis an
d sec
ti
on
4 desc
r
ibes c
on
cl
us
io
n
.
2.
PROP
OSE
D MET
HO
DOL
OGY
The
pro
posed
blo
c
k
diag
ram
of
m
ov
in
g
ob
j
ect
detect
i
on
and
trac
king
i
s
sho
wn
in
Fi
gure
1.
T
he
flo
wch
a
rt
of
pr
opos
e
d
m
et
ho
dolo
gy
is
as
s
hown
in
F
ig
ure
2.
I
niti
al
ly
the
input
vid
e
o
is
ta
ken
f
r
om
database
.
Fr
am
es
can
be
extracte
d
f
ro
m
input
vid
e
o,
t
hen
m
od
ifie
d
BGS
is
a
ppli
ed
to
ide
ntify
the
obj
ect
m
ov
em
ent
by
us
in
g
a
dap
ti
ve
thres
ho
l
d.
Dete
ct
ion
of
obje
ct
s
is
reali
zed
by
subtract
in
g
ba
ckgr
ound
im
age
from
ever
y
f
r
a
m
e
o
f
v
i
deo u
si
ng
m
od
ifie
d
BGS
.
Figure
1. Pro
pose
d
blo
c
k diagr
am
u
sin
g
m
od
ifie
d
bac
kgr
ound s
ubtract
io
n
Fo
rm
ula for
Im
age subtracti
on
:
Subtrac
te
d Im
age =
I(x,
y)
–
B(
x,
y
).
=
{
;
≥
ℎ
0
;
<
ℎ
(
1)
Wh
e
re
I(x,
y)
i
s the
fore
groun
d fr
am
e and
B
is t
he
Ba
ck
gro
und fr
am
e. Ad
pt
th
is an
ad
a
ptive th
res
ho
l
d.
Wh
e
n
|
I(x,
y
)
–
B(x,
y)
|>
A
dpt
th
,
pix
el
s
are
de
fine
d
as
fore
gro
und
ob
j
ect
s
.
Si
m
il
arly
,
w
hen
|I
(x,y)
-
Bt
(x
,y)
|<
Adpt
th
,
p
i
xels
are
def
i
ned
as
ba
ckgr
ound.
Fi
na
ll
y,
the
resu
lt
is
a
detect
ed
obj
ect
in
f
or
e
gro
und
fr
am
e.
In
t
his
pa
pe
r,
the
bac
kgr
ound
im
age
al
ways
cha
ng
e
s
due
to
the
c
ha
ng
e
s
in
ob
j
ect
m
ov
em
ent
and
oth
e
r
env
i
ron
m
ent
dis
turb
a
nc
e.
So
we
nee
d
co
ntinu
al
ly
update
the
bac
kgr
ound
i
m
age.
Using
an
adap
ti
ve
thres
ho
l
d,
t
he
m
ov
e
m
ent
of
ob
j
ect
s
ca
n
be
detect
ed
correct
ly
unde
r
sud
de
n
cha
ng
e
in
il
lum
i
nation.
Mor
phologica
l
operati
on
is
a tec
hn
iq
ue
of
e
xtracti
ng o
bjec
t
from
fr
am
e
that
ar
e u
sef
ul
i
n
the obj
ect
sh
a
pe
a
nd
represe
ntati
on
and
re
gion
s
ha
pe
descr
i
ption,
su
c
h
as
boun
dar
ie
s,
sk
el
et
ons,
a
nd
s
o
on.
It
rem
ov
es
no
ise
in
detect
ed o
bj
ect
b
y
dilat
ion
a
nd e
ro
si
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Detect
ion
an
d t
ra
cki
ng
o
f m
ov
ing
obje
ct
u
si
ng mo
difi
ed
b
ac
kgr
ou
nd
…
(
J
eevi
th S
.
H.
)
219
Figure
2. Flo
w
char
t
of ob
j
ect
detect
ion an
d
t
rack
i
ng
2.1. K
alma
n fi
lt
er
Af
te
r
the
m
orpholo
gical
ope
rati
on,
ta
rg
et
obj
ect
can
be
tracke
d
us
in
g
Kalm
an
filt
er
in
F
igure
2
.
It
can
be
a
st
r
ai
gh
t
ideal
sta
nd
i
ng
est
i
m
at
i
on
ap
proac
h
t
hat
is
th
ought
al
ong
a
bout
th
e
pr
eem
inent
r
eal
iz
ed
hyp
oth
esi
s
c
ha
nn
el
s
pec
ulati
on
s
[
2][7
]
.
Sta
nd
i
ng
co
ndit
ion
ca
n
be
a
str
ai
gh
t
ou
tl
ine
of
w
k
,
u
k−1
a
nd
x
k−1
.
Ob
se
r
vation
c
onditi
on
m
igh
t
be
a
direct
de
li
neati
on
of
x
k
a
nd
v
k
.
A
powe
rful
m
od
el
is
giv
e
n
sta
nd
i
ng
conditi
on and
per
ce
ptio
n
c
onditi
on
t
hro
ugh t
he
so
li
d
est
im
at
ion
redresse
d by est
im
ati
ons
[2
].
Kalm
an
channel
(m
et
ho
dical
)
st
and
i
ng con
diti
on is
pr
inte
d
as
pur
s
ue:
11
xu
x
A
B
kk
kk
w
(2)
Ob
se
r
vation eq
uation i
s
def
i
ne
d
as
foll
ow
:
k
k
x
k
z
H
v
(3)
In
the
highe
r
than
recipes:
v
k
,
z
k
,
A,
H,
w
k
,
u
k−1
,
v
k
is
the
sta
nd
i
ng
vect
or,
the
pe
rce
ption
vecto
r,
the
sta
ndin
g
c
hange
netw
ork
,
the
perce
ptio
n
f
ram
ewo
r
k,
the
f
ram
ewo
r
k
com
m
otion
ve
ct
or
,
the
fr
am
ewor
k
the
boar
d
vector,
the
per
ce
pt
ion
cl
am
or
ve
ct
or
,
se
ve
rall
y.
w
k
an
d
v
k
zo
ne
unit
ex
pect
ed
to
fu
l
fill
po
sit
iv
e
un
m
ist
akab
le
,
par
al
le
l
an
d
ra
ndom
,
zero
m
ean
re
searc
her
rack
et
vecto
r;
k
ca
n
be
a
s
ubscri
pt;
w
k
a
nd
z
_k
te
rr
it
or
y
unit
happ
y:
T
E
w
0
,
c
o
v
w
E
(
w
w
)
Q
(4)
TT
()
E
v
0
,
c
o
v
v
E
v
v
(
R,
E
w
v
0
)
(5)
k
R
n
is
sk
et
che
d
out
beca
us
e
of
the
past
sta
ndin
g
est
im
a
ti
on
got
f
r
om
st
and
i
ng
pro
gre
ss
conditi
on
rig
ht
now
of
k
-
1,
x
k
is
pr
inted
be
cause
of
the
ba
ck
sta
ndin
g
est
i
m
ation
joins
the
est
i
m
a
ti
on
s
rig
ht
now of k
. T
he deviat
io
ns
unit
of esti
m
a
ti
on
app
ea
re
d
in
cond
it
io
n (6) a
nd cond
it
io
n (7):
k
kk
e
x
x
(6)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
2
1
7
-
2
2
3
220
k
kk
e
x
x
(7)
The
pr
io
ri
an
d
back
est
i
m
at
i
on
de
viati
on
c
o
-
var
ia
nce
c
onditi
on
s
ca
n
be
char
act
e
rized
a
s
co
nd
it
io
n
(8)
a
nd con
diti
on (9)
:
kk
k
T
p
[
e
e
]
E
(8)
kk
T
k
[e
e
pE
]
(
9)
We
s
hould
se
e
sta
nd
i
ng
|a
sta
nd
i
ng
}
est
i
m
at
ion
conditi
on
x
k
that
as
certai
n
the
ba
ck
sta
ndin
g
est
i
m
ation
in
this
m
ann
er
on
prom
pt
the
Kalm
an
channel
conditi
ons.
It
ne
eds
the
com
pu
t
at
ion
eq
uatio
n
of
x
k
is
that
the
strai
gh
t
m
ix
of
f
rom
the
earli
er
est
i
m
ation
an
d
weig
hted
qu
al
i
ficat
ion
betwe
en
ob
vious
est
im
at
ion
s
and
est
im
at
ed
figure
est
ee
m
.
the
fo
ll
ow
ing
f
oreca
st
a
nd
update
c
onditi
ons
f
r
om
the
Kalm
an
channel
hypothesis
unit
of esti
m
a
ti
on
go
t.
F
or
ecast
c
onditi
ons
un
it
of estim
at
ion
m
ade open as
purs
ues:
k
k
1
k
1
x
A
x
B
u
(10)
T
k
k1
p
A
P
A
Q
(11)
Eq
uations de
fi
ned h
e
re:
kk
1
TT
k
K
p
H
H
p
H
R
(12)
k
k
1
k
k
k
x
x
)
KH
(
z
(13)
kk
k
p
I
K
H
(
)
p
(
14)
w
he
re
K
k
, x
k
, p
k
, I
is t
hat the
Kalm
an
gain n
et
work, ideal c
hannel
worth
, c
hannel
dev
ia
ti
on fram
ewo
r
k,
corner
to cor
ner
lat
ti
ce.
3.
RESU
LT
S
AND A
N
ALYSIS
The
p
r
opos
e
d
work
is
im
plem
ented
in
Ma
tl
ab.
Th
e
tw
o
vi
deo
dataset
s
are
us
e
d
in
the
pro
po
se
d
work.
It
prov
i
des
gr
ound
-
tr
ut
h
m
ark
ing
s
a
nd
a
nnotati
on
with
va
rio
us
a
tt
ribu
te
s
li
ke
oc
cl
us
io
n,
backgro
und
cl
utter,
r
otati
on,
S
cene
il
lum
inati
on
c
ha
ng
e
et
c.
on
vid
e
o
seq
uen
ce
s.
In
this
pa
per,
A
naly
sis
of
dif
f
eren
t
detect
ion
al
gorithm
is
do
ne
for
di
ff
e
ren
t
da
ta
set
s
us
in
g
Kalm
an
filt
er.
From
co
m
par
at
ive
res
ult
Table
1
,
accuracy
of m
od
i
fied
B
GS
m
et
hod
is m
or
e t
han o
t
her 2
m
e
thods.
Table
1.C
om
par
at
ive r
e
su
lt
of
al
l detec
ti
on
m
et
hods
for wal
king
dataset
Fra
m
e
Dif
f
erence
Op
tical
Flo
w
Back
g
rou
n
d
Su
b
traction
Mod
if
ied
Back
g
ro
u
n
d
Su
b
traction
Accurac
y
(
%)
1
5
.13
1
6
.77
9
2
.70
9
9
.27
IoU
72
66
74
76
Fra
m
e
per
seco
n
d
4
.38
e
-
01
4
.74
e
-
01
3
.79
e
-
01
3
.88
e
-
01
Proces
sin
g
ti
m
e
(s
ec)
1
8
0
.36
1
9
5
.5
1
5
6
.32
1
5
1
.3
Total n
u
m
b
er
o
f
Fr
a
m
es
412
412
412
412
False f
ra
m
e
r
ate
164
20
65
57
We
hav
e
quantit
at
ively
evaluated
our
detect
io
n
al
gorithm
s
us
in
g
par
am
et
ers:
Acc
ur
acy
,
In
te
r
sect
ion
ov
er
U
nion
(IoU)
,
Fr
am
es
per
s
econd
(F
P
S)
,
False
fr
am
e
D
et
ect
ion
(FD
)
r
at
e
.
Fig
ur
e
3
s
how
is
the
segm
ented
i
m
age
us
i
ng
Fr
am
e
diff
ere
nc
ing
te
ch
nique
,
Figure
4
s
ho
wn
in
t
he
seg
m
ented
i
m
age
us
in
g
op
ti
cal
flo
w
te
chn
i
qu
e
an
d
Fi
gure
5
s
how
n
in
the
se
gm
ented
im
age
us
in
g
m
od
ifie
d
BGS
te
chn
iq
ue.
F
or
night
dataset
,
this
propose
d
work
i
s
i
m
ple
m
ented,
w
hich
c
ould
extract
m
ov
ing
obj
ect
s
from
t
he
lo
w
il
lum
in
at
io
n
vid
e
o.
From
Tab
le
2
res
ult a
na
ly
sis
of
B
GS
i
s also d
on
e
usi
ng K
al
m
an
filt
er in ni
gh
t t
im
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Detect
ion
an
d t
ra
cki
ng
o
f m
ov
ing
obje
ct
u
si
ng mo
difi
ed
b
ac
kgr
ou
nd
…
(
J
eevi
th S
.
H.
)
221
Groun
dtr
uth
Track
i
ng im
age
Segm
ented
im
age
Figure
3. Se
gme
nted res
ult o
f f
ram
e d
iffe
rence
Groun
dtr
uth
Track
i
ng im
age
Segm
ented
im
age
Figure
4. Se
gme
nted res
ult o
f o
ptica
l fl
ow
Groun
d
tr
uth
50 Tim
e fr
am
e
of
vid
e
o
Segm
ented
im
age
Figure
5. Se
gme
nted res
ult o
f m
od
ifie
d
bac
kgr
ound s
ubtrac
ti
on
Table2
.
Com
pa
rati
ve
r
esult
of all
d
et
ect
io
n m
et
ho
ds f
or n
i
gh
t
dataset
Nig
h
t Video
datas
et
Fra
m
e
Dif
f
erence
Op
tical
Flo
w
Back
g
rou
n
d
Su
b
traction
Mod
if
ied
Back
g
ro
u
n
d
Su
b
traction
Fra
m
e
per
seco
n
d
1
.47
e+0
0
2
.47
e
-
01
3
.22
e
-
01
2
.60
e
-
01
Ov
erall
p
rocess
in
g
Ti
m
e
4
.05
e+0
2
1
.39
e+0
2
1
.82
e+.02
1
.47
e+0
2
Nu
m
b
e
r
o
f
Fr
a
m
e
378
1281
565
565
Nu
m
b
e
r
o
f
Fr
a
m
e
5
0
0
x
5
0
0
5
0
0
x
5
0
0
5
0
0
x
5
0
0
5
0
0
x
5
0
0
False f
ra
m
e
r
ate
22
29
80
180
4.
CONCL
US
I
O
N
Mov
i
ng
ob
j
ect
detect
ion
&
trackin
g
for
the
dynam
ic
env
iron
m
ents
espec
ia
ll
y
in
design
i
ng
of
vi
de
o
su
r
veill
ance
sy
stem
is
a
challen
gi
ng
ta
s
k
in
CV.
I
n
this
pa
per,
m
od
ifie
d
BGS
is
pro
po
sed
us
in
g
ada
ptive
thres
ho
l
d
for
detect
ion
w
hic
h
is
e
ff
ic
ie
ntl
y
identify
t
he
m
ov
ing
ob
j
ec
t
in
f
or
e
gro
un
d
us
in
g
bac
kgrou
nd.
In this
pro
po
se
d w
ork
, Kal
m
a
n fil
te
r
is use
d for
trac
king
of
m
ov
ing
obj
ect
.
Fr
om
resu
lt
e
va
luati
on
,
it
is
m
anifest
that
the
acc
ur
acy
of
pro
po
se
d
m
et
ho
d
is
99.
2%
,
w
hich
is
bette
r
than
e
xisti
ng
te
chn
iq
ues
a
nd
al
so
proces
s
ing
ti
m
e
is
1.
47
e+
02
w
hic
h
is
le
ss
than
existi
ng
te
c
hn
i
qu
e
s.
Ther
e
f
or
e
t
he
pro
po
se
d
m
et
h
od
is
rob
us
t
an
d
ada
ptable
to
work
in
day
and
night
tim
e
eff
ic
ie
ntly
.
In
fu
t
ur
e
pro
po
se
d
m
et
ho
d ha
s to
m
eet
the r
eal
-
ti
m
e req
ui
rem
ents.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
2
1
7
-
2
2
3
222
REFERE
NCE
[1]
S.
R.
Ba
la
j
i,
S.
Kart
he
y
an
,
“
A
surve
y
on
Movin
g
Objec
t
Tracki
ng
Us
ing
Im
age
Proce
ss
ing,
”
11
th
international
conf
ere
n
ce on
in
te
lligent syste
ms
and
con
trol
,
pp.
469
-
474,
2017.
[2]
Shridevi
.
S.
Va
seka
r,
Sanj
iva
ni
K.
Shah,
“
Background
Subtrac
ti
on
and
Ka
lman
Filt
er
Algor
i
thm
for
Objec
t
Tra
ck
ing,”
Int
.
c
onfe
renc
e
on
Rece
nt
Tr
ends
in
I
mage
Proce
ss
in
g
and
Pa
tt
ern
Recogni
ti
on
,
pp
.
19
4
-
202,
2019
.
[3]
H.S.G.
Supree
th · Cha
ndra
s
heka
r
M.
Pati
l,
”
Eff
ic
i
ent
m
ult
ipl
e
m
oving
obje
ct
de
tecti
on
and
tr
ac
kin
g
using c
om
bine
d
bac
kground
subt
rac
t
ion
and
c
lust
eri
ng”
,
Sign
al
,
I
m
age
and
Video
Proce
ss
ing,
vol
.
12,
no.
6
,
pp
.
109
7
-
1105
,
2018
[4]
A.
A.
Mali
k,
A.
Khali
l
,
and
H.
U.
Khan,
“
Obj
ec
t
Det
ec
t
ion
a
nd
Tra
cki
ng
usi
ng
Bac
kground
Subtrac
ti
on
and
Connec
t
ed
Com
ponent
La
be
li
ng
,”
Int
ernati
onal
J
ournal
of
Comp
ute
r A
pp
li
ca
ti
on
s,
vol. 75,
no.
13
,
2013
[5]
W
enc
hao
Li
u,
He
Chen
,
Long
Ma
,
“
Moving
obje
c
t
det
e
ct
ion
and
tra
cki
ng
b
ase
d
on
ZYNQ
FPGA
and
AR
M
SOC
,”
pp.
1
-
4
,
2
015.
[6]
Trupt
i
A.
Chop
kar
,
Shashikan
t
La
had
e,
”Real
Ti
m
e
Detect
io
n
of
Moving
Object
Based
On
FP
GA
,”
IOSR
J.
El
e
ct
ron.
Comm
un.
Eng
,
vol
.
11
,
no
.
1
,
pp
.
37
-
4
1,
2016
.
[7]
S.
Shanta
i
y
a
,
et
al
,
“
Multi
pl
e
obje
ct
tracki
ng
usin
g
kal
m
an
fil
te
r
a
nd
opti
ca
l
flow,
”
European
Journal
of
Adv
ances
in
Eng
ine
ering
a
nd
Technol
og
y,
vol.
2
,
no
.
2
,
pp
.
34
-
39,
2015
.
[8]
Za
ki
Y.
Abid,
T
hamir
R.
Saee
d,
and
Sam
ei
r
A.
Azie
z
,
”
Sel
ec
t
io
n,
Dete
c
ti
on
,
an
d
Tra
cki
ng
of
Vi
deo
obje
c
ts
Based
on
FP
GA
,
”
Internat
ion
al
jou
rna
l
of
Com
pute
r,
C
om
m
unic
a
ti
on
,
Control
and
s
y
st
em
Engi
ne
eri
n
g,
v
ol
.
15,
n
o.
1
,
2015
.
[9]
G.
Sindhura
Bharga
vi
,
B.
Prave
e
n
Kum
ar,
and
T
.
Kal
y
an,
“
Fast
Bac
kground
Subtraction
Algori
thm
for
Moving
Objec
t
Det
ec
t
ion
& Tracking in
F
PG
A
,”
Int. J.
So
f
tware
Har
dware
Re
s.
Eng
,
vo
l.
2
,
no
.
6
,
pp
.
65
-
7
0,
2014
.
[10]
Poonam
Gujra
th
i
R,
Arokia
Pri
ya,
P.
Mala
th
i,
“
Detect
ing
Moving
Objec
t
Us
ing
B
ac
kground
Subtrac
t
ion
Algorit
h
m
in
FP
GA
,”
Fourt
h
Inte
rnat
ional
Confe
renc
e
on
A
dvanc
es
in
Com
puti
ng
and
Comm
unic
ati
ons,
pp.
117
-
120,
2014.
[11]
Abutal
eb
M.
M,
Ham
dy
.
A,
Abu
el
wafa
.
M.
E.,
Saad
E.
M,
“
FPGA
-
base
d
obje
c
t
ext
ra
ct
ion
b
ase
d
on
m
ult
imodal
sigm
a
-
del
ta
b
ackground
esti
m
ation
,”
2
nd
Inte
rn
ati
onal
Conf
ere
nce
on
Comput
er,
Control
and
Comm
unic
ati
on
,
pp.
1
-
7
,
2009
.
[12]
M.
Saad,
A.
Ha
m
dy
and
M.
M.
Abutal
eb
,
“
FP
GA
-
Based
implementa
t
ion
of
a
Low
Cost
and
Area
Re
al
-
Ti
m
e
Motion
Det
ec
t
io
n
,”
15
th
In
te
rnat
ional
Conf
ere
nc
e
of
Mi
xe
d
D
esig
n
MIX
DES
,
pp
.
249
-
254,
2008
.
[13]
Butl
er
.
D,
Srid
har
an.
S,
Bove
V.
M.,
“
Real
-
ti
m
e
ada
p
ti
ve
bac
kground
seg
m
ent
at
ion
,”
I
E
EE
Int
ernati
on
al
Confe
renc
e
on
A
cousti
cs,
Spe
ec
h
,
and
Signa
l Proc
essing,
vol
.
3
,
pp
.
349
-
352
,
2003
.
[14]
Li
Q
-
Z
.
,
H
e
D
-
X.,
and
W
ang
B,
”
Eff
e
ct
iv
e
m
oving
objects
d
e
te
c
ti
on
b
ase
d
on
cl
uster
ing
b
ac
k
ground
m
odel
f
or
vide
o
surve
il
l
an
ce
,
”
Proceedi
ng
s of
th
e
2008
Co
ngress
on
Image
and
Signa
l Proc
ess
ing,
vol
.
3
,
pp
.
656
-
660
,
2008
.
[15]
Appiah
K.,
Hunter
A,
“
A
single
-
ch
ip
FP
G
A
implementa
tion
of
rea
l
-
ti
m
e
ada
pt
ive
background
m
odel
,”
IEE
E
Inte
rnat
io
nal
Conf
ere
nce
on
Fi
eld
-
Progra
mm
able
Technology,
pp
.
95
-
102
,
2005.
[16]
Picc
ard
i,
M.”
Ba
ckgr
ound
subtra
ct
ion
techniqu
es:
a
rev
ie
w
,
”
IE
E
E
Int
ernati
onal
Confe
renc
e
on
S
yste
ms
,
Man
an
d
Cybe
rnet
ic
s,
vol
.
4,
pp.
3099
–
310
4
,
2004
.
[17]
Cucc
hi
ara,
R.
,
Grana
,
C
.
,
Pi
ccardi
,
M
.
,
Pra
ti
,
A.
,
“
Detect
ing
m
oving
obje
cts
,
ghost
s,
and
s
hadows
in
video
strea
m
s,
”
IEEE
Tr
ansacti
ons
on
Pat
te
rn
Anal
ysi
s.
Mac
hine
Intel
li
gen
ce
,
vol
.
25,
no.
10
,
pp.
1337
–
1442
,
October
2003
.
[18]
Vladi
m
ir
Y.
Maria
no
et.
al.
"P
erf
orm
anc
e
Evaluation
of
obje
ct
Det
ec
t
ion
Algorit
h
m
,
"
16th
int
ernati
onal
conferen
c
e
proce
edi
ng
on
p
att
ern
re
cogni
t
io
n
,
vol
.
3
,
2002
.
[19]
Olive
r,
N
.
M.,
R
osario,
B.
,
Pent
l
and,
A.P,
“
A
Ba
y
esi
an
computer
vision
s
y
stem
f
or
m
odel
ing
hu
m
an
int
eract
ions
,
”
IEE
E
Tr
ans.
Pattern
Ana
ly
sis.
M
achi
ne
Intelli
g
en
ce
,
vol
.
22
,
no
.
8
,
pp
.
831
–
843
,
A
ugust
2000
.
[20]
Ya
Li
u
,
Ha
iz
hou
Ai,
and
Guang
-
y
ou
Xu
,
“
Moving
object
d
etec
t
io
n
and
tracki
ng
b
ase
d
on
B
ac
kgro
und
subtraction
,
”
Inte
rnational
So
ci
e
ty
for
Optic
s
and
Phot
onic
s
In
Mult
ispec
tr
al
Image
Proces
sing
and
Pat
te
rn
Re
cogni
t
ion,
pp.
62
-
66
,
2001
.
[21]
Stauffe
r
C,
Gri
m
son
W
.
E.
L
,
“
Adapti
ve
background
m
ixt
ure
m
odel
s
for
re
al
-
ti
m
e
tra
ck
ing
,”
IEEE
Comput
er
Soci
e
ty
Con
fe
re
nce
on
Compute
r V
ision
and
Pa
t
te
rn R
ec
ogni
ti
on
,
vol
.
2
,
pp
.
637
-
663
,
1999
.
[22]
Coll
ins,
R
.
,
L
ipton,
A.,
Kan
ade,
T.
,
“
A s
y
st
em for
vi
deo
survei
ll
a
nce
and
m
onit
or
i
ng,
”
Int
ernati
on
al
Proceedi
ngs
of
Ame
rican
Nuc
lear
Soci
ety
(
ANS)
Ei
ghth
I
n
te
rnat
i
onal
Topical
M
e
et
ing
on
Robotic
and
Re
mot
e
Sys
te
ms
,
pp.
25
–
29
,
1999
.
[23]
Alan
J
.
L
ipt
on,
Hironobu
Fuji
y
o
shi,
and
R
aj
u
S
.
Pati
l
.
,
“
Moving
ta
rge
t
c
la
ss
ifi
c
a
ti
on
and
tracki
n
g
from
rea
l
-
t
ime
vide
o
,”
Proce
ed
i
ngs f
ourth
IE
EE
workshop on
ap
pli
cations of
co
mputer
vi
sion
,
p
p.
8
-
14
,
1998
.
[24]
Hari
ta
og
lu
I,
Harwood
D,
Davi
s
L.
S,
“
W
4:
W
ho
?
when
?
wh
e
re
?
wha
t
?
A
r
eal
ti
m
e
s
y
s
te
m
f
or
det
e
cting
and
tra
ck
ing
people,”
Proceedi
ngs
Thir
d
IEE
E
Int
ernati
onal
Conf
ere
nce
on
Au
tomati
c
Fa
ce
and
Ge
sture
Re
cognitio
n
,
pp.
222
-
227
,
19
98.
[25]
W
ren
C.
R,
Aza
rba
y
e
ja
ni
A,
Darre
l
l
T,
Pentla
nd
A.P,”
Pfinder
:
rea
l
-
ti
m
e
tra
ck
ing
of
the
huma
n
bod
y
,
”
IEEE
Tr
ansacti
ons on Patt
ern
Analysis and
Mac
h
ine In
te
lligen
ce
, v
ol
.
1
9,
no
7
,
pp
.
780
–
785
,
Jul.
1997
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Detect
ion
an
d t
ra
cki
ng
o
f m
ov
ing
obje
ct
u
si
ng mo
difi
ed
b
ac
kgr
ou
nd
…
(
J
eevi
th S
.
H.
)
223
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Mr.
J
ee
vith
S
.
H
.
is
a
Resea
r
ch
Schola
r
at
VTU
,
Bel
ag
avi
,
Karn
at
ak
a,
Indi
a.
He
has
pursed
B.
E
(
Te
l
ec
om
m
unic
a
ti
on
Engi
n
ee
rin
g
)
and
M.T
ec
h
(
VLSI
a
nd
Emb
edde
d
S
y
st
em
)
from
VTU
in
the
y
ea
r
2009
an
d
2012
respe
ct
iv
ely
.
He
is
worki
ng
as
As
si
stant
profe
ss
or
,
Dept
.
of
El
ectroni
cs
&
Telec
om
m
unic
a
ti
o
n
Eng
ine
e
ring
,
SS
IT
,
Tu
m
akur
u,
Karna
t
aka
,
Ind
ia
.
He
has
working
expe
ri
ence
of
m
ore
tha
n
7
y
e
a
rs
.
His
ar
ea
o
f
r
ese
arc
h
intere
st
are
Im
ag
e
pro
ces
sing
and
VLSI.
He
is t
h
e
m
embe
r
of
AM
IE
.
Dr.
Lak
sh
mikan
th.
S
.
has
purs
ed
his
B
.
E
(
EEE)
and
M.T
ec
h
(CAID
)
from
V
TU,
Be
la
g
avi
,
Karna
ta
k
a
,
Indi
a
,
in
the
y
ea
r
20
02
and
2007
res
pec
t
ive
l
y
.
He
ha
s
bee
n
awa
rde
d
Ph.D
in
Jain
unive
rsit
y
,
B
en
gal
uru,
Karna
taka
in
the
b
y
ea
r
2015.
He
is
c
urre
ntly
w
orkin
g
as
As
oci
at
e
profe
ss
or
,
Dept
.
of
ECE
,
Ach
a
r
y
a
Instit
u
te
of
Te
chno
log
y
,
B
enga
luru
,
Karna
ta
ka
.
He
has
working
expe
ri
e
nce
of
m
ore
th
a
n
18
y
e
ars.
His
a
rea
of
r
ese
ar
ch
i
nte
rest
are
Sign
al
proc
essing
,
Signal
D
eno
isin
g
,
Pow
er
El
e
ct
ro
nic
s a
nd
Control s
y
st
em.He
is
th
e
m
ember
of
M
IE
and
MIS
TE
.
Evaluation Warning : The document was created with Spire.PDF for Python.