Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
2
,
A
pr
il
2020
, p
p. 18
49
~
1858
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
2
.
pp1849
-
18
58
1849
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Foregro
un
d
algorithms for
detect
ion and
extractio
n of an
ob
jec
t
in m
ulti
media
Rekha
V
.
1
,
Na
ta
r
ajan K
2
,
In
nil
a
R
os
e
J.
3
1,2
Depa
rt
m
ent of
Com
pute
r
Sci
en
ce
and
Engi
ne
ering
,
Facu
lty
of E
ngine
er
ing
,
Chri
st (
Dee
m
ed to
b
e
Univer
sit
y
)
,
Ind
ia
3
Cent
er
for
d
igi
t
al
innova
t
ion
(C
DI),
Facu
lty
of
Engi
ne
eri
ng,
Ch
rist
(Dee
m
ed
to
be
Univer
si
t
y
), I
ndia
.
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
pr
1
, 2
019
Re
vised
Oct
23
,
20
19
Accepte
d
Nov
4
, 2
0
19
Bac
kground
Subtraction
of
a
f
ore
ground
obje
c
t
in
m
ult
imedia
is
one
of
the
m
aj
or
p
rep
r
oce
ss
ing
steps
i
nvolve
d
in
m
any
v
ision
-
base
d
appl
i
ca
t
ions
.
The
m
ai
n
logi
c
for
det
ecting
m
oving
obje
ct
s
fro
m
the
vide
o
is
d
iffe
ren
ce
of
the
cur
r
ent
fr
ame
and
a
ref
ere
n
c
e
fra
m
e
which
is
ca
l
le
d
“
bac
kg
ro
und
image”
and
thi
s
m
et
ho
d
is
known
as
fra
m
e
diff
ere
n
ci
ng
m
et
hod
.
Bac
kground
Subtrac
ti
on
is
widely
us
ed
for
rea
l
-
t
ime
m
oti
on
gesture
rec
o
gnit
ion
to
b
e
used
in
gesture
ena
bl
ed
it
ems
like
vehi
c
le
s
or
aut
om
at
ed
gadget
s.
It
is
al
so
used
in
con
te
nt
-
base
d
vid
eo
c
oding,
t
raf
fi
c
m
onit
oring,
obj
ec
t
tracki
n
g
,
digi
tal
fore
nsi
cs
and
hum
an
-
computer
intera
ct
ion
.
Now
-
a
-
da
y
s
du
e
to
adv
en
t
in
t
ec
hnolog
y
i
t
is
no
ti
c
ed
that
m
ost
of
th
e
conf
ere
n
ce
s,
m
e
et
ings
and
int
erv
i
ews
are
d
one
on
vide
o
cal
ls.
It’s
quite
obv
ious
tha
t
a
conf
e
ren
ce
room
li
ke
at
m
osphere
is
not
al
wa
y
s
rea
d
ily
available
at
an
y
poi
nt
of
t
ime
.
To
era
d
icate
this
issue,
an
eff
icient
al
gori
t
hm
for
fore
ground
e
xtra
c
ti
on
i
n
a
m
ult
imedia
on
vide
o
c
al
ls
is
ver
y
m
uch
nee
d
e
d.
Thi
s
pap
er
is
not
to
jus
t
buil
d
Ba
ckgr
ou
nd
Subtrac
t
ion
appl
i
ca
t
ion
fo
r
Mobile
Pl
at
f
orm
bu
t
to
opti
m
iz
e
the
exis
ti
ng
OpenCV
a
lgori
thm
to
wor
k
on
li
m
ited
res
ourc
es
on
m
obil
e
pl
at
form
without
r
educ
ing
the
per
form
ance
.
In
thi
s
pap
er,
c
om
par
ison
of
var
ious
fore
ground
det
ection,
ext
racti
on
and
f
ea
tur
e
det
e
ction
al
gorit
hm
s
are
done
on
m
o
bil
e
pl
at
form
using
OpenCV.
T
he
set
of
exp
erim
ent
s
were
conduc
t
ed
to
a
ppra
ise
th
e
eff
i
ci
en
c
y
of
e
ac
h
al
gori
thm
over
the
oth
er.
The
over
all
per
f
orm
anc
es
of
th
e
se
al
gori
thms
were
compare
d
on
the
b
asis
of
exe
cu
ti
on
ti
m
e, r
esolut
ion
and res
ourc
es
r
equi
red
.
Ke
yw
or
d
s
:
Ba
ckgrou
nd
s
ubtract
io
n
Con
t
ours
Fo
r
eg
rou
nd s
ubtract
io
n
G
ra
b
c
ut
Haar
-
casca
de
HOG
Im
age seg
m
entat
ion
Stat
ic
b
ack
gro
und
Vab cut
Waters
hed
Copyright
©
202
0
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Innila
Rose
J
.
,
Ce
nter fo
r digi
ta
l i
nn
ovat
io
n (
CDI),
Faculty
of E
ngineerin
g,
Christ (
Deem
e
d
to
b
e
Unive
rs
it
y), I
ndia
.
Em
a
il
:
Rose.inn
il
a.10@
gm
ai
l
.co
m
1.
INTROD
U
CTION
Ba
ckgrou
nd
subtract
io
n
is
th
e
m
eans
of
det
achin
g
the
for
egro
und
ob
j
ect
s
from
the
back
gr
ound
in
a
pro
gr
essi
on
of
vid
e
o
fr
am
e
s
.
I
n
the
m
os
t
r
ecent
tw
o
deca
des
the
re
hav
e
been
a
num
ero
us
,
im
pr
ovem
e
nts
i
n
m
et
ho
ds
of
doing
bac
kgr
ound
su
bt
racti
on.
T
his
al
gorith
m
has
broa
d
us
a
ge
in
var
i
ous
cr
uc
ia
l
app
li
cat
ion
s
li
ke
visu
al
m
on
it
or
i
ng,
spo
rts
vid
e
o
ju
dg
em
ent,
act
ion
sei
zu
re,
a
nd
s
o
on
[
1
-
6]
.
In
this
paper
a
com
par
at
ive
study
of
al
l
the
foregrou
nd
de
te
ct
ion
an
d
extracti
on
al
gorithm
s
is
pr
ese
nted.
F
or
detect
ion
Haa
r
-
casca
de,
c
on
t
ours
,
waters
hed
a
nd
H
OG
m
et
ho
d
are
com
par
ed
.
F
or
ext
racti
on
va
bcu
t
a
nd
gr
ab
-
c
ut
al
gori
thm
s
are
com
par
e
d.
The
m
ai
n
m
otive
of
this
c
om
par
ison
is
to
ens
ur
e
w
hich
al
gorithm
i
s
a
bette
r
on
e
and
will
prov
i
de
a
com
par
at
ively
le
ss
t
i
m
e
f
or
e
xecu
ti
on
t
han
oth
e
rs
sin
ce
the
m
a
in
i
de
a
of
this
pr
oj
ect
is
to
e
xe
cute
the
ext
racti
on
process
on
a
m
ob
il
e
platform
and
f
or
that
the
e
xec
ution
tim
e
in
des
kt
op
s
hould
be
m
ini
m
al
.
In
t
his
pap
e
r,
f
or
e
gro
und
seg
m
entat
ion
in
a
vid
e
o
is
do
ne
us
in
g
O
pen
C
V.
It
wa
s
f
ound
that
f
or
f
or
e
gro
und
segm
entat
ion
in
O
pen
C
V
the
re
is
an
inbuil
t
m
et
ho
d
of
bac
kgr
ound
subtra
ct
ion
.
All
the
ba
sic
inb
uilt
m
e
thods
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.
10
, No
.
2
,
A
pr
i
l 202
0
:
1849
-
1858
1850
for
foregr
ound
segm
entat
ion
[7]
wer
e
im
ple
m
ented
to
ena
ble
a
basic
under
sta
nd
i
ng
of
how
the
c
ode
w
orks
and
to
c
hec
k
wh
et
her
m
aking
an
y
m
ino
r
changes
in
it
af
fects
the
outp
ut
var
ia
nce.
T
he
aim
was
to
find
an
eff
ic
ie
nt al
gorithm
f
or
foregr
ound e
xtracti
on in
a m
ultim
e
dia.
The
m
ai
n
obj
e
ct
ive
of
this
w
ork
i
s
no
t
t
o
j
ust
bu
il
d
Ba
ck
gro
und
Subtract
ion
a
ppli
cat
ion
for
M
ob
il
e
Plat
fo
rm
bu
t
t
o
opt
im
iz
e
the
existi
ng
O
penC
V
al
gorithm
to
w
ork
on
li
m
it
ed
resour
ce
s
on
m
ob
il
e
platfor
m
without
reduci
ng the
pe
rfor
m
ance.
The
m
ai
n
obj
ect
iv
es
of the
p
ape
r
wer
e
as foll
ows:
To dete
ct
the
f
or
e
gro
und o
bj
e
ct
and
get t
he o
utli
ne
of t
he fo
regrou
nd ob
j
ec
t.
To
e
xtract
t
he
f
or
e
gro
und o
bj
e
ct
eff
ect
ively
a
nd
va
ryres
olu
ti
on
s
of t
he vide
o
acc
ordin
gly.
Ena
ble b
ac
kgr
ound
rep
la
cem
ent
an
d
Re
duc
e the e
xec
ution t
i
m
e.
Bl
end
in
g
t
he n
ew back
gro
und wit
h
t
he
e
xtra
ct
ed
f
or
e
gro
un
d object.
In
paper
[
8
]
CNN
m
et
ho
d
is
us
ed
for
obta
ining
th
e
f
oregro
und
m
ask.
They
us
e
d
th
e
dataset
of
CDn
et
2014
a
nd
are
us
in
g
Subsense
al
go
rithm
to
pr
eve
nt
over
-
fitt
in
g.
In
pa
per
[
9
]
sel
ect
ive
bac
kgr
ound
Subtract
io
n
is
us
e
d.
Ba
c
kground
m
od
el
li
ng
is
us
e
d
to
de
duct
tw
o
co
ns
ec
utive
f
ram
es
to
get
the
sta
ti
c
pix
el
s
.
The
pr
opos
e
d
te
chn
iq
ue
help
s
in
with
ho
l
din
g
unwa
nted
backg
rou
nd
from
the
fo
refront
an
d
the
bac
kdr
op
scene
in
ste
ad
of
subtract
in
g
the
w
hole
ba
ckgr
ound.
Pa
pe
r
[
10
]
focuse
s
on
re
duci
ng
the
ho
le
s
w
hi
ch
is
ob
ta
ine
d
wh
il
e
getti
ng
a
f
ore
gro
und
m
ask.
This
pa
pe
r
ex
pl
ai
ns
an
d
im
pl
e
m
ents
wh
y
th
ree
fram
e
diff
eren
ci
ng
are
us
ed
instea
d
of
tw
o
f
ram
e
dif
fer
e
ncin
g
.
Pape
r
[
1
]
f
oc
use
s
on
M
OG
m
et
ho
d
f
or
det
ect
ion
of
ob
j
ec
t.
They
fo
c
us
e
d
on
fi
ve
crit
eria
of
MOG
pa
rtic
ularly
segm
ented
backg
rou
nd
weig
ht
th
resho
ld,
sta
nd
a
rd
de
viat
ion
scal
ing
factor,
us
er
de
fine
d
l
earn
i
ng
rate,
t
otal
num
ber
of
Ga
us
sia
n
co
m
po
nen
ts
a
nd
m
axi
m
u
m
num
ber
of
com
po
ne
nts
in
the
backgro
und
m
od
el
.
Pape
r
[
11
-
1
6
]
is
on
vid
eo
se
gm
entat
ion
for
extra
ct
ing
the
f
or
e
gro
und
obj
ect
in
a
vide
o.
I
n
this
pa
pe
r
they
a
re
us
i
ng
S
URF
al
go
r
it
h
m
fo
r
feat
ure
detect
io
n
of
t
he
foregr
ound
obj
ect
and
e
ve
ntu
al
ly
segm
enting
them
eff
ect
ively
.
Pape
r
[
1
7
]
ta
lks
ab
ou
t
G
r
a
b
Cut
Algo
rithm
wh
ic
h
is
one
of
the
fam
i
li
ar
app
r
oac
hes
f
or
f
or
e
gro
und
e
xtr
act
ion
in
the
dom
ai
n
of
Im
a
ge
proce
ssin
g.
In
p
a
pe
r
[
1
]
FCFN
m
echan
ism
is
us
e
d
f
or
bac
kgr
ound
s
ub
tr
act
ion
.
FCFN
m
eans
fu
zzy
n
ear
ness
de
gree
w
hich
use
s
f
uzzy
c
-
m
eans clusterin
g
al
gorit
hm
for
ide
ntifyi
ng
wh
et
her the
pi
xel sele
ct
ed
is
backg
rou
nd or
foregr
ound.
2.
RESEA
R
CH MET
HO
D
2.1
.
F
r
ame
w
ork
m
odel
of
b
ackgr
ou
n
d
s
ubt
r
act
i
on
of
an
o
b
ject
in
m
ultime
dia
Ba
ckgrou
nd
S
ub
t
racti
on
of
a
n
obj
ect
i
n
m
ultim
edia
(BS
O
M)
is
on
e
of
t
he
pr
im
ary
ste
ps
i
n
var
i
ou
s
view
-
base
d
operati
on
s
.
A
s
the
nam
e
su
gg
est
s
backg
rou
nd
s
ubtract
io
n
is
the
m
echan
ism
of
deducti
ng
foref
ront
s
ubsta
nces
from
the
en
vir
on
m
ent
in
a
se
ries
of
vi
deo
blo
c
ks.
T
he
m
ai
n
ph
il
os
ophy
f
or
ide
ntifyi
ng
dynam
ic
su
bs
ta
nces
f
r
om
the
vid
e
o
is
to
s
ubtract
bet
ween
the
prese
nt
fra
m
e
and
a
re
fe
ren
ce
fr
am
e
wh
ic
h
i
s
cal
le
d
“bac
kgr
ound
im
age”
a
nd
t
his
schem
e
is
kno
w
n
as
fra
m
e
diff
ere
nci
ng
m
et
ho
d.
Ba
ckgr
ound
Subt
racti
on
is
extensi
vely
us
e
d
in
tra
ff
ic
su
r
veyi
ng
(
det
ect
ing
veh
ic
le
s
),
hu
m
an
m
ov
em
ent
identific
a
ti
on
,
hum
an
-
m
achine
co
-
operati
on,
obj
ect
m
ov
em
ent
traci
ng,
di
gital
fo
re
ns
ic
s
et
c.
I
n
belo
w
Figure
1
Firstl
y,
the
vi
de
o
is
div
id
e
d
into
fr
am
es.
Aft
er
that
su
btrac
ti
on
of
the
est
i
m
at
ed
back
gr
ound
i.e
the
pr
e
vious
fr
am
e
fr
om
the
cur
re
nt
fr
am
e
is
done
.
Lat
er
on,
a
t
hr
es
hold
is
ap
plied
t
o
ge
t
a
cl
ear
foregrou
nd
m
ask.
Ba
sed
on
the
thr
esh
old
value
w
e
will
get
a
cl
ear
f
or
e
g
r
ound
m
ask.
I
t
can
be
done
ei
ther
m
anu
al
ly
by
giv
in
g
a
t
hr
es
hold
value
or
by
us
i
ng
va
rio
us
thres
ho
l
d
al
go
r
it
h
m
s.
Figure
1. Me
th
odology
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
Fore
groun
d algo
rit
hm
s
for
det
ect
ion
and
ex
tracti
on o
f
an
ob
je
ct
…
(
Rekh
a
V
.
)
1851
The
f
our
pri
m
ary
ste
ps
in
Ba
ckgrou
nd
Subtract
io
n
ar
e
Pr
e
-
proces
sing,
bac
kgr
ound
m
od
el
ing,
foregr
ound
de
te
ct
ion
an
d
da
ta
validat
ion.
In
t
he
fir
st
ste
p,
a
giv
e
n
vi
deo
is
bro
ke
n
dow
n
into
f
r
a
m
es.
The
n
the
in
put
fr
am
e
is
conver
te
d
int
o
num
py
arr
ay
for
f
ur
the
r
proce
ssing.
T
he
n
f
r
om
this
nu
m
py
arr
a
y
the
backg
rou
nd
m
od
el
is
fo
rm
ed.
If
it
’s
a
vid
e
o
then
tw
o
con
sec
utive
f
r
a
m
es
are
ta
ken
and
de
duct
ed
to
get
the
sta
ti
c
pix
el
s.
These
sta
ti
c
pix
el
s
co
ns
ti
tut
e
the
backgro
und
a
nd
th
us
ba
ckgr
ound
m
od
el
is
fo
rm
ed.
Fu
rt
her
al
l
the
pix
el
s
in
a
f
ram
e
are
com
par
ed
with
the
bac
kgr
ound
m
od
el
th
us
form
ed
wh
ic
h
helps
i
n
ide
ntifyi
ng
the
sign
i
ficant
forefro
nt
pi
xels
eff
ect
ively
.
L
ast
ly
the
un
wa
nted
pix
el
s
w
hi
ch
neither
c
on
sti
tutes
the
fo
r
efront
nor
t
he back
groun
d
a
re r
em
ov
ed
to get a
pr
op
e
r fore
groun
d
m
ask
as s
ho
wn in t
he
la
st s
te
p
of
F
ig
ur
e
2
.
Figure
2
.
Bl
oc
k Diag
ram
o
f B
ackgrou
nd S
ub
t
rac
ti
on Proc
ess
.
2.2
.
Te
ch
niq
ues use
d for B
SOM
In
this
pa
per
,
va
rio
us
te
c
hniq
ues
a
re
use
d
for
BS
O
M
for
f
or
e
groun
d
obj
ect
detect
ion
a
nd
extracti
on
[
18
-
2
1
]
.
Fo
ll
ow
i
ng secti
on
of
th
e
pap
e
r discu
sse
s
te
ch
niques a
nd it
s im
ple
m
en
ta
ti
on
:
2.2.1
.
Hist
og
r
am of
g
r
ad
ie
n
ts
(
HO
G)
In
F
i
gure
3
th
e
HOG
al
gorithm
is
sh
own
wh
ic
h
us
es
S
VMDetect
or
f
or
detect
ing
th
e
fo
re
gro
un
d
obj
ect
in
an
im
age
or
a
vi
deo
even
wh
e
n
the
y
are
of
diff
e
re
nt
scal
es.
Af
te
r
th
e
hu
m
an
fo
r
egro
und
is
detect
ed,
a rectan
gle is
f
or
m
ed
ar
ound i
t and dis
p
la
ye
d as
ou
t
pu
t.
Figure
3
.
H
OG algori
thm
.
2.2.2
.
Haar
-
ca
scad
e
a
l
go
ri
t
h
m
In
F
ig
ur
e
4
H
aar
casca
de
is
al
gorithm
is
sh
ow
n
ste
p
-
wise.
S
uppose
fa
ce
detect
ion
is
ta
ken
i
nt
o
consi
der
at
io
n,
then
t
he
in
put
dataset
will
hav
e
im
ages
of
face
s
in
al
l
it
s
diff
e
ren
t
vi
ews.
The
n
t
he
haa
r
featur
e
s
a
re
id
entifi
ed
i
n
t
hose
im
ages
and
an
i
nteg
ral
i
m
age
is
f
or
m
ed
as
sho
wn
in
F
ig
ur
e
5
.
T
he
n
the
se
integral
im
ages
are
ta
ke
n
a
nd
giv
e
n
t
o
vari
ou
s
cl
assifi
ers
us
i
ng
the
ada
pti
ve
boos
ti
ng
te
ch
nique.
Finall
y
,
the m
achine is
trai
ned to i
den
t
ify
f
aces i
n
a c
ertai
n
im
age o
r
v
ide
o.
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.
10
, No
.
2
,
A
pr
i
l 202
0
:
1849
-
1858
1852
Figure
4
.
Haa
r
-
casca
de
al
gorithm
[
2
2
]
Figure
5
.
I
nteg
ral Im
age [
2
2
]
2.2.3
.
Gra
b
-
c
u
t alg
orithm
In
Fig
ur
e
6
G
r
ab
c
ut
al
gorith
m
is
sh
ow
n
w
her
e
firstly
a
m
ask
in
a
ppli
ed
to
the
i
nputted
im
age
for
getti
ng
the
f
oreg
rou
nd
a
nd
backg
rou
nd
obj
ect
s
.
Her
e
i
f
the
m
ask
value
is
2
then
it
is
con
side
red
as
backg
rou
nd
a
nd
if
the
m
ask
value
is
1
th
en
it
is
con
si
der
e
d
as
f
or
e
gro
und.
Finall
y,
this
m
ask
va
lue
is
m
ul
ti
plied w
it
h t
he fram
e to get
a for
e
gro
und m
ask.
Figure
6
.
G
rab
-
cut al
gorithm
[
2
3
]
2.2.4
.
C
on
t
ou
r
a
lg
orit
hm
In
Fig
ure
7
the
al
go
rithm
of
con
t
o
urs
is
giv
e
n
w
her
e
resizi
ng
of
im
age
is
done
an
d
the
n
a
pyr
m
ean
sh
ift
filt
er
is
a
pp
li
ed
t
o
it
.
T
hen
t
he
im
age
is
conver
te
d
to
gray
scal
e
and
s
om
e
thres
ho
l
d
val
ue
is
giv
e
n.
An
i
nbuilt
functi
on
of
fi
ndCo
ntour
s
is
us
e
d
to
ide
ntify
the
co
nt
ours
in
t
he
i
m
age
a
nd
by
us
in
g
the dra
wConto
ur
s
func
ti
on th
e iden
ti
fie
d
c
onto
ur is dra
wn
and d
is
play
ed
i
n
the
outp
ut.
Figure
7
.
Co
nto
ur al
gorithm
[
2
4
]
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
Fore
groun
d algo
rit
hm
s
for
det
ect
ion
and
ex
tracti
on o
f
an
ob
je
ct
…
(
Rekh
a
V
.
)
1853
2.2.5
.
Waters
hed
algorithm
In
F
i
gure
8
,
w
at
ersh
e
d
al
gori
thm
is
giv
en
.
The
in
putt
ed
im
age
is
co
nve
rted
int
o
gray
s
cal
e
and
the
n
a
su
it
able
thr
esh
old
val
ue
is
app
li
ed
to
it
.
A
su
it
abl
e
kernel
siz
e
is
ta
ken
an
d
the
m
or
phol
og
ic
a
l
trans
form
ations
are
app
li
e
d
to
it
.
The
su
r
e
backgro
und
and
foregr
ound
are
obta
ine
d
and
the
n
the
y
are
su
bt
racted
t
o
ge
t
the
un
know
n
re
gion
w
hich
is
t
he
bo
unda
ry
of
the
f
or
e
fron
t
obj
ect
i
n
the
in
putt
ed
i
m
age.
Fr
om
the
a
bove
t
he
Haa
r
-
cas
cade
is
us
e
d
t
o
detect
the
f
or
egro
und
obj
ect
an
d
t
he
Gr
a
b
-
cut
is
us
e
d
to
extract
the fore
gro
und t
hus
detect
ed.
Figure
8
.
W
at
e
rsh
e
d
al
gorith
m
2.3
.
A
na
l
ys
is
of vari
ous
algorithm
s for
B
SOM
In
this
sect
i
on
of
the
pa
per
giv
es
detai
le
d
descr
i
ption
of
com
par
ison
of
var
i
ou
s
al
gori
thm
s.
These
al
gorithm
s ar
e i
m
ple
m
ented
f
or the BS
OM.
Com
par
ison
of
Histogram
of
G
rad
ie
nts
(HOG)
a
nd
Haar
-
casca
de
for
detect
ion
of
foregr
ound:
HOG
m
et
ho
d
is
us
ed
to
detect
the
fu
ll
body
of
a
hu
m
an
unli
ke
Haa
r
-
ca
scade
al
gorith
m
.
Haar
-
casca
de
al
gorithm
has
var
i
ou
s
xm
l
fi
le
s
f
or
dif
fer
e
nt
par
ts
of
the
body
li
ke
hal
f
body,
face,
f
ull
body, s
m
il
e etc.
[
2
2
].
Com
par
ison
of
Gr
a
b
-
c
ut
an
d
Va
b
-
c
ut:
Grab
Cu
t
and
V
ab
cut
both
are
us
e
d
in
the
fiel
d
of
i
m
age
segm
entat
ion
.
Both
the
al
gor
it
h
m
helps
in
extracti
ng
the
foregr
ound
obje
ct
.
The
dr
a
w
ba
ck
of
the
gr
a
b
-
cut alg
or
it
hm
is that i
t i
s not
optim
iz
ed.
V
a
b C
ut is an exte
nsi
on of
Gr
a
b
C
ut
[
2
3
]
.
Com
par
ison
of
Co
ntou
rs
a
nd
W
at
e
rsh
e
d:
Con
to
urs
an
d
water
sh
e
d
are
us
e
d
f
or
the
detect
io
n
of
the
outl
ine
of
the
fore
ground
ob
j
ect
f
r
om
an
i
m
age
or
a
vid
e
o.
C
on
t
ours
us
es
the
functi
on
“
f
in
d
Con
t
ours”
to
identify
the
co
ntours
in
the
gi
ven
im
age
or
vid
e
o.
I
n
wate
rsh
e
d
al
gorith
m
,
m
or
phologi
cal
trans
form
s ar
e d
one
on the
im
age
first a
nd th
en
s
ur
e
for
e
gro
und
a
nd
backg
rou
nd is ide
ntif
ie
d
[
2
4
,
2
5
]
.
3.
RESU
LT
S
A
ND
DI
SCUS
S
ION
S
In
t
his
w
ork
O
pen
C
V
to
ol
w
as
us
e
d
w
hich
is
an
ope
n
s
ource
li
brary
th
at
approves
use
of
var
i
ous
com
pu
te
r
la
ng
uag
e
s
an
d
is
app
li
cable
on
s
e
ver
al
platf
orm
s
.
From
the
above
F
ig
ur
e
s
9
-
1
1
it
can
be
c
oncl
ud
e
d
that
Haar
-
casc
ade
is
bette
r
de
te
ct
ion
al
gori
thm
as
the
input
i
m
age
or
vid
eo
doesn
’t
al
ways
ha
ve
to
be
f
ull
body.
From
F
igure
s
1
2
,
1
3
i
t
can
be
co
ncl
ud
e
d
that
Gr
a
b
-
c
ut
is
n’
t
e
ff
i
ci
ent
in
e
xtrac
ti
ng
t
he
foregr
ound
obj
ect
.
Fig
ur
e
1
4
cl
early
sho
ws
how
m
anu
a
l
m
asking
the
i
nput
im
age
helps
in
the
e
ffec
ti
ve
ext
racti
on
us
in
g
Gr
a
b
-
c
ut
[
2
2
]
.
Figures
1
5
a
nd
1
6
s
how
how
co
ntour
s
are
dr
a
wn
i
n
an
i
m
age
and
vid
e
o.
Fi
gure
s
17
and
18
sh
ows
ho
w wa
te
rsh
e
d
is
dif
fe
ren
t
from
co
nt
ours.
Figure
9
.
Non
-
Ma
xim
u
su
ppr
ession i
n H
OG
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.
10
, No
.
2
,
A
pr
i
l 202
0
:
1849
-
1858
1854
Figure
10
.
Haa
r
-
casca
de of
21
60p
Figure
1
1
.
Haa
r
-
casca
de of
14
4p
Figure
1
2
.
Gr
a
b
-
c
ut in
a
vid
e
o
in
48
0p
Figure
13
.
Garb
-
c
ut in
a
vid
e
o
in
72
0p
Figure
1
4
.
Gr
a
b
-
c
ut
us
in
g
m
a
nu
al
m
askin
g o
n
a
n
i
m
age
[
23
]
Figure
1
5
.
C
ontour in
a
vid
e
o
Figure
1
6
.
C
ontour in
an i
m
ag
e
Figure
1
7
.
Wat
ersh
e
d
i
n
a
n
im
age
[
2
4
]
In
F
ig
ur
es
1
8
a
nd
19
a
diff
e
re
nce
in
t
he
ou
t
put
is
seen
w
he
n
a
certai
n
vi
de
o
of
a
s
pa
n
of
30
seco
nds
is
dow
nlo
a
ded
in
dif
fer
e
nt
r
esolutio
ns
,
t
o
check
t
he
ef
fe
ct
iveness
of
t
he
outp
ut.
Ta
bu
la
r
val
ues
of
haa
r
-
casca
de
with
grab
-
cut
f
or
a
vi
deo
do
wn
l
oaded
at
diff
ere
nt
reso
luti
ons
ar
e
sh
own
in
Ta
ble
1
an
d
gr
a
phic
al
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
Fore
groun
d algo
rit
hm
s
for
det
ect
ion
and
ex
tracti
on o
f
an
ob
je
ct
…
(
Rekh
a
V
.
)
1855
represe
ntati
on
are
sho
wn
in F
igure
20.
A
vi
de
o
of
30
sec
onds
with
a
ce
rtai
n
res
olu
ti
on
is t
aken
,
a
nd
a
ta
ble
of
com
par
ison
of
it
s
execu
ti
on
tim
e
of
each
of
the
al
gorith
m
(u
sed
in
t
he
cod
e
)
is
sho
wn
i
n
T
able
2
with
diff
e
re
nt
res
ol
ution
va
lues
(
change
d
by
th
e
co
de)
an
d
grap
hical
re
pr
e
s
entat
ion
of
T
able
2
is
s
ho
wn
i
n
Figure 21.
A
vi
deo
of 30
sec
onds
wit
h
a
ce
rt
ai
n
res
olu
ti
on
is
ta
ke
n,
a
nd
a
t
able
of
c
om
pari
so
n
of
it
s
e
xec
utio
n
tim
e
of
each
of
th
e
al
gorith
m
(u
sed
i
n
th
e
cod
e
)
is
sho
wn
with
diff
e
r
en
t
res
olu
ti
on
values
(ch
a
ng
ed
by
the
cod
e
)
in
T
able
3
an
d
Fig
ur
e
22
s
hows
t
h
e
grap
hical
re
pr
ese
natat
io
n
of
T
a
ble
3.
A
f
te
r
the
su
r
vey
of
al
l
these
res
ults
it
was
fou
nd
tha
t
Haar
-
casca
de
with
Gr
a
b
-
c
ut
gav
e
a
bette
r
resu
lt
than
a
ny
oth
er
c
om
bin
at
ion
.
The res
ults can
b
e cle
a
rly
see
n
in
Fig
ure
23
wh
e
re a
r
es
olu
t
ion
of
144p is t
aken into
consi
der
at
io
n.
Figure
1
8
.
Haa
r
-
casca
de wit
h Gr
a
b
-
c
ut
on a
vid
e
o dow
nlo
a
ded w
it
h
a
720p
reso
l
ution
Figure
19
.
Haa
r
-
casca
de wit
h Gr
a
b
-
c
ut
on a
vid
e
o
dow
nlo
a
ded w
it
h a
1080p res
olu
ti
on
Table
1
.
T
ab
ul
ar
values o
f
ha
ar
-
casca
de wit
h gr
a
b
-
c
ut
for a
v
ide
o d
ownl
oad
e
d
at
dif
fere
nt r
es
olu
ti
on
s
Res
o
lu
tio
n
(
In Pix
els)
Haar
-
C
ascad
e Alg
o
rith
m
w
ith
Grab
-
Cu
t
Alg
o
rith
m
(
In
Seco
n
d
s)
1
0
8
0
p
7
9
.05
720p
6
0
.91
360p
3
1
.34
144p
3
7
8
.13
Figure
20. Gra
ph
ic
al
represe
nt
at
ion
of t
he
Ta
ble 1 in
Fig
ur
e
19
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.
10
, No
.
2
,
A
pr
i
l 202
0
:
1849
-
1858
1856
Table
2.
T
ab
ul
ar
values o
f ha
ar
-
casca
de wit
h gr
a
b
-
c
ut
for a
v
ide
o
of cert
ai
n
res
olu
ti
on
Res
o
lu
tio
n
(
In Pix
els)
Haar
-
C
ascad
e Alg
o
rith
m
(I
n
Secon
d
s)
Grab
-
Cu
t Algo
rith
m
(I
n
Secon
d
s)
Haar
-
C
ascad
e Alg
o
rith
m
w
ith
Grab
-
Cu
t Algo
rith
m
(
In
Seco
n
d
s)
2
1
6
0
p
5
9
.05
7
1
.40
1
2
7
.47
1
0
8
0
p
5
8
.13
7
0
.16
1
2
5
.19
720p
5
9
.57
6
7
.54
1
4
2
.20
480p
5
9
.16
6
7
.75
1
4
4
.76
360p
5
8
.92
6
6
.54
1
2
4
.87
240p
5
8
.02
6
5
.42
1
2
9
.29
144p
5
9
.05
6
6
.58
1
2
9
.51
Figure
2
1
.
Gr
a
ph
ic
al
represe
nt
at
ion
of t
he
T
a
ble 2
Table
3
.
T
ab
ul
ar
values o
f ha
ar
-
casca
de wit
h gr
a
b
-
c
ut
for a
v
ide
o o
f
cert
ai
n
res
olu
ti
on
Res
o
lu
tio
n
(
In Pix
els)
Haar
-
C
ascad
e Alg
o
rith
m
(I
n
Secon
d
s)
Grab
-
Cu
t Algo
rith
m
(I
n
Secon
d
s)
Haar
-
C
ascad
e Alg
o
r
ith
m
w
ith
Grab
-
Cu
t Algo
rith
m
(
In
Seco
n
d
s)
2
1
6
0
p
5
9
.05
7
1
.40
1
2
7
.47
1
0
8
0
p
5
8
.13
7
0
.16
1
2
5
.19
720p
5
9
.57
6
7
.54
1
4
2
.20
480p
5
9
.16
6
7
.75
1
4
4
.76
360p
5
8
.92
6
6
.54
1
2
4
.87
240p
5
8
.02
6
5
.42
1
2
9
.29
144p
5
9
.05
6
6
.58
1
2
9
.51
Figure
2
2
.
Gr
a
ph
ic
al
represe
nt
at
ion
of t
he
T
a
ble 3
’
Figure
23. Fi
na
l ou
t
pu
t
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
Fore
groun
d algo
rit
hm
s
for
det
ect
ion
and
ex
tracti
on o
f
an
ob
je
ct
…
(
Rekh
a
V
.
)
1857
4.
CONCL
US
I
O
N
In
this
pa
pe
r,
var
io
us
f
or
e
gro
und
detect
ion
an
d
extracti
on
al
go
rithm
s
are
com
par
ed
.
The
m
ai
n
obj
ect
ive o
f
ou
r
pro
j
ect
is
to
i
den
ti
fy v
ario
us
inbuil
t
m
et
ho
ds
for
ge
tt
ing
a b
est
f
or
e
gro
und
m
ask.
O
ut
of
these
al
gorithm
s
we
fou
nd
t
hat
the
com
bin
at
ion
of
Gr
a
b
c
ut
an
d
Haar
casc
ade
i
s
best
f
or
extra
ct
ing
the
fore
gro
und
obj
ect
.
As
cl
early
seen
in
F
ig
ur
e
2
3
the
re
is
a
li
ve
vid
eo
ta
ken
from
the
c
a
m
era
it
sel
f
at
a
reso
luti
on
of
144p
and
gr
a
b
-
cut
is
app
li
ed
t
o
it
.
The
only
dr
a
wb
ac
k
of
this
m
et
ho
d
is
that
the
li
gh
t
com
i
ng
from
the
window
(as
sho
wn
is
the
f
ig
ure)
ca
nnot
be
c
ut
out
us
in
g
grab
-
cu
t.
Fo
r
furthe
r
work
we
can
fo
c
us
on
op
ti
m
iz
ing
the gra
b
c
ut algorit
hm
f
or ext
racti
on of
fore
gro
und o
bj
ect
even w
he
n
a li
gh
t
source
is
ne
ar it.
REFERE
NCE
S
[1]
Shahriz
a
t
Shaik
Moham
ed,
N.
M.
“
Bac
kgroun
d
Modell
ing
an
d
Bac
kground
Subtraction
P
erf
o
rm
anc
e
for
Obje
c
t
Dete
c
ti
on
,
”
2010
6th
Inte
rnatio
nal
Coll
oquium
on
Signal
Proces
sing
&
it
s
App
li
cations,
Malla
ca
Cit
y
,
pp
.
1
-
6.
2010.
[2]
Sen
-
Ching
S.
C
heung
and
Chandri
ka
Kam
at
h,
“
Rob
ust
Te
chni
q
ues
for
Bac
kgrou
nd
Subtrac
ti
on
in
Urban
Tra
ffi
c
Video,
”
EUR
ASI
P
Journal
on
Ap
pli
ed
Signal P
ro
ce
ss
ing
,
vo
l. 200
5,
pp
.
2330
-
234
0,
Jan
.
2005
.
[3]
F.
El
Baf
,
T.
Bouwm
ans,
and
B.
Vac
hon
.
“
Co
m
par
ison
of
ba
ckgr
ound
subtr
a
ct
ion
m
et
hods
f
or
a
m
ultim
edia
appl
i
ca
t
ion,”
Int
ernati
onal
Conf
ere
nce
on
syste
ms
,
Signal
s
and
Image
Proce
ss
ing,
IWSSIP
2007
,
page
s
385
–
388,
June
2007.
[4]
Donovan
H.
Parks
and
Sidne
y
S.
Fels,
“
Eva
lua
t
ion
of
Bac
kground
Subtrac
ti
on
Alg
orit
hm
with
Post
-
Proce
ss
ing,
”
in
IEE
E
F
if
th
In
te
rnational
Co
nfe
renc
e
on
A
dvanc
ed
V
ide
o
&
Signal
Based
Surv
ei
l
lance
,
p.
192
-
199
,
2008
.
[5]
Mass
imo
Picc
ar
di,
”
Ba
ckgr
ound
Subtrac
ti
on
Tec
hnique
s:
A
Revi
ew,
”
IEEE
Inte
r
nati
onal
Journal
on
Syste
ms
,
Ma
n
and
Cybe
rne
ti
cs
,
Vol.
2,
(5), pp.
05
-
25,
2004
.
[6]
Gourav
Ta
khar
,
Chandra
Praka
sh,
Nam
it
a
Mitt
a
l,
Ra
je
sh
Kum
ar,
”Compara
ti
v
e
Anal
y
s
is
of
Bac
kground
Subtrac
ti
on
Tec
hnique
s
and
Applicati
ons,
”
IEEE
Inte
rnational
Co
nfe
renc
e
on
Rec
ent
Ad
vanc
es
an
d
Innov
ati
ons
in
Engi
ne
ering
(
ICRA
IE
-
2016)
,
pp
.
1
-
8,
2016
.
[7]
S.
Ho
re,
et
al.,
“
An
int
egr
a
te
d
int
er
ac
t
ive
tech
nique
for
image
segm
ent
ation
using
stac
k
b
ase
d
see
d
ed
reg
io
n
growing
and
thre
sholding,”
Inte
rnational
Jour
nal
of
El
ectric
al
and
Comput
er
Engi
nee
ring
,
vol/
issue:
6(6),
pp.
2773
–
2780
,
2016.
[8]
Chum
ing
Li
n,
B.
Y,
“
Foregroun
d
Dete
ct
ion
in
Surveil
l
ance
Vide
o
with
Full
y
Convolut
ional
Sem
ant
i
c
Network,
”
2018
25th
I
EEE
Inte
rnational
Co
nfe
renc
e
on
Image
P
roc
essing
(
ICIP)
,
Athens,
pp
.
4118
-
4122
,
20
18.
[9]
Neha
S.
Sakpal,
M.
S,
“
Adapti
ve
B
ac
kground
S
ubtra
c
ti
on
in
Im
age
s”
,
I
EE
E
Tr
a
nsacti
ons
on
Mult
imedi
a
,
18(10),
pp.
2093
–
2103
,
2018.
[10]
Rahul
Dutt
Shar
m
a,
S.
L.
“
Optimize
d
D
y
namic
Bac
kground
Sub
tra
c
ti
on
Techni
q
ue
for
Moving
Objec
t
Detect
io
n
and
Tracki
ng,
”
2017
2nd
Inte
rnational
Conf
ere
nce
on
Tel
ec
o
mm
unic
ati
on
an
d
Net
works
(
TE
L
-
NET)
,
Noida,
pp.
1
-
3.
2017
.
[11]
Yanxin
Sun,
G
.
Y.,
“
The
Foregr
ound
Segm
ent
ation
Based
on
Su
rf
Algorit
hm
an
d
B
ac
kground
Subtraction,”
201
5
Sev
en
th
Int
ernat
ional
Conf
ere
nc
e
on
Adv
anc
ed
Comm
u
nic
ati
on
and
Net
working
(
ACN
)
,
Kota
Ki
naba
lu
,
pp
.
24
-
2
7.
2015.
[12]
Carste
n
Roth
er
and
Vlad
imir
Kolm
ogorov
and
Andrew
Bla
ke
,
“
"G
rab
Cut":
Int
era
c
ti
ve
Fore
gro
und
Ext
r
ac
t
ion
Us
ing
Ite
rat
ed
Graph
Cuts,”
ACM
Tra
nsac
ti
on
s
on
Graphi
cs
(TOG)
TOG,
Vo
lume
23
Iss
ue
3
,
Pages
309
-
314
,
Augus
t
2004
.
[13]
Ravi
ndra
S.
He
gadi
,
Basava
r
a
j
A
Goudann
ava
r,
"Inte
r
ac
t
iv
e
Segm
ent
a
ti
o
n
of
Medical
Im
age
s
Us
ing
GrabCut",
I
JM
I
,
Volum
e:
3
,
Iss
ue:
3
,
Pag
e
s:
168
-
171,
2011
.
[14]
Y.
W
ang,
et
al.
,
“
Com
pre
ss
ive
bac
kground
m
o
del
ing
for
fore
g
round
ext
r
ac
t
ion
,
”
Journal
o
f
El
e
ct
rica
l
and
Computer
Engi
n
ee
ring
,
vol
.
201
5,
pp
.
1
–
8
,
2015
.
[15]
Bouwm
ans,
“
Re
ce
nt
Advan
ce
d
Stat
isti
ca
l
B
ac
kg
round
Modeli
ng
for
Foreground
Dete
c
ti
on:
A
S
ystemati
c
Surve
y
.
Rec
en
t
Pat
ent
s o
n
Com
pute
r
Sci
e
nce
,
”
RP
CS
2
01
1
,
4(3):147
–
176
,
Septe
m
ber
2011
.
[16]
T.
Bouwm
ans,
“
Tra
dit
ional
and
rec
en
t
ap
proa
che
s
in
b
ac
kground
m
odel
ing
for
for
e
ground
det
ectio
n:
An ove
rvie
w
,
”
Computer
Scienc
e
Revie
w
,
11(3
1
-
6
6),
Ma
y
2014.
[17]
Yubing
Li
,
J.
Z
.
“
Grab
Cut
Im
a
ge
Segm
ent
at
ion
Based
on
Im
age
Regi
on,
”
Inte
r
nati
on
al,
Con
fe
r
enc
e
on
Image
,
Vi
sion
and
Computing
(
ICIVC)
,
June
2018.
[18]
M.
Mous
sa,
et
a
l.
,
“
Com
par
at
iv
e
stud
y
of
statisti
ca
l
ba
ckgr
ound
m
odel
ing
and
subtr
action,”
Indo
nesian
Journal
of
El
e
ct
rica
l
Eng
in
ee
ring a
nd
Computer
Sc
ie
nc
e
,
v
ol/
issue:
8
(2), pp
.
287
–
295
,
2017
.
[19]
Boren
L
i
and
Mao
Pan,
“
An
I
m
prove
d
Segm
ent
at
ion
of
High
Spatial
R
esolu
t
ion
Remote
Sen
sing
Im
age
usin
g
Marke
r
-
base
d
W
at
ershe
d
Algor
ithm
,”
2012
20th
Inte
rnational
Co
nfe
renc
e
on
Geo
inf
orm
ati
cs
,
I
EEE,
Hong
Kong
,
pp.
1
-
5.
pp
.
98
-
1
04,
2012
.
[20]
Vac
ava
n
t,
A.;
Chat
ea
u
,
T
.
;
W
il
hel
m
,
A.;
Le
qu
ie
vr
e,
L.
,
“
A
Benc
hm
ark
Datas
et
fo
r
Outdoor
Foreground/Ba
c
kground
Ext
r
ac
t
i
on
,”
In
Proc
ee
d
i
ngs
of
th
e
11th
Asian
Conf
ere
nc
e
on
Comput
er
Vi
sion
,
Dae
je
on,
Korea
,
pp.
291
–
300
,
5
–
9
Novem
ber
2012
.
[21]
Bashir,
F.;
Pori
kli
,
F.
,
“
Perform
anc
e
Evaluati
on
of
Objec
t
Detect
ion
and
Tracki
ng
S
y
s
t
ems
,”
In
Proce
edi
ngs
of
the
9
th
IEE
E
Int
ernati
onal
Work
shop
on
Pe
rfor
mance
Ev
alua
ti
o
n
of
Tr
ack
ing
Su
rve
il
la
n
ce
,
New
York,
NY
,
USA
,
18
June
2006
.
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.
10
, No
.
2
,
A
pr
i
l 202
0
:
1849
-
1858
1858
[22]
Open
CV
Open
Source
Com
puter
Vision.
(n
.
d.
)
.
Ret
ri
eve
d
from
I
nte
ra
ct
iv
e
Foreg
round
Ext
r
ac
t
ion
using
Grab
Cut
Algorit
hm
:
ht
tps:/
/doc
s
.
openc
v
.
o
rg/3.
1.
0
/d8/
d83/
t
utori
al
_p
y
_g
rab
cut
.
h
tml
.
[23]
Open
CV
Open
Source
Com
pute
r
Vision
.
(
n.
d.
)
.
Ret
rie
v
ed
from
C
ontours:
htt
ps://
do
cs.
open
cv.
org/3
.
3.
1
/d4/d73/tut
ori
al
_p
y
_
cont
ours_begin
.
html
.
[24]
OpenCV
Open
Source
Com
pute
r
Vision.
(n
.
d.
)
.
Ret
r
ie
ved
fro
m
Im
age
Segm
e
nta
ti
on
wi
th
wat
ershe
d
al
gor
it
h
m
:
htt
ps://
do
cs.
open
cv.
org/3
.
1.
0
/d3/db4/tut
ori
al
_p
y
_
wate
rshed.
h
tml
.
[25]
OpenCV
Open
Source
Com
p
ute
r
Vision.
(n.
d.
)
.
Ret
r
i
eve
d
from
Face
Detect
ion
Us
ing
Haa
r
-
Casca
d
es:
htt
ps://
do
cs.
open
cv.
org/3
.
3.
0
/d7/d8b/tut
ori
al
_p
y
_
fac
e_d
etec
t
ion.
h
tml
.
Evaluation Warning : The document was created with Spire.PDF for Python.