TELKOM
NIKA
, Vol. 11, No. 8, August 2013, pp. 43
7
2
~4
378
e-ISSN: 2087
-278X
4372
Re
cei
v
ed Ma
rch 1, 2
013;
Re
vised
Ma
y 12, 2013; Accepted Ma
y 22
, 2013
Resear
ch of CamShift Algorithm to Track Mo
tion
Objects
Li Zhu*, Tao Hu
F
a
cult
y
of Imformation En
gin
e
e
rin
g
, Hube
i U
n
iversit
y
for Na
tiona
lities
No. 39th
Xu
e
y
uan ro
ad, Ensh
i,
Hube
i, P.R.
Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
:
lier_z
hu@
16
3.com*, huta
o
_
505
@hotma
il.c
o
m
A
b
st
r
a
ct
At prese
n
t, obj
ects detecti
on,
ide
n
tificati
on
and
tr
ackin
g
a
r
e the v
e
ry
po
pul
ar res
earch
are
a
s
i
n
computing vis
i
on. A real-time objects
tracking system
was
proposed
in th
is paper. This paper focuses
on
improve
Ca
mShift al
gorith
m
, w
h
ich can
track the
moti
o
n
ob
jects w
i
th
hig
h
a
acc
u
racy. An Ad
apt
i
v
e
Gaussia
n
Bac
k
grou
nd Mod
e
l
is propos
ed in
the system,
which w
a
s estab
lishe
d in or
der
to automatical
l
y
upd
ate
the ba
ckgrou
nd and
detect
the
outli
ne of
mov
i
n
g
obj
ects. By an
aly
z
i
n
g d
i
ffere
nt alg
o
rith
ms,
this
pap
er bri
ngs
o
u
t appr
oac
hes
to pro
m
ote
th
e perfor
m
anc
e
.
T
he Ca
mSh
i
ft algor
ith
m
to
compl
e
te
moti
on
detectio
n
a
n
d
obj
ects trackin
g
, w
h
ich
app
li
e
d
for stat
ic
bac
kgrou
nd v
i
d
eo
sequ
enc
es. An
d the
exp
e
ri
me
nta
l
results show
that this algor
ith
m
can
detect al
l moti
on o
b
ject
s and track al
most moti
on ob
je
cts.
Ke
y
w
ords
: motion detection,
object tr
acking,
Cam
S
hift, adaptive
gaussian backg
round model
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Dete
ction an
d trackin
g
of moving obj
ects refe
rs to a
widely u
s
ed
technol
ogy
use
d
for
automatically re
co
gnizi
ng,
extr
a
c
ting
a
nd tracking
obje
c
ts f
r
om
a
seq
uen
ce
of video
im
age
informatio
n which
combi
n
e
s
with imag
e pro
c
e
ssi
ng, a
u
tomatic cont
rol, informatio
n sci
ence an
d
other
discipli
nes [1]. Det
e
ction
and
trackin
g
of
moving o
b
je
cts te
ch
nolo
g
y is o
ne
o
f
the
techn
o
logie
s
whi
c
h a
r
e i
m
porta
nt to reali
z
e
the so-called ubiq
u
itous so
ciet
y,
which ca
n
be
applie
d t
o
variou
s ind
u
st
ries
su
ch a
s
t
r
an
spo
r
tat
i
on, astro
n
o
m
ical ob
se
rvation, scie
n
tific
resea
r
ch, national defe
n
se
, and medi
cin
e
, to ac
hieve
advances
of prod
uctivity and safety, and
improve the q
uality of human life [2].
The pu
rpo
s
e
of object dete
c
tion is to det
ect
if there are the obje
c
ts
moving occu
rred in
the su
rveillan
c
e a
r
ea. If th
at, then dete
r
mine the in
fo
rmation
of location an
d si
ze of the obj
e
c
ts.
Obje
ct tracki
ng is also a
n
important
appli
c
ati
on b
r
an
ch of Co
mputer Visi
o
n
[4]. The goal o
f
tracking
is to
recogni
ze ta
rget obje
c
ts f
r
om t
he b
a
ckg
r
oun
d an
d ex
tract featu
r
e
s
, then to de
pi
ct
the moving l
o
cu
s of the
m
. How to t
r
a
c
k the o
b
je
cts pre
c
isely an
d fast is th
e
key p
r
oble
m
o
f
obje
c
t
s
t
r
a
cki
ng sy
st
e
m
.
In this paper,
the main obje
c
tive is to design
a real
-time monitorin
g
system for th
e object
intere
sted. After the imag
e pretre
atme
nt, the sy
stem firstly reco
gnizes the m
o
ving target
s in
video
seq
uen
ce
s by
ba
ckg
r
oun
d diffe
re
nce
meth
o
d
and upd
ates the
ba
ckgrou
nd
ima
ge usi
ng
Adaptive Gau
ssi
an Ba
ckground M
odel.
Acco
rdi
ng to
the outline a
n
d
colo
r of the
moving targ
ets,
the system
complete
s the tracking of
th
e obje
c
ts via CamShift alg
o
rithm.
2. Rese
arch
Metho
d
2.1. Mathem
atical Morph
o
logical Filtering
The
prin
cipl
e
of math
emat
ical
morphol
o
g
y re
fe
rs to
measure a
n
d
extra
c
t the
shape
of
obje
c
ts in ima
ges a
nd it ha
s an excellent
mathem
atica
l
foundation
whi
c
h is
set theory. The
r
e
are
four ba
si
c op
erato
r
s
of ma
thematical
m
o
rph
o
l
ogy, in
cludi
ng dilati
on, ero
s
io
n, openi
ng, cl
osing,
whi
c
h a
r
e u
s
ually defined
via set theo
ry name
s
[3, 6]. With the p
a
rallel implem
e
n
tation struct
ure,
mathemati
c
al
morp
hology
filtering can
easily
a
c
hi
eve morphol
ogical analy
s
is
and p
a
rallel
pro
c
e
ssi
ng, thus it greatly im
prove
s
th
e spe
ed of i
m
age a
nalysi
s
and
pro
c
e
s
sing.
When it
is
used in image pre-
processing, mathem
atical
morphology filter
ing, with
sm
ooth contours,
filling
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Res
e
arch of CamShift Alg
o
rithm to
Track M
o
t
i
on Obj
e
ct
s
(Li Zhu
)
4373
hole
s
, conne
cted fractu
re
zo
ne
cha
r
a
c
teri
stics,
ca
n sim
p
lify image
data to
maintain
obj
ects’
basi
c
shap
e chara
c
te
risti
c
s and rem
o
ve extraneo
us
structu
r
e
s
.
2.2. Foregro
und-b
ackg
ro
und Subtra
c
t
ion Meth
od
In this pa
per,
fore
gro
und
-b
ackgroud
sub
t
ract
ion
meth
od i
s
a
dopte
d
for target d
e
tectio
n
and extra
c
tio
n
based on th
e differen
c
e b
e
twee
n t
he current frame i
n
the video seque
nce ima
g
e
and the
ba
ckgroud. T
h
e
flowcha
r
t is sho
w
n
in
Figure 1. According
to the meth
od,
the
backg
rou
nd i
m
age
ado
pte
d
is first sto
r
e
d
in th
e
s
y
s
t
em. Be
c
a
u
s
e
th
e
r
e ar
e
d
i
ffe
r
e
nc
es
be
tw
ee
n
moving obj
ects and
ba
ckg
r
oun
d in
gray
scale o
r
co
lo
r, the target
can be
distin
g
u
ish
ed from t
h
e
backg
rou
nd i
m
age by doi
n
g
sub
s
tractio
n
. Each pixel
value in the result will
be compa
r
ed
with
a
pre
-
set threshold value. If the value of the pixed
is g
r
eater than th
e threshold, this poi
nt belo
ngs
to the foregro
und, otherwi
se
it belong
s to backg
rou
n
d
.
Figure 1. The
Flow of Fore
grou
nd
-ba
c
kg
roun
d Subtra
ction Metho
d
2.3. Adap
tiv
e
Gaus
sian Bac
k
grou
nd Model
Ho
wever, the
backg
rou
nd i
s
relatively se
nsitiv
e to external
co
nditio
n
s
su
ch a
s
ill
umition.
There are m
any false target point whi
c
h will
redu
ce the corre
c
t
rate. So, we adopte
d
the
Adaptive G
a
ussian
Ba
ckgrou
nd
Mod
e
l to
adapt
to ch
ang
es i
n
the
ba
ckg
r
oun
d. Ada
p
tive
Gau
ssi
an Ba
ckgro
und M
o
del applie
s to
the case of
the cam
e
ra st
ationary. It firstly establi
s
h
ed
a
backg
rou
nd
model fo
r a
stationa
ry ba
ckgro
und, a
n
d
then auto
m
atic up
date
s
the b
a
ckg
r
ound
model a
c
cord
ing to the cha
nge of the en
vironme
n
t.
The mo
del e
s
tabli
s
he
s th
e ori
g
in b
a
ckgrou
nd m
ode
l
0
B
via Equatio
n 1. The
0
ref
e
rs
to the pixel average valu
e and
2
0
refers
to pixel variance.
]
[
B
2
0
0
0
，
.
(1)
It defines a
p
a
ram
e
ter
to pre
s
ent the
u
pdate rate u
s
i
ng Equatio
n
2. In the form
ula, K
is
a
co
nst
b
e
long
s to
[0,
1
] and
the
ch
ara
c
ter t p
r
e
s
ents th
e fram
e t. The
)
,
|
(
1
1
t
t
t
f
means the Gaussian
Probability Densit
y Function i
n
whi
c
h
1
-
t
refers
to the
avera
g
e
value
an
d
1
-
t
refers
to the varianc
e
.
)
,
|
(
K
1
1
t
t
t
f
(2)
The Equatio
n
3 sho
w
s the
method to ca
culate
the pix
e
l averag
e value of the fra
m
e t.
t
t
f
1
t
)
1
(
(3)
The varia
n
ce of the frame t is ca
culate
d usin
g Equatio
n 4.
)
(
)
(
)
1
(
2
1
2
t
t
t
T
t
t
t
f
f
(4)
The equ
ation
5 discrib
e
s h
o
w to esta
blishes the b
a
ckgrou
nd mo
del
of frame t.
]
[
B
2
t
t
t
，
(5)
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 8, August 2013: 4372 –
4378
4374
3. Ke
y
Algorithm
3.1. MeanShift Algo
rithm
MeanShift is non-pa
ramet
r
ic method ba
sed on d
e
n
s
ity function est
i
mation. On basi
s
of
the pro
babilit
y distributio
n
gradi
ent, M
eanShift
find
s the n
earest extremum
and impl
eme
n
ts
efficiently tracking throug
h color probability distri
bution of objects [5]
.
3.2. CamShift Algorithm
On ba
si
s of MeanShift al
gorithm,
Cam
S
hift Algor
ithm achi
eves
o
b
ject
s tra
cki
n
g
throu
gh
iterative cal
c
ulation, whi
c
h mean
s Ca
mShift defi
nes the initial value of sea
r
ch
wind
ow for
MeanShift al
gorithm
usi
n
g the re
sult
s gotten in th
e previo
us frame [7]. The
formula
of the
algorithm is as follows.
Ca
culate th
e
se
con
d
-m
on
ment of the
ma
tch
wind
o
w
. The Eq
ua
tion 6 di
scrib
e
s the
se
con
d
-m
om
ent of X direction. The equ
ation 7 di
scri
bes the
se
co
nd-m
o
ment o
f
Y direction
and
the Equation
8 discrib
e
s th
e se
con
d
-m
o
m
ent of XY directio
n.
I(x,y)
x
M
xy
2
20
(6)
xy
02
I(x,y)
y
M
2
(7)
xy
11
y
x
xyI
M
)
,
(
(8)
Ca
culate thre
e operational
para
m
eters u
s
ing Equ
a
tion
9.
2
c
00
20
x
M
M
a
(9-a)
)
y
x
M
M
2(
b
c
c
00
11
(9-b)
c
y
M
M
c
00
02
(9-c
)
Ca
culate the l
ength an
d wi
dth of t
he match wi
ndo
w u
s
ing Equation
10.
2
)
(
)
(
2
2
c
a
b
c
a
length
(10-a)
2
)
(
)
(
2
2
c
a
b
c
a
width
(10-b)
3.2. Compari
s
on
MeanShift algorithm
can
match the tra
cki
ng ra
nge,
but the initial value of the search
wind
ow
ca
n
not be a
d
ju
sted. Whe
n
the tra
cki
ng t
a
rget
s move
towards
or
away fro
m
the
aqui
siton dev
ice
s
, the targ
et’s are
a
will
enlarg
e
or redu
ce in vari
ous d
egree
s. If
the search
wind
ow rema
ins the
same
size, the ta
rget’s a
r
ea
wil
l
essily exce
ed the search wind
ow o
r
be
within
a n
a
rrow
ran
ge.
Howeve
r, Cam
S
hift Algorith
m
complete
s the
real
-tim
e an
d a
c
curacy
tracking tho
u
gh the ada
ptive adjus
tme
n
t
of the search wind
ow.
CamShift alg
o
rithm
sea
r
ch
the obje
c
ts a
l
ong t
he
aradi
ent direction,
con
s
id
erin
g o
n
ly the
colo
r di
stribut
ion rath
er tha
n
obje
c
t co
ntours. Th
e
r
efo
r
e, CamS
hift can
well ad
a
p
t to the external
environ
ment.
The time
complexity of CamShift is
O(N2),
whi
c
h
is a fa
st, re
al-time tracki
ng
algorith
m
an
d ca
n re
du
ce the tra
cki
n
g
cycl
e. Bea
c
au
se
CamS
hift sho
w
s
g
ood a
n
ti-jam
ming
ability, a considerable degree of noi
se, partial oc
clusi
o
n of disruptors an
d the light changes can
Evaluation Warning : The document was created with Spire.PDF for Python.
TEL
K
be i
g
been
4.Sy
s
4.1.
S
movi
n
back
g
su
cc
e
movi
n
the
v
influ
e
math
enha
colo
r
HSV
imag
e
throu
accu
r
4.2. I
platf
o
Wind
the
d
follo
w
1.
S
reali
z
K
OM
NIKA
g
no
red. Ca
m
us
ed
in
th
e
s
tem
De
sig
n
S
ys
t
e
m
S
t
r
u
The p
r
o
p
n
g obje
c
ts
t
g
rou
nd met
h
e
ss
rate. In
n
g obje
c
ts r
a
Firstly, t
h
v
ideo i
m
ag
e
e
nc
es
s
u
c
h
emati
c
al
m
o
nce the stru
c
RGB col
o
similarit
y
e
a
colo
r s
p
ac
e
e
s int
o
col
o
gh
Me
an
Sh
i
r
acy tra
c
kin
g
mplement
a
t
A
came
r
a
o
r
m
s
ar
e c
o
m
ows
XP, V
S
d
ifferent ex
p
w
s.
S
ingle movin
g
The exp
e
z
ed m
o
ving
o
m
Shift s
h
ow
s
sy
st
em t
o
i
m
n
and Imple
u
cture
p
os
e
d
sy
st
e
m
t
rac
k
in
g. In
t
h
od wa
s us
e
the other p
a
a
ther than th
e
h
e sy
st
em e
s
e
s to sepa
r
as illuminat
o
rphology filt
e
c
ture.
o
r sp
ace
is
c
a
s
ily. Then,
e
. After that,
o
r pro
babi
lit
y
i
ft algorithm
g
.
t
ion and
Ex
p
a
w
a
s set
in
m
p
o
s
e
d of
a
S
2010, Op
en
p
erim
ental
e
n
g
obje
c
t det
e
e
rim
ental re
o
b
j
ec
ts
dete
c
e-I
Re
sea
r
c
h
o
s
excell
en
t
e
m
prove tra
c
k
i
menta
t
io
n
m
is
co
m
p
os
e
t
he part of
e
d in o
r
de
r
a
rt of the sy
s
e
whole o
n
e
Figure
2
s
tabli
s
hes th
e
r
ate the
ba
c
ion. After t
h
e
r is used t
o
c
onverted in
t
build a col
o
acco
rdi
n
g
t
o
y
distributio
n
a
nd Cam
S
h
p
eriments
a stat
ionar
y
a
comp
uter
a
CV2.3 and
v
n
vironment,
e
ction an
d tr
a
sults ar
e
s
h
c
tion an
d tra
c
SSN: 2087
-
2
o
f CamShift
A
e
ffic
i
enc
y, a
c
i
ng a
c
curac
y
e
d
of two p
a
moving obj
e
to reduce t
h
s
tem, CamS
. The flow o
f
2
. Sys
t
em P
r
e
Adaptive
G
c
kg
rou
nd a
n
h
at, in orde
r
o
eliminate t
h
t
o HSV
col
o
o
r hi
stog
ram
o
the color
h
n
s imag
es,
w
h
ift algorith
m
y
pl
ace t
o
m
o
a
nd
a
we
bc
a
v
ideo parse
r
three
grou
p
a
ck
ing
h
own i
n
Fig
u
c
king with
A
d
2
78
X
A
lg
orithm
to
c
cur
a
cy
a
n
d
y
.
a
rts th
at are
e
ct
s det
e
c
t
i
n
h
e extern
al
hift applied
t
f
the pro
c
es
s
r
ocess Flow
G
aussian
Ba
c
n
d foregrou
r
to obtain
h
e noise, si
m
r sp
ace so
t
throug
h de
t
h
is
togram, t
h
w
hi
ch mea
n
m
，
t
he sy
st
e
m
o
nitor a
sce
n
a
m. And t
h
e
xvid and Vi
r
p
s of
expe
ri
u
re 3,
whi
c
h
d
aptive Gau
s
Track M
o
t
i
o
r
o
bu
s
t
ne
ss
m
o
ving obj
e
g,
the auto
m
i
n
terferen
ce
t
o tracking t
s
es
is
sh
ow
n
c
kgro
und M
o
nd and
re
d
the conto
u
r
m
plify the sh
a
t
ha
t h
u
m
an
e
t
ac
h
i
ng
H
p
a
h
e sy
st
em
c
s r
e
ver
s
e
p
m
completes
n
e. In this
p
a
e
s
o
ftw
ar
e
p
r
turalDub 1.
1
mental re
su
h
p
r
ove
tha
s
sian B
a
ck
g
r
o
n Obj
e
ct
s
(
L
, therefore,
e
ct
s d
e
t
e
c
t
in
m
at
ically up
d
an
d
impr
o
v
h
e
cent
roid
in Figure
2.
o
del after in
p
d
u
c
e the e
x
more effe
c
a
pe
of objec
t
e
ye c
a
n
jud
g
a
ra
meter fr
o
c
o
n
verts the
p
roj
e
ction.
A
the real
-tim
a
pe
r, the ha
r
d
p
latf
orm
s
co
n
1
0.
1.
A
cco
r
d
lts
ar
e sh
o
w
t the s
y
s
t
e
m
r
o
und Mod
e
l
L
i Zhu)
4375
it
has
g and
d
ating
v
e the
of the
p
utting
x
te
rnal
c
tively,
t
s
and
g
e th
e
m t
h
e
or
ig
in
A
t las
t,
e a
nd
d
ware
n
c
l
ude
d
ing to
w
n a
s
m
ha
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 8, August 2013: 4372 –
4378
4376
Figure 3. Single Moving O
b
ject Detectio
n and
Trackin
g
in Different
Weath
e
r Con
d
ition
In different test scena
rio
s
, we can get th
e experim
ent
al results sho
w
n in Tabl
e 1
.
Table 1. Experime
n
tal Re
sults in
Different
Weath
e
r Con
d
ition
scene
Object detection
Object tracking
sunn
y
√
√
r
a
ining
√
√
2.
Two movin
g
obje
c
ts dete
c
tion and tra
c
king
Whe
n
the m
o
ving target
s i
n
crea
se to t
w
o and th
e two
target
s are cl
ose,
we find t
hat the
search window will
automatically
expand and the cont
our
of two targets
detected will
be joined
together. Th
e
rea
s
on fo
r this is th
at mathematic
al
morp
holo
g
y filter ca
n fill some sm
all h
o
les
throug
h cl
osi
ng op
erato
r
.
In Figure 4,
if the
distan
ce between t
he targ
ets i
s
not clo
s
e, t
h
e
system h
a
s
d
e
tected t
w
o o
b
ject
s; ho
wev
e
r, it has
t
r
a
c
ked
only one
target a
c
cordi
ng to the o
r
d
e
r
scanni
ng fro
m
top down.
Figure 4. Two
Moving Obje
cts De
tectin
g and Trackin
g
Re
sults
3.
Multiple movi
ng obje
c
ts d
e
t
ection an
d tracking
Whe
n
all th
e
target
s a
r
e
clo
s
e, the
se
ar
ch wind
ow will
auto
m
ati
c
ally
expa
nd and
th
e
conto
u
r of th
e target
s det
ected
will be j
o
ined tog
e
the
r
. In addition,
the system
h
a
s d
e
tecte
d
two
obje
c
ts; ho
wever, it has tracked o
n
ly one target
a
ccordin
g to the orde
r sca
nnin
g
from top do
wn.
Figure 5 sho
w
s th
at the result of movi
ng obj
e
c
ts d
e
tection
a
nd tracking ba
se
d
on
CamShi
ft
algorith
m
.
Figure 5. Multiple Moving O
b
ject
s Det
c
tio
n
and Trackin
g
Re
sults
The re
sult for different num
ber of moving
targets i
s
sh
own in Ta
ble
2.
Table 2. Experime
n
tal Re
sults fo
r Diffe
rent Nu
mbe
r
of Target
s
The numbe
r of ta
rgets
Object detection
Object tracking
single
√
√
tw
o
√
Distant
objects
multiple
√
Distant
objects
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Res
e
arch of CamShift Alg
o
rithm to
Track M
o
t
i
on Obj
e
ct
s
(Li Zhu
)
4377
4.
Accu
ra
cy rate
analysi
s
In ord
e
r t
o
de
tect the
accu
racy of th
e alg
o
rithm,
we u
s
ed 5
0
sets
of video
scene
whi
c
h
are re
sp
ectiv
e
ly compo
s
e
d
by a different numbe
r of moving objects. The re
sult is sh
own
in
Figure 6.
A: Target det
ected a
nd tra
c
ked in the
s
a
me
ar
ea
B:Target dete
c
ted an
d tracked in the
different area
s
Figure 6. Accura
cy Rate A
nalysi
s
From
the ab
ove
results, we ca
n sp
eculate
that
th
e alg
o
rithm
i
s
a
b
le
to d
e
t
ect the
moving o
b
je
cts in th
e m
oni
tored
area. B
u
t the a
c
cu
ra
cy rate of th
e
algo
rithm
de
cre
a
ses line
a
r
ly
with the incre
a
se in the n
u
m
ber of movi
ng obje
c
ts.
5.
Dete
ction an
d tracking u
n
der continu
o
u
s
video sce
n
e
For fu
rthe
r a
nalyzin
g the
accuracy
rat
e
of
the
algo
rithm, we ha
ve com
p
leted
the te
st
unde
r so
me sets of comtin
uou
s video scene. The results are
sho
w
n in Figure 7.
A: detecting a
nd trackin
g
al
l target
B:
detecting all target, tracking the mo
st
C: detectin
g
all target, tracking the mo
st
D: detectin
g
all target, tracking the mo
st
E: detecting a
ll, trackin
g
the remote ta
rg
et
F: detecting a
nd trackin
g
al
l target
Figure 7. Analysis of Multip
le Moving Obj
e
cts T
e
a
ckin
g
0%
50%
100%
150%
Accuarcy
Trace
people
0%
50%
100%
150%
Accuarcy
Trace
people
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 8, August 2013: 4372 –
4378
4378
From
the ab
ove
results, we ca
n sp
eculate
that
th
e alg
o
rithm
i
s
a
b
le
to d
e
t
ect the
moving obj
ects und
er
co
mtinuou
s vid
eo sce
ne, ho
wever, the
a
c
cura
cy rate
of the algo
rithm
decrea
s
e
s
wit
h
the in
crea
se in the
comp
lexity of
the chang
es i
n
the
monite
red
area. Tho
ugh
the
algorith
m
ca
n track the
remote target
well, it
is no
t effective to track the
clo
s
e-ra
nge m
o
ving
target, whi
c
h
is one of the follow-up
work in the proje
c
t.
5. Conclusio
n
In this pape
r, we have pro
posed a mon
i
tori
ng sy
ste
m
for the static backg
rou
n
d
video
seq
uen
ce
s.
T
h
is s
u
rv
eilla
n
c
e sy
st
em m
a
inly
com
b
in
es two
su
b-systems, moti
on dete
c
tion
and
obje
c
t tra
cki
n
g
. Ho
w to
de
tect an
d track the
obj
ect
s
pre
c
i
s
ely a
n
d
fast i
s
th
e
key p
r
obl
em
of
obje
c
ts
dete
c
tion an
d tra
c
king
sy
stem.
By analyzi
n
g
the d
e
tectin
g an
d tra
c
kin
g
alg
o
rithm, t
h
is
pape
r b
r
in
gs out m
e
thod
s to
promote
the p
e
rf
o
r
m
ance. Experi
m
ents show
our metho
d
can
achi
eve the pre
-
dete
r
min
ed target
s. Furthe
r improv
ement of the system is
o
u
r cu
rr
ent wo
rk to
deal with p
o
rt
rait and shad
ow overl
ap.
Ackn
o
w
l
e
dg
ements
This wo
rk
wa
s
fina
ncially suppo
rted by
the
Program o
f
Educatio
n
Commissio
n of
Hu
bei
Province (B2
0122
905
) an
d the Progra
m
of Ethni
c and Reli
giou
s Affairs Co
mmission of Hub
e
i
Province (HBMW20
120
06
). Moreove
r
, we would li
ke to thank T
ang Yulei
for her work in
this
term.
Referen
ces
[1]
Z
hang h
eng
ju
a
n
. Research of
target detectio
n
and trackin
g
base
d
on bloc
ked Gaussia
n
backgr
oun
d.
T
i
anjin N
o
rmal
Univers
i
t
y
. 20
0
8
.
[2]
Z
heng j
i
an
gb
in
g, zhang
ya
nn
i
ng, F
eng da
ji
n
g
, Z
hao
rongc
h
un. Movin
g
Obj
e
ct Detection a
nd
T
r
acking
Algorit
hm for V
i
deo Mo
nitori
ng
.
Systems Engi
neer
ing a
nd El
ectronics
. 20
02
; 10(24): 34-3
7
．
[3]
Dorin Comaniciu, V
i
sva
n
a
tha
n
Ramesh, Pet
e
r Meer
. Real-
T
ime
T
r
acking of Non-Ri
gid O
b
jects usin
g
MeanS
hift.
IEEE Confere
n
ce
on Co
mputer V
i
sion
and P
a
ttern Reco
gniti
on
.
2000; 2: 14
2-1
49.
[4]
D Coma
nici
u, P
Meer
.
Mea
n
S
hift Analysis
and Ap
plic
atio
ns
. Proc. Int'l Conf. Computer V
i
sion. 1999:
1
1
97
-1
20
3
.
[5]
MJ Sw
ain, DH Ballard.
Index
i
ng via col
o
r hi
stogra
m
s
. Proc. of
the
T
h
ird Internati
o
n
a
l C
onfere
n
ce on
Comp
uter V
i
si
o
n
. 1990; 3
90-3
93.
[6]
Gao Ch
ong. I
n
vestig
atio
n a
nd
Ana
l
ysis of
Ar
ithmetic of Human M
o
ti
on Det
e
ction
and
T
r
acki
ng.
Engi
neer
in
g C
ontrol C
o
mput
ers
. 2008; 2
1
(1
0): 49-50.
[7]
Cha
o
-Ch
i
ng
H
o
, Chin
g-L
ong
Shih.
A
Real-T
i
m
e F
u
zz
y
R
e
a
s
oni
ng Bas
ed
Contro
l S
y
ste
m
for Catchin
g
a Moving Goldf
i
sh.
Internatio
n
a
l Jour
nal of C
ont
rol,
Aut
o
mat
i
on, an
d Syste
m
s
. 20
09; 7(5):
755-7
63.
Evaluation Warning : The document was created with Spire.PDF for Python.