I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
,
p
p
.
4
2
9
2
~
4
3
0
3
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
5
.
p
p
4
2
9
2
-
4
3
0
3
4292
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Dy
na
mic spo
rts e
nv
iro
nments: au
t
o
ma
tic
featur
e pl
a
y
ers
tracking
and r
e
-
i
dentif
i
ca
tion wi
th
simila
r appea
ra
n
ces
Adil Ab
du
lh
ur
Abu
s
ha
na
1
,
Riy
a
dh
Als
a
ee
di
2
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
E
d
u
c
a
t
i
o
n
,
U
n
i
v
e
r
s
i
t
y
o
f
K
u
f
a
,
N
a
j
a
f
,
I
r
a
q
2
D
e
p
a
r
t
me
n
t
o
f
S
t
a
t
i
s
t
i
c
s,
C
o
l
l
e
g
e
o
f
A
d
mi
n
i
s
t
r
a
t
i
o
n
a
n
d
Ec
o
n
o
mi
c
s
,
U
n
i
v
e
r
si
t
y
o
f
W
a
si
t
,
W
a
si
t
,
I
r
a
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
y
2
4
,
2
0
2
5
R
ev
is
ed
J
u
l 2
2
,
2
0
2
6
Acc
ep
ted
Au
g
1
2
,
2
0
2
6
Trac
k
in
g
m
u
lt
i
p
le
p
lay
e
rs
i
n
s
p
o
rts
v
i
d
e
o
s
re
m
a
in
s
c
h
a
ll
e
n
g
i
n
g
d
u
e
to
fre
q
u
e
n
t
o
c
c
lu
sio
n
s,
sim
il
a
r
a
p
p
e
a
ra
n
c
e
s
a
m
o
n
g
p
lay
e
rs,
a
n
d
ra
p
i
d
m
o
ti
o
n
.
Th
is
p
a
p
e
r
p
re
se
n
ts
a
c
o
m
p
re
h
e
n
siv
e
p
la
y
e
r
trac
k
in
g
fra
m
e
wo
rk
th
a
t
in
teg
ra
tes
d
e
e
p
p
lay
e
r
re
-
id
e
n
ti
fi
c
a
ti
o
n
(DPRI),
a
n
in
d
iv
id
u
a
l
p
r
o
b
a
b
i
li
stic
o
v
e
rlap
p
in
g
b
o
u
n
d
in
g
b
o
x
o
c
c
u
p
a
n
c
y
m
a
p
p
in
g
(IP
OM)
f
o
r
r
o
b
u
st
sp
a
ti
a
l
lo
c
a
li
z
a
ti
o
n
u
n
d
e
r
se
v
e
re
o
v
e
rlap
,
a
n
d
a
n
id
e
n
t
it
y
-
a
wa
re
k
-
s
h
o
r
tes
t
p
a
th
o
p
ti
m
iza
ti
o
n
a
lg
o
rit
h
m
(KSP
-
ID
)
fo
r
imp
r
o
v
e
d
traje
c
to
ry
a
ss
o
c
i
a
ti
o
n
.
Th
e
DPRI
m
o
d
u
le
c
o
m
b
i
n
e
s
p
o
se
-
b
a
se
d
e
m
b
e
d
d
in
g
s,
jers
e
y
a
p
p
e
a
ra
n
c
e
c
u
e
s,
a
n
d
fa
c
ial
fe
a
tu
re
s
t
o
g
e
n
e
ra
t
e
d
isc
rimin
a
ti
v
e
id
e
n
ti
ty
re
p
re
se
n
tatio
n
s.
Ex
p
e
rime
n
tal
e
v
a
lu
a
ti
o
n
o
n
t
h
e
a
u
to
n
o
m
o
u
s
p
ro
d
u
c
ti
o
n
o
f
ima
g
e
s
b
a
se
d
o
n
d
istri
b
u
ted
a
n
d
i
n
telli
g
e
n
t
se
n
si
n
g
(
APIDIS
)
a
n
d
S
h
a
n
t
o
u
Un
i
v
e
rsity
(
S
TU
)
d
a
tas
e
ts
d
e
m
o
n
stra
tes
stro
n
g
p
e
rf
o
rm
a
n
c
e
,
a
c
h
iev
in
g
8
5
.
5
%
m
u
lt
i
p
le
o
b
jec
t
trac
k
in
g
a
c
c
u
ra
c
y
(M
OTA)
,
6
9
.
0
%
g
e
n
e
ra
li
z
e
d
m
u
lt
i
p
le
o
b
jec
t
trac
k
in
g
a
c
c
u
ra
c
y
(G
M
OTA),
a
n
d
9
0
.
2
%
ra
n
k
-
1
re
-
id
e
n
ti
fica
ti
o
n
a
c
c
u
ra
c
y
.
F
u
rth
e
rm
o
re
,
t
h
e
p
r
o
p
o
se
d
fra
m
e
wo
rk
si
g
n
ifi
c
a
n
tl
y
re
d
u
c
e
s
id
e
n
ti
t
y
sw
it
c
h
e
s
(IDS)
c
o
m
p
a
re
d
wi
th
c
o
n
v
e
n
ti
o
n
a
l
k
-
sh
o
rtes
t
p
a
th
(
K
S
P
)
a
p
p
ro
a
c
h
e
s,
p
ro
b
a
b
il
isti
c
o
c
c
u
p
a
n
c
y
m
o
d
e
ls,
a
n
d
re
c
e
n
t
d
e
e
p
lea
rn
in
g
-
b
a
se
d
trac
k
in
g
m
e
th
o
d
s.
Th
e
se
re
su
lt
s
d
e
m
o
n
stra
te
th
a
t
in
teg
ra
ti
n
g
se
m
a
n
ti
c
id
e
n
ti
t
y
in
fo
rm
a
ti
o
n
wit
h
sp
a
ti
a
l
re
a
so
n
in
g
e
n
a
b
les
ro
b
u
st
a
n
d
e
fficie
n
t
p
la
y
e
r
trac
k
in
g
in
h
i
g
h
l
y
d
y
n
a
m
ic sp
o
rts
e
n
v
ir
o
n
m
e
n
ts.
K
ey
w
o
r
d
s
:
3
D
lo
ca
lizatio
n
Dee
p
p
lay
er
r
e
-
id
en
tific
atio
n
KSP
-
I
D
m
o
d
el
Play
er
tr
ac
k
in
g
SS
D
alg
o
r
ith
m
an
d
o
b
ject
in
ter
ac
tio
n
s
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ad
il Ab
d
u
lh
u
r
Ab
u
s
h
an
a
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
,
Facu
lty
o
f
E
d
u
ca
tio
n
,
Un
i
v
er
s
ity
o
f
Ku
f
a
Naja
f
,
I
r
aq
E
m
ail:
ad
elsh
an
n
a3
0
0
0
@
g
m
ai
l.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
MO
T
o
r
m
u
lti
-
o
b
ject
tr
ac
k
in
g
in
s
p
o
r
ts
is
n
ec
ess
ar
y
f
o
r
a
u
to
m
atic
an
aly
tics
,
tactica
l
p
e
r
f
o
r
m
a
n
ce
u
n
d
er
s
tan
d
i
n
g
,
a
n
d
s
m
ar
t
s
p
o
r
ts
b
r
o
ad
ca
s
tin
g
.
K
-
s
h
o
r
test
p
a
th
(
KSP)
o
p
tim
izatio
n
h
as
b
ee
n
ex
ten
s
iv
ely
u
s
ed
with
tr
ac
k
in
g
-
by
-
d
etec
tio
n
f
r
a
m
ewo
r
k
s
d
u
e
to
th
eir
a
b
ilit
y
t
o
ex
p
r
ess
th
e
id
en
tity
ass
o
ciatio
n
as
a
c
o
n
s
tr
ain
ed
f
lo
w
p
r
o
b
lem
,
wh
ich
allo
ws
th
e
g
lo
b
al
o
p
tim
izatio
n
o
f
tr
ac
k
in
g
tr
ajec
to
r
ies
to
b
e
v
er
y
ef
f
icien
t
[
1
]
.
Ho
wev
er
,
ev
en
with
th
ese
ad
v
an
tag
es,
id
en
tity
s
witch
es
(
I
DS)
ar
e
s
til
l
a
m
ajo
r
d
r
awb
ac
k
in
th
e
ca
s
e
o
f
s
p
o
r
ts
s
ettin
g
s
th
at
ar
e
v
er
y
d
y
n
am
ic
s
in
ce
t
h
e
co
n
s
tan
t
in
ter
ac
tio
n
b
etwe
en
p
lay
er
s
,
m
o
tio
n
b
lu
r
,
o
cc
lu
s
io
n
s
,
ca
m
er
a
an
g
le
ch
an
g
es,
an
d
s
im
ilar
ity
in
ap
p
ea
r
an
ce
am
o
n
g
th
e
p
lay
e
r
s
m
ak
e
it
v
er
y
d
if
f
ic
u
lt
to
k
ee
p
th
e
id
en
tity
co
n
s
is
ten
t
f
o
r
lo
n
g
p
e
r
io
d
s
o
f
tim
e
[
2
]
,
[
3
]
.
T
o
h
elp
m
itig
ate
t
h
ese
p
r
o
b
lem
s
,
s
o
m
e
r
ec
en
t
wo
r
k
s
h
av
e
p
r
o
p
o
s
ed
le
v
er
ag
in
g
b
o
t
h
ap
p
e
ar
an
ce
an
d
o
n
lin
e
lear
n
in
g
m
eth
o
d
s
t
o
h
elp
d
if
f
er
en
tiate
p
la
y
er
s
in
cr
o
wd
ed
s
ce
n
a
r
io
s
[
2
]
,
[
3
]
.
I
D
-
Net
is
o
n
e
s
u
ch
m
eth
o
d
b
ased
o
n
th
e
u
s
e
o
f
d
ee
p
r
e
p
r
esen
tatio
n
s
o
f
th
e
p
ar
ts
o
f
h
u
m
an
b
o
d
ies
to
im
p
r
o
v
e
r
o
b
u
s
tn
ess
in
id
en
tity
p
r
eser
v
atio
n
,
u
s
in
g
s
tr
u
ctu
r
al
cu
es
an
d
lo
ca
lized
b
o
d
y
p
a
r
ts
r
ath
er
th
an
r
ely
in
g
s
o
lely
o
n
h
o
lis
tic
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dyn
a
mic
s
p
o
r
ts
en
viro
n
men
ts
:
a
u
to
ma
tic
fea
tu
r
e
p
l
a
ye
r
s
tr
a
ck
in
g
a
n
d
…
(
A
d
il A
b
d
u
lh
u
r
A
b
u
s
h
a
n
a
)
4293
cu
es
[
4
]
.
On
th
e
o
th
e
r
h
a
n
d
,
j
er
s
ey
n
u
m
b
er
d
etec
tio
n
h
as
b
e
co
m
e
an
o
th
er
p
o
wer
f
u
l
id
en
tit
y
cu
e,
aid
ed
b
y
th
e
ad
v
an
ce
m
e
n
ts
o
f
co
ar
s
e
-
to
-
f
in
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
ap
p
r
o
ac
h
es
an
d
s
o
p
h
is
ticated
s
eg
m
en
tatio
n
ar
ch
itectu
r
es
th
at
ar
e
ab
le
to
is
o
late
jer
s
ey
n
u
m
b
er
d
ig
its
u
n
d
er
p
a
r
tial
o
cc
lu
s
io
n
s
[
5
]
,
[
6
]
.
Ad
d
itio
n
ally
,
p
r
o
b
ab
ilis
tic
o
cc
u
p
an
cy
m
a
p
p
in
g
,
m
u
lti
-
v
iew
f
u
s
io
n
,
p
o
s
e
esti
m
atio
n
,
an
d
ca
m
er
a
ca
lib
r
atio
n
h
av
e
im
p
r
o
v
ed
s
p
atial
lo
ca
lizatio
n
an
d
3
D
s
ce
n
e
in
ter
p
r
etatio
n
[
7
]
,
[
8
]
.
W
h
ile
th
er
e
h
av
e
b
ee
n
m
aj
o
r
ad
v
an
ce
m
e
n
ts
in
s
p
o
r
ts
p
lay
er
tr
ac
k
in
g
in
r
ec
en
t
y
ea
r
s
,
m
an
y
o
f
th
ese
m
eth
o
d
s
f
ac
e
ch
allen
g
es
in
p
r
eser
v
in
g
id
en
tity
in
ca
s
es
wh
er
e
th
er
e
is
s
im
ilar
ity
in
v
is
u
al
ap
p
ea
r
an
ce
,
p
ar
tial
je
r
s
ey
n
u
m
b
er
s
v
is
ib
ilit
y
,
o
r
o
v
er
lap
p
i
n
g
p
lay
er
s
.
O
n
e
o
f
th
e
wea
k
n
ess
es
o
f
p
r
ev
io
u
s
m
eth
o
d
s
is
th
e
s
ep
ar
ate
co
n
s
id
er
atio
n
o
f
ea
ch
o
f
th
e
av
ailab
le
id
en
tity
c
u
es
–
jer
s
ey
r
ec
o
g
n
itio
n
,
ap
p
ea
r
an
ce
f
ea
tu
r
es
o
r
f
ac
ial
in
f
o
r
m
atio
n
,
an
d
s
p
atial
lo
ca
lizatio
n
–
wh
ich
u
s
u
ally
r
esu
lts
in
lo
s
s
o
f
id
en
tity
in
c
o
m
p
lex
in
ter
ac
tio
n
s
.
I
n
lig
h
t
o
f
th
e
ab
o
v
e
p
r
o
b
lem
,
a
co
m
p
r
eh
en
s
iv
e
ap
p
r
o
ac
h
is
n
ee
d
ed
th
at
in
c
o
r
p
o
r
ates
ap
p
e
ar
an
ce
an
d
tr
ajec
to
r
y
r
ep
r
esen
tatio
n
a
n
d
i
d
en
tity
p
r
eser
v
atio
n
in
3
D
s
p
atial
lo
ca
lizatio
n
,
in
o
n
e
co
n
s
is
ten
t
o
p
tim
izatio
n
f
r
am
ewo
r
k
.
T
h
is
wo
r
k
p
r
esen
ts
a
m
u
lti
-
p
lay
er
tr
ac
k
in
g
m
eth
o
d
,
wh
ich
is
m
ad
e
u
p
o
f
th
r
ee
tig
h
tly
co
n
n
ec
ted
m
o
d
u
les.
Firstl
y
,
th
e
d
ee
p
p
lay
er
r
e
-
i
d
en
tific
atio
n
(
DPR
I
)
m
o
d
u
le
u
s
es
p
o
s
e
g
u
id
ed
em
b
ed
d
in
g
s
,
jer
s
ey
s
em
an
tics
,
an
d
f
ac
ial
in
f
o
r
m
atio
n
to
g
e
n
er
ate
h
i
g
h
ly
d
is
cr
im
in
ativ
e
id
en
tity
f
ea
tu
r
es
u
n
d
er
o
cc
lu
s
io
n
an
d
am
b
ig
u
o
u
s
co
n
d
itio
n
s
.
Seco
n
d
ly
,
in
d
iv
id
u
al
p
r
o
b
ab
ilis
tic
o
v
er
lap
p
i
n
g
b
o
u
n
d
in
g
b
o
x
o
cc
u
p
an
cy
m
ap
p
in
g
(
I
POM
)
–
th
e
in
d
iv
id
u
al
p
r
o
b
a
b
ilis
tic
o
v
er
lap
p
in
g
b
o
u
n
d
in
g
b
o
x
f
o
r
o
cc
u
p
an
cy
m
a
p
p
in
g
–
m
o
d
u
le
allo
ws
f
o
r
im
p
r
o
v
e
d
s
p
atial
co
n
s
is
ten
cy
th
r
o
u
g
h
ac
cu
r
ate
3
D
s
ce
n
e
g
r
o
u
n
d
in
g
f
r
o
m
m
u
ltip
le
v
i
ews,
esp
ec
ially
in
cr
o
wd
ed
in
ter
ac
tio
n
s
ce
n
ar
io
s
.
Fin
ally
,
th
e
id
en
tity
-
awa
r
e
k
-
s
h
o
r
test
p
ath
o
p
tim
izatio
n
m
eth
o
d
(
KSP
-
I
D)
u
tili
ze
s
an
id
en
tity
co
r
r
elatio
n
co
ef
f
icien
t
in
th
e
ass
o
ciatio
n
g
r
ap
h
,
wh
ich
s
ig
n
if
ican
tly
lo
wer
s
th
e
id
e
n
tity
s
witch
r
ate
co
m
p
ar
ed
t
o
co
n
v
e
n
tio
n
al
KSP.
2.
RE
L
AT
E
D
WO
RK
Ov
er
th
e
last
d
ec
ad
e
o
r
s
o
,
p
lay
er
d
etec
tio
n
,
id
en
tific
atio
n
,
an
d
tr
ac
k
in
g
in
s
p
o
r
ts
v
id
eo
s
h
av
e
u
n
d
er
g
o
n
e
s
u
b
s
tan
tial
p
r
o
g
r
es
s
to
war
d
m
o
r
e
ac
cu
r
ate
s
o
lu
tio
n
s
.
Par
t
an
d
lo
ca
l
s
tr
u
ctu
r
e
r
ep
r
esen
tatio
n
s
h
av
e
b
ee
n
am
o
n
g
th
e
ea
r
ly
m
eth
o
d
s
f
o
r
ad
d
r
ess
in
g
s
u
ch
task
s
as
r
e
-
id
en
tific
atio
n
an
d
co
p
in
g
with
ap
p
ea
r
an
ce
ch
an
g
es
[
4
]
.
I
n
itially
,
p
lay
er
id
en
tific
atio
n
r
elied
o
n
h
an
d
cr
a
f
ted
f
ea
tu
r
e
s
ets,
co
lo
r
,
an
d
m
o
m
en
t
r
ep
r
esen
tatio
n
s
in
v
a
r
ian
t
to
m
u
ltip
le
co
n
d
itio
n
s
b
ef
o
r
e
t
r
an
s
itio
n
in
g
to
s
ce
n
e
p
e
r
ce
p
tio
n
s
y
s
tem
s
with
in
teg
r
ated
s
eg
m
e
n
tatio
n
,
d
et
ec
tio
n
,
an
d
r
ec
o
g
n
itio
n
p
r
o
c
ed
u
r
es
[
9
]
.
I
n
t
h
e
p
r
o
ce
s
s
o
f
en
h
an
cin
g
team
id
en
tific
atio
n
r
o
b
u
s
tn
ess
,
th
er
e
h
as
b
ee
n
a
g
r
o
win
g
i
n
ter
est
in
u
tili
zin
g
team
m
em
b
er
s
'
j
er
s
ey
s
with
u
n
iq
u
e
id
en
tifie
r
s
b
ased
o
n
th
e
tech
n
o
lo
g
ies
r
an
g
i
n
g
f
r
o
m
o
p
tical
ch
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
R
)
s
y
s
tem
s
to
th
o
s
e
u
tili
zin
g
d
ee
p
lear
n
in
g
,
C
NNs,
r
eg
io
n
-
b
ased
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
R
-
C
NN
)
with
h
u
m
an
p
o
s
es,
m
u
lti
-
task
tr
ain
in
g
,
s
p
atial
tr
a
n
s
f
o
r
m
atio
n
,
an
d
v
is
u
al
-
tex
t c
o
m
b
in
atio
n
s
[
1
0
]
–
[
1
3
]
.
Mu
lti
-
tar
g
et
tr
ac
k
in
g
was
a
s
i
m
u
ltan
eo
u
s
p
r
o
ce
s
s
f
r
o
m
th
e
b
eg
in
n
in
g
,
r
ely
in
g
o
n
th
e
g
lo
b
al
o
p
tim
al
g
r
ap
h
o
p
tim
izatio
n
,
p
r
o
b
a
b
ilis
tic
r
ea
s
o
n
in
g
,
g
r
ee
d
y
s
tr
ateg
i
es,
an
d
KSP
s
o
lu
tio
n
s
am
o
n
g
o
th
er
s
to
k
ee
p
th
e
tem
p
o
r
al
id
e
n
tity
ass
o
ciatio
n
s
tab
le
[
1
4
]
–
[
1
7
]
.
T
h
e
d
ee
p
lear
n
in
g
r
e
v
o
lu
tio
n
th
en
ca
m
e
f
o
r
th
e
o
n
lin
e
t
r
ac
k
in
g
wh
ich
,
am
o
n
g
o
th
e
r
s
,
r
esu
lte
d
in
th
e
d
ee
p
s
im
p
le
o
n
lin
e
an
d
r
e
al
-
tim
e
tr
ac
k
i
n
g
(
Dee
p
SOR
T
)
m
o
d
el
th
at
m
er
g
ed
d
ee
p
ap
p
ea
r
an
ce
em
b
ed
d
in
g
s
with
Kalm
an
f
ilter
in
g
to
i
n
cr
ea
s
e
th
e
id
en
tity
co
n
tin
u
ity
[
1
5
]
f
o
llo
wed
b
y
ad
v
an
ce
d
o
p
tim
izatio
n
a
n
d
g
r
ap
h
ical
m
o
d
elin
g
a
cr
o
s
s
d
if
f
er
en
t sp
o
r
ts
co
n
te
x
ts
[
1
6
]
,
[
1
7
]
.
T
h
e
d
ee
p
lear
n
in
g
r
ev
o
lu
tio
n
ev
en
tu
ally
a
f
f
ec
ted
th
e
v
er
y
d
etec
tio
n
:
th
e
c
o
n
v
o
lu
tio
n
al
ar
ch
itectu
r
es
lik
e
f
aster
R
-
C
NN,
m
ask
R
-
C
NN,
y
o
u
o
n
ly
l
o
o
k
o
n
ce
(
YOL
O
)
,
an
d
f
ea
tu
r
e
p
y
r
am
i
d
n
etwo
r
k
s
(
FP
N)
to
o
k
t
h
e
lea
d
in
ac
cu
r
ac
y
,
s
p
ee
d
,
a
n
d
m
u
lti
-
s
ca
le
r
ea
s
o
n
i
n
g
in
th
e
d
y
n
am
ic
s
p
o
r
ts
en
v
ir
o
n
m
en
t
b
y
a
b
ig
m
ar
g
in
[
1
8
]
–
[
2
1
]
.
Glo
b
al
co
n
t
ex
tu
al
m
o
d
elin
g
alo
n
g
with
tr
an
s
f
o
r
m
er
-
b
ased
d
etec
tio
n
b
o
o
s
ted
th
e
s
tr
en
g
th
i
n
d
en
s
e
an
d
co
m
p
lex
s
ce
n
es
in
d
ir
ec
tly
[
2
1
]
,
th
e
s
am
e
way
th
at
th
e
b
en
ch
m
ar
k
in
g
in
itiativ
es
tak
in
g
p
lace
d
id
b
y
m
ak
in
g
p
e
r
f
o
r
m
an
ce
ev
al
u
atio
n
a
s
tan
d
ar
d
ac
r
o
s
s
th
e
d
at
asets
s
u
ch
as
m
u
ltip
le
o
b
ject
tr
ac
k
in
g
ch
allen
g
e
(
MO
T
C
h
allen
g
e
)
,
au
to
n
o
m
o
u
s
p
r
o
d
u
ctio
n
o
f
im
ag
es
b
ased
o
n
d
is
tr
ib
u
te
d
an
d
in
tellig
en
t
s
en
s
in
g
(
API
DI
S)
,
So
cc
er
Net,
an
d
o
th
e
r
s
in
clu
d
ed
in
th
e
lar
g
e
-
s
ca
le
p
r
o
to
co
l
s
[
2
2
]
.
I
d
en
tity
s
ig
n
als
h
elp
in
g
o
n
e
an
o
th
er
,
with
f
ac
e
r
ec
o
g
n
itio
n
a
n
d
s
p
atial
lo
ca
lizatio
n
cu
es
b
ei
n
g
th
e
m
o
s
t
p
r
o
m
in
e
n
t,
wo
r
k
ed
to
g
eth
er
f
o
r
th
e
r
eliab
le
id
en
tity
in
f
er
e
n
ce
u
n
d
er
th
e
m
o
s
t d
if
f
icu
lt scen
ar
io
s
[
2
3
]
–
[
2
5
]
.
C
las
s
ical
d
escr
ip
tio
n
m
eth
o
d
s
h
av
e
alr
ea
d
y
d
e
m
o
n
s
tr
ated
h
ig
h
ef
f
icien
cy
:
e
v
en
with
th
e
m
o
d
er
ate
r
o
b
u
s
tn
ess
o
f
h
u
m
an
d
etec
ti
o
n
b
y
h
is
to
g
r
am
o
f
o
r
ie
n
ted
g
r
ad
ien
t
(
HOG
)
,
s
p
ar
s
e
h
is
to
g
r
am
en
co
d
i
n
g
,
co
r
r
elatio
n
f
ilter
s
,
an
d
m
u
ltip
le
in
s
tan
ce
lear
n
i
n
g
,
it
was
e
asy
to
f
o
o
l
s
u
ch
s
y
s
tem
s
wit
h
o
cc
lu
s
io
n
,
m
o
tio
n
b
lu
r
r
in
g
,
a
n
d
s
ig
n
if
ican
t
s
ca
le
ch
an
g
e
[
2
6
]
,
[
2
7
]
.
T
h
e
jer
s
ey
r
ec
o
g
n
itio
n
an
d
id
e
n
tity
m
o
d
eli
n
g
p
r
o
ce
s
s
es
wer
e
g
o
in
g
th
r
o
u
g
h
th
e
ar
ch
itectu
r
al
ad
v
an
ce
m
e
n
ts
o
f
C
NNs,
v
is
u
al
-
tex
tu
al
d
ata
f
u
s
io
n
,
an
d
r
o
b
u
s
t
r
ec
o
g
n
itio
n
tech
n
iq
u
es
s
u
itab
le
f
o
r
s
o
lv
i
n
g
th
e
p
r
o
b
lem
o
f
v
ar
ia
b
ilit
y
in
m
atch
es
[
2
8
]
.
Me
an
wh
ile,
n
ew
tr
ac
k
in
g
ap
p
r
o
ac
h
es
wer
e
b
ein
g
d
esig
n
ed
to
ac
h
iev
e
g
lo
b
al
o
p
tim
ality
in
ter
m
s
o
f
ass
o
ciatio
n
,
p
r
o
b
ab
ilis
tic
g
r
ap
h
ical
m
o
d
els,
an
d
e
f
f
icien
t
id
en
tit
y
s
tab
ilit
y
[
2
9
]
–
[
3
4
]
;
h
en
ce
,
th
ey
b
en
e
f
ited
f
r
o
m
th
e
s
t
r
u
ctu
r
ed
tem
p
o
r
al
r
ea
s
o
n
in
g
ad
v
an
ce
s
.
Po
s
e
-
b
as
ed
p
lay
er
d
escr
ip
to
r
s
alo
n
g
with
th
eir
s
tr
u
ctu
r
al
an
d
p
ar
t
-
b
ased
ch
ar
ac
ter
is
tics
h
av
e
b
ee
n
o
n
e
o
f
th
e
k
ey
r
e
-
id
en
tific
atio
n
im
p
r
o
v
em
en
ts
in
a
lm
o
s
t id
en
tical
s
p
o
r
ts
d
o
m
ain
s
[
3
5
]
–
[
4
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
2
9
2
-
4
3
0
3
4294
Su
ch
f
u
r
th
e
r
im
p
r
o
v
em
en
ts
h
a
v
e
h
elp
ed
in
r
ea
ch
in
g
h
ig
h
p
e
r
f
o
r
m
a
n
ce
lev
els
in
th
e
f
o
llo
w
in
g
ar
ea
s
:
h
ig
h
-
o
r
d
er
o
p
tim
izatio
n
in
g
r
ap
h
s
,
h
eu
r
is
tic
s
o
lu
tio
n
s
f
o
r
b
etter
ass
o
ciatio
n
,
an
d
r
esil
ien
t
jer
s
ey
id
e
n
tity
p
er
s
is
ten
ce
in
ch
allen
g
in
g
s
ce
n
es
[
4
1
]
,
[
4
2
]
.
Stan
d
a
r
d
d
atasets
,
b
en
ch
m
ar
k
s
,
an
d
e
v
alu
ati
o
n
p
r
o
to
co
ls
h
elp
ed
in
p
r
o
v
id
in
g
u
n
b
iased
c
o
m
p
ar
is
o
n
b
etwe
en
co
m
p
etin
g
alg
o
r
ith
m
s
[
4
3
]
.
Par
t
-
b
ased
r
e
-
id
e
n
tific
atio
n
ap
p
r
o
ac
h
es
wer
e
cr
ea
ted
to
u
s
e
m
u
lti
-
s
ca
le
s
tr
u
ctu
r
al
f
ea
tu
r
es
an
d
lo
ca
l
d
is
cr
im
in
ativ
e
p
atch
es;
th
er
ef
o
r
e
,
jer
s
ey
d
ig
its
h
av
e
b
ec
o
m
e
less
s
en
s
itiv
e
in
ca
s
e
o
f
m
ask
in
g
o
r
d
is
to
r
tio
n
[
4
4
]
.
T
h
e
jer
s
ey
r
e
co
g
n
itio
n
m
eth
o
d
s
wer
e
g
r
a
d
u
ally
ex
p
a
n
d
in
g
,
an
d
v
is
u
al
p
atter
n
s
an
d
n
u
m
er
ical
id
en
tifie
r
s
b
ec
am
e
i
n
cr
ea
s
in
g
ly
r
elate
d
s
em
an
tically
;
h
en
ce
,
th
e
r
elia
b
ilit
y
o
f
g
am
e
p
lay
an
al
y
s
is
b
ec
am
e
h
ig
h
er
[
4
5
]
–
[
4
7
]
.
B
esi
d
es,
b
en
ch
m
a
r
k
in
g
an
d
s
tr
u
ctu
r
ed
e
v
alu
atio
n
s
h
av
e
f
in
ally
cr
ea
ted
a
s
o
lid
b
ase
f
o
r
o
b
jectiv
e
ass
ess
m
en
t
in
s
p
o
r
ts
an
aly
tics
r
esear
ch
[
4
8
]
.
W
ith
th
e
d
ev
elo
p
m
en
t
o
f
m
u
lti
-
ca
m
e
r
a
tr
ac
k
in
g
in
d
en
s
ely
p
o
p
u
lated
s
p
o
r
ts
s
ce
n
ar
io
s
,
ad
v
an
ce
d
KSP
o
p
tim
izatio
n
,
p
r
o
b
ab
ilis
tic
f
o
r
m
u
latio
n
s
,
an
d
id
e
n
tity
p
er
s
is
ten
ce
m
ec
h
an
is
m
s
ca
m
e
in
to
p
lay
[
4
9
]
,
[
5
0
]
.
T
h
e
u
s
e
o
f
p
r
o
b
ab
ilis
tic
o
cc
u
p
an
c
y
m
a
p
p
in
g
,
v
o
l
u
m
etr
ic
r
ea
s
o
n
in
g
,
d
ep
t
h
cu
es,
an
d
m
ar
k
ed
p
o
in
t
p
r
o
ce
s
s
es
b
r
o
u
g
h
t
to
lo
c
aliza
tio
n
a
r
em
ar
k
ab
le
r
esil
ien
ce
ev
en
u
n
d
er
s
ev
er
e
o
cc
lu
s
io
n
[
5
1
]
–
[
5
6
]
,
an
d
all
o
f
th
ese
ad
v
an
ce
d
tech
n
iq
u
es
r
esu
lted
in
th
e
cr
ea
tio
n
o
f
s
t
ate
-
of
-
th
e
-
a
r
t
m
u
lti
-
ca
m
er
a
tr
ac
k
in
g
p
ip
elin
es
in
wh
ich
s
p
atial
r
e
-
id
en
tific
atio
n
,
ca
lib
r
atio
n
,
an
d
id
e
n
tity
f
u
s
io
n
s
tr
ateg
ies
to
g
eth
er
p
r
o
v
id
e
d
ac
cu
r
ate
m
u
lti
-
p
lay
er
tr
ac
k
in
g
p
er
f
o
r
m
an
ce
[
5
2
]
,
[
5
7
]
.
Ho
wev
er
,
n
o
twith
s
tan
d
in
g
t
h
is
im
p
r
ess
iv
e
ad
v
an
ce
m
en
t
,
m
o
s
t
o
f
th
e
p
r
ev
io
u
s
m
e
th
o
d
s
s
till
co
n
s
id
er
jer
s
ey
r
ec
o
g
n
itio
n
,
f
ac
ial
cu
es,
s
p
atial
lo
ca
lizati
o
n
,
an
d
id
en
tity
ass
o
ciatio
n
as
lo
o
s
ely
r
elate
d
m
o
d
u
les,
wh
ich
m
ak
es
id
en
tit
y
v
er
y
wea
k
in
s
itu
atio
n
s
wh
en
p
lay
er
s
f
r
eq
u
en
tly
o
v
e
r
lap
,
h
av
e
v
er
y
s
im
ilar
ap
p
ea
r
an
ce
s
,
o
r
m
o
v
e
th
r
o
u
g
h
th
e
v
iew
o
f
m
u
ltip
le
ca
m
er
as.
As
a
r
esu
lt
o
f
th
ese
d
ef
ici
en
cies,
o
u
r
r
esear
ch
in
teg
r
ates
p
o
s
e
-
g
u
i
d
ed
jer
s
ey
r
ec
o
g
n
itio
n
,
d
ee
p
r
e
-
i
d
en
tifi
ca
tio
n
,
p
r
o
b
ab
ilis
tic
m
u
lti
-
v
ie
w
lo
ca
lizatio
n
,
an
d
id
en
tity
-
awa
r
e
KSP
o
p
tim
izat
io
n
in
to
o
n
e
co
h
er
e
n
t
f
r
am
ew
o
r
k
th
at
is
s
p
ec
if
ically
d
esig
n
ed
f
o
r
th
e
r
eliab
le
id
en
tity
p
r
eser
v
atio
n
in
h
i
g
h
ly
d
y
n
am
ic
s
p
o
r
ts
en
v
ir
o
n
m
en
ts
.
3.
M
E
T
H
O
D
3
.
1
.
Da
t
a
s
et
T
h
e
d
ee
p
p
lay
e
r
m
o
d
el
is
tr
ai
n
ed
in
th
r
ee
s
tag
es
f
o
llo
win
g
th
e
im
ag
e
-
ce
n
tr
ic
s
am
p
lin
g
s
tr
ateg
y
[
2
1
]
with
th
e
s
to
ch
asti
c
g
r
ad
ien
t d
escen
t (
SGD
)
s
o
lv
er
[
2
2
]
.
T
h
e
r
esid
u
al
n
etwo
r
k
5
0
la
y
er
s
(
R
esNet
-
50
)
b
ac
k
b
o
n
e
is
p
r
e
-
tr
ain
ed
,
wh
ile
th
e
r
e
m
a
in
in
g
lay
er
s
ar
e
r
an
d
o
m
ly
in
itialized
with
a
Gau
s
s
ian
d
is
tr
ib
u
tio
n
.
T
h
e
lo
s
s
is
d
ef
in
ed
as (
1
)
.
=
1
+
2
(
1
)
W
ea
k
ly
lab
eled
p
lay
er
b
o
u
n
d
in
g
b
o
x
es
co
n
tr
ast
with
f
u
lly
lab
eled
jer
s
ey
n
u
m
b
e
r
b
o
x
es
.
API
DI
S
o
f
f
er
s
1
0
,
0
0
0
wea
k
ly
an
d
2
,
0
0
0
f
u
lly
lab
eled
s
am
p
les,
with
em
p
ir
ical
weig
h
ts
λ
₁
=
0
.
5
5
a
n
d
λ
₂
=
0
.
5
5
.
I
POM
an
d
KSP
-
I
D
p
ar
a
m
eter
s
ar
e
d
ep
en
d
en
t
o
n
d
ata
s
ets:
g
r
id
s
o
f
s
ize
1
2
8
×
7
2
f
o
r
API
DI
S,
an
d
1
1
2
×
6
0
f
o
r
Sh
an
to
u
Un
iv
e
r
s
ity
(
STU
)
.
API
DI
S
[
2
3
]
co
n
tain
s
1
,
5
0
0
f
r
am
es
p
er
s
ec
o
n
d
at
2
5
f
p
s
f
o
r
s
ev
en
ca
m
er
as
(
8
0
0
×
6
0
0
)
with
1
2
p
la
y
er
s
,
wh
ile
STU
f
ea
tu
r
es
2
,
5
0
0
f
r
am
es
p
er
s
ec
o
n
d
at
2
4
f
p
s
f
o
r
eig
h
t
ca
m
er
as
(
1
2
8
0
×
7
2
0
)
with
1
6
p
lay
e
r
s
,
an
d
h
as
an
e
x
p
an
d
ed
1
1
-
p
e
r
io
d
v
er
s
io
n
in
d
ev
el
o
p
m
en
t
.
Hu
e,
s
atu
r
atio
n
,
a
n
d
v
alu
e
(
HSV
)
is
p
o
s
s
ib
le
f
o
r
an
aly
zin
g
jer
s
ey
n
u
m
b
er
s
an
d
f
ac
e
r
ec
o
g
n
itio
n
in
b
o
th
d
atasets
.
Fig
u
r
e
1
is
illu
s
tr
atio
n
o
f
th
e
d
ictio
n
ar
y
c
r
ea
tio
n
p
r
o
ce
s
s
in
v
o
l
v
in
g
th
r
e
e
co
m
p
lem
en
tar
y
co
m
p
o
n
en
ts
th
at
u
s
e
co
lo
r
an
d
tex
tu
r
e
f
ea
tu
r
es
o
b
tain
ed
i
n
H
SV.
3
.
2
.
Ro
bu
s
t
s
eg
m
ent
a
t
io
n
T
h
is
s
tu
d
y
co
n
s
id
er
s
a
s
tep
-
by
-
s
tep
ap
p
r
o
ac
h
to
im
ag
e
s
eg
m
en
tatio
n
b
eg
in
n
in
g
with
a
n
in
i
tial
im
ag
e
f
r
o
m
th
e
s
eq
u
e
n
ce
.
Sev
er
al
h
u
m
an
d
etec
tio
n
m
eth
o
d
s
,
wh
i
ch
in
v
o
lv
e
b
aselin
e
C
NN
m
o
d
els,
Fas
ter
R
-
C
NN,
an
d
HOG
alg
o
r
ith
m
,
wer
e
co
m
p
ar
ed
f
o
r
th
eir
ab
ilit
y
to
r
ec
o
g
n
ize
p
lay
er
s
[
2
4
]
–
[
2
6
]
.
Alth
o
u
g
h
s
ca
le
-
in
v
a
r
ian
t
f
ea
tu
r
e
tr
a
n
s
f
o
r
m
(
SIFT
)
d
es
cr
ip
to
r
h
as
b
ee
n
u
s
ed
m
o
s
tly
f
o
r
f
ac
e
r
ec
o
g
n
itio
n
,
o
th
er
m
en
tio
n
ed
d
etec
tio
n
m
eth
o
d
s
also
in
v
o
lv
e
s
u
ch
c
u
es
as
jer
s
ey
n
u
m
b
er
s
an
d
o
v
er
lay
tex
ts
[
2
7
]
.
T
h
e
s
eg
m
en
ted
r
eg
io
n
s
ar
e
r
ep
r
esen
ted
as (
2
)
.
=
(
,
,
,
,
,
ℎ
,
,
)
(
2
)
W
h
er
e
,
,
,
d
en
o
te
th
e
lef
t,
r
ig
h
t,
b
o
tto
m
,
a
n
d
to
p
co
o
r
d
i
n
ates
o
f
b
o
u
n
d
in
g
r
ec
tan
g
le;
,
ℎ
r
ep
r
ese
n
t
its
wid
th
an
d
h
eig
h
t;
is
th
e
class
lab
el;
an
d
co
n
tain
s
ad
d
itio
n
al
attr
ib
u
tes f
o
r
p
la
y
er
id
en
tity
.
T
h
e
s
eg
m
en
ted
r
eg
i
o
n
ca
p
tu
r
e
s
th
e
p
ip
e
-
lik
e
s
tr
u
ctu
r
e
o
f
jer
s
ey
n
u
m
b
e
r
s
u
s
in
g
a
m
o
r
p
h
o
lo
g
ical
o
p
en
o
p
er
atio
n
,
lev
e
r
ag
in
g
th
e
o
b
s
er
v
atio
n
th
at
d
ig
ital
ch
ar
ac
te
r
s
co
n
s
is
t
o
f
p
ip
e
-
lik
e
co
m
p
o
n
en
ts
.
T
o
m
an
ag
e
o
cc
lu
s
io
n
s
,
m
er
g
e
–
s
p
lit
am
b
i
g
u
ities
,
an
d
s
p
atial
o
v
er
lap
s
,
a
MO
T
s
y
s
tem
i
s
ap
p
lied
u
s
i
n
g
n
ea
r
est
-
n
eig
h
b
o
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dyn
a
mic
s
p
o
r
ts
en
viro
n
men
ts
:
a
u
to
ma
tic
fea
tu
r
e
p
l
a
ye
r
s
tr
a
ck
in
g
a
n
d
…
(
A
d
il A
b
d
u
lh
u
r
A
b
u
s
h
a
n
a
)
4295
ass
ig
n
m
en
t,
co
n
s
id
er
in
g
b
lo
b
s
ize,
s
p
atial
r
elatio
n
s
h
ip
s
,
an
d
r
eliab
le
f
ea
tu
r
e
d
escr
ip
to
r
s
[
2
8
]
–
[
3
0
]
.
T
h
r
esh
o
ld
s
an
d
f
ea
tu
r
e
weig
h
ts
ar
e
ad
ap
tiv
ely
tu
n
ed
f
o
r
o
p
tim
al
p
er
f
o
r
m
an
ce
in
d
y
n
a
m
ic
s
p
o
r
ts
e
n
v
ir
o
n
m
en
ts
[
3
0
]
.
Alg
o
r
ith
m
1
s
u
m
m
a
r
izes
th
is
s
eg
m
en
tatio
n
p
r
o
ce
d
u
r
e,
f
r
o
m
b
ac
k
g
r
o
u
n
d
-
b
ased
ca
n
d
id
at
e
d
etec
tio
n
th
r
o
u
g
h
co
lo
r
-
b
ased
r
eg
io
n
m
er
g
i
n
g
.
Alg
o
r
ith
m
1
.
R
o
b
u
s
t
s
eg
m
en
ta
tio
n
o
f
jer
s
ey
s
i)
Initialize
parameters
s
et
thresholds:
color
s
im
il
ar
it
y
th
re
sh
ol
d
ℎ
and
si
ze
threshold for candidates
ℎ
.
ii)
Capture
vi
de
o
fr
am
es
:
op
e
n
th
e
vi
de
o
fi
le
an
d
in
it
ia
li
ze
a
li
st
to
st
or
e
fr
am
es
,
wh
il
e
the video is still open read each frame
and append it to the list of frames.
iii)
Create
background
model
by
convert
the
list
of
f
ra
me
s
in
to
an
ar
ra
y
F.
C
al
cu
la
te
th
e
median background model using: B = median(F) (along
the
temporal
axis)
.
iv)
Process
each frame for each frame
in the captured frames
a)
Segment
players compute the absolute
difference:
D
n
=
∣
F
n
−B
∣
convert
to grayscale:
‒
G
n
=
cvtColor(
D
n
,COLOR_BGR2GRAY)
‒
Apply binary thresholding:
=
{
255
>
300
0
ℎ
‒
Ap
pl
y
mo
rp
ho
lo
gi
ca
l
op
en
i
ng
O
n
=
mo
rp
ho
lo
gy
Ex
(
T
n
,
MO
R
PH
_O
PE
N,
K)
wh
er
e
K
i
s
a
structuring element.
b)
Find
identify
contours C in
O
n
:
‒
C
=
findContours(
O
n
,
RETR_EXTERNAL, CHAIN_APPROX_SIMPLE)
.
‒
Initialize an empty list for candidate blobs.
c)
Filter
candidates
fo
r
ea
ch
co
nt
ou
r
c
∈
C
as
fo
ll
ow
s
compute
area
A
c:
A
=
contourArea(c)
:
‒
If
A
=>
ℎ
:
compute
the bounding rectangle (x,y,w,h)=bounding Rect(c)
.
‒
Add (x,y,w,h) to the candidates list.
‒
4.5
merge adjacent regions
.
‒
Initialize a list for merged candidates for each pair of candidates (
R
i
,
R
j
)
.
‒
Check color similarity:
C
i
=mean(
R
i
)and
C
j
=mean(
R
j
)
.
‒
Compute the color distance:d=
∣∣
C
i
−
C
j
∣
|
<
t
C
.
‒
If
d
is
tr
ue
,
me
rg
e
th
e
bo
un
di
ng
re
ct
an
gl
es
:
M=
(m
in
(
x
i
,
x
j
),
min(
y
i
,
y
j
),
max(
x
i
+
w
i
,
x
j
+
w
j
)
−
min(
x
i
,
x
j
),
max(
y
i
,
+
h
i
,
y
j
+
h
j
)
−
min(
y
i
,
y
j
))
.
Ad
d
M
to
th
e
me
rg
e
d
candidates list.
v)
Draw
r
ectangles for each merged
candidate (
F
n
,(x,y),(x+w,y+h) ,(0,255,0),2)
.
Fig
u
r
e
1
.
C
r
ea
tin
g
d
ictio
n
ar
y
with
th
r
ee
co
m
p
o
n
en
ts
co
l
o
r
t
ex
tu
r
e
f
ea
tu
r
es HSV
3
.
3
.
Co
l
o
r
t
ex
t
ure
ex
t
ra
c
t
io
n f
o
r
f
a
cia
l e
x
press
io
n
T
h
e
p
r
o
p
o
s
ed
d
ictio
n
ar
y
s
y
s
tem
u
s
es
co
lo
r
–
tex
tu
r
e
f
ea
tu
r
es
f
r
o
m
R
GB
f
ac
e
im
ag
es.
His
to
g
r
am
-
b
ased
f
ea
t
u
r
es
ar
e
ex
tr
ac
ted
ac
r
o
s
s
m
u
ltip
le
co
lo
r
b
a
n
d
s
to
en
h
an
ce
s
p
a
r
s
e
r
ep
r
esen
tatio
n
f
ac
e
r
ec
o
g
n
itio
n
[
3
1
]
,
[
3
2
]
,
wh
ile
n
o
v
el
co
lo
r
v
ec
to
r
d
if
f
er
en
c
e
p
atter
n
s
im
p
r
o
v
e
s
k
in
tex
t
u
r
e
d
is
cr
im
in
atio
n
,
co
m
p
u
ted
u
s
in
g
a
cir
c
u
lar
n
ei
g
h
b
o
r
h
o
o
d
o
f
r
ad
i
u
s
R
an
d
P p
ix
els
[
3
3
]
.
3
.
4
.
Aut
o
m
a
t
ic
f
a
ce
a
nd
j
er
s
ey
f
ind
er
B
u
ild
in
g
o
n
p
r
io
r
wo
r
k
[
3
4
]
,
[
3
5
]
th
e
s
y
s
tem
in
teg
r
ates
an
au
to
m
atic
f
ac
e
an
d
jer
s
ey
f
in
d
er
u
s
in
g
a
h
ier
ar
ch
ical
to
p
–
d
o
wn
p
ar
s
in
g
tr
ee
to
d
etec
t
jer
s
ey
n
u
m
b
er
s
an
d
f
ac
ial
f
ea
tu
r
es.
Fra
m
es
ar
e
r
ec
u
r
s
iv
ely
Color
spac
e
c
on
v
e
r
s
ion
Ext
r
a
ction
of
color
t
e
xt
ur
e
f
ea
t
ur
es
Dimensi
ona
l
i
t
y
r
ed
uction
/
Con
c
a
t
ena
t
i
on
F
e
a
t
ur
e
v
e
ct
o
r
F
e
a
t
ur
e
v
e
ct
o
r
F
e
a
t
ur
e
v
e
ct
o
r
Col
or
t
e
xt
ur
e
f
ea
t
ur
es
F
us
ed
f
ea
t
ur
e
Dic
t
i
onar
y
wi
t
h
Col
or
t
e
xt
ur
e
f
ea
t
ur
es
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
2
9
2
-
4
3
0
3
4296
s
eg
m
en
ted
to
r
ed
u
ce
co
m
p
u
t
atio
n
,
wh
ile
k
er
n
elize
d
co
r
r
elatio
n
f
ilter
s
an
d
m
u
ltip
le
in
s
tan
ce
lear
n
in
g
[
3
7
]
en
ab
le
f
ast
tr
ac
k
in
g
an
d
ad
a
p
t
to
ap
p
ea
r
an
ce
v
ar
iatio
n
s
.
T
h
ese
m
ec
h
an
is
m
s
to
g
eth
e
r
im
p
r
o
v
e
ac
c
u
r
ac
y
an
d
r
ea
l
-
tim
e
p
lay
er
m
o
n
ito
r
i
n
g
.
3
.
5
.
K
SP
-
I
D
m
o
del w
it
h dee
p lea
rning
t
ec
hn
iq
ues
S
ubs
ec
tio
n
3
.
2
o
u
tp
u
ts
f
ee
d
in
to
th
e
KSP
-
I
D
m
o
d
el,
wh
er
e
ef
f
icien
t
s
eg
m
en
tatio
n
en
s
u
r
es
p
r
ec
is
e
p
lay
er
id
e
n
tific
atio
n
.
R
ec
o
g
n
itio
n
is
d
o
n
e
th
r
o
u
g
h
m
u
ltip
le
ca
m
er
a
a
n
d
cl
o
s
e
-
u
p
v
iew
s
,
u
s
in
g
b
o
th
f
ac
e
d
etec
tio
n
an
d
jer
s
ey
n
u
m
b
er
r
ec
o
g
n
itio
n
[
3
8
]
,
[
3
9
]
.
So
m
e
ap
p
r
o
ac
h
es
u
s
e
d
ir
ec
t
je
r
s
ey
n
u
m
b
er
r
ec
o
g
n
itio
n
th
r
o
u
g
h
alter
n
ativ
e
d
escr
ip
to
r
s
[
4
1
]
.
T
h
e
MO
T
tech
n
iq
u
e
u
s
ed
is
b
ased
o
n
tr
ac
k
-
to
-
d
et
ec
tio
n
ass
o
ciatio
n
b
ased
o
n
KSP
o
p
tim
izatio
n
[
4
2
]
,
b
u
t
s
witch
in
g
id
en
titi
es
tak
es
p
lace
in
t
h
e
ab
s
en
ce
o
f
s
tr
o
n
g
ap
p
ea
r
an
ce
m
o
d
els
[
4
3
]
;
o
n
lin
e
MO
T
b
ased
o
n
MO
T
ca
n
o
n
ly
h
ap
p
en
wh
e
n
c
u
es
ar
e
p
r
esen
t
in
m
o
tio
n
a
n
d
ap
p
ea
r
an
ce
[
4
4
]
.
I
n
ter
m
s
o
f
r
o
b
u
s
tn
ess
,
th
e
d
ee
p
lear
n
in
g
tech
n
iq
u
es
h
av
e
b
ee
n
u
s
ef
u
l:
I
D
-
Net
u
s
es
b
o
d
y
attr
ib
u
tes
wh
ile
d
ee
p
r
eg
r
ess
io
n
tr
ac
k
er
s
[
4
5
]
u
s
e
r
esid
u
al
co
n
n
ec
tio
n
s
a
n
d
h
ar
d
-
s
am
p
le
m
in
in
g
o
n
C
NNs.
KSP
-
I
D
f
u
s
es
f
ac
e
an
d
jer
s
ey
tr
ac
k
in
g
b
y
u
s
in
g
s
p
atial
co
l
o
r
f
ea
tu
r
es
an
d
l
o
ca
l
b
in
ar
y
p
atter
n
(
L
B
P)
with
d
ee
p
p
lay
er
s
'
id
en
tific
atio
n
[
4
6
]
.
Ap
p
ea
r
a
n
ce
s
im
ilar
ity
s
co
r
es
(
,
)
ass
e
s
s
e
s
v
is
u
al
ch
ar
ac
ter
is
tics
u
s
in
g
m
etr
ics
lik
e
co
lo
r
h
is
to
g
r
am
s
an
d
tex
tu
r
e
d
escr
ip
to
r
s
.
Mo
tio
n
s
im
ilar
ity
(
,
)
is
d
er
iv
e
d
f
r
o
m
th
e
p
r
ed
icted
p
o
s
itio
n
̂
u
s
in
g
a
m
o
tio
n
m
o
d
el,
ca
lcu
latin
g
th
e
d
is
tan
ce
b
etwe
en
th
e
d
etec
tio
n
an
d
p
r
ed
icted
p
o
s
itio
n
.
Pro
x
im
ity
s
im
ilar
ity
(
,
)
m
ea
s
u
r
es
th
e
clo
s
en
ess
o
f
th
e
d
etec
tio
n
to
t
h
e
p
r
e
d
icted
p
o
s
itio
n
.
T
o
tal
s
im
ilar
ity
s
co
r
e
co
m
b
in
e
s
th
ese
m
etr
ics as in
(
3
)
.
S(
,
)=
⋅
(
,
)+
⋅
(
,
)+
⋅
(
,
)
(
3
)
A
co
s
t
m
atr
ix
C
is
f
o
r
m
ed
w
h
er
e
=
1
−S(
,
)
d
ef
in
ed
as
ass
ig
n
m
en
t
p
r
o
b
lem
.
T
h
e
Hu
n
g
a
r
ia
n
a
lg
o
r
ith
m
o
p
tim
ally
ass
ig
n
s
d
etec
tio
n
s
to
tr
ac
k
s
,
co
n
f
ir
m
i
n
g
a
tr
ac
k
as
n
o
r
m
al
if
it
m
ain
tain
s
co
n
s
is
ten
t
ass
o
ciatio
n
s
ac
r
o
s
s
f
r
am
es.
I
n
ca
s
es
o
f
ab
s
en
ce
,
tr
ac
k
s
m
a
y
t
r
an
s
itio
n
to
"
L
o
s
t"
o
r
"Oc
clu
d
ed
,
"
in
f
l
u
en
ce
d
b
y
th
e
d
u
r
atio
n
o
f
th
e
a
b
s
en
ce
,
w
h
ile
id
en
tity
r
ec
o
v
er
y
u
tili
ze
s
ap
p
ea
r
an
ce
m
o
d
els
f
o
r
r
ea
l
-
tim
e
s
u
r
v
eillan
ce
an
d
ef
f
ec
tiv
e
m
u
lti
-
tar
g
et
m
o
n
ito
r
in
g
t
h
r
o
u
g
h
p
r
o
b
ab
ilis
tic
d
ata
ass
o
ciatio
n
.
T
o
r
e
d
u
ce
d
r
if
tin
g
a
n
d
m
is
m
atch
in
g
af
ter
o
cc
lu
s
io
n
,
th
r
esh
o
ld
s
an
d
o
ccl
u
d
ed
ar
e
ap
p
lied
t
o
n
o
r
m
al
tr
ac
k
s
with
o
u
t
ass
ig
n
ed
m
ea
s
u
r
em
en
ts
.
No
r
m
al
tr
ac
k
s
with
o
u
t
ass
ig
n
ed
m
ea
s
u
r
em
e
n
ts
m
ay
tr
a
n
s
itio
n
to
"
L
o
s
t"
o
r
"Oc
clu
d
ed
"
b
ased
o
n
ab
s
en
ce
d
u
r
atio
n
,
wh
ile
id
en
tity
r
ec
o
v
er
y
u
s
es
ap
p
ea
r
a
n
ce
m
o
d
els
an
d
p
r
o
b
a
b
ilis
tic
d
ata
ass
o
ciatio
n
to
en
ab
le
r
ea
l
-
tim
e
s
u
r
v
eillan
ce
o
f
lo
w
-
r
eso
lu
tio
n
co
lo
r
tar
g
ets
in
d
y
n
am
ic
en
v
ir
o
n
m
en
ts
,
r
ef
er
r
e
d
to
as
th
e
r
e
-
id
en
tific
atio
n
m
ec
h
an
is
m
[
4
2
]
,
[
4
4
]
,
[
4
5
]
.
Alg
o
r
ith
m
2
d
etails
th
is
p
r
o
ce
d
u
r
e
,
co
v
er
i
n
g
s
im
ilar
ity
-
s
co
r
e
co
m
p
u
tatio
n
,
th
e
Hu
n
g
ar
ian
as
s
ig
n
m
en
t step
,
an
d
th
e
tr
ac
k
-
s
tate
tr
an
s
itio
n
r
u
les d
escr
ib
ed
a
b
o
v
e.
Alg
o
r
ith
m
2
.
Similar
ity
s
co
r
es c
o
m
p
u
tatio
n
an
d
tr
ac
k
m
a
n
ag
em
en
t
i)
Similarity
sc
or
es
co
mp
ut
a
ti
on
fo
r
ea
ch
de
te
ct
io
n
an
d
ex
is
ti
ng
tr
ac
k
,
co
mp
ut
e
th
e
similarity scores
:
‒
Appearance
s
imilarity
(
,
):
t
hi
s
fu
nc
ti
on
ca
n
in
c
or
po
ra
te
me
tr
ic
s
li
ke
co
lo
r
histograms, texture descriptors, and shape features.
‒
Motion
s
imilarity
(
,
)
:
p
redicted
position
̂
us
in
g
a
mo
ti
on
mo
de
l
(e
.g
.
,
Ka
lm
an
fi
lt
er
):
̂
= f(
,
) where
is
the
control
input.
‒
Compute the distance from the detection:
(
,
)
=exp
(
|
|
−
̂
|
|
2
2
)
2
.
‒
Proximity
similarity
:
(
,
)
=
1
1
+
|
|
−
̂
|
|
.
‒
Total
similarity score
S(
,
): combine the scores using weights
,
,
:
S(
,
)=
⋅
(
,
)+
⋅
(
,
)+
⋅
(
,
)
ii)
Assignment
problem: given similarity
scores S(
,
), construct a cost matrix C:
‒
Total
similarity score
:
= 1−S(
,
)
.
‒
Us
e
th
e
Hu
ng
ar
ia
n
al
go
ri
th
m
to
fi
nd
th
e
op
ti
ma
l
as
si
gn
me
nt
th
at
mi
ni
mi
z
es
to
ta
l
cost:
minimize
∑
M
j
=
1
,
w
here
is
a
bi
na
ry
va
ri
ab
l
e
in
di
ca
ti
ng
wh
et
he
r
detection k is assigned to track j
.
‒
Track
state
transition conditions
.
‒
Tentative
track confirm as normal track
if associated across multiple frames:
I
f
∑
=
1
(
,
,
)≥
⇒
normal track
.
iii)
Transition to
lost track:
‒
Transition to
lost track if
no association for
frames:
I
f
association (
,
,
)
=0
for
frames
⇒
lost track
.
‒
Occluded
tr
ac
k
ha
nd
li
ng
:
re
ma
in
in
oc
cl
ud
ed
tr
ac
k
if
pa
rt
ia
l
de
te
c
ti
on
is
associated: if
partial_association(
,
,
)
⇒
occluded track
.
‒
Termination of
tracks: terminate if in lost track
for
frames
I
f
not
detected
for
frames
⇒
terminate track
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dyn
a
mic
s
p
o
r
ts
en
viro
n
men
ts
:
a
u
to
ma
tic
fea
tu
r
e
p
l
a
ye
r
s
tr
a
ck
in
g
a
n
d
…
(
A
d
il A
b
d
u
lh
u
r
A
b
u
s
h
a
n
a
)
4297
iv)
Re
-
identification
m
echanism use appearance models for identity recovery:
‒
(
,
,
)
=
similarity
(
ap
pe
a
ra
nc
e
embedding
(
),
ap
pe
ar
an
ce
em
be
dd
in
g
(
))
.
A
re
al
ti
me
ad
ap
ti
ve
vi
s
ua
l
su
rv
ei
ll
an
ce
sy
st
em
f
or
tr
ac
ki
ng
lo
w
-
resolution
color
targets in dynamically changing scenes.
3
.
6
.
E
v
a
lua
t
i
o
n
m
et
rics
T
o
im
p
r
o
v
e
an
n
o
tatio
n
q
u
ali
ty
,
th
e
in
itial
lab
els
ar
e
r
ef
in
ed
b
y
s
u
p
p
r
ess
in
g
f
alse
p
o
s
itiv
es
an
d
in
co
r
p
o
r
atin
g
m
is
s
ed
d
etec
tio
n
s
,
p
r
o
d
u
ci
n
g
co
r
r
ec
te
d
lab
els
y
_
co
r
r
.
T
h
ese
ar
e
u
s
ed
to
f
in
e
-
tu
n
e
th
e
teac
h
e
r
n
etwo
r
k
,
y
ield
in
g
th
e
f
in
al
t
ea
ch
er
n
etwo
r
k
(
FTN)
with
im
p
r
o
v
e
d
an
n
o
tatio
n
s
FTN(
X)
.
Per
f
o
r
m
an
ce
is
m
ea
s
u
r
ed
u
s
in
g
t
h
e
m
u
ltip
le
o
b
ject
tr
ac
k
in
g
ac
c
u
r
ac
y
(
MO
T
A)
m
etr
ic
,
as d
ef
in
e
d
in
(
4
)
[
4
7
]
.
=
1
−
(
∑
(
+
+
)
)
(
∑
)
(
4
)
W
h
er
e
,
,
an
d
r
ep
r
esen
t
f
al
s
e
n
eg
ativ
es
(
FN)
,
f
alse
p
o
s
itiv
es
(
FP
)
,
an
d
I
DS
,
r
esp
ec
t
iv
ely
,
with
as
g
r
o
u
n
d
tr
u
th
.
T
o
ad
d
r
ess
s
en
s
itiv
ity
to
I
DS
,
th
e
g
en
er
alize
d
m
u
ltip
le
o
b
ject
tr
ac
k
in
g
ac
cu
r
a
cy
(
GM
OT
A)
[
2
]
r
e
p
lace
s
m
m
e
with
g
m
m
e,
ac
co
u
n
tin
g
f
o
r
co
r
r
ec
tly
lab
eled
tr
ajec
to
r
ies
,
as d
ef
in
ed
in
(
5
)
.
=
1
−
(
∑
(
+
+
)
)
(
∑
)
(
5
)
Alo
n
g
s
id
e
MO
T
A
an
d
GM
O
T
A,
tr
u
e
p
o
s
itiv
es
(
TP
)
,
FP
,
FN,
an
d
I
DS
wer
e
r
ep
o
r
ted
.
Kn
o
wled
g
e
d
is
till
atio
n
f
o
r
r
ea
l
-
tim
e
d
etec
t
io
n
to
e
n
s
u
r
e
r
ea
l
-
tim
e
d
ep
l
o
y
m
en
t,
k
n
o
wled
g
e
is
d
is
till
ed
f
r
o
m
th
e
R
esNet
-
50
teac
h
er
n
etwo
r
k
in
t
o
a
f
aster
R
esNet
-
1
8
s
tu
d
en
t
n
etwo
r
k
,
b
o
th
en
h
an
ce
d
with
an
FP
N
[
4
8
]
.
T
h
e
p
ip
elin
e
is
tr
ain
ed
o
n
u
n
lab
eled
So
cc
er
N
et
b
r
o
a
d
ca
s
t
v
id
eo
s
[
4
9
]
,
im
p
r
o
v
e
d
etec
tio
n
o
f
f
ac
es,
jer
s
ey
s
,
an
d
p
lay
er
b
o
d
ies,
b
alan
cin
g
ac
c
u
r
ac
y
with
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
f
o
r
r
ea
l
-
ti
m
e
s
p
o
r
ts
an
aly
tics
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Resul
t
s
pa
ce
,
ev
a
lua
t
io
n
po
s
s
ibi
litie
s
,
a
nd
a
lg
o
rit
hm
ic
beha
v
io
r
T
h
e
s
u
g
g
ested
m
o
d
el
was
th
o
r
o
u
g
h
ly
test
ed
in
r
ea
l
-
wo
r
ld
o
p
er
atio
n
s
to
ass
ess
its
d
u
r
ab
ilit
y
an
d
s
tab
ilit
y
.
T
h
e
ca
s
es
u
s
ed
in
v
o
lv
ed
th
e
s
ca
tter
in
g
o
f
p
la
y
e
r
s
in
an
u
n
ev
en
m
a
n
n
er
,
in
te
r
ac
tio
n
at
m
ed
iu
m
lev
els,
an
d
d
en
s
ely
p
ac
k
ed
p
l
ay
in
g
s
ess
io
n
s
,
wh
er
e
th
e
k
e
y
p
r
o
b
lem
was
o
f
ten
o
v
er
la
p
p
in
g
o
f
th
e
p
la
y
er
s
'
im
ag
es.
I
n
ad
d
itio
n
,
th
e
test
was
p
er
f
o
r
m
ed
in
s
tatic
an
d
h
ig
h
ly
d
y
n
am
ic
m
o
tio
n
en
v
ir
o
n
m
en
ts
in
v
o
l
v
in
g
s
h
o
r
t
an
d
lo
n
g
-
ter
m
ac
tiv
ity
m
o
n
ito
r
in
g
.
T
h
is
ev
alu
atio
n
w
as
ca
r
r
ied
o
u
t
in
b
o
th
s
in
g
le
-
an
d
m
u
ltip
le
-
ca
m
e
r
a
co
n
f
ig
u
r
atio
n
s
allo
win
g
f
o
r
th
e
co
m
p
ar
is
o
n
o
f
s
in
g
le
-
v
iew
s
p
atial
p
er
ce
p
tio
n
an
d
m
u
ltip
le
-
v
iew
r
ed
u
n
d
an
cy
.
So
m
e
o
f
th
e
f
ac
to
r
s
in
v
o
lv
e
d
in
th
e
ev
al
u
atio
n
in
clu
d
e
jer
s
ey
d
is
to
r
tio
n
f
r
o
m
q
u
ick
b
o
d
y
m
o
v
e
m
en
ts
,
o
cc
lu
s
io
n
s
o
win
g
to
p
lay
er
s
o
v
er
lap
p
in
g
ea
ch
o
th
er
,
r
esem
b
lan
ce
b
etwe
en
p
lay
e
r
s
,
an
d
im
ag
e
b
lu
r
b
ec
au
s
e
o
f
v
ar
y
in
g
lig
h
t c
o
n
d
itio
n
s
.
I
n
f
ac
t,
th
e
DPR
I
b
lo
ck
s
u
cc
e
ed
ed
in
m
ai
n
tain
in
g
a
n
ex
ce
ll
en
t
d
is
cr
im
in
ativ
e
p
o
we
r
am
o
n
g
v
is
u
ally
s
im
ilar
p
lay
er
s
ac
r
o
s
s
all
c
o
n
d
itio
n
s
m
en
tio
n
ed
ab
o
v
e,
wh
ile
th
e
in
d
iv
id
u
al
p
r
o
b
a
b
ilis
tic
o
v
er
lap
p
in
g
b
o
u
n
d
in
g
b
o
x
co
n
tr
ib
u
te
d
s
ig
n
if
ican
tly
to
3
D
g
r
o
u
n
d
in
g
s
in
t
h
e
p
r
esen
ce
o
f
n
o
is
e.
W
ith
th
e
in
tr
o
d
u
ctio
n
o
f
a
n
i
d
e
n
t
i
t
y
-
a
wa
r
e
K
S
P
-
I
D
m
e
t
h
o
d
,
a
c
o
n
t
i
n
u
o
u
s
i
d
e
n
t
i
t
y
f
l
o
w
f
o
r
t
r
a
c
k
i
n
g
-
by
-
d
e
t
e
c
t
i
o
n
c
o
u
l
d
b
e
a
c
h
i
e
v
e
d
w
i
t
h
o
u
t
f
a
c
i
n
g
t
h
e
i
s
s
u
e
o
f
f
r
a
g
m
e
n
t
a
tio
n
a
n
d
i
d
e
n
t
i
t
y
d
r
i
f
t
,
a
t
y
p
i
ca
l
f
e
a
t
u
r
e
o
f
K
SP
a
p
p
r
o
a
c
h
e
s
[
4
9
]
,
[
5
0
]
.
G
e
n
e
r
a
l
l
y
s
p
e
a
k
i
n
g
,
t
h
e
m
o
d
e
l
wa
s
h
i
g
h
ly
r
e
s
i
l
ie
n
t
i
n
a
ll
s
ce
n
a
r
i
o
s
,
g
e
n
er
a
t
i
n
g
c
o
n
s
is
t
e
n
t
t
r
a
c
k
s
w
i
t
h
m
in
i
m
a
l
i
d
e
n
t
it
y
l
o
s
s
.
4
.
2
.
Q
ua
ntit
a
t
iv
e
e
v
a
lua
t
io
n
Qu
an
titativ
e
ev
alu
atio
n
o
n
A
PID
I
S
an
d
co
m
p
ar
ativ
e
an
aly
s
is
with
li
ter
atu
r
e.
E
v
alu
atio
n
r
esu
lts
o
n
th
e
API
DI
S
d
ataset
ar
e
s
h
o
w
n
in
T
a
b
le
1
.
T
h
e
tr
ad
itio
n
al
KSP
-
b
ased
alg
o
r
ith
m
s
KSP
[
4
9
]
a
n
d
c
o
n
s
tr
ain
ed
k
-
s
h
o
r
test
p
ath
s
(
C
-
KSP)
p
er
f
o
r
m
well
in
ter
m
s
o
f
m
o
d
er
ate
tr
ac
k
in
g
ac
c
u
r
ac
y
(
MO
T
A
=
4
9
.
0
%
f
o
r
KSP
an
d
5
8
.
9
%
f
o
r
C
-
KSP)
b
u
t
ex
p
er
ien
ce
h
ig
h
id
en
tity
f
r
ag
m
en
t
atio
n
(
I
DS
=
6
.
5
f
o
r
KSP
an
d
4
.
6
f
o
r
C
-
KSP),
m
ain
ly
d
u
e
to
wea
k
ap
p
ea
r
an
ce
r
ea
s
o
n
in
g
ca
p
a
b
ilit
y
.
P
r
o
b
ab
ilis
tic
an
d
o
cc
u
p
a
n
cy
-
b
ased
s
o
lu
tio
n
s
lik
e
d
y
n
am
ic
p
r
o
g
r
am
m
in
g
(
DP)
[
5
1
]
an
d
v
o
lu
m
etr
ic
m
ass
d
en
s
ities
(
VM
D
)
/
v
o
lu
m
etr
i
c
m
ass
d
en
s
itie
s
co
n
s
tr
ain
ed
(
VM
Dc
)
[
5
2
]
ca
n
en
h
an
ce
s
p
atial
g
r
o
u
n
d
in
g
a
n
d
lo
wer
I
DS,
alb
eit
in
d
e
n
s
e
s
ce
n
es.
Mo
r
e
ad
v
an
ce
d
s
y
s
tem
s
s
u
ch
as
tem
p
o
r
al
m
o
tio
n
an
d
s
p
atia
l
ass
o
ciatio
n
(
T
MO
SA)
[
5
3
]
,
tr
ajec
to
r
y
m
atr
ix
ap
p
r
o
x
im
atio
n
v
ia
ex
p
ec
tatio
n
-
m
ax
im
izatio
n
(
T
MA
P
-
EM)
[
5
4
]
,
a
n
d
m
u
lti
-
s
en
s
o
r
p
ed
estrian
tr
ac
k
in
g
(
MSPT)
[
5
5
]
in
co
r
p
o
r
ate
r
ich
er
tem
p
o
r
al
an
d
s
tr
u
ctu
r
al
co
n
s
tr
ain
ts
,
ac
h
iev
in
g
h
ig
h
er
M
OT
A
(
7
9
.
9
–
8
5
.
5
%)
an
d
s
u
b
s
tan
tially
r
ed
u
ce
d
I
DS.
Me
an
wh
ile,
m
u
lti
-
ca
m
er
a
d
ee
p
f
r
am
ewo
r
k
s
s
u
ch
as
m
u
lti
-
ca
m
er
a
m
u
lti
-
p
er
s
o
n
tr
ac
k
in
g
an
d
id
en
tific
atio
n
(
MCMPTI
)
[
5
8
]
ex
p
lo
it
s
p
atial
r
ed
u
n
d
an
c
y
to
im
p
r
o
v
e
g
lo
b
al
tr
ac
k
in
g
p
er
f
o
r
m
a
n
ce
,
o
b
tain
in
g
h
ig
h
GM
OT
A
(
7
2
.
3
%)
an
d
th
e
lo
west
r
ep
o
r
ted
I
D
S
(
0
.
6
)
.
T
h
e
g
i
v
e
n
f
r
am
ewo
r
k
,
in
d
ir
ec
t
co
m
p
ar
is
o
n
,
s
co
r
ed
MO
T
A
8
4
.
5
%,
G
MO
T
A
6
7
%,
an
d
I
DS
1
.
0
,
t
h
u
s
s
h
o
win
g
ac
cu
r
ac
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
2
9
2
-
4
3
0
3
4298
th
at
is
eith
er
o
n
p
ar
o
r
b
etter
th
an
m
o
s
t
r
ec
en
t
to
p
-
n
o
tc
h
s
y
s
tem
s
an
d
at
th
e
s
am
e
tim
e
en
s
u
r
in
g
id
en
tity
s
tab
ilit
y
th
at
is
a
s
g
o
o
d
as
o
r
b
etter
th
an
th
e
b
est
p
r
o
b
a
b
ilis
tic
an
d
h
y
b
r
id
ass
o
ciatio
n
m
o
d
els
[
5
3
]
–
[
5
7
]
.
I
t
s
h
o
u
ld
b
e
n
o
te
d
th
at
th
e
p
r
o
p
o
s
ed
m
eth
o
d
is
n
o
t
lik
e
T
MO
SA
[
5
3
]
th
at
r
eq
u
ir
es
th
e
im
p
o
s
itio
n
o
f
p
r
e
-
s
tr
u
ctu
r
e
d
co
n
s
tr
ain
ts
o
r
MCMPTI
[
5
7
]
th
at
ca
n
o
n
ly
wo
r
k
u
n
d
er
ce
r
tain
m
u
lti
-
ca
m
er
a
co
n
d
itio
n
s
,
b
u
t
r
ath
er
it is
a
s
tr
o
n
g
p
er
f
o
r
m
er
i
n
th
e
ar
ea
o
f
f
u
lly
au
to
m
ate
d
a
n
d
g
en
er
aliza
b
le
p
ip
elin
es.
T
ab
le
1
.
Qu
a
n
titativ
e
co
m
p
ar
i
s
o
n
o
f
p
r
o
p
o
s
ed
m
eth
o
d
ag
ain
s
t state
-
of
-
th
e
-
ar
t tr
ac
k
er
s
o
n
t
h
e
API
DI
S d
ataset
M
e
t
h
o
d
M
O
TA
↑
G
M
O
TA
↑
TP↑
F
P
↓
F
N
↓
I
D
S
↓
K
S
P
[
4
9
]
4
9
.
0
3
8
.
0
8
6
.
2
1
3
.
8
3
1
.
3
6
.
5
C
-
K
S
P
[
5
0
]
5
8
.
9
4
3
.
3
8
8
.
1
1
1
.
9
2
2
.
5
4
.
6
D
P
[
5
1
]
4
9
.
5
3
9
.
1
—
—
—
—
V
M
D
[
5
2
]
5
9
.
7
—
8
8
.
7
1
7
.
2
2
8
.
6
1
.
4
V
M
D
c
[
5
2
]
6
7
.
5
—
9
3
.
0
1
2
.
5
2
4
.
4
1
.
3
TM
O
S
A
[
5
3
]
8
5
.
5
6
8
.
7
1
0
4
.
5
*
2
.
6
1
3
.
5
1
.
2
TM
A
P
-
E
M
[
5
4
]
7
9
.
9
—
1
0
1
.
6
*
6
.
6
1
5
.
4
1
.
6
M
S
P
T
[
5
5
]
7
8
.
6
—
9
8
.
7
1
0
.
2
2
2
.
2
4
.
3
H
y
b
r
i
d
f
u
l
l
[
5
6
]
6
2
.
6
—
—
—
—
—
M
C
M
P
TI
[
5
7
]
8
1
.
1
7
2
.
3
1
0
3
.
2
*
6
.
2
1
4
.
6
0
.
6
Th
i
s
w
o
r
k
8
4
.
5
6
7
.
0
1
0
1
.
0
*
4
.
2
1
2
.
9
1
.
0
4
.
3
.
Abla
t
io
n
a
na
ly
s
is
o
n AP
I
DIS
T
h
e
s
tu
d
y
c
o
n
d
u
cted
a
n
ab
lati
o
n
an
aly
s
is
to
d
eter
m
in
e
th
e
i
m
p
ac
t
o
f
ea
ch
n
ew
co
m
p
o
n
en
t
with
th
e
h
elp
o
f
T
ab
le
2
.
T
h
e
b
asic
p
r
o
b
ab
ilis
tic
o
cc
u
p
an
c
y
m
ap
p
in
g
co
m
b
in
e
d
with
KSP
-
I
D
s
h
o
w
s
th
at
h
ig
h
r
ates
o
f
p
er
f
o
r
m
an
ce
ar
e
im
p
o
s
s
ib
le
(
MO
T
A
4
9
.
0
%,
GM
OT
A
3
8
.
0
%,
an
d
I
DS
6
.
5
)
,
th
e
r
esu
lts
ar
e
p
r
etty
m
u
ch
th
e
s
am
e
as
th
o
s
e
p
r
ev
io
u
s
ly
r
ep
o
r
ted
f
o
r
g
r
ap
h
-
o
p
tim
izatio
n
f
r
am
ewo
r
k
s
th
at
ar
e
class
ical
[
4
9
]
,
[
5
0
]
.
Ad
d
in
g
c
ascad
e
m
ask
R
-
C
NN
to
th
e
p
r
o
ce
s
s
n
o
t
o
n
ly
g
r
ea
tly
i
n
cr
e
ases
th
e
in
teg
r
ity
o
f
d
etec
tio
n
b
u
t
also
l
o
wer
s
th
e
I
DS
to
4
.
7
,
wh
ich
in
d
icate
s
th
e
s
ig
n
if
ican
ce
o
f
r
eliab
le
s
eg
m
en
tatio
n
p
r
i
o
r
s
f
o
u
n
d
ea
r
lier
in
p
lay
er
th
ick
en
v
ir
o
n
m
en
ts
[
1
8
]
,
[
2
4
]
.
T
ab
le
2
.
Ab
latio
n
ex
p
e
r
im
en
t
r
esu
lts
o
n
th
e
API
DI
S d
ataset
M
e
t
h
o
d
M
O
TA
↑
G
M
O
TA
↑
TP ↑
F
P
↓
F
N
↓
I
D
S
↓
P
O
M
+
K
S
P
4
9
.
0
3
8
.
0
8
6
.
2
2
2
.
1
3
1
.
3
6
.
5
C
a
sc
a
d
e
m
a
sk
R
C
N
N
-
P
+
P
O
M
+
K
S
P
6
1
.
3
5
1
.
4
9
2
.
4
1
4
.
5
2
6
.
9
4
.
7
O
u
r
s DP
R
I
7
2
.
0
6
0
.
0
9
7
.
8
1
1
.
3
2
1
.
4
3
.
6
O
u
r
s I
P
O
B
B
7
5
.
0
6
2
.
0
9
9
.
2
1
0
.
6
2
0
.
0
2
.
8
O
u
r
s
K
S
P
-
ID
7
8
.
0
6
5
.
0
1
0
1
.
3
*
8
.
5
1
8
.
6
2
.
1
O
u
r
s (f
u
l
l
)
8
5
.
5
6
9
.
0
1
0
4
.
8
*
4
.
2
1
2
.
2
0
.
7
T
h
e
ad
d
itio
n
o
f
th
e
DPR
I
m
o
d
u
le
h
as
a
d
r
asti
c
ef
f
ec
t
o
n
th
e
tr
ac
k
in
g
co
n
s
is
ten
cy
th
at
n
o
w
r
ea
ch
es
(
MO
T
A
7
2
.
0
%)
an
d
f
u
r
th
er
b
r
in
g
s
th
e
I
DS
d
o
wn
to
3
.
6
.
T
h
is
s
h
o
ws
th
e
n
ee
d
f
o
r
s
tr
o
n
g
ap
p
ea
r
a
n
ce
an
d
id
en
tity
em
b
e
d
d
in
g
s
wh
ich
a
r
e
alr
ea
d
y
co
r
r
o
b
o
r
ated
b
y
[
3
5
]
,
[
4
4
]
to
b
e
th
e
ca
s
e
with
d
e
ep
id
en
tity
lear
n
in
g
s
y
s
tem
s
.
T
h
en
ca
m
e
th
e
ad
d
iti
o
n
o
f
I
POM
an
d
KSP
-
I
D
wh
ich
lo
wer
ed
I
DS
f
u
r
th
er
to
2
.
1
b
y
m
ea
n
s
o
f
b
etter
3
D
lo
ca
lizatio
n
an
d
id
en
tity
-
a
war
e
ass
o
ciatio
n
,
wh
ic
h
is
s
i
m
ilar
to
th
e
r
esu
lts
ea
r
lier
o
b
tain
ed
in
[
1
4
]
,
[
5
1
]
with
m
u
lti
-
v
iew
o
cc
u
p
a
n
cy
tr
ac
k
in
g
f
r
am
ewo
r
k
s
an
d
[
4
1
]
,
[
5
2
]
with
id
e
n
tity
-
co
r
r
elate
d
g
r
ap
h
o
p
tim
izatio
n
.
T
h
e
co
m
p
lete
f
r
am
ewo
r
k
at
last
ac
h
iev
es
MO
T
A
8
5
.
5
%,
GM
OT
A
6
9
.
0
%,
I
DS
0
.
7
,
t
h
u
s
ev
id
e
n
cin
g
th
e
in
ter
d
ep
en
d
en
ce
o
f
jer
s
ey
r
ec
o
g
n
itio
n
,
f
ac
ial
cu
es,
d
ee
p
em
b
ed
d
in
g
s
,
an
d
KSP
-
b
ased
id
en
tity
r
ein
f
o
r
ce
m
e
n
t
th
r
o
u
g
h
th
eir
c
o
m
p
lem
e
n
tar
y
i
n
ter
ac
tio
n
.
4
.
4
.
E
v
a
lua
t
i
o
n o
n t
he
ST
U
d
a
t
a
s
et
E
v
alu
atin
g
th
e
s
y
s
tem
with
th
e
STU
d
ataset
o
n
ce
m
o
r
e
is
an
ass
u
r
an
ce
o
f
its
g
en
er
aliza
tio
n
as sh
o
wn
in
T
ab
le
3
.
T
h
e
v
e
r
y
f
ir
s
t
b
aselin
e
r
esu
lts
(
MO
T
A
4
1
.
4
%,
GM
OT
A
3
4
.
9
%,
an
d
I
DS
1
3
.
6
)
i
n
d
icate
th
e
ex
tr
em
e
d
i
f
f
icu
lty
o
f
th
e
d
ataset
wh
ich
is
m
a
r
k
ed
b
y
v
e
r
y
h
i
g
h
p
lay
er
d
en
s
ity
a
n
d
f
r
eq
u
en
t
s
ev
er
e
o
cc
lu
s
io
n
s
.
T
h
o
u
g
h
MO
T
A
is
escalate
d
t
o
5
5
.
1
%
b
y
ca
s
ca
d
e
m
ask
R
-
a
s
f
o
r
C
NN,
th
e
d
eg
r
ee
o
f
i
d
e
n
tity
f
r
ag
m
en
tatio
n
r
em
ain
s
r
ath
er
h
ig
h
s
in
ce
th
e
in
ter
s
ec
tio
n
o
f
all
s
ce
n
ar
io
s
le
ad
s
to
a
lo
t
o
f
am
b
ig
u
ities
.
At
th
at,
th
e
in
clu
s
io
n
o
f
DPR
I
,
I
POM,
an
d
KSP
-
I
D
m
o
d
u
les in
s
eq
u
en
ce
co
n
tr
ib
u
t
es to
a
co
n
s
is
ten
t in
cr
ea
s
e
in
MO
T
A
u
p
to
8
0
.
0
%
f
r
o
m
7
0
.
0
%
an
d
in
GM
OT
A
u
p
to
6
5
.
0
%
f
r
o
m
5
5
.
0
%,
wh
ile
at
th
e
s
am
e
t
im
e
lead
in
g
to
a
r
ed
u
ctio
n
in
I
DS
as
well.
Su
ch
a
ten
d
en
cy
in
p
er
f
o
r
m
an
ce
ca
n
b
e
r
e
g
ar
d
e
d
as
an
ex
am
p
le
o
f
wh
at
was
s
tated
p
r
ev
io
u
s
ly
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dyn
a
mic
s
p
o
r
ts
en
viro
n
men
ts
:
a
u
to
ma
tic
fea
tu
r
e
p
l
a
ye
r
s
tr
a
ck
in
g
a
n
d
…
(
A
d
il A
b
d
u
lh
u
r
A
b
u
s
h
a
n
a
)
4299
r
eg
ar
d
in
g
th
e
p
r
o
b
lem
o
f
id
en
tity
p
r
eser
v
atio
n
in
d
en
s
e
s
p
o
r
ts
tr
ac
k
in
g
[
5
2
]
,
[
5
4
]
.
At
th
e
s
am
e
tim
e,
C
NN
is
p
r
o
v
e
d
m
o
r
e
s
tab
le
th
an
KSP
-
b
ased
p
ip
elin
es a
n
d
ap
p
r
o
ac
h
e
s
.
T
ab
le
3
.
Ab
latio
n
ex
p
e
r
im
en
t
r
esu
lts
o
n
th
e
STU
d
ataset
M
e
t
h
o
d
M
O
TA
↑
G
M
O
TA
↑
TP ↑
F
P
↓
F
N
↓
I
D
S
↓
P
O
M
+
K
S
P
4
1
.
4
3
4
.
9
7
7
.
9
2
5
.
6
3
6
.
2
1
3
.
6
C
a
sc
a
d
e
m
a
sk
R
C
N
N
-
P
+
P
O
M
+
K
S
P
5
5
.
1
4
2
.
5
8
6
.
8
2
2
.
3
3
1
.
4
9
.
9
O
u
r
s DP
R
I
6
7
.
0
5
4
.
0
9
7
.
3
1
2
.
9
2
5
.
7
–
O
u
r
s I
P
O
M
6
9
.
0
5
6
.
0
1
0
0
.
3
*
1
2
.
1
1
2
.
1
2
4
.
3
O
u
r
s
K
S
P
-
ID
7
2
.
0
5
8
.
0
1
0
3
.
2
*
1
0
.
0
1
0
.
0
2
3
.
0
O
u
r
s (f
u
l
l
)
8
0
.
0
6
5
.
0
1
0
7
.
6
*
5
.
7
5
.
7
1
7
.
2
4
.
5
.
Re
-
identif
ica
t
io
n per
f
o
r
m
a
nce
Play
er
re
-
id
en
tific
atio
n
p
er
f
o
r
m
an
ce
is
s
u
m
m
ar
ized
in
T
ab
l
e
4
.
B
aselin
e
p
er
f
o
r
m
a
n
ce
r
e
m
ain
s
wea
k
d
u
e
to
lim
ited
a
p
p
ea
r
an
ce
d
i
s
cr
im
in
atio
n
,
wh
ile
t
h
e
ca
s
ca
d
e
m
ask
R
-
C
NN
m
o
d
el
co
n
s
id
er
ab
ly
im
p
r
o
v
es
m
atch
co
n
s
is
ten
cy
.
T
h
e
d
ee
p
p
lay
er
m
o
d
u
le
s
u
r
p
ass
es
b
o
th
,
ac
h
iev
in
g
r
an
k
-
1
=
9
0
.
2
%,
t
o
p
-
3
=
9
3
.
5
%,
an
d
m
ea
n
av
er
a
g
e
p
r
ec
is
io
n
(
m
AP
)
=
8
0
.
0
%,
co
n
s
is
ten
t
with
p
r
io
r
e
v
id
en
ce
th
at
d
ee
p
s
tr
u
ctu
r
ed
id
en
tity
r
ep
r
esen
tatio
n
s
u
b
s
tan
tially
en
h
an
ce
s
p
lay
er
r
ec
o
g
n
itio
n
r
o
b
u
s
tn
ess
in
s
p
o
r
ts
an
aly
tics
[
3
5
]
,
[
5
2
]
.
T
ab
le
4
.
Ab
latio
n
s
tu
d
ies
o
n
p
lay
er
r
e
-
id
e
n
tific
atio
n
o
n
th
e
S
T
U
d
ataset
M
e
t
h
o
d
t
o
p
-
1
↑
t
o
p
-
3
↑
mA
P
↑
P
O
M
+
K
S
P
5
8
.
4
6
8
.
1
9
.
8
C
a
sc
a
d
e
m
a
sk
R
C
N
N
-
P
+
P
O
M
+
K
S
P
8
5
.
9
9
0
.
9
7
5
.
0
D
e
e
p
p
l
a
y
e
r
9
0
.
2
9
3
.
5
8
0
.
0
4
.
6
.
Dis
cus
s
io
n
T
h
e
ex
p
er
i
m
en
tal
r
esu
lts
clea
r
ly
s
h
o
w
th
at
co
m
b
in
in
g
th
e
d
ee
p
r
e
-
i
d
en
tific
atio
n
a
p
p
r
o
ac
h
,
th
e
p
r
o
b
a
b
ilis
tic
m
u
lti
-
v
iew
lo
ca
lizatio
n
,
an
d
t
h
e
id
en
tity
-
awa
r
e
g
r
ap
h
o
p
tim
izatio
n
h
as
a
tr
em
en
d
o
u
s
ef
f
ec
t
o
n
th
e
r
eliab
ilit
y
o
f
m
u
ltip
le
p
lay
er
s
’
tr
ac
k
in
g
in
d
y
n
a
m
ic
s
p
o
r
ts
en
v
ir
o
n
m
e
n
ts
.
T
h
e
class
ical
K
SP
tr
ac
k
er
s
[
4
9
]
,
[
5
0
]
,
o
n
th
e
o
t
h
er
h
an
d
,
wh
ich
m
ain
l
y
r
ely
o
n
k
ee
p
i
n
g
tr
ac
k
o
f
th
e
tr
ajec
to
r
ies
an
d
u
s
e
v
er
y
litt
le
ap
p
ea
r
an
ce
cu
es,
d
o
n
o
t
h
av
e
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
th
at
in
tr
o
d
u
ce
s
d
is
cr
im
in
ativ
e
id
en
tity
s
em
an
tics
an
d
s
p
atial
g
r
o
u
n
d
i
n
g
,
wh
o
s
e
r
o
le
is
to
s
et
u
p
s
tab
ilizin
g
c
o
n
s
tr
ain
ts
in
th
e
tim
es
o
f
am
b
ig
u
o
u
s
ass
o
ciatio
n
s
tates.
T
h
is
is
esp
ec
ially
tr
u
e
f
o
r
t
h
e
h
i
g
h
-
d
e
n
s
ity
s
ce
n
es
wh
er
e
th
er
e
ar
e
f
r
eq
u
e
n
t
in
te
r
-
p
lay
er
in
ter
ac
tio
n
s
an
d
o
v
er
lap
p
in
g
b
o
u
n
d
in
g
r
e
g
io
n
s
th
at
h
av
e
b
ee
n
lead
in
g
to
f
r
ag
m
en
tatio
n
an
d
id
en
tity
d
r
if
t
tr
ad
itio
n
ally
,
p
h
en
o
m
en
a
th
at
h
av
e
b
ee
n
ex
ten
s
iv
ely
r
ep
o
r
ted
in
th
e
p
r
ev
i
o
u
s
tr
ac
k
in
g
liter
atu
r
e
[
1
4
]
,
[
5
1
]
,
[
5
2
]
,
[
5
4
]
.
T
h
e
ev
alu
atio
n
r
esu
lts
th
at
s
h
o
w
a
m
ar
k
ed
r
ed
u
ctio
n
o
f
I
DS
in
o
u
r
ca
s
e
co
n
f
i
r
m
th
at
d
ee
p
f
ea
t
u
r
e
em
b
e
d
d
in
g
s
a
n
d
jer
s
ey
-
g
u
id
ed
r
ec
o
g
n
itio
n
r
ei
n
f
o
r
ce
m
e
n
t
o
f
id
en
tity
p
r
o
v
id
e
a
co
r
r
ec
tiv
e
in
f
l
u
en
ce
o
v
er
th
e
co
n
v
en
tio
n
a
l
m
o
tio
n
-
o
n
ly
r
ea
s
o
n
in
g
.
On
e
o
f
th
e
m
o
s
t sig
n
if
ican
t r
esu
lts
i
s
th
e
D
PR
I
m
o
d
u
le's su
p
p
o
r
t o
f
lo
n
g
-
ter
m
id
en
tity
co
n
s
is
ten
cy
in
a
s
ig
n
if
ican
t
way
.
T
h
e
m
o
d
el,
lik
e
p
r
ev
io
u
s
id
en
tity
-
ce
n
tr
ic
s
y
s
tem
s
[
3
5
]
,
[
4
4
]
,
is
aid
ed
b
y
a
h
y
b
r
i
d
ap
p
ea
r
an
ce
r
ep
r
esen
tatio
n
t
h
at
co
m
p
r
is
es
g
lo
b
al
p
lay
er
s
tr
u
ctu
r
e,
jer
s
ey
s
em
an
t
ics,
an
d
l
o
ca
lized
d
is
cr
im
in
ativ
e
em
b
e
d
d
in
g
s
,
b
u
t
it
is
n
o
t
as
p
r
ev
io
u
s
w
o
r
k
s
wh
er
e
jer
s
ey
r
ec
o
g
n
itio
n
was
an
is
o
lated
r
ec
o
g
n
itio
n
task
[
1
0
]
–
[
1
3
]
,
[
3
8
]
,
[
3
9
]
a
n
d
r
ec
o
g
n
itio
n
d
o
n
e
s
ep
ar
ately
,
th
u
s
,
th
e
cu
r
r
en
t
f
r
a
m
ewo
r
k
co
n
n
ec
ts
jer
s
ey
cu
es
to
th
e
ass
o
ciatio
n
p
r
o
ce
s
s
an
d
tu
r
n
in
g
jer
s
ey
d
ig
its
f
r
o
m
a
p
ass
iv
e
f
e
atu
r
e
in
to
an
ac
tiv
e
id
en
tity
co
n
s
tr
ain
t.
T
h
is
ac
co
u
n
ts
f
o
r
th
e
s
tr
o
n
g
co
r
r
elatio
n
b
etwe
en
im
p
r
o
v
ed
r
e
-
id
e
n
tific
atio
n
m
etr
ics
an
d
r
ed
u
ce
d
I
DS
v
alu
es,
wh
ich
f
u
r
th
er
s
u
p
p
o
r
ts
th
e
r
ec
en
t
co
n
clu
s
io
n
s
o
f
m
u
lti
-
ca
m
e
r
a
a
n
d
s
p
o
r
ts
tr
ac
k
in
g
r
esear
ch
[
1
3
]
,
[
5
2
]
,
[
5
7
]
.
T
h
e
p
r
o
b
ab
ilis
tic
o
cc
u
p
a
n
cy
an
d
I
POM
m
eth
o
d
s
g
iv
e
a
n
ex
tr
a
s
tr
u
ctu
r
al
b
en
ef
it
in
p
a
r
ticu
lar
to
g
am
ep
lay
s
itu
atio
n
s
th
at
a
r
e
h
ea
v
ily
b
lo
c
k
ed
.
E
a
r
ly
wo
r
k
s
u
s
in
g
p
r
o
b
ab
ilis
tic
an
d
v
o
lu
m
etr
ic
m
eth
o
d
s
[
1
4
]
,
[
5
1
]
,
[
5
3
]
,
[
5
6
]
h
av
e
s
h
o
wn
th
at
3
D
r
ea
s
o
n
in
g
with
o
cc
u
p
an
c
y
r
estrictio
n
s
ca
n
s
u
b
s
tan
tially
lo
wer
lo
ca
lizatio
n
am
b
ig
u
ity
.
T
h
e
cu
r
r
en
t
s
y
s
tem
g
o
es
o
n
e
s
tep
f
u
r
th
er
b
y
in
teg
r
atin
g
o
cc
u
p
a
n
cy
in
to
id
en
tity
-
awa
r
e
g
r
ap
h
o
p
tim
iza
tio
n
in
s
tead
o
f
u
s
in
g
it
as
a
s
e
p
ar
ate
lo
ca
lizatio
n
s
tag
e.
B
y
d
o
in
g
s
o
,
th
e
s
y
s
tem
m
ak
es
s
u
r
e
th
at
th
e
s
p
atial
a
n
d
id
en
tity
d
ec
is
io
n
s
ar
e
m
ad
e
co
n
cu
r
r
en
tly
th
r
o
u
g
h
o
u
t
th
e
tr
ac
k
in
g
p
r
o
ce
s
s
,
wh
ich
av
o
i
d
s
th
e
c
o
m
m
o
n
"i
d
en
tity
s
n
ap
"
p
h
e
n
o
m
en
o
n
h
a
p
p
en
in
g
wh
en
a
s
y
s
tem
r
elies
o
n
ly
o
n
g
eo
m
etr
ic
alig
n
m
en
t
o
r
a
v
is
u
al
s
im
ilar
it
y
.
T
h
e
id
e
n
tical
p
er
f
o
r
m
a
n
ce
o
f
API
DI
S
an
d
STU
i
n
d
icate
s
th
at
th
e
n
ew
m
o
d
el
h
as
th
e
ca
p
ab
ilit
y
o
f
m
ain
tai
n
in
g
its
r
o
b
u
s
tn
ess
o
v
er
d
ata
s
ets
wi
th
v
er
y
d
if
f
er
e
n
t
lev
els
o
f
co
n
g
esti
o
n
an
d
v
is
u
al
co
m
p
lex
ity
.
T
h
is
h
as b
e
en
a
m
ajo
r
p
r
o
b
lem
in
s
p
o
r
ts
an
aly
tics
h
is
to
r
ically
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
2
9
2
-
4
3
0
3
4300
T
h
e
ev
alu
atio
n
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
was
ag
ain
s
t
T
MO
SA
[
5
3
]
,
T
MA
P
-
EM
[
5
4
]
,
M
SP
T
[
5
5
]
,
h
y
b
r
id
s
to
ch
asti
c
–
d
eter
m
i
n
is
tic
tr
ac
k
er
s
[
1
7
]
,
an
d
MCMPTI
[
5
7
]
a
n
d
s
h
o
wed
th
at
th
e
p
r
o
p
o
s
ed
s
y
s
tem
h
as
a
well
-
b
alan
ce
d
p
er
f
o
r
m
an
ce
p
r
o
f
ile.
MCMPTI
,
with
it
s
ex
t
en
s
iv
e
m
u
lti
-
ca
m
er
a
r
ed
u
n
d
a
n
cy
,
an
d
T
MO
SA,
with
its
s
tr
o
n
g
s
tr
u
ctu
r
ed
co
n
s
tr
ain
ts
,
h
av
e
ac
h
iev
ed
b
etter
p
er
f
o
r
m
an
ce
th
an
th
e
p
r
o
p
o
s
ed
m
o
d
el.
Ho
wev
er
,
th
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
is
a
b
le
to
p
er
f
o
r
m
at
a
s
im
ilar
lev
el
an
d
ev
en
s
h
o
w
s
o
m
e
c
o
m
p
etitiv
e
tr
aits
in
ter
m
s
o
f
id
en
tity
s
tab
ilit
y
with
o
u
t
a
g
g
r
ess
iv
e
s
ce
n
e
ass
u
m
p
tio
n
s
o
r
co
m
p
le
x
co
n
s
tr
ain
t
tu
n
i
n
g
.
T
h
is
in
d
icate
s
th
at
th
e
p
r
o
p
o
s
ed
a
r
ch
itectu
r
e
h
as
a
m
o
r
e
ad
a
p
tab
le
s
o
lu
tio
n
f
o
r
d
if
f
er
en
t
t
y
p
es
o
f
s
p
o
r
tin
g
s
itu
atio
n
s
,
s
u
ch
as
th
o
s
e
with
in
co
n
s
is
ten
t
ca
m
er
a
ca
lib
r
atio
n
o
r
with
o
u
t
an
y
i
n
f
r
astru
ctu
r
e
at
all,
wh
er
e
it
m
ay
n
o
t
b
e
alwa
y
s
g
u
ar
an
teed
th
at
th
e
ca
m
e
r
a
ca
l
ib
r
atio
n
is
th
e
s
am
e
o
r
th
at
th
e
in
f
r
astru
ctu
r
e
is
p
r
esen
t.
I
n
s
p
ite
o
f
th
ese
b
en
ef
its
,
th
e
r
e
ar
e
s
till
a
f
ew
lim
itatio
n
s
th
at
n
ee
d
t
o
b
e
p
o
i
n
ted
o
u
t.
O
v
er
ex
tr
em
e
an
d
p
r
o
lo
n
g
ed
o
cc
lu
s
io
n
,
o
r
w
h
en
jer
s
ey
d
ig
its
s
tay
c
o
m
p
let
ely
in
v
is
ib
le
f
o
r
lo
n
g
p
e
r
io
d
s
o
f
tim
e,
th
e
s
y
s
tem
p
er
f
o
r
m
an
ce
s
till
g
iv
es
way
.
T
h
is
is
co
n
s
i
s
ten
t
wi
th
p
ast
r
esear
ch
o
n
jer
s
ey
-
d
r
iv
e
n
s
y
s
tem
s
[
1
0
]
–
[
1
3
]
,
[
3
8
]
.
Similar
ly
,
it
co
u
ld
b
e
th
at
v
is
u
al
em
b
ed
d
in
g
s
u
s
ed
f
o
r
id
en
t
ity
r
ely
to
o
m
u
c
h
o
n
th
e
d
o
m
a
in
,
an
d
tr
an
s
f
er
r
i
n
g
ac
r
o
s
s
s
ig
n
if
ican
tly
d
if
f
e
r
en
t
v
is
u
al
en
v
ir
o
n
m
en
ts
,
lig
h
tin
g
co
n
d
itio
n
s
,
o
r
team
u
n
if
o
r
m
d
esig
n
s
m
ay
b
ec
o
m
e
an
is
s
u
e.
T
h
e
s
y
s
tem
also
s
u
f
f
er
s
f
r
o
m
h
ig
h
co
m
p
u
tatio
n
al
co
s
t
d
u
e
to
th
e
m
u
lti
-
s
tag
e
p
r
o
ce
s
s
in
g
an
d
h
ig
h
-
d
im
e
n
s
io
n
al
em
b
ed
d
in
g
ex
tr
ac
tio
n
,
wh
ich
is
a
ch
ar
ac
ter
is
tic
s
h
ar
ed
b
y
ad
v
a
n
ce
d
d
ee
p
tr
ac
k
in
g
ar
ch
itectu
r
es
[
2
1
]
,
[
3
5
]
.
B
esid
es,
ev
en
th
o
u
g
h
th
e
s
tab
ilit
y
is
en
h
an
ce
d
b
y
m
u
lti
-
ca
m
er
a
co
n
d
itio
n
s
,
th
e
p
er
f
o
r
m
an
ce
in
c
h
allen
g
in
g
m
o
n
o
cu
lar
en
v
ir
o
n
m
en
ts
co
u
ld
s
till
b
e
im
p
r
o
v
ed
b
y
s
tr
o
n
g
er
tem
p
o
r
al
r
ea
s
o
n
in
g
o
r
lear
n
e
d
m
o
tio
n
p
r
io
r
s
.
I
n
s
u
m
m
ar
y
,
th
e
p
r
esen
ted
r
e
s
u
lts
b
o
ls
ter
th
e
ar
g
u
m
en
t
th
at
th
e
co
m
b
in
atio
n
o
f
jer
s
ey
s
em
an
tics
,
d
ee
p
r
e
-
id
en
tific
atio
n
,
p
r
o
b
ab
i
lis
tic
s
p
atial
r
ea
s
o
n
in
g
,
an
d
KSP
-
b
ased
id
en
tity
-
awa
r
e
o
p
tim
izatio
n
is
th
e
m
o
s
t
ef
f
ec
tiv
e
ap
p
r
o
ac
h
to
s
p
o
r
ts
m
u
lti
-
p
lay
er
tr
ac
k
in
g
f
r
o
m
a
s
tr
u
ctu
r
al
p
o
in
t
o
f
v
iew.
T
h
e
s
h
o
wca
s
ed
ad
v
an
tag
es
o
v
er
b
asic
KSP
f
r
am
ewo
r
k
s
[
4
9
]
,
[
5
0
]
,
p
r
o
b
ab
ilis
tic
o
cc
u
p
an
cy
m
o
d
els
[
1
4
]
,
[
5
1
]
,
h
y
b
r
id
o
p
tim
izatio
n
tr
ac
k
er
s
[
1
7
]
,
[
4
1
]
,
an
d
co
n
te
m
p
o
r
ar
y
d
ee
p
lear
n
i
n
g
an
d
m
u
lti
-
ca
m
er
a
s
o
lu
tio
n
s
[
5
2
]
,
[
5
4
]
,
[
5
5
]
,
[
5
7
]
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
m
eth
o
d
is
a
s
ig
n
if
ican
t
an
d
s
ca
lab
le
s
tep
f
o
r
war
d
in
t
h
e
d
ir
ec
tio
n
o
f
d
e
p
en
d
ab
le
au
to
m
ated
s
p
o
r
ts
an
aly
tics
.
5.
CO
NCLU
SI
O
N
I
n
th
is
p
ap
er
,
a
co
m
p
r
eh
e
n
s
iv
e
p
lay
er
tr
ac
k
in
g
m
eth
o
d
is
s
u
g
g
ested
wh
ich
co
m
b
in
es
DP
R
I
,
m
u
lti
-
v
iew
o
cc
u
p
an
cy
m
o
d
elin
g
,
a
n
d
i
d
en
tity
-
awa
r
e
KSP
o
p
tim
izatio
n
f
o
r
h
a
n
d
lin
g
id
en
t
ity
ch
an
g
es
an
d
th
e
in
s
tab
ilit
y
o
f
p
ath
s
in
d
y
n
am
i
c
s
p
o
r
ts
s
ce
n
es.
T
h
e
tes
ts
d
o
n
e
o
n
API
DI
S
an
d
STU
d
atase
ts
s
h
o
wed
th
at
th
e
p
r
o
p
o
s
ed
m
eth
o
d
p
r
o
v
i
d
es
co
n
s
id
er
ab
le
b
en
ef
its
in
ter
m
s
o
f
tr
ac
k
in
g
ac
cu
r
ac
y
,
g
lo
b
al
s
tab
ilit
y
,
an
d
id
en
tity
co
n
s
is
ten
cy
wh
en
co
m
p
ar
e
d
with
tr
ad
itio
n
al
KSP
s
y
s
tem
s
,
p
r
o
b
ab
ilis
tic
o
cc
u
p
a
n
cy
m
eth
o
d
s
,
an
d
cu
ttin
g
-
ed
g
e
d
ee
p
an
d
h
y
b
r
id
t
r
ac
k
er
s
.
T
h
e
a
b
latio
n
s
tu
d
y
co
n
f
ir
m
ed
t
h
at
th
e
r
o
b
u
s
tn
ess
ac
h
iev
ed
is
th
e
r
esu
lt
o
f
th
e
co
llab
o
r
atio
n
b
etwe
en
j
er
s
ey
-
g
u
id
e
d
s
em
an
tics
,
d
ee
p
em
b
ed
d
in
g
s
,
an
d
s
p
atial
g
r
o
u
n
d
in
g
.
T
h
e
a
p
p
r
o
a
ch
th
at
is
b
ein
g
p
r
esen
te
d
s
till
r
ev
ea
ls
a
s
tr
o
n
g
lev
el
o
f
g
en
er
aliza
tio
n
an
d
s
ca
lab
ilit
y
ev
en
th
o
u
g
h
ex
tr
e
m
e
o
cc
lu
s
io
n
s
an
d
d
o
m
ain
s
h
if
ts
ar
e
s
till
is
s
u
e
s
im
p
ac
tin
g
p
er
f
o
r
m
an
ce
.
T
h
e
f
u
tu
r
e
r
esear
c
h
will
th
en
p
o
in
t
at
im
p
r
o
v
in
g
co
m
p
u
tatio
n
al
ef
f
i
cien
cy
,
ca
r
r
y
in
g
o
u
t
e
v
alu
ati
o
n
s
in
a
wid
er
v
ar
iety
o
f
s
p
o
r
ts
s
ce
n
ar
io
s
,
an
d
in
teg
r
atin
g
th
e
latest
tem
p
o
r
al
an
d
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
tech
n
iq
u
es
to
f
u
r
th
er
co
n
f
i
r
m
lo
n
g
-
ter
m
id
en
tity
p
r
eser
v
atio
n
as th
e
m
o
s
t c
r
itical
p
r
o
b
lem
t
o
b
e
s
o
lv
e
d
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
A
d
i
l
A
b
d
u
l
h
u
r
A
b
u
s
h
a
n
a
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
R
iy
ad
h
Als
ae
ed
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dyn
a
mic
s
p
o
r
ts
en
viro
n
men
ts
:
a
u
to
ma
tic
fea
tu
r
e
p
l
a
ye
r
s
tr
a
ck
in
g
a
n
d
…
(
A
d
il A
b
d
u
lh
u
r
A
b
u
s
h
a
n
a
)
4301
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
I
NF
O
RM
E
D
CO
NS
E
N
T
T
h
is
s
ec
tio
n
n
o
t
ap
p
licab
le.
T
h
is
s
tu
d
y
d
id
n
o
t
in
v
o
lv
e
h
u
m
an
p
ar
ticip
an
ts
,
h
u
m
a
n
d
a
ta,
o
r
an
y
p
er
s
o
n
ally
id
en
tifia
b
le
in
f
o
r
m
atio
n
.
All d
ata
u
s
ed
wer
e
eith
e
r
p
u
b
licly
a
v
ailab
le,
f
u
lly
an
o
n
y
m
ized
,
o
r
d
e
r
iv
ed
f
r
o
m
n
o
n
‑
h
u
m
a
n
s
o
u
r
ce
s
,
an
d
th
er
ef
o
r
e
n
o
in
f
o
r
m
e
d
co
n
s
en
t w
as r
eq
u
ir
ed
f
r
o
m
i
n
d
iv
id
u
als.
E
T
H
I
CAL
AP
P
RO
V
AL
T
h
is
s
ec
tio
n
n
o
t
a
p
p
licab
le
.
T
h
is
r
esear
ch
d
id
n
o
t
in
v
o
lv
e
h
u
m
a
n
s
u
b
jects,
h
u
m
an
b
io
lo
g
ical
m
ater
ials
,
o
r
ex
p
er
im
en
tal
p
r
o
ce
d
u
r
es
o
n
an
im
als.
T
h
e
wo
r
k
was
co
n
d
u
cted
s
o
lely
o
n
co
m
p
u
tatio
n
al
m
o
d
els,
p
u
b
licly
av
ailab
le
d
atasets
,
o
r
n
o
n
‑
s
en
s
itiv
e
d
ata
th
at
d
i
d
n
o
t
r
eq
u
i
r
e
in
ter
v
en
tio
n
with
liv
in
g
o
r
g
a
n
is
m
s
.
T
h
er
ef
o
r
e,
eth
ical
ap
p
r
o
v
al
f
r
o
m
an
in
s
titu
tio
n
al
r
ev
iew
b
o
ar
d
o
r
a
n
im
al
eth
ics
c
o
m
m
itte
e
was
n
o
t
n
ec
ess
ar
y
f
o
r
th
is
s
tu
d
y
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
i
n
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
a
v
ailab
le
f
r
o
m
th
e
c
o
r
r
esp
o
n
d
in
g
au
t
h
o
r
,
[
AA
]
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
H
.
P
i
r
s
i
a
v
a
s
h
,
D
.
R
a
ma
n
a
n
,
a
n
d
C
.
C
.
F
o
w
l
k
e
s,
“
G
l
o
b
a
l
l
y
-
o
p
t
i
mal
g
r
e
e
d
y
a
l
g
o
r
i
t
h
ms
f
o
r
t
r
a
c
k
i
n
g
a
v
a
r
i
a
b
l
e
n
u
mb
e
r
o
f
o
b
j
e
c
t
s,
”
i
n
I
EEE
C
o
m
p
u
t
e
r V
i
si
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
(
C
VP
R)
2
0
1
1
,
J
u
n
.
2
0
1
1
,
p
p
.
1
2
0
1
–
1
2
0
8
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
1
.
5
9
9
5
6
0
4
.
[
2
]
N
.
W
o
j
k
e
,
A
.
B
e
w
l
e
y
,
a
n
d
D
.
P
a
u
l
u
s,
“
S
i
m
p
l
e
o
n
l
i
n
e
a
n
d
r
e
a
l
t
i
me
t
r
a
c
k
i
n
g
w
i
t
h
a
d
e
e
p
a
ss
o
c
i
a
t
i
o
n
me
t
r
i
c
,
”
i
n
2
0
1
7
I
E
EE
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
I
m
a
g
e
P
ro
c
e
ssi
n
g
,
S
e
p
.
2
0
1
7
,
p
p
.
3
6
4
5
–
3
6
4
9
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
P
.
2
0
1
7
.
8
2
9
6
9
6
2
.
[
3
]
A
.
M
i
l
a
n
,
L.
L.
-
T
a
i
x
e
,
I
.
R
e
i
d
,
S
.
R
o
t
h
,
a
n
d
K
.
S
c
h
i
n
d
l
e
r
,
“
M
O
T
1
6
:
a
b
e
n
c
h
mark
f
o
r
m
u
l
t
i
-
o
b
j
e
c
t
t
r
a
c
k
i
n
g
,
”
2
0
1
6
,
a
rXi
v
:
1
6
0
3
.
0
0
8
3
1
.
[
4
]
H
.
W
u
,
Y
.
H
u
,
K
.
W
a
n
g
,
H
.
L
i
,
L.
N
i
e
,
a
n
d
H
.
C
h
e
n
g
,
“
I
n
st
a
n
c
e
-
a
w
a
r
e
r
e
p
r
e
se
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
a
n
d
a
ss
o
c
i
a
t
i
o
n
f
o
r
o
n
l
i
n
e
mu
l
t
i
-
p
e
r
s
o
n
t
r
a
c
k
i
n
g
,
”
P
a
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
v
o
l
.
9
4
,
p
p
.
2
5
–
3
4
,
O
c
t
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
c
o
g
.
2
0
1
9
.
0
4
.
0
1
8
.
[
5
]
S
.
G
e
r
k
e
,
K
.
M
u
l
l
e
r
,
a
n
d
R
.
S
c
h
a
f
e
r
,
“
S
o
c
c
e
r
j
e
r
se
y
n
u
m
b
e
r
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
i
n
2
0
1
5
I
E
E
E
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r V
i
s
i
o
n
Wo
rks
h
o
p
,
D
e
c
.
2
0
1
5
,
p
p
.
7
3
4
–
7
4
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
W
.
2
0
1
5
.
1
0
0
.
[
6
]
Z.
C
a
i
a
n
d
N
.
V
a
s
c
o
n
c
e
l
o
s,
“
C
a
s
c
a
d
e
R
-
C
N
N
:
d
e
l
v
i
n
g
i
n
t
o
h
i
g
h
q
u
a
l
i
t
y
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
1
8
I
E
EE/
C
V
F
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
J
u
n
.
2
0
1
8
,
p
p
.
6
1
5
4
–
6
1
6
2
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
8
.
0
0
6
4
4
.
[
7
]
A
.
K
h
o
t
a
n
z
a
d
a
n
d
Y
.
H
.
H
o
n
g
,
“
I
n
v
a
r
i
a
n
t
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
b
y
Ze
r
n
i
k
e
m
o
me
n
t
s
,
”
I
E
EE
T
r
a
n
sa
c
t
i
o
n
s
o
n
P
a
t
t
e
rn
A
n
a
l
y
s
i
s
a
n
d
Ma
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
2
,
n
o
.
5
,
p
p
.
4
8
9
–
4
9
7
,
M
a
y
1
9
9
0
,
d
o
i
:
1
0
.
1
1
0
9
/
3
4
.
5
5
1
0
9
.
[
8
]
S
.
H
.
Le
e
,
J.
Y
.
C
h
o
i
,
Y
.
M
.
R
o
,
a
n
d
K
.
N
.
P
l
a
t
a
n
i
o
t
i
s
,
“
L
o
c
a
l
c
o
l
o
r
v
e
c
t
o
r
b
i
n
a
r
y
p
a
t
t
e
r
n
s
f
r
o
m
mu
l
t
i
c
h
a
n
n
e
l
f
a
c
e
i
m
a
g
e
s
f
o
r
f
a
c
e
r
e
c
o
g
n
i
t
i
o
n
,
”
I
EEE
T
r
a
n
sa
c
t
i
o
n
s
o
n
I
m
a
g
e
Pr
o
c
e
ssi
n
g
,
v
o
l
.
2
1
,
n
o
.
4
,
p
p
.
2
3
4
7
–
2
3
5
3
,
A
p
r
.
2
0
1
2
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
P
.
2
0
1
1
.
2
1
8
1
5
2
6
.
[
9
]
Z.
T
u
,
X
.
C
h
e
n
,
Y
u
i
l
l
e
,
a
n
d
Z
h
u
,
“
I
mag
e
p
a
r
si
n
g
:
u
n
i
f
y
i
n
g
s
e
g
m
e
n
t
a
t
i
o
n
,
d
e
t
e
c
t
i
o
n
,
a
n
d
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
N
i
n
t
h
I
EEE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
,
2
0
0
3
,
p
p
.
1
8
–
25
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
.
2
0
0
3
.
1
2
3
8
3
0
9
.
[
1
0
]
H
.
Li
,
H
.
Zh
a
n
g
,
a
n
d
Y
.
C
h
e
n
,
“
D
u
a
l
-
TSS
T:
a
d
u
a
l
-
b
r
a
n
c
h
t
e
m
p
o
r
a
l
–
sp
e
c
t
r
a
l
–
s
p
a
t
i
a
l
t
r
a
n
sf
o
r
mer
mo
d
e
l
f
o
r
EEG
d
e
c
o
d
i
n
g
,
”
I
EEE
J
o
u
r
n
a
l
o
f
B
i
o
m
e
d
i
c
a
l
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
9
,
n
o
.
9
,
p
p
.
6
5
2
4
–
6
5
3
7
,
S
e
p
.
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
J
B
H
I
.
2
0
2
5
.
3
5
7
7
6
1
1
.
[
1
1
]
H
.
Li
u
a
n
d
B
.
B
h
a
n
u
,
“
P
o
se
-
g
u
i
d
e
d
R
-
C
N
N
f
o
r
j
e
r
se
y
n
u
m
b
e
r
r
e
c
o
g
n
i
t
i
o
n
i
n
s
p
o
r
t
s,
”
i
n
2
0
1
9
I
E
EE/
C
V
F
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
W
o
rks
h
o
p
s
,
J
u
n
.
2
0
1
9
,
p
p
.
2
4
5
7
–
2
4
6
6
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
W
.
2
0
1
9
.
0
0
3
0
1
.
[
1
2
]
K
.
V
a
t
s
,
M
.
F
a
n
i
,
D
.
A
.
C
l
a
u
si
,
a
n
d
J.
Ze
l
e
k
,
“
M
u
l
t
i
t
a
sk
l
e
a
r
n
i
n
g
f
o
r
j
e
r
se
y
n
u
m
b
e
r
r
e
c
o
g
n
i
t
i
o
n
i
n
i
c
e
h
o
c
k
e
y
,
”
i
n
4
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
W
o
rks
h
o
p
o
n
Mu
l
t
i
m
e
d
i
a
C
o
n
t
e
n
t
A
n
a
l
y
si
s
i
n
S
p
o
r
t
s
,
O
c
t
.
2
0
2
1
,
p
p
.
1
1
–
15
,
d
o
i
:
1
0
.
1
1
4
5
/
3
4
7
5
7
2
2
.
3
4
8
2
7
9
4
.
[
1
3
]
A
.
C
i
o
p
p
a
,
A
.
D
e
l
i
è
g
e
,
S
.
G
i
a
n
c
o
l
a
,
B
.
G
h
a
n
e
m,
a
n
d
M
.
V
.
D
r
o
o
g
e
n
b
r
o
e
c
k
,
“
S
c
a
l
i
n
g
u
p
S
o
c
c
e
r
N
e
t
w
i
t
h
mu
l
t
i
-
v
i
e
w
s
p
a
t
i
a
l
l
o
c
a
l
i
z
a
t
i
o
n
a
n
d
r
e
-
i
d
e
n
t
i
f
i
c
a
t
i
o
n
,
”
S
c
i
e
n
t
i
f
i
c
D
a
t
a
,
v
o
l
.
9
,
n
o
.
1
,
J
u
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
7
-
022
-
0
1
4
6
9
-
1.
[
1
4
]
F
.
F
l
e
u
r
e
t
,
J.
B
e
r
c
l
a
z
,
R
.
Le
n
g
a
g
n
e
,
a
n
d
P
.
F
u
a
,
“
M
u
l
t
i
c
a
mer
a
p
e
o
p
l
e
t
r
a
c
k
i
n
g
w
i
t
h
a
p
r
o
b
a
b
i
l
i
s
t
i
c
o
c
c
u
p
a
n
c
y
m
a
p
,
”
I
EE
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
A
n
a
l
y
si
s
a
n
d
Ma
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
3
0
,
n
o
.
2
,
p
p
.
2
6
7
–
2
8
2
,
F
e
b
.
2
0
0
8
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
0
7
.
1
1
7
4
.
[
1
5
]
Q
.
W
a
n
g
,
Z.
Li
u
,
Z
.
W
u
,
J.
W
a
n
g
,
C
.
Q
u
,
a
n
d
J
.
L
i
,
“
D
r
i
f
t
s
u
p
p
r
e
ss
i
o
n
m
e
t
h
o
d
b
a
s
e
d
o
n
si
g
n
a
l
st
a
b
i
l
i
t
y
d
e
t
e
c
t
i
o
n
a
n
d
a
d
a
p
t
i
v
e
K
a
l
ma
n
f
i
l
t
e
r
f
o
r
N
M
R
s
e
n
s
o
r
,
”
D
i
g
i
t
a
l
S
i
g
n
a
l
Pro
c
e
ss
i
n
g
,
v
o
l
.
1
5
6
,
Ja
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
d
sp
.
2
0
2
4
.
1
0
4
8
1
2
.
[
1
6
]
W
.
L
u
,
X
.
W
a
n
g
,
W
.
A
n
,
C
.
X
i
a
o
,
Q
.
Y
i
n
,
a
n
d
G
.
Zh
a
n
g
,
“
O
I
Tr
a
c
k
:
mu
l
t
i
-
o
b
j
e
c
t
t
r
a
c
k
i
n
g
f
o
r
sm
a
l
l
t
a
r
g
e
t
s
i
n
s
a
t
e
l
l
i
t
e
v
i
d
e
o
v
i
a
o
n
l
i
n
e
t
r
a
j
e
c
t
o
r
y
c
o
m
p
l
e
t
i
o
n
a
n
d
i
t
e
r
a
t
i
v
e
e
x
p
a
n
s
i
o
n
o
v
e
r
u
n
i
o
n
,
”
Re
m
o
t
e
S
e
n
s
i
n
g
,
v
o
l
.
1
7
,
n
o
.
1
2
,
Ju
n
.
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s
1
7
1
2
2
0
4
2
.
[
1
7
]
R
.
T.
C
o
l
l
i
n
s
a
n
d
P
.
C
a
r
r
,
“
H
y
b
r
i
d
st
o
c
h
a
st
i
c
/
d
e
t
e
r
m
i
n
i
st
i
c
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
t
r
a
c
k
i
n
g
s
p
o
r
t
s
p
l
a
y
e
r
s
a
n
d
p
e
d
e
s
t
r
i
a
n
s,”
i
n
C
o
m
p
u
t
e
r
Vi
si
o
n
–
E
C
C
V
2
0
1
4
,
C
h
a
m,
S
w
i
t
z
e
r
l
a
n
d
:
S
p
r
i
n
g
e
r
,
2
0
1
4
,
p
p
.
2
9
8
–
313
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
1
0
6
0
5
-
2
_
2
0
.
[
1
8
]
K
.
H
e
,
G
.
G
k
i
o
x
a
r
i
,
P
.
D
o
l
l
a
r
,
a
n
d
R
.
G
i
r
sh
i
c
k
,
“
M
a
sk
R
-
C
N
N
,
”
I
E
EE
T
ra
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
An
a
l
y
s
i
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
4
2
,
n
o
.
2
,
p
p
.
3
8
6
–
3
9
7
,
F
e
b
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
1
8
.
2
8
4
4
1
7
5
.
[
1
9
]
J.
R
e
d
m
o
n
,
S
.
D
i
v
v
a
l
a
,
R
.
G
i
r
sh
i
c
k
,
a
n
d
A
.
F
a
r
h
a
d
i
,
“
Y
o
u
o
n
l
y
l
o
o
k
o
n
c
e
:
u
n
i
f
i
e
d
,
r
e
a
l
-
t
i
me
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
1
6
I
E
E
E
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
J
u
n
.
2
0
1
6
,
p
p
.
7
7
9
–
7
8
8
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
6
.
9
1
.
[
2
0
]
T.
-
Y
.
Li
n
,
P
.
D
o
l
l
a
r
,
R
.
G
i
r
s
h
i
c
k
,
K
.
H
e
,
B
.
H
a
r
i
h
a
r
a
n
,
a
n
d
S
.
B
e
l
o
n
g
i
e
,
“
F
e
a
t
u
r
e
p
y
r
a
mi
d
n
e
t
w
o
r
k
s
f
o
r
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
1
7
I
EEE
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
,
Ju
l
.
2
0
1
7
,
p
p
.
9
3
6
–
9
4
4
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
7
.
1
0
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.