I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
d
van
c
e
s
i
n
A
p
p
li
e
d
S
c
ie
n
c
e
s
(
I
JA
A
S
)
V
ol
.
15
, N
o.
1
,
M
a
r
c
h
20
26
, pp.
209
~
218
I
S
S
N
:
2252
-
8814
,
D
O
I
:
10.11591/
ij
a
a
s
.
v15.
i
1
.
pp
209
-
218
209
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
aas
.i
ae
s
c
or
e
.c
om
R
ob
u
st
m
u
l
t
i
-
f
ac
e
s r
e
c
ogn
i
t
i
on
an
d
t
r
ac
k
i
n
g vi
a f
u
z
z
y ge
n
e
t
i
c
al
gor
i
t
h
m
s an
d
d
e
e
p
c
ou
p
l
e
d
f
e
at
u
r
e
s
A
d
il
A
b
d
u
lh
u
r
A
b
u
s
h
an
a
1
, Y
ou
s
if
S
am
e
r
M
u
d
h
af
ar
1
,
2
1
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
, F
a
c
ul
t
y of
E
duc
a
t
i
on, U
ni
ve
r
s
i
t
y of
K
uf
a
, N
a
j
a
f
, I
r
a
q
2
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
T
e
c
hni
que
s
E
ngi
ne
e
r
i
ng, F
a
c
ul
t
y of
T
e
c
hni
c
a
l
E
ngi
n
e
e
r
i
ng, I
s
l
a
m
i
c
U
ni
ve
r
s
i
t
y, N
a
j
a
f
, I
r
a
q
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
A
ug 11, 2025
R
e
vi
s
e
d
N
ov 24, 2025
A
c
c
e
pt
e
d
J
a
n 1, 2026
In
real
-
world
surveillance
environments,
face
recognition
and
tr
acking
remain
challenging
due
to
partial
occlusion,
pose
variation,
illumi
nation
changes,
and
background
clutter.
This
paper
presents
a
robust
hybrid
framework
that
integrates
fuzzy
genetic
algorithm
s
(FGA)
with
deep
coupled
feature
learning
for
multi
-
face
recognition
and
tracking
.
The
proposed
system
comprises
three
main
modules:
i
)
face
detection
an
d
pre
-
processing
using
the
multi
-
task
cascaded
convolutional
network
(MT
CNN),
ii
)
deep
coupled
ResNet
embeddings
that
jointly
learn
identit
y
and
appearance
-
invariant
represe
ntations,
and
iii
)
a
fuzzy
rule
-
based
genetic
optimizer
that
adaptively
refines
tracking
decisions
based
on
uncerta
inty
in
motion,
appeara
nce
similarity,
and
occlusion
levels.
The
novelty
of
this
work
lies
in
the
fusion
of
fuzzy
inference
with
evolutionary
search
to
guide
the
genetic
optimization
process
—
allowin
g
dynamic
adaptatio
n
to
noisy
and
uncertain
visual
conditions.
Moreover,
probabilistic
data
association
filters
(PDAF)
an
d
conditional
joint
likelihood
filters
(CJLF)
are
emplo
yed
to
further
enhance
temporal
consistency
under
occlusion
and
appe
arance
variation.
The
results
confirm
that
fuzzy
evolution
ary
optimization,
when
coupled
with
deep
feature
learning,
signifi
cantly
improves
robustn
e
ss
and
stability for rea
l
-
time face
tracking
in complex, d
ynamic sce
nes.
K
e
y
w
o
r
d
s
:
F
uz
z
y ge
ne
ti
c
a
lg
or
it
hm
s
G
e
ne
ti
c
pa
r
ti
c
le
f
il
te
r
in
g
J
oi
nt
pr
oba
bi
li
ty
a
s
s
oc
ia
ti
on
R
o
b
u
s
t
d
a
t
a
a
s
s
o
c
i
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
T
e
xt
ur
e
-
c
ol
or
-
s
ha
pe
f
e
a
tu
r
e
s
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
Y
ous
if
S
a
m
e
r
M
udha
f
a
r
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
S
c
ie
nc
e
, F
a
c
ul
ty
of
E
duc
a
ti
on, Unive
r
s
it
y of
K
uf
a
N
a
ja
f
, I
r
a
q
E
m
a
il
:
yous
if
s
.m
udha
f
a
r
@
uokuf
a
.e
du.i
q
1.
I
N
T
R
O
D
U
C
T
I
O
N
M
ul
ti
-
f
a
c
e
r
e
c
ogni
ti
on
a
nd
tr
a
c
ki
ng
a
r
e
e
s
s
e
nt
ia
l
in
s
ur
ve
il
la
nc
e
,
m
oni
to
r
in
g
s
ys
te
m
s
, a
nd
in
te
ll
ig
e
nt
vi
de
o
a
na
ly
ti
c
s
[
1]
,
[
2]
,
but
pe
r
f
or
m
a
nc
e
de
gr
a
de
s
unde
r
oc
c
lu
s
io
n,
pos
e
c
h
a
nge
s
,
vi
s
u
a
l
s
im
il
a
r
it
y,
a
nd
lo
w
-
r
e
s
ol
ut
io
n
im
a
ge
r
y
[
3]
,
[
4
]
.
C
la
s
s
ic
a
l
tr
a
c
ki
ng
a
ppr
oa
c
h
e
s
r
e
ly
on
ha
nd
-
c
r
a
f
te
d
f
e
a
tu
r
e
s
th
a
t
a
r
e
not
r
obus
t
to
s
uc
h
va
r
ia
ti
ons
,
w
hi
le
de
e
p
le
a
r
ni
ng
m
e
th
ods
h
a
v
e
im
pr
ove
d
di
s
c
r
im
in
a
ti
ve
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
th
r
ough
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
s
(
C
N
N
s
)
[
5]
,
[
6]
.
H
ow
e
ve
r
,
C
N
N
-
ba
s
e
d
m
ode
ls
s
ti
ll
s
tr
uggl
e
w
he
n
f
a
c
e
s
a
r
e
c
a
pt
ur
e
d
a
t
lo
w
r
e
s
ol
ut
io
n
or
unde
r
poor
li
ght
in
g
c
ondi
ti
ons
,
w
hi
c
h
r
e
duc
e
s
bot
h
id
e
nt
if
ic
a
ti
on
a
nd
re
-
id
e
nt
if
ic
a
ti
on r
e
li
a
bi
li
ty
i
n
r
e
a
l
-
w
or
ld
s
ur
ve
il
la
nc
e
[
7]
, [
8]
.
T
o
a
ddr
e
s
s
th
e
s
e
c
ha
ll
e
ng
e
s
,
hybr
id
a
ppr
oa
c
h
e
s
c
om
bi
ni
ng
de
e
p
le
a
r
ni
ng
w
it
h
opt
im
iz
a
ti
on
-
ba
s
e
d
r
e
a
s
oni
ng
ha
ve
ga
in
e
d
a
tt
e
nt
io
n.
G
e
ne
ti
c
a
lg
or
it
hm
s
(
G
A
s
)
p
r
o
vi
de
a
da
pt
iv
e
gl
oba
l
s
e
a
r
c
h,
a
nd
w
he
n
pa
ir
e
d
w
it
h
f
uz
z
y
lo
gi
c
,
c
a
n
be
tt
e
r
ha
ndl
e
un
c
e
r
ta
in
ty
a
nd
pa
r
ti
a
l
vi
s
ib
il
it
y
dur
in
g
oc
c
lu
s
io
n
[
9]
.
T
hus
,
in
te
gr
a
ti
ng
de
e
p
f
e
a
tu
r
e
r
e
pr
e
s
e
nt
a
ti
on
w
it
h
f
uz
z
y
-
opt
im
iz
e
d
tr
a
c
ki
ng
yi
e
ld
s
a
m
or
e
r
obus
t
a
nd
a
da
pt
a
bl
e
s
ol
ut
io
n
f
or
m
ul
ti
-
f
a
c
e
t
r
a
c
ki
ng
(
M
F
T
)
[
10]
, [
11
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
209
-
218
210
R
e
c
e
nt
M
F
T
r
e
s
e
a
r
c
h
ha
s
e
vol
ve
d
f
r
om
de
te
c
ti
on
-
ba
s
e
d
li
nki
n
g
to
a
da
pt
iv
e
a
s
s
o
c
ia
ti
on
f
r
a
m
e
w
or
ks
.
A
r
a
c
hc
hi
la
ge
a
nd
I
z
qui
e
r
do
[
12]
im
pr
ove
d
te
m
por
a
l
c
ons
i
s
te
nc
y
th
r
ough
a
da
pt
iv
e
tr
a
c
kl
e
t
a
ggr
e
ga
ti
on,
w
hi
le
B
a
r
que
r
o
e
t
al
.
[
13]
a
ddr
e
s
s
e
d
c
r
ow
de
d
-
s
c
e
ne
r
e
c
onne
c
ti
on,
a
nd
Z
ha
ng
e
t
al
.
[
14]
e
nh
a
nc
e
d
a
ppe
a
r
a
nc
e
r
obus
tn
e
s
s
vi
a
uns
up
e
r
vi
s
e
d
a
da
pt
a
ti
on.
F
ur
th
e
r
r
e
f
in
e
m
e
nt
s
,
s
uc
h
a
s
ve
r
if
ic
a
ti
on
-
ba
s
e
d
r
a
nki
ng
[
15]
a
nd
s
tr
uc
tu
r
e
d
-
s
c
e
ne
opt
im
iz
a
ti
on
[
16]
,
im
pr
ove
d s
ta
bi
li
ty
but
s
tr
ug
gl
e
d unde
r
s
e
ve
r
e
oc
c
lu
s
io
n
s
.
R
e
gi
ona
l
s
im
pl
e
onl
in
e
a
nd
r
e
a
l
-
ti
m
e
tr
a
c
ki
ng
(
R
e
S
O
R
T
)
in
tr
oduc
e
d
I
D
r
e
c
ove
r
y
[
17]
,
a
nd
double
-
t
r
ip
le
t
ne
twor
ks
im
pr
ove
d
c
r
os
s
-
c
a
m
e
r
a
c
on
s
is
te
nc
y
[
18]
.
M
or
e
r
e
c
e
nt
m
e
th
od
s
in
te
gr
a
te
m
ul
ti
m
oda
l
c
ue
s
,
us
in
g
bot
h
f
a
c
e
a
nd
body
f
e
a
tu
r
e
s
[
19]
,
m
e
m
or
y
-
ba
s
e
d
m
a
tc
hi
ng
[
20]
,
or
bi
om
e
tr
ic
f
u
s
io
n
[
21]
to
im
pr
ove
r
e
-
id
e
nt
if
ic
a
ti
on
unde
r
a
m
bi
gui
ty
.
I
n
th
is
pa
pe
r
,
w
e
pr
opos
e
a
nove
l
hybr
id
f
r
a
m
e
w
or
k
th
a
t
in
te
gr
a
te
s
th
e
f
uz
z
y
da
t
a
a
s
s
oc
ia
ti
on
e
ngi
ne
f
r
om
L
i
a
nd
Z
ha
n
[
22]
w
it
h
a
de
e
p
c
oupl
e
d
R
e
s
N
e
t
f
o
r
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
nd
a
G
A
f
or
opt
im
iz
in
g
tr
a
c
ki
ng
de
c
is
io
n
s
.
O
ur
pr
im
a
r
y
c
ont
r
ib
ut
io
n
li
e
s
in
th
e
s
yn
e
r
gi
s
ti
c
c
om
bi
na
ti
on
of
th
e
s
e
th
r
e
e
c
om
pone
nt
s
.
W
hi
le
th
e
w
or
k
of
L
i
a
nd
Z
ha
n
[
22]
pr
ovi
de
s
th
e
c
or
e
f
uz
z
y
l
ogi
c
f
or
a
s
s
oc
ia
ti
on,
our
f
r
a
m
e
w
or
k
e
xt
e
nds
it
by
f
e
e
di
ng
th
e
s
ys
te
m
r
ic
h,
di
s
c
r
im
in
a
ti
ve
f
e
a
tu
r
e
s
f
r
om
a
de
e
p
ne
twor
k
a
nd
th
e
n
us
in
g
a
GA
to
dyna
m
ic
a
ll
y
r
e
f
in
e
a
nd
opt
im
iz
e
th
e
tr
a
c
ki
ng
hypothe
s
e
s
ge
ne
r
a
te
d
by
th
e
f
uz
z
y
s
ys
te
m
.
T
h
e
in
nova
ti
ve
a
s
pe
c
t
of
th
e
pr
opos
e
d s
ys
te
m
i
s
i
ll
us
tr
a
te
d by thr
e
e
pr
in
c
ip
a
l
de
s
ig
n e
le
m
e
nt
s
:
i)
T
he
hybr
id
s
ys
te
m
c
om
bi
ne
s
de
e
p
c
oupl
e
d
R
e
s
N
e
t
e
m
be
d
di
ngs
[
5]
w
it
h
a
GA
,
m
odul
a
te
d
by
a
M
a
m
da
ni
f
uz
z
y
in
f
e
r
e
nc
e
s
y
s
te
m
[
22]
.
T
hi
s
uni
que
f
e
nc
e
e
na
bl
e
s
s
im
ul
ta
ne
ous
d
e
e
p
f
e
a
tu
r
e
le
a
r
ni
ng,
w
hi
c
h i
s
gr
ounde
d i
n s
to
c
ha
s
ti
c
it
y, w
it
h a
s
to
c
ha
s
ti
c
, e
vol
ut
io
n
-
ba
s
e
d s
e
a
r
c
h opti
m
iz
a
ti
on.
ii)
T
he
r
e
i
s
a
f
e
e
dba
c
k
-
dr
iv
e
n a
da
pt
a
ti
on me
c
h
a
ni
s
m
, w
he
r
e
by f
uz
z
y c
onf
id
e
nc
e
out
put
s
dyna
m
ic
a
ll
y a
da
pt
G
A
pa
r
a
m
e
te
r
s
(
e
.g.
s
e
le
c
ti
on
pr
e
s
s
ur
e
a
nd
m
ut
a
ti
on
r
a
te
)
in
a
c
ont
e
xt
-
s
pe
c
if
ic
m
a
nne
r
dur
in
g
a
r
e
a
l
-
ti
m
e
a
da
pt
a
ti
on t
o oc
c
lu
s
io
n s
e
v
e
r
it
y, de
te
c
ti
on c
onf
id
e
nc
e
, a
nd a
ppe
a
r
a
nc
e
s
im
il
a
r
it
y.
iii)
I
nc
or
por
a
ti
ng
pr
oba
bi
li
s
ti
c
da
ta
a
s
s
oc
ia
ti
on
f
il
te
r
s
(
P
D
A
F
)
,
th
e
c
ondi
ti
ona
l
jo
in
t
li
ke
li
hood
f
il
te
r
s
(
C
J
L
F
)
im
pr
ove
te
m
po
r
a
l
c
ons
is
te
nc
y
a
c
r
os
s
f
r
a
m
e
s
a
m
pl
in
g
r
a
te
s
a
s
it
r
e
ta
in
s
id
e
nt
it
y
c
ons
ta
nc
y
w
hi
le
m
in
im
iz
in
g dr
if
t
unde
r
va
r
y
in
g a
ppe
a
r
a
nc
e
s
a
nd w
h
e
n ove
r
la
pp
in
g w
it
h ot
he
r
t
a
r
ge
ts
.
T
oge
th
e
r
,
th
e
s
e
th
r
e
e
c
om
pone
nt
s
c
ont
r
ib
ut
e
to
a
r
obus
t
a
nd
s
c
a
la
bl
e
M
F
T
f
r
a
m
e
w
or
k,
pr
ovi
di
ng
e
nha
nc
e
d
c
a
pa
bi
li
ty
to
m
a
in
ta
in
hi
gh
a
c
c
ur
a
c
y
a
nd
s
ta
bi
li
ty
in
dyna
m
ic
a
nd
s
om
e
ti
m
e
s
c
ha
ll
e
ngi
ng
r
e
a
l
-
w
or
ld
c
ondi
ti
ons
.
2.
M
E
T
H
O
D
2.1
.
P
r
op
os
e
d
f
r
a
m
e
w
or
k
ar
c
h
it
e
c
t
u
r
e
T
he
s
ugg
e
s
te
d
M
F
T
s
y
s
te
m
in
c
lu
de
s
f
our
pr
im
a
r
y
pr
oc
e
s
s
in
g
s
ta
g
e
s
:
f
a
c
e
d
e
te
c
ti
on
a
nd
s
e
gm
e
nt
a
ti
on,
de
e
p
c
oupl
e
d
R
e
s
N
e
t
-
ba
s
e
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
f
uz
z
y
ge
ne
ti
c
opt
im
iz
a
ti
on,
a
nd
pr
oba
bi
li
s
ti
c
da
ta
a
s
s
oc
ia
ti
on.
T
h
e
s
ta
g
e
s
,
il
lu
s
tr
a
te
d
in
F
ig
ur
e
1,
pr
oc
e
s
s
in
c
om
in
g
vi
de
o
f
r
a
m
e
s
th
r
ough
th
e
va
r
io
us
c
om
pone
nt
s
, pr
oduc
in
g s
ta
bl
e
i
de
nt
it
y t
r
a
c
ki
ng output
s
.
F
a
c
e
de
te
c
ti
on oc
c
ur
s
f
ir
s
t
th
r
ough
H
S
V
-
ba
s
e
d s
ki
n
c
ol
or
c
la
s
s
if
ic
a
ti
on
a
nd
th
e
n
oc
c
lu
s
io
n
-
a
w
a
r
e
r
e
gi
on
s
e
gm
e
nt
a
ti
on
us
in
g
M
a
r
kov
r
a
ndom
f
ie
ld
s
(
M
R
F
)
. T
he
s
e
gm
e
nt
e
d
r
e
gi
ons
r
e
pr
e
s
e
nt
in
g
th
e
de
t
e
c
te
d
s
ubj
e
c
ts
’
f
a
c
e
s
a
r
e
e
nc
ode
d
u
s
in
g
a
de
e
p
c
oupl
e
d
R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
t
une
d f
or
l
ow
-
r
e
s
ol
ut
io
n s
ur
ve
il
la
nc
e
vi
de
o i
m
a
ge
r
y.
F
ig
ur
e
1. P
r
opos
e
d
M
F
T
f
r
a
m
e
w
or
k a
r
c
hi
te
c
tu
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
R
obus
t
m
ul
ti
-
fa
c
e
s
r
e
c
ogni
ti
on and tr
ac
k
in
g v
ia
f
uz
z
y
ge
ne
ti
c
al
gor
it
hm
s
and
…
(
A
di
l
A
bdul
hur
A
bus
hana
)
211
2.2
.
D
e
e
p
c
ou
p
le
d
R
e
s
N
e
t
f
or
lo
w
-
r
e
s
ol
u
t
io
n
f
ac
e
r
e
c
ogn
it
io
n
T
he
de
e
p
c
oupl
e
d
R
e
s
N
e
t
a
ddr
e
s
s
e
s
lo
w
-
r
e
s
ol
ut
io
n
f
a
c
e
r
e
c
ogni
ti
on
by
jo
in
tl
y
le
a
r
ni
ng
f
e
a
tu
r
e
m
a
ppi
ngs
be
twe
e
n
lo
w
-
a
nd
hi
gh
-
r
e
s
ol
ut
io
n
dom
a
i
ns
th
r
ough
two
c
oupl
e
d
ne
twor
k
br
a
nc
he
s
.
T
he
lo
w
-
r
e
s
ol
ut
io
n
pa
th
pr
oc
e
s
s
e
s
th
e
de
gr
a
de
d
in
put
,
w
hi
le
th
e
hi
gh
-
r
e
s
ol
ut
io
n
br
a
nc
h
p
r
ovi
de
s
s
upe
r
vi
s
or
y
gui
da
nc
e
f
or
f
e
a
tu
r
e
a
li
gnm
e
nt
.
T
hi
s
c
oupl
e
d
r
e
pr
e
s
e
nt
a
ti
on
im
pr
ove
s
di
s
c
r
im
in
a
bi
li
ty
unde
r
qua
li
ty
lo
s
s
by
le
ve
r
a
gi
ng
s
ta
ti
s
ti
c
a
l
s
ki
n
-
c
ol
or
m
ode
li
ng,
w
he
r
e
hom
oge
ne
ous
r
e
gi
ons
a
r
e
r
e
pr
e
s
e
nt
e
d
u
s
in
g
a
3D
G
a
us
s
ia
n
P
D
F
i
n R
G
B
/HS
V
s
pa
c
e
a
s
i
n (
1)
.
(
)
=
1
(
2
)
3
/
2
|
|
1
/
2
(
1
2
(
1
−
)
−
1
(
1
−
)
)
(
1)
H
e
r
e
,
I
de
not
e
t
he
c
ol
or
ve
c
to
r
, μ
t
he
m
e
a
n, a
nd
th
e
c
ova
r
ia
nc
e
m
a
tr
ix
, a
nd bina
r
y s
ki
n m
a
s
ks
a
r
e
ge
ne
r
a
te
d
vi
a
a
da
pt
iv
e
th
r
e
s
hol
di
ng
of
th
e
G
a
us
s
ia
n
P
D
F
.
F
ig
ur
e
2
s
how
s
th
e
de
e
p
c
oupl
e
d
R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
,
c
ons
is
ti
ng
of
a
tr
unk
ne
twor
k
f
or
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
(
w
it
h
c
onvolut
io
na
l
la
ye
r
s
of
32
-
512
c
ha
nne
ls
)
a
nd
br
a
nc
h
ne
twor
ks
th
a
t
c
oupl
e
lo
w
-
a
nd
hi
gh
-
r
e
s
ol
ut
io
n
f
e
a
t
ur
e
s
pa
c
e
s
.
T
he
m
ode
l
in
te
gr
a
te
s
s
of
tm
a
x,
c
e
nt
e
r
,
a
nd
c
r
o
s
s
-
m
oda
li
ty
(
CM
)
lo
s
s
e
s
to
jo
in
tl
y
e
nf
or
c
e
c
la
s
s
s
e
pa
r
a
bi
li
ty
a
nd
c
r
os
s
-
r
e
s
ol
ut
io
n
c
ons
is
t
e
nc
y.
R
e
s
id
ua
l
c
onne
c
ti
ons
(
1
,
2
,
3
,
5
)
s
uppor
t
gr
a
di
e
nt
s
ta
bi
li
ty
a
nd
hi
e
r
a
r
c
hi
c
a
l
f
e
a
tu
r
e
r
e
us
e
.
T
hi
s
c
oupl
e
d
de
s
ig
n
im
pr
ove
s
r
e
c
ogni
ti
on
a
c
c
ur
a
c
y
f
or
lo
w
-
r
e
s
ol
ut
io
n
s
ur
ve
il
la
nc
e
im
a
ge
r
y
c
om
pa
r
e
d
to
s
in
gl
e
-
pa
th
m
ode
ls
.
F
ig
ur
e
2. I
ll
us
tr
a
ti
on of
a
de
e
p
c
oupl
e
d R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
2.3
.
F
u
z
z
y
ge
n
e
t
ic
al
gor
it
h
m
op
t
im
iz
at
io
n
T
he
f
uz
z
y
ge
ne
ti
c
opt
im
iz
a
ti
on
m
odul
e
a
ddr
e
s
s
e
s
lo
c
a
l
m
in
im
a
a
nd
unc
e
r
ta
in
ty
in
da
ta
a
s
s
oc
ia
ti
on
by
in
c
or
por
a
ti
ng
th
e
M
a
m
d
a
ni
f
uz
z
y
in
f
e
r
e
nc
e
f
r
a
m
e
w
or
k
of
L
i
a
nd
Z
ha
n
[
22]
,
w
hi
c
h
e
s
ti
m
a
te
s
a
s
s
o
c
ia
ti
on
c
onf
id
e
nc
e
us
in
g
m
ot
io
n
a
nd
a
ppe
a
r
a
nc
e
a
f
f
in
it
ie
s
.
W
e
a
dopt
th
e
s
a
m
e
f
uz
z
y
s
tr
uc
tu
r
e
but
r
e
la
be
l
th
e
f
iv
e
in
put
s
e
ts
to
r
e
f
le
c
t
s
e
m
a
nt
ic
c
onf
id
e
nc
e
le
ve
ls
,
e
.g.
z
e
r
o
c
o
nf
id
e
nc
e
(
Z
C
)
a
nd
lo
w
c
onf
id
e
nc
e
(
L
C
)
.
O
ur
c
ont
r
ib
ut
io
n
li
e
s
in
in
te
gr
a
ti
ng
th
e
s
e
f
uz
z
y
out
put
s
in
to
a
G
A
-
ba
s
e
d
opt
im
iz
e
r
to
im
pr
ove
a
s
s
ig
nm
e
nt
s
ta
bi
li
ty
.
T
he
G
A
us
e
s
e
li
ti
s
m
w
it
h
r
ou
le
tt
e
-
w
he
e
l
s
e
le
c
ti
on,
a
c
r
os
s
ove
r
pr
oba
bi
li
ty
of
pc
=
0.5,
a
nd
a
m
ut
a
ti
on
pr
oba
bi
li
ty
o
f
pm
=
0.1,
w
it
h
B
ha
tt
a
c
ha
r
yya
di
s
ta
nc
e
us
e
d
a
s
th
e
f
it
ne
s
s
m
e
tr
ic
.
A
ppe
a
r
a
nc
e
a
f
f
in
it
y
is
c
om
put
e
d
u
s
in
g
c
ol
or
hi
s
to
gr
a
m
s
in
R
G
B
/HS
V
s
pa
c
e
c
om
bi
ne
d
w
it
h
lo
c
a
l
bi
n
a
r
y
pa
tt
e
r
n
(
L
B
P
)
te
xt
ur
e
f
e
a
tu
r
e
s
to
im
pr
ove
di
s
c
r
im
in
a
ti
on
unde
r
oc
c
lu
s
io
n
a
nd
il
lu
m
in
a
ti
on
c
ha
nge
s
.
U
nl
ik
e
e
a
r
li
e
r
m
e
th
ods
th
a
t
r
e
li
e
d
s
ol
e
ly
on
ha
ndc
r
a
f
te
d
hi
s
to
gr
a
m
s
,
th
is
hybr
id
r
e
pr
e
s
e
nt
a
ti
on
s
tr
e
ngt
he
ns
a
s
s
oc
ia
ti
on
r
e
li
a
bi
li
ty
in
c
lu
tt
e
r
e
d
e
nvi
r
onm
e
nt
s
by
jo
in
tl
y
e
xpl
oi
ti
ng
c
ol
or
a
nd
te
xt
ur
e
c
ue
s
.
A
s
a
r
e
s
ul
t,
th
e
f
r
a
m
e
w
or
k
m
a
in
ta
in
s
r
obus
t
tr
a
c
kl
e
t
c
ont
in
ui
ty
di
r
e
c
tl
y
f
r
om
r
a
w
vi
de
o
w
it
hout
r
e
qui
r
in
g
pr
e
-
f
il
te
r
e
d
de
te
c
ti
ons
or
m
a
nua
l
f
a
ls
e
-
pos
it
iv
e
r
e
m
ova
l
.
2.3.1
.
M
e
m
b
e
r
s
h
ip
f
u
n
c
t
io
n
s
T
he
r
ul
e
s
of
th
e
f
uz
z
y
in
f
e
r
e
nc
e
s
ys
te
m
f
or
f
uz
z
y
w
e
ig
ht
.
I
n
th
is
pa
pe
r
,
two
in
put
s
a
nd
one
out
put
a
r
e
r
e
pr
e
s
e
nt
e
d a
s
s
how
n i
n F
ig
ur
e
3.
T
w
o
in
put
va
r
ia
bl
e
s
:
i)
=
(
,
)
,
=
(
,
)
,
=
(
,
)
m
e
a
ns
m
e
m
b
e
r
s
hi
p
de
not
e
th
e
m
ot
io
n,
s
h
a
pe
,
a
nd
a
ppe
a
r
a
nc
e
a
f
f
in
it
ie
s
be
twe
e
n
obj
e
c
t
f
a
c
e
s
i
a
nd
obs
e
r
va
ti
on
j
,
r
e
s
pe
c
ti
ve
ly
m
e
a
ns
m
e
m
be
r
s
hi
p
de
not
e
th
e
m
ot
io
n, s
ha
pe
, a
nd a
ppe
a
r
a
nc
e
a
f
f
in
it
ie
s
be
twe
e
n obje
c
t
f
a
c
e
s
i
a
nd obs
e
r
va
ti
on
j
, r
e
s
p
e
c
ti
ve
ly
.
ii)
̂
.
̂
,
̂
m
e
a
ns
non
-
m
e
m
be
r
s
hi
p.
T
he
s
ha
pe
a
f
f
in
it
y
(
i
,
j
)
.
B
e
twe
e
n
obj
e
c
t
i
a
nd
obs
e
r
va
ti
on
is
de
f
in
e
d
a
s
(
2)
.
(
−
(
ℎ
−
ℎ
)
2
2
2
+
(
−
)
2
2
2
)
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
209
-
218
212
W
he
r
e
ℎ
a
nd
ℎ
de
not
e
th
e
he
ig
ht
s
of
obj
e
c
t
i
a
nd
obs
e
r
va
ti
on
,
r
e
s
pe
c
ti
ve
ly
,
a
nd
de
not
e
th
e
w
id
th
s
of
obj
e
c
t
i
a
nd
obs
e
r
va
ti
on
,
r
e
s
pe
c
ti
ve
ly
,
2
,
2
de
not
e
th
e
v
a
r
ia
nc
e
f
or
th
e
he
ig
ht
a
nd
w
id
th
,
r
e
s
pe
c
ti
ve
ly
.
T
he
a
f
f
in
it
y
be
twe
e
n
th
e
pr
e
di
c
te
d
s
t
a
te
of
f
a
c
e
i
a
nd
obs
e
r
va
ti
on
j
is
nor
m
a
li
z
e
d
to
a
va
lu
e
be
twe
e
n
0
a
nd
1.
T
he
s
e
nor
m
a
li
z
e
d
va
lu
e
s
a
r
e
th
e
n
m
a
ppe
d
to
c
or
r
e
s
ponding
f
uz
z
y
s
e
ts
w
it
hi
n
th
e
f
uz
z
y
in
f
e
r
e
nc
e
s
ys
te
m
.
I
n
ge
ne
r
a
l,
in
c
r
e
a
s
in
g
th
e
num
be
r
of
f
uz
z
y
s
e
ts
c
a
n
le
a
d
to
hi
ghe
r
a
c
c
ur
a
c
y,
th
ough
it
a
ls
o
r
a
is
e
s
c
om
put
a
ti
ona
l
c
om
pl
e
xi
ty
.
T
h
e
r
e
f
or
e
,
th
e
num
be
r
of
f
uz
z
y
s
e
ts
is
of
te
n
de
te
r
m
in
e
d e
m
pi
r
ic
a
ll
y ba
s
e
d on the
t
r
a
de
-
of
f
be
twe
e
n pr
e
c
is
io
n a
nd e
f
f
ic
ie
nc
y.
F
ig
ur
e
3. C
onf
id
e
nc
e
m
e
m
be
r
s
hi
p f
unc
ti
on,
a
da
pt
e
d f
r
om
L
i
a
nd Z
ha
n
[
22]
I
n
th
is
pa
pe
r
,
w
e
c
hoos
e
f
iv
e
f
uz
z
y
s
e
ts
to
de
s
c
r
ib
e
a
f
f
in
it
y
i
n
th
e
f
uz
z
y
in
f
e
r
e
nc
e
s
ys
te
m
,
w
he
r
e
e
a
c
h
e
xpl
ic
it
in
put
da
ta
(
=
(
i
,
j
)
,
=
(
i
,
j
)
,
=
(
i
,
j
)
)
is
c
a
te
gor
iz
e
d
in
to
Z
C
,
L
C
,
m
e
d
iu
m
c
onf
id
e
nc
e
(
M
C
)
, hi
gh c
onf
id
e
nc
e
(
H
C
)
, a
nd ve
r
y hi
gh c
onf
id
e
nc
e
(
V
H
C
)
;
f
e
a
tu
r
e
a
f
f
in
it
y va
lu
e
s
l
e
s
s
t
ha
n or
e
qua
l
to
0.1
in
di
c
a
te
unr
e
li
a
bl
e
f
a
c
e
f
e
a
tu
r
e
s
,
w
hi
le
va
lu
e
s
gr
e
a
te
r
th
a
n
or
e
qua
l
to
0.9
s
ig
ni
f
y
ve
r
y
r
e
li
a
bl
e
f
e
a
tu
r
e
s
.
C
ons
e
que
nt
ly
,
e
a
c
h
f
uz
z
y
r
ul
e
in
T
a
bl
e
s
1
to
3
ut
il
iz
e
s
c
onf
id
e
nc
e
le
ve
ls
f
r
om
a
ppe
a
r
a
nc
e
,
m
ot
io
n,
a
nd
s
ha
p
e
to
m
a
na
g
e
obj
e
c
t
m
e
r
gi
ng,
s
pl
it
ti
ng,
a
nd
oc
c
lu
s
io
n
h
a
ndl
in
g
e
f
f
e
c
ti
ve
ly
.
A
dr
op
in
m
ot
io
n
a
f
f
in
it
y
unde
r
th
r
e
s
hol
d
α
r
e
duc
e
s
th
e
in
f
lu
e
nc
e
of
a
pp
e
a
r
a
nc
e
a
f
f
in
it
y,
m
it
ig
a
ti
ng
f
a
ls
e
obs
e
r
va
ti
ons
,
a
nd
c
a
us
e
s
a
ll
a
ppe
a
r
a
nc
e
w
e
ig
ht
s
.
be
in
g s
e
t
to
V
H
C
.
T
a
bl
e
1. F
uz
z
y
r
ul
e
s
b
a
s
e
w
e
ig
ht
M
VHC
HC
MC
LC
ZC
VHC
HC
HC
VHC
VHC
ZC
̂
M
VHC
HC
HC
HC
VHC
LC
UK
VHC
MC
HC
VHC
MC
UK
VHC
MC
MC
VHC
HC
T
a
bl
e
2. F
uz
z
y
r
ul
e
s
b
a
s
e
w
e
ig
ht
μ
ij
S
W
M
K
VHC
HC
MC
LC
ZC
LC
LC
ZC
ZC
ZC
ZC
μ
̂
ij
S
MC
LC
LC
ZC
ZC
LC
HC
MC
MC
LC
ZC
MC
VHC
HC
MC
MC
MC
HC
UK
HC
HC
HC
HC
VHC
T
a
bl
e
3. F
uz
z
y
r
ul
e
s
b
a
s
e
w
e
ig
ht
A
VHC
HC
MC
LC
ZC
MC
MC
LC
LC
ZC
ZC
̂
A
HC
HC
MC
MC
LC
LC
VHC
HC
MC
MC
LC
MC
UK
VHC
HC
MC
HC
HC
UK
VHC
HC
HC
VHC
VHC
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
R
obus
t
m
ul
ti
-
fa
c
e
s
r
e
c
ogni
ti
on and tr
ac
k
in
g v
ia
f
uz
z
y
ge
ne
ti
c
al
gor
it
hm
s
and
…
(
A
di
l
A
bdul
hur
A
bus
hana
)
213
T
he
li
ngui
s
ti
c
va
r
ia
bl
e
s
ha
ve
be
e
n
r
e
la
be
le
d
to
r
e
pr
e
s
e
nt
c
onf
id
e
nc
e
le
ve
ls
in
our
f
r
a
m
e
w
or
k.
W
he
n
pr
e
di
c
ti
ons
a
r
e
a
c
c
ur
a
te
,
th
e
m
ot
io
n
a
f
f
in
it
y
f
or
e
a
c
h
f
a
c
e
g
a
in
s
im
por
ta
nc
e
,
a
nd
th
e
w
e
ig
ht
of
th
e
a
pp
e
a
r
a
nc
e
a
f
f
in
it
y
s
houl
d
in
c
r
e
a
s
e
a
s
̂
.
R
is
e
s
,
w
it
h
th
e
f
uz
z
y
r
ul
e
s
in
th
e
f
our
th
c
ol
um
n
a
dj
us
te
d
to
L
C
,
V
H
C
,
a
nd
unknown
(
U
K
)
,
r
e
s
pe
c
ti
ve
ly
.
T
he
f
ir
s
t
a
nd
s
e
c
ond
f
uz
z
y
r
ul
e
s
in
th
e
f
if
th
c
ol
um
n
a
ddr
e
s
s
th
e
c
ha
ll
e
nge
s
pos
e
d
by
oc
c
lu
de
d
f
a
c
e
s
or
c
lu
tt
e
r
e
d
e
nvi
r
onm
e
nt
s
,
w
he
r
e
di
s
ti
ngui
s
hi
ng
di
f
f
e
r
e
nc
e
s
in
th
e
ir
a
ppe
a
r
a
nc
e
s
be
c
om
e
s
di
f
f
ic
ul
t;
t
hus
, t
he
w
e
ig
ht
s
.
th
e
y a
r
e
s
e
t
to
V
H
C
,
w
hi
le
ot
he
r
r
ul
e
s
a
r
e
de
s
ig
na
te
d a
s
U
K
.
A
ddi
ti
ona
ll
y,
th
e
f
uz
z
y
r
ul
e
s
in
th
e
s
e
c
ond
a
nd
th
ir
d
c
ol
um
n
s
ty
pi
c
a
ll
y
m
a
na
ge
s
c
e
n
a
r
io
s
w
he
r
e
pr
e
di
c
ti
on
pos
it
io
ns
f
or
m
ul
ti
pl
e
f
a
c
e
s
la
c
k
a
c
c
ur
a
c
y.
A
s
i
nc
r
e
a
s
e
s
th
e
a
ppe
a
r
a
n
c
e
a
f
f
in
it
y,
in
im
por
ta
nc
e
,
a
dj
us
ti
ng.
to
L
C
,
M
C
,
a
nd
Z
C
,
r
e
s
pe
c
ti
ve
ly
.
I
n
T
a
bl
e
s
2
a
nd
3,
th
e
f
our
th
a
nd
f
if
th
f
uz
z
y
r
ul
e
s
in
th
e
f
ir
s
t
c
ol
um
n
a
ddr
e
s
s
oc
c
lu
s
io
ns
,
e
m
ph
a
s
iz
in
g
a
ppe
a
r
a
nc
e
a
f
f
in
it
y
w
he
n
obj
e
c
t
pos
it
io
n
s
a
r
e
c
lo
s
e
to
obs
e
r
va
ti
ons
,
w
it
h
w
e
ig
ht
s
a
nd
s
e
t
to
H
C
a
nd
V
H
C
,
r
e
s
pe
c
ti
ve
ly
,
w
hi
le
th
e
a
ppe
a
r
a
nc
e
a
f
f
in
it
y
in
c
r
e
a
s
e
s
a
s
,
.
A
n
upw
a
r
d
tr
e
nd
in
tr
ue
pos
it
iv
e
s
(
T
P
)
a
nd
tr
ue
ne
g
a
ti
ve
s
(
T
N
)
,
a
lo
ng
w
it
h
r
e
duc
e
d
f
a
ls
e
pos
it
iv
e
s
(
F
P
)
a
nd
f
a
ls
e
ne
ga
ti
ve
s
(
F
N
)
,
r
e
f
le
c
ts
im
pr
ove
d
id
e
nt
if
ic
a
ti
on r
e
li
a
bi
li
ty
ove
r
ti
m
e
.
A
s
s
how
n
in
T
a
bl
e
1,
th
e
pr
opos
e
d
m
e
th
od
out
pe
r
f
or
m
s
a
lt
e
r
na
ti
ng
di
r
e
c
ti
on
m
e
th
od
o
f
m
ul
ti
pl
ie
r
s
(
A
D
M
M
)
a
c
r
os
s
r
e
c
a
ll
,
pr
e
c
is
io
n
,
F1
-
s
c
or
e
,
m
ul
ti
pl
e
obj
e
c
ts
tr
a
c
ki
ng
a
c
c
ur
a
c
y
(
M
O
T
A
)
,
a
nd
m
ul
ti
pl
e
obj
e
c
ts
tr
a
c
ki
ng
pr
e
c
i
s
io
n
(
M
O
T
P
)
.
T
he
M
O
T
A
-
ba
s
e
d
e
va
lu
a
ti
o
n
,
w
hi
c
h
in
c
or
por
a
te
s
m
o
s
t
tr
a
c
ke
d
(
M
T
)
,
m
os
t
lo
s
t
(
M
L
)
,
f
r
a
gm
e
nt
a
ti
on
(
F
G
)
,
a
nd
bounding
-
box ove
r
la
p vi
a
M
O
T
P
, c
onf
ir
m
s
s
upe
r
io
r
t
r
a
c
ki
ng e
f
f
e
c
ti
ve
ne
s
s
f
or
m
ul
ti
pl
e
f
a
c
e
s
a
s
i
n (
3)
.
M
O
T
P
=
∑
∑
(
3
)
W
he
r
e
r
e
pr
e
s
e
nt
s
t
he
t
ot
a
l
num
be
r
of
a
s
s
oc
ia
te
d obj
e
c
ts
a
t
th
e
t
i
m
e
. T
he
s
e
e
qua
ti
ons
,
a
s
de
f
in
e
d by M
O
T
A
a
nd M
O
T
P
, pr
ovi
de
m
a
th
e
m
a
ti
c
s
f
or
e
va
lu
a
t
in
g t
r
a
c
ki
ng pe
r
f
or
m
a
nc
e
.
2.3.2
.
F
u
z
z
y s
ys
t
e
m
(
in
p
u
t
s
, ou
t
p
u
t
s
,
an
d
r
u
le
s
)
L
in
gui
s
ti
c
va
r
ia
bl
e
s
a
nd me
m
be
r
s
hi
p f
unc
ti
ons
i
nc
lu
de
:
i)
I
nput
s
(
pe
r
a
s
s
oc
ia
ti
on hypothes
i
s
be
twe
e
n t
r
a
c
k
a
nd de
te
c
ti
on
:
–
O
c
c
=
oc
c
lu
s
io
n l
e
ve
l
∈
[
0
,
1
]
M
F
:
{
lo
w
, m
e
d, hi
gh
}
vi
a
t
r
ia
ngul
a
r
/t
r
a
pe
z
oi
da
l
s
e
ts
lo
w
:
[
0
,
0
,
0
.
3
]
,
m
e
d
:
[
0
.
2
,
0
.
5
,
0
.
8
]
,
hi
gh
:
[
0
.
6
,
1
,
1
]
.
–
S
im
=
a
ppe
a
r
a
nc
e
s
im
il
a
r
it
y (
de
e
p
-
c
oupl
e
d c
os
in
e
)
∈
[
0, 1]
M
F
:
{
lo
w
, m
e
d, hi
gh
}
→
lo
w
:
[
0
,
0
,
0
.
4
]
,
m
e
d
:
[
0
.
3
,
0
.
6
,
0
.
8
]
,
hi
gh
:
[
0
.
7
,
1
,
1
]
.
–
C
onf
=
de
te
c
to
r
c
onf
id
e
nc
e
∈
[
0
,
1
]
M
F
:
{
lo
w
, m
e
d, hi
gh
}
→
lo
w
:
[
0
,
0
,
0
.
4
]
,
m
e
d
:
[
0
.
3
,
0
.
6
,
0
.
85
]
,
hi
gh
:
[
0
.
75
,
1
,
1
]
.
–
M
ot
=
m
ot
io
n c
ons
is
t
e
nc
y (
M
a
ha
l
a
nobi
s
/KF
r
e
s
id
ua
l
nor
m
a
li
z
e
d)
∈
[
0
,
1
]
, hi
ghe
r
i
s
be
tt
e
r
M
F
:
{
poor
, f
a
ir
, good
}
→
poor
:
[
0
,
0
,
0
.
4
]
,
f
a
ir
:
[
0
.
3
,
0
.
6
,
0
.
85
]
,
good
:
[
0
.
75
,
1
,
1
]
.
ii)
O
ut
put
s
(
de
f
uz
z
if
ie
d by c
e
nt
r
oi
d)
:
–
_a
pp
∈
[
0
,
1
]
—
w
e
ig
ht
f
or
a
ppe
a
r
a
nc
e
t
e
r
m
i
n a
s
s
oc
ia
ti
on
.
–
_m
ot
∈
[
0
,
1
]
—
w
e
ig
ht
f
or
m
ot
io
n t
e
r
m
(
e
nf
or
c
e
α_a
pp +
α_mot
=
1 a
f
te
r
de
f
uz
z
)
.
–
_a
s
s
oc
∈
[
0
,
1
]
—
a
da
pt
iv
e
a
s
s
oc
ia
ti
on t
hr
e
s
hol
d
.
–
G
A
hype
r
pa
r
a
m
e
te
r
s
pe
r
ge
ne
r
a
ti
on:
_
∈
[
0
.
01
,
0
.
3
]
,
_
∈
[
0
.
6
,
0
.
95
]
,
_
∈
[
1
.
2
,
2
.
0
]
.
2.3.3
.
S
ix
c
or
e
f
u
z
z
y r
u
le
s
U
s
e
M
a
m
d
a
ni
r
ul
e
s
;
a
c
om
pa
c
t,
hi
gh
-
im
pa
c
t
s
ubs
e
t:
−
R
1:
ℎ
ℎ
ℎ
,
,
ℎ
,
,
ℎ
.
−
R
2:
ℎ
,
,
,
↑
,
ℎ
(
)
.
−
R
3:
,
ℎ
,
−
ℎ
.
−
R
4:
,
,
,
ℎ
(
)
.
−
R
5:
ℎ
ℎ
,
,
−
ℎ
.
−
R
6:
ℎ
,
−
ℎ
,
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
209
-
218
214
2.3.4
.
C
h
r
om
os
o
m
e
e
n
c
od
in
g
E
a
c
h c
hr
om
os
om
e
e
nc
ode
s
pe
r
-
f
r
a
m
e
a
s
s
oc
ia
ti
on a
nd glob
a
l
knobs
:
i)
G
lo
ba
l
ge
ne
s
:
−
∈
[
0
,
1
]
:
pr
io
r
w
e
ig
ht
on a
ppe
a
r
a
nc
e
(
be
f
or
e
f
uz
z
y a
dj
us
tm
e
nt
)
.
−
=
1
−
.
−
∈
[
0
,
1
]
:
ba
s
e
a
s
s
oc
ia
ti
on t
hr
e
s
hol
d
.
−
∈
[
0
,
1
]
:
I
oU
ga
ti
ng w
e
ig
ht
.
−
ℎ
∈
[
0
,
1
]
:
tr
a
je
c
to
r
y s
m
oot
hne
s
s
r
e
gul
a
r
iz
a
ti
on
.
−
∈
[
0
,
1
]
:
C
J
L
F
bl
e
ndi
ng w
it
h P
D
A
F
.
ii)
P
e
r
-
tr
a
c
k ge
ne
s
(
opt
io
na
l
c
om
pa
c
t
f
or
m
us
in
g s
ha
r
e
d pa
r
a
m
s
by
c
lu
s
te
r
s
)
:
−
∈
[
0
.
1
,
0
.
9
]
:
ga
ti
ng r
a
di
us
s
c
a
li
ng
.
−
∈
{
1
,
…
,
}
:
m
e
m
or
y l
e
ngt
h f
or
f
e
a
tu
r
e
ga
ll
e
r
y
.
iii)
A
s
s
oc
ia
ti
on
g
e
ne
s
(
f
or
to
p
-
k
c
a
ndi
da
te
p
a
ir
s
pe
r
f
r
a
m
e
)
:
bi
na
r
y
ve
c
to
r
A
w
it
h
{
,
}
∈
{
0
,
1
}
unde
r
1
-
to
-
1
c
ons
tr
a
in
ts
(
H
unga
r
ia
n
-
c
om
pa
ti
bl
e
)
.
I
n
pr
a
c
ti
c
e
,
G
A
s
e
a
r
c
he
s
ove
r
th
r
e
s
hol
ds
/we
ig
ht
s
;
th
e
f
in
a
l
1
-
to
-
1 i
s
pr
oduc
e
d by Hunga
r
ia
n on the
G
A
-
w
e
ig
ht
e
d c
os
t
m
a
tr
ix
.
2.3.5
.
F
it
n
e
s
s
f
u
n
c
t
io
n
(
s
in
gl
e
-
ob
j
e
c
t
iv
e
, f
as
t
M
O
T
A
p
r
oxy)
F
or
a
va
li
da
ti
on c
hunk (
e
.g., 200
-
500 f
r
a
m
e
s
)
, c
om
put
e
:
−
A
ppe
a
r
a
nc
e
c
o
s
t:
=
1−
S
im
(
c
os
in
e
)
.
−
M
ot
io
n c
os
t:
=
nor
m
a
li
z
e
d K
F
/J
P
D
A
r
e
s
id
ua
l
.
−
I
oU
pe
na
lt
y:
=
1
–
I
oU
.
P
e
r
hypothe
s
is
c
os
t
a
s
i
n (
4)
.
=
∗
+
∗
+
∗
(
4
)
A
f
te
r
a
s
s
ig
nm
e
nt
(
H
unga
r
ia
n)
, a
c
c
um
ul
a
te
:
−
F
N
, F
P
, I
D
S
, F
r
a
g (
onl
in
e
e
s
ti
m
a
te
s
)
.
−
S
m
oot
hne
s
s
:
=
|
−
{
−
1
}
|
ove
r
t
r
a
c
ks
(
v =
ve
lo
c
it
y)
.
−
R
unt
im
e
pr
oxy:
R
=
#ops
/f
r
a
m
e
(
e
s
ti
m
a
te
d f
r
om
a
c
ti
ve
t
r
a
c
ks
,
ga
ll
e
r
y s
iz
e
)
.
F
it
ne
s
s
t
o m
a
xi
m
iz
e
(
c
onve
r
t
to
m
in
im
iz
a
ti
on a
s
ne
e
d
e
d)
a
s
i
n (
5)
.
=
1
×
(
1
−
)
+
2
×
(
1
−
)
+
3
×
(
1
−
)
+
4
×
(
1
−
)
+
5
×
−
6
×
−
7
×
(
5
)
T
ypi
c
a
l
w
e
ig
ht
s
a
s
i
n (
6)
.
1
=
0
.
25
,
2
=
0
.
15
,
3
=
0
.
2
,
4
=
0
.
15
,
5
=
0
.
15
,
6
=
0
.
05
,
7
=
0
.
05
(
6
)
2.4
.
P
r
ob
ab
il
is
t
ic
d
at
a as
s
oc
ia
t
io
n
f
il
t
e
r
(
P
D
A
F
/J
P
D
A
F
)
T
he
a
lg
or
it
hm
m
e
r
ge
s
bot
h
th
e
P
D
A
F
a
nd
th
e
jo
in
t
pr
oba
bi
li
s
ti
c
da
ta
a
s
s
oc
ia
ti
on
f
il
te
r
(
J
P
D
A
F
)
f
or
m
ul
ti
-
ta
r
ge
t
e
nvi
r
onm
e
nt
s
ta
bl
e
da
ta
a
s
s
oc
ia
ti
on.
P
D
A
F
i
s
a
p
pl
ie
d
f
or
tr
a
c
ki
ng
a
s
in
gl
e
ta
r
ge
t
w
it
h
c
lu
tt
e
r
,
w
hi
le
J
P
D
A
F
is
th
e
m
ul
ti
-
ta
r
ge
t
e
xt
e
ns
io
n
w
it
h
pot
e
nt
ia
l
in
te
r
a
c
ti
ons
be
twe
e
n
ta
r
ge
ts
.
A
s
s
o
c
ia
ti
on
pr
oba
bi
li
ti
e
s
a
r
e
c
om
put
e
d a
s
(
7)
.
(
∣
∣
)
=
∑
∑
(
;
,
)
=
1
(
7)
W
he
r
e
is
t
he
pr
oba
bi
li
ty
t
ha
t
m
e
a
s
ur
e
m
e
nt
i
is
of
t
he
t
a
r
ge
t
obj
e
c
t,
H
is
t
he
obs
e
r
va
ti
on ma
tr
ix
, a
nd
R
is
t
he
m
e
a
s
ur
e
m
e
nt
noi
s
e
c
ov
a
r
ia
nc
e
.
J
oi
nt
li
ke
li
hoods
a
r
e
e
s
ti
m
a
te
d
by
J
P
D
A
F
f
or
m
ul
ti
pl
e
ove
r
la
ppi
ng
f
a
c
e
s
to
m
a
in
ta
in
t
r
a
c
k i
nt
e
gr
it
y dur
in
g c
lo
s
e
i
nt
e
r
a
c
ti
on.
2.5
.
C
on
d
it
io
n
al
j
oi
n
t
l
ik
e
li
h
ood
f
il
t
e
r
T
he
C
J
L
F
c
om
pone
nt
im
pr
ove
s
tr
a
c
ki
ng
a
c
c
ur
a
c
y
by
m
ode
li
ng
jo
in
t
obj
e
c
ts
a
nd
s
ta
te
s
pr
oba
bi
li
ty
of
c
or
r
e
la
te
d
obj
e
c
ts
w
it
h
c
ons
tr
a
in
ts
on
s
pa
ti
a
l
a
nd
te
m
por
a
l
lo
c
a
ti
ons
.
T
he
f
il
te
r
a
ddr
e
s
s
e
s
th
e
oc
c
lu
s
io
n
a
nd
c
lu
tt
e
r
pr
obl
e
m
s
us
in
g
de
pt
h
or
de
r
in
g
a
nd
vi
s
ib
il
it
y
-
c
ons
tr
a
in
e
d
tr
a
c
k
li
ke
li
hood
upda
te
.
T
r
a
c
k
e
r
pa
r
a
m
e
te
r
s
a
r
e
upda
te
d
upon
oc
c
lu
s
io
n
by
m
odi
f
yi
ng
th
e
s
ys
t
e
m
us
in
g
gr
a
di
e
nt
a
s
c
e
nt
opt
im
iz
a
ti
on
w
it
h
de
r
iv
a
ti
ve
-
f
r
e
e
P
ow
e
ll
'
s
opt
im
iz
a
ti
on me
th
od.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
R
obus
t
m
ul
ti
-
fa
c
e
s
r
e
c
ogni
ti
on and tr
ac
k
in
g v
ia
f
uz
z
y
ge
ne
ti
c
al
gor
it
hm
s
and
…
(
A
di
l
A
bdul
hur
A
bus
hana
)
215
2.6
.
F
e
at
u
r
e
e
xt
r
ac
t
io
n
an
d
s
k
in
d
e
t
e
c
t
io
n
T
he
s
ys
te
m
ut
il
iz
e
d
m
ul
ti
-
m
oda
l
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
ba
s
e
d
on
c
ol
or
,
te
xt
ur
e
,
a
nd
s
ha
pe
f
e
a
tu
r
e
s
.
S
ki
n pi
xe
l
c
la
s
s
if
ic
a
ti
on i
s
done
u
s
in
g a
pr
oba
bi
li
s
ti
c
m
ode
l
a
s
i
n (
8)
.
(
s
kin
∣
,
,
)
=
(
,
,
∣
∣
s
kin
)
⋅
(
sk
i
n
)
(
,
,
)
(
8
)
W
he
r
e
C
,
T
, a
nd
S
de
not
e
c
ol
or
,
t
e
xt
ur
e
, a
nd s
ha
pe
f
e
a
tu
r
e
s
, r
e
s
pe
c
ti
ve
ly
. L
B
P
de
s
c
r
ip
to
r
s
a
r
e
us
e
d t
o c
a
pt
ur
e
te
xt
ur
e
de
ta
il
s
,
a
nd
H
S
V
c
ol
or
hi
s
to
gr
a
m
s
a
r
e
us
e
d
f
or
in
s
e
ns
it
iv
it
y
to
c
ol
or
r
e
pr
e
s
e
nt
a
ti
on
unde
r
c
ha
ngi
ng
il
lu
m
in
a
ti
on c
ondi
ti
ons
.
2.7
.
E
val
u
at
io
n
m
e
t
r
ic
s
P
e
r
f
or
m
a
nc
e
a
s
s
e
s
s
m
e
nt
e
m
pl
oys
s
ta
nda
r
d
m
e
a
s
ur
e
s
f
or
m
ul
ti
-
obj
e
c
t
tr
a
c
ki
ng,
in
c
lu
di
ng
M
O
T
A
,
M
O
T
P
,
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
.
M
O
T
A
c
om
put
e
s
ove
r
a
ll
tr
a
c
ki
ng
p
r
e
c
is
io
n
c
ons
id
e
r
in
g
FP
,
FN
,
a
nd i
de
nt
it
y s
w
it
c
he
s
(
I
D
S
)
.
M
O
T
A
=
1
−
∑
(
FP
+
FN
+
ID
)
∑
(
9
)
W
he
r
e
F
P
, F
N
, a
nd I
D
r
e
pr
e
s
e
nt
f
a
ls
e
pos
it
iv
e
s
, f
a
ls
e
ne
ga
ti
v
e
s
,
a
nd
I
D
S
, r
e
s
pe
c
ti
ve
ly
, a
nd mₜ
i
s
t
he
numb
e
r
of
gr
ound tr
ut
h obje
c
ts
a
t
ti
m
e
t
.
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
E
xpe
r
im
e
nt
s
w
e
r
e
c
onduc
te
d
on
th
e
m
us
ic
vi
de
o
da
ta
s
e
t
in
tr
oduc
e
d
by
Z
ha
ng
e
t
al
.
[
11
]
,
w
hi
c
h
c
ont
a
in
s
20
m
a
nua
ll
y
a
nnot
a
te
d
m
ul
ti
-
f
a
c
e
vi
de
o
s
e
que
nc
e
s
w
it
h
va
r
ia
ti
ons
in
il
lu
m
in
a
ti
on,
oc
c
lu
s
io
n,
a
nd
pos
e
.
A
s
r
e
por
te
d
in
T
a
bl
e
4,
th
e
pr
opos
e
d
f
uz
z
y
g
e
ne
ti
c
d
e
e
p
-
c
oupl
e
d
f
r
a
m
e
w
or
k
a
c
hi
e
ve
s
a
n
a
v
e
r
a
ge
F1
-
s
c
or
e
of
86.1±1.2%
a
nd
M
O
T
A
of
66.5±1.7%
,
out
pe
r
f
or
m
in
g
pr
io
r
m
ode
ls
w
it
h
lo
w
va
r
ia
nc
e
a
c
r
os
s
s
e
que
nc
e
s
,
in
di
c
a
ti
ng
s
tr
ong
ge
n
e
r
a
li
z
a
ti
on.
T
he
s
ys
te
m
a
ls
o
a
c
hi
e
ve
s
31.8
F
P
S
,
de
m
ons
tr
a
ti
ng
r
e
a
l
-
ti
m
e
c
a
pa
bi
li
ty
f
or
s
ur
ve
il
la
nc
e
a
nd e
dge
-
ba
s
e
d de
pl
oym
e
nt
.
T
a
bl
e
4. S
ta
ti
s
ti
c
a
l
pe
r
f
or
m
a
nc
e
of
t
he
pr
opos
e
d
f
uz
z
y g
e
ne
ti
c
de
e
p
-
c
oupl
e
d f
r
a
m
e
w
or
k dur
in
g t
r
a
in
in
g a
nd
te
s
ti
ng pha
s
e
s
P
ha
s
e
P
r
e
c
i
s
i
on
(%)
R
e
c
a
l
l
(%)
F1
-
s
c
or
e
(%)
T
r
a
c
ki
ng
a
c
c
ur
a
c
y
(
M
O
T
A
, %
)
T
r
a
c
ki
ng
pr
e
c
i
s
i
on
(
M
O
T
P
, %
)
I
D
S
F
r
a
gm
e
nt
a
t
i
ons
(F
G
)
R
unt
i
m
e
(
F
P
S
)
T
r
a
i
ni
ng
91.2
±
0.8
82.3
±
1.1
86.6
±
0.9
67.4
±
1.3
81.2
±
0.7
642
±
28
1705
±
45
32.4
±
1.1
T
e
s
t
i
ng
90.5
±
1.3
81.0
±
1.5
86.1
±
1.2
66.5
±
1.7
80.3
±
1.0
669
±
31
1745
±
52
31.8
±
1.2
T
a
bl
e
5
a
d
di
t
io
n
a
l
ly
d
e
m
on
s
tr
a
t
e
s
th
a
t
th
e
s
ug
ge
s
t
e
d
f
u
z
z
y
ge
ne
ti
c
d
e
e
p
-
c
ou
pl
e
d
f
r
a
m
e
w
or
k
out
p
e
r
f
or
m
s
b
ot
h
c
o
nv
e
n
ti
o
na
l
a
nd
hybr
id
t
r
a
c
k
e
r
s
.
O
n
a
v
e
r
a
g
e
,
th
i
s
le
a
d
s
t
o
3
-
5%
h
ig
h
e
r
F
1
-
s
c
or
e
a
n
d
2
-
3%
hi
gh
e
r
m
e
a
n
a
v
e
r
a
g
e
tr
a
c
ki
ng
pr
e
c
i
s
io
n
(
M
O
T
A
)
s
c
o
r
e
s
,
w
h
il
e
m
a
in
ta
in
i
ng
a
r
a
t
e
>
30
F
P
S
.
D
e
e
pS
O
R
T
is
de
p
e
n
d
e
nt
on
ly
o
n
m
ot
i
on
a
n
d
r
e
-
I
D
w
e
ig
ht
i
ng
,
w
hi
c
h
a
r
e
f
ix
e
d.
I
n
c
o
nt
r
a
s
t,
th
e
f
u
z
z
y
-
ge
ne
ti
c
l
a
y
e
r
dyn
a
m
i
c
a
ll
y
c
ha
ng
e
s
a
s
s
o
c
ia
ti
o
n
c
onf
i
de
n
c
e
a
lo
ng
d
yn
a
m
i
c
u
s
e
r
s
'
not
io
n
of
c
onf
i
de
nc
e
(
tr
a
c
k
e
r
s
'
r
a
ti
o
s
of
r
e
s
po
n
s
e
s
)
. T
h
e
f
uz
z
y
g
e
n
e
t
ic
_d
e
e
p
c
o
upl
i
ng'
s
go
a
l
i
s
t
o
r
e
d
u
c
e
I
D
s
w
it
c
h
e
s
a
nd
FG
w
he
n
o
c
c
lu
s
i
on
s
oc
c
ur
. A
s
c
om
pa
r
e
d
to
R
e
ti
na
F
a
c
e
+
K
a
lm
a
n
[
23]
,
t
he
pr
o
po
s
e
d
m
e
t
hod
'
s
d
e
e
p
c
o
upl
e
d
tr
a
c
k
e
r
de
ve
lo
p
s
m
or
e
s
ta
bl
e
tr
a
c
ki
ng
p
e
r
f
or
m
a
nc
e
be
c
a
u
s
e
it
c
ou
pl
e
s
th
e
de
e
p
R
e
s
N
e
t
e
m
be
ddi
ng
s
w
i
th
f
uz
z
y
o
pt
im
iz
a
t
io
n
to
en
s
ur
e
c
on
te
xt
-
a
w
a
r
e
a
s
s
oc
ia
ti
o
n
s
.
T
a
bl
e
6
r
e
ve
a
l
s
th
a
t
th
e
f
u
z
z
y
ge
ne
ti
c
de
e
p
-
c
o
up
le
d
f
r
a
m
e
w
or
k
pr
op
os
e
d
i
n
th
i
s
p
a
p
e
r
ou
tp
e
r
f
or
m
s
r
e
c
e
n
t
M
F
T
t
e
c
h
ni
q
ue
s
w
it
h
t
he
hi
gh
e
s
t
F
1
-
s
c
or
e
(
8
6.1
%
)
,
a
n
d
a
c
om
p
e
t
it
i
ve
M
O
T
A
of
66.
5%
,
e
xc
e
e
di
ng
r
e
s
ul
ts
of
d
e
e
p
m
e
tr
i
c
-
l
e
a
r
ni
ng
ba
s
e
li
ne
s
(
s
i
a
m
e
s
e
,
tr
i
pl
e
t
,
a
n
d
S
ym
T
r
i
pl
e
t)
a
nd
opt
im
i
z
a
ti
o
n
-
b
a
s
e
d
m
o
de
li
n
g
r
e
s
ul
ts
s
u
c
h
a
s
A
D
M
M
[
24]
a
nd
it
e
r
a
ti
ve
H
a
nk
e
l
to
ta
l
le
a
s
t
s
qu
a
r
e
s
(
I
H
T
L
S
)
[
25]
.
T
he
l
ow
f
r
e
q
ue
nc
y
of
f
a
ls
e
a
l
a
r
m
s
(
F
A
F
=
0
.15)
a
l
s
o
in
di
c
a
te
s
r
e
l
ia
bl
e
tr
a
c
k c
on
ti
n
ui
t
ie
s
i
n c
ha
o
ti
c
c
on
di
ti
on
s
.
W
hi
l
e
a
l
s
o bui
lt
o
n
t
he
s
a
m
e
da
t
a
s
e
t,
t
he
pr
ob
a
b
il
i
s
t
ic
i
nt
e
gr
a
t
e
d
tr
a
c
ki
n
g
a
nd
d
e
t
e
c
ti
o
n f
r
a
m
e
w
or
k
(
P
I
T
A
D
F
)
[
26]
r
e
p
or
t
e
d
F
1
-
s
c
or
e
=
85
.3%
a
nd
M
O
T
A
=
69
.2%
.
T
he
pr
op
o
s
e
d
a
p
pr
o
a
c
h
a
c
h
ie
ve
d
a
h
ig
h
e
r
pr
e
c
is
io
n
(
90.
5%
)
w
it
h
s
ig
ni
f
ic
a
n
tl
y
f
e
w
e
r
I
D
S
=
66
9
a
nd
FG
=
1
74
5,
w
hi
c
h
is
a
r
e
s
ul
t
of
th
e
s
t
a
bi
li
ty
pr
o
vi
d
e
d
by t
he
f
uz
z
y r
ul
e
–
gui
d
e
d G
A
o
pt
i
m
i
z
e
r
.
T
h
e
de
e
p
c
ou
pl
e
d
R
e
s
N
e
t
e
m
b
e
d
di
n
gs
a
u
gm
e
nt
th
i
s
be
ha
vi
or
b
y
m
a
i
nt
a
in
i
ng
id
e
n
ti
t
y c
o
he
r
e
nc
y ov
e
r
l
o
w
-
r
e
s
ol
u
ti
o
n
a
nd
o
c
c
lu
de
d
f
a
c
e
s
.
T
h
i
s
i
s
a
ls
o
s
upp
or
t
e
d
b
y
o
ur
M
O
T
P
s
c
or
e
of
80.
3%
,
a
s
s
p
a
ti
a
l
c
on
s
i
s
t
e
n
c
y
i
s
r
e
ta
in
e
d
th
r
o
ugh
P
D
A
F
a
nd
C
J
L
F
,
w
hi
c
h
r
e
du
c
e
d
dr
if
t
i
n
t
he
tr
a
j
e
c
to
r
y.
O
v
e
r
a
ll
,
th
e
s
e
q
ua
nt
i
ta
ti
v
e
m
e
tr
i
c
s
d
e
m
on
s
tr
a
t
e
th
a
t
th
e
pr
opo
s
e
d
f
r
a
m
e
w
or
k
c
on
s
is
te
nt
l
y
pr
o
vi
de
s
h
ig
h
a
c
c
ur
a
c
y
a
nd
t
e
m
por
a
l
c
on
s
is
te
nc
y;
t
he
r
e
f
o
r
e
, i
t
m
a
y
b
e
im
pl
e
m
e
n
te
d i
n r
e
a
l
-
ti
m
e
s
ur
v
e
i
ll
a
n
c
e
-
dr
i
ve
n
a
pp
li
c
a
ti
o
ns
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
209
-
218
216
T
a
bl
e
5. C
om
pa
r
a
ti
ve
e
va
lu
a
ti
on of
t
he
pr
opos
e
d f
r
a
m
e
w
or
k a
g
a
in
s
t
ba
s
e
li
ne
m
od
e
ls
a
c
r
os
s
m
ul
ti
pl
e
da
ta
s
e
ts
M
e
t
hod
D
a
t
a
s
e
t
P
r
e
c
i
s
i
on
(%)
R
e
c
a
l
l
(%)
F1
-
s
c
or
e
(%)
M
O
T
A
(%)
M
O
T
P
(%)
I
D
S
F
r
a
g
R
unt
i
m
e
(
F
P
S
)
D
e
e
pS
O
R
T
[
3]
W
I
D
E
R
-
F
a
c
e
88.1±
1.7
74.6±
2.1
80.7±
1.8
61.5±
2.3
79.2±
1.1
1021
1830
28.4±
0.9
Y
T
F
87.4±
1.9
75.8±
2.0
81.2±
1.7
60.8±
2.4
78.6±
1.3
986
1775
27.9±
1.0
I
J
B
-
S
86.8±
2.2
73.3±
2.4
79.4±
1.9
59.6±
2.6
77.8±
1.4
1125
1921
26.8±
1.1
R
e
t
i
na
F
a
c
e
+K
a
l
m
a
n
[
23
]
W
I
D
E
R
-
F
a
c
e
90.4±
1.3
80.5±
1.6
85.1±
1.4
65.9±
2.0
81.6±
1.0
744
1610
25.6±
1.2
Y
T
F
89.9±
1.4
80.7±
1.5
85.0±
1.3
65.4±
2.1
81.2±
1.1
759
1634
25.0±
1.1
I
J
B
-
S
89.5±
1.5
79.8±
1.7
84.3±
1.5
64.8±
2.2
80.9±
1.1
771
1659
24.5±
1.0
P
r
opos
e
d (
f
uz
z
y
ge
ne
t
i
c
+de
e
p c
oupl
e
d)
W
I
D
E
R
-
F
a
c
e
91.0±
1.1
82.4±
1.3
86.5±
1.2
68.2±
1.8
82.5±
0.9
698
1562
31.4±
1.0
Y
T
F
90.5±
1.2
81.3±
1.5
86.1±
1.2
66.5±
1.7
80.3±
1.0
669
1745
31.8±
1.2
I
J
B
-
S
90.2±
1.3
80.9±
1.4
85.8±
1.2
66.1±
1.9
79.9±
1.1
701
1792
30.9±
1.1
T
a
bl
e
6. Q
ua
nt
it
a
ti
ve
c
om
p
a
r
is
ons
w
it
h t
he
s
t
a
te
-
of
-
th
e
-
a
r
t
tr
a
c
ki
ng me
th
ods
on t
he
vi
de
o da
ta
s
e
t
M
e
t
hod
R
e
c
a
l
l
P
r
e
c
i
s
i
on
F1
-
s
c
or
e
F
A
F
MT
I
D
S
F
G
M
O
T
A
M
O
T
P
A
D
M
M
[
24]
75.5
61.8
68.0
0.50
23
2382
2959
51.7
63.7
I
H
T
L
S
[
25]
75.5
68.0
71.6
0.41
23
2013
2880
56.2
63.7
P
r
e
-
t
r
a
i
ne
d
[
11]
60.1
88.8
71.7
0.17
5
931
2140
51.5
79.5
M
ul
t
i
-
t
a
r
ge
t
l
e
a
r
ni
ng a
nd
de
t
e
c
t
i
on (
m
T
L
D
)
[
11]
69.1
88.1
71.4
0.21
14
1914
2786
57.7
80.1
P
I
T
A
D
F
[
26]
81.7
90.2
85.3
0.27
32
624
1645
69.2
86.0
O
ur
81.0
90.5
86.1
0.15
30
669
1745
66.5
80.30
T
a
bl
e
7
s
how
s
th
a
t
th
e
pr
opos
e
d
f
uz
z
y
ge
ne
ti
c
+
de
e
p
c
oupl
e
d
R
e
s
N
e
t
f
r
a
m
e
w
or
k
is
s
upe
r
io
r
to
th
e
hybr
id
tr
a
c
ke
r
s
of
C
N
N
+
pa
r
ti
c
le
s
w
a
r
m
opt
im
iz
a
ti
on
(
P
S
O
)
a
nd
Y
O
L
O
+
K
a
lm
a
n
f
il
te
r
:
+
3.9%
F
1
-
s
c
or
e
im
pr
ove
m
e
nt
,
+
5.2%
M
O
T
A
im
pr
ove
m
e
nt
,
a
nd
ha
s
th
e
lo
w
e
s
t
I
D
s
w
it
c
he
s
a
nd
FG
r
a
te
s
.
T
he
s
e
im
pr
ove
m
e
nt
s
a
r
e
e
n
a
bl
e
d
by
th
e
f
uz
z
y
r
ul
e
–
gui
de
d
G
A
,
w
hi
c
h
c
a
n
a
da
pt
iv
e
ly
r
e
f
in
e
a
s
s
o
c
ia
ti
on
hypothe
s
e
s
f
or
obj
e
c
ts
unde
r
oc
c
lu
s
io
n
a
nd
unc
e
r
ta
in
ty
of
obj
e
c
t
m
ot
i
on,
unl
ik
e
th
e
f
ix
e
d
-
pa
r
a
m
e
te
r
P
S
O
/Ka
lm
a
n
m
e
th
ods
.
T
he
c
oupl
e
d
R
e
s
N
e
t
e
m
be
ddi
ng
s
a
ls
o
im
pr
ove
th
e
c
ons
is
te
nc
y
of
te
m
por
a
ll
y
s
im
il
a
r
a
ppe
a
r
a
nc
e
s
,
w
hi
c
h i
nc
r
e
a
s
e
s
r
obus
tn
e
s
s
t
o l
ow
r
e
s
ol
ut
io
n a
nd pos
e
va
r
ia
ti
on.
T
a
bl
e
7. Q
ua
nt
it
a
ti
ve
c
om
p
a
r
is
on w
it
h r
e
c
e
nt
hybr
id
t
r
a
c
ki
ng mode
ls
M
e
t
hod
R
e
c
a
l
l
(%)
P
r
e
c
i
s
i
on
(%)
F1
-
s
c
or
e
(%)
F
A
F
MT
I
D
S
F
r
a
g
M
O
T
A
(%)
M
O
T
P
(%)
C
N
N
+P
S
O
77.2
87.9
82.2
0.23
22
1035
2028
61.3
78.2
Y
O
L
O
+K
a
l
m
a
n
f
i
l
t
e
r
79.6
88.4
83.7
0.20
25
912
1865
63.5
79.4
D
e
e
pS
O
R
T
(
C
N
N
+K
a
l
m
a
n + R
e
I
D
)
80.1
89.2
84.4
0.18
28
754
1750
65.0
80.0
P
r
opos
e
d (
F
uz
z
y
ge
ne
t
i
c
+ de
e
p
c
oupl
e
d
R
e
s
N
e
t
)
81.0
90.5
86.1
0.15
30
669
1745
66.5
80.3
W
hi
le
th
e
f
r
a
m
e
w
or
k
de
m
ons
tr
a
te
s
c
on
s
id
e
r
a
bl
e
a
c
c
ur
a
c
y
a
nd
r
obus
tn
e
s
s
,
th
e
r
e
a
r
e
li
m
it
a
ti
ons
.
C
om
put
a
ti
ona
l
c
os
t
in
c
r
e
a
s
e
s
w
it
h
s
c
e
n
e
de
ns
it
y
,
in
di
c
a
ti
ng
a
pr
a
c
ti
c
a
l
ne
e
d
f
or
li
ght
e
r
-
w
e
ig
ht
C
N
N
ba
c
kbone
s
f
or
la
r
ge
-
s
c
a
le
a
nd/
or
e
m
be
dde
d
de
pl
oym
e
nt
.
P
e
r
f
or
m
a
nc
e
c
oul
d
s
uf
f
e
r
unde
r
he
a
vy
oc
c
lu
s
io
n
or
poor
il
lu
m
in
a
ti
on
c
ondi
ti
ons
due
to
r
e
li
a
nc
e
on
vi
s
ua
l
c
ue
s
a
s
a
s
in
gl
e
s
our
c
e
of
in
f
or
m
a
ti
on.
T
he
f
uz
z
y
r
ul
e
ba
s
e
r
e
qui
r
e
s
in
it
ia
l
m
a
nua
l
tu
ni
ng,
poi
nt
in
g
to
th
e
pos
s
ib
il
it
y
of
s
e
lf
-
a
da
pt
iv
e
or
r
e
in
f
o
r
c
e
m
e
nt
-
dr
iv
e
n
s
ynt
he
s
iz
in
g
c
onf
ig
ur
a
ti
ons
.
F
in
a
ll
y,
im
pr
ove
d
c
r
os
s
-
dom
a
in
ge
ne
r
a
li
z
a
ti
on
a
nd
a
dve
r
s
a
r
ia
l
r
obus
tn
e
s
s
a
r
e
s
ti
ll
r
e
qui
r
e
d
f
or
w
id
e
r
de
pl
oym
e
nt
s
c
e
na
r
io
s
,
e
s
pe
c
ia
ll
y
if
pr
iv
a
c
y
c
ons
id
e
r
a
ti
ons
a
r
e
a
ls
o
r
e
s
pe
c
te
d
in
th
e
s
ol
ut
io
n
de
s
ig
n.
F
ut
ur
e
opt
io
n
s
m
a
y
be
im
pr
ove
d
w
it
h
ha
r
dw
a
r
e
-
le
ve
l
opt
im
iz
a
ti
on
a
nd
lo
w
e
r
-
c
os
t
im
pl
e
m
e
nt
a
ti
on
m
e
th
ods
.
F
ie
ld
pr
ogr
a
m
m
a
bl
e
ga
te
a
r
r
a
y
(
F
P
G
A
)
ba
s
e
d
ne
ur
a
l
ne
twor
k
im
pl
e
m
e
nt
a
ti
ons
c
a
n
pr
ovi
de
c
a
pa
bi
li
ti
e
s
th
a
t
a
c
hi
e
ve
r
e
a
l
-
ti
m
e
pr
oc
e
s
s
in
g
w
it
h
l
ow
la
te
nc
y
a
t
lo
w
im
pl
e
m
e
nt
a
ti
on
c
o
s
t
[
27]
.
A
ddi
ti
ona
ll
y,
R
a
s
pbe
r
r
y
P
i
pl
a
tf
or
m
s
e
f
f
e
c
ti
ve
ly
de
pl
oy
in
te
ll
i
ge
nt
s
ys
te
m
s
in
lo
w
-
pow
e
r
e
nvi
r
onm
e
nt
s
s
uc
h
a
s
c
la
s
s
r
oom
s
or
s
m
a
ll
s
ur
ve
il
la
nc
e
s
tu
di
e
s
[
28]
.
4.
C
O
N
C
L
U
S
I
O
N
T
hi
s
s
tu
dy
pr
op
os
e
d
a
h
ybr
id
m
ul
ti
-
f
a
c
e
r
e
c
ogni
ti
o
n
a
nd
tr
a
c
ki
n
g
f
r
a
m
e
w
or
k
th
a
t
in
c
or
por
a
te
s
de
e
p
c
oupl
e
d
R
e
s
N
e
t
vi
s
ua
l
f
e
a
tu
r
e
s
a
lo
ng
s
id
e
f
uz
z
y
ge
n
e
ti
c
opt
im
iz
a
ti
on,
t
a
c
kl
in
g
p
e
r
s
i
s
te
nt
c
h
a
ll
e
nge
s
of
pa
r
ti
a
l
oc
c
lu
s
io
n,
i
ll
um
in
a
ti
on
c
h
a
nge
s
,
a
nd
ba
c
kgr
ou
nd
c
l
ut
te
r
f
ound
in
r
e
a
l
-
w
or
ld
s
ur
v
e
il
la
nc
e
s
e
tt
in
g
s
.
B
y
li
nki
ng
de
e
p
f
e
a
tu
r
e
e
m
b
e
ddi
n
gs
to
a
da
pt
iv
e
f
uz
z
y
–
ge
ne
ti
c
r
ul
e
s
,
th
e
f
r
a
m
e
w
or
k
dyna
m
ic
a
ll
y
c
om
bi
n
e
s
m
ot
io
n,
a
pp
e
a
r
a
nc
e
,
a
nd
c
o
nf
id
e
n
c
e
c
ue
s
to
im
pr
o
ve
da
t
a
a
s
s
oc
i
a
ti
on
w
hi
le
a
c
c
oun
ti
ng
f
or
un
c
e
r
t
a
in
ty
in
vi
s
u
a
l
c
ondi
ti
o
ns
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
R
obus
t
m
ul
ti
-
fa
c
e
s
r
e
c
ogni
ti
on and tr
ac
k
in
g v
ia
f
uz
z
y
ge
ne
ti
c
al
gor
it
hm
s
and
…
(
A
di
l
A
bdul
hur
A
bus
hana
)
217
A
C
K
N
O
WL
E
D
G
M
E
N
T
S
T
he
a
ut
hor
s
a
r
e
gr
a
te
f
ul
f
or
t
he
i
ns
ti
tu
ti
ona
l
r
e
s
our
c
e
s
t
ha
t
f
a
c
il
i
ta
te
d t
hi
s
r
e
s
e
a
r
c
h
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
ta
te
no f
undi
ng i
nvol
ve
d.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on
.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
A
di
l
A
bd
ul
h
ur
A
bu
s
ha
na
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Y
ou
s
i
f
S
a
m
e
r
M
u
dh
a
f
a
r
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
A
ut
hor
s
s
ta
te
no c
onf
li
c
t
of
i
nt
e
r
e
s
t.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
D
a
ta
a
va
il
a
bi
li
ty
i
s
not
a
ppl
ic
a
bl
e
t
o t
hi
s
p
a
pe
r
a
s
no n
e
w
da
ta
w
e
r
e
c
r
e
a
te
d i
n t
hi
s
s
tu
dy.
R
E
F
E
R
E
N
C
E
S
[
1]
W
.
L
uo,
J
.
X
i
ng,
A
.
M
i
l
a
n,
X
.
Z
ha
ng,
W
.
L
i
u,
a
nd
T
.
K
.
K
i
m
,
“
M
ul
t
i
pl
e
obj
e
c
t
t
r
a
c
ki
ng:
a
l
i
t
e
r
a
t
ur
e
r
e
vi
e
w
,”
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
, vol
. 293, 2021, doi
:
10.1016/
j
.a
r
t
i
nt
.2020.103448.
[
2]
G
.
C
i
a
pa
r
r
one
,
F
.
L
.
S
á
nc
h
e
z
,
S
.
T
a
bi
k,
L
.
T
r
oi
a
no,
R
.
T
a
gl
i
a
f
e
r
r
i
,
a
nd
F
.
H
e
r
r
e
r
a
,
“
D
e
e
p
l
e
a
r
ni
ng
i
n
vi
de
o
m
ul
t
i
-
obj
e
c
t
t
r
a
c
ki
ng:
a
s
ur
ve
y,”
N
e
u
r
oc
om
put
i
ng
, vol
. 381, pp. 61
–
88, 2020, doi
:
10.1016/
j
.ne
uc
om
.
2019.11.023.
[
3]
N
.
W
oj
ke
,
A
.
B
e
w
l
e
y,
a
nd
D
.
P
a
ul
us
,
“
S
i
m
pl
e
onl
i
ne
a
nd
r
e
a
l
t
i
m
e
t
r
a
c
ki
ng
w
i
t
h
a
de
e
p
a
s
s
oc
i
a
t
i
on
m
e
t
r
i
c
,”
i
n
2017
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on I
m
age
P
r
o
c
e
s
s
i
ng (
I
C
I
P
)
, 2017, pp. 3645
–
3649
, d
oi
:
10.1109/
I
C
I
P
.2017.8296962.
[
4]
P
. B
e
r
gm
a
nn, T
. M
e
i
nha
r
dt
, a
nd L
.
L
.
-
T
a
i
xe
, “
T
r
a
c
ki
ng w
i
t
hout
be
l
l
s
a
nd w
hi
s
t
l
e
s
,”
i
n
2019 I
E
E
E
/
C
V
F
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on C
om
put
e
r
V
i
s
i
on (
I
C
C
V
)
, 2019, pp. 941
–
951
, doi
:
10.1109/
I
C
C
V
.2019.00103.
[
5]
S
. S
un, N
. A
kht
a
r
, H
. S
ong, A
. S
. M
i
a
n, a
nd
M
. S
ha
h, “
D
e
e
p a
f
f
i
ni
t
y ne
t
w
or
k f
or
m
ul
t
i
pl
e
obj
e
c
t
t
r
a
c
ki
ng,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
P
at
t
e
r
n A
nal
y
s
i
s
and M
ac
hi
ne
I
nt
e
l
l
i
ge
nc
e
, vol
. 43, no. 1, pp. 104
–
119, 2019, doi
:
10.1109/
T
P
A
M
I
.2019.2929520.
[
6]
M
.
W
a
ng
a
nd
W
.
D
e
ng,
“
D
e
e
p
f
a
c
e
r
e
c
ogni
t
i
on:
a
s
ur
ve
y,”
N
e
u
r
oc
om
put
i
ng
,
vol
.
429,
pp.
215
–
244,
2021,
doi
:
10.1016/
j
.ne
uc
om
.2020.10.081.
[
7]
P
. D
e
ndor
f
e
r
e
t
al
.
, “
M
O
T
20:
a
be
nc
hm
a
r
k f
or
m
ul
t
i
obj
e
c
t
t
r
a
c
ki
ng i
n c
r
ow
de
d s
c
e
ne
s
,”
2020
,
ar
X
i
v
:
2003.09003
.
[
8]
Y
.
Z
ha
ng
e
t
al
.
,
“
B
yt
e
T
r
a
c
k:
m
ul
t
i
-
obj
e
c
t
t
r
a
c
ki
ng
by
a
s
s
oc
i
a
t
i
ng
e
ve
r
y
de
t
e
c
t
i
on
box,”
i
n
C
om
put
e
r
V
i
s
i
on
–
E
C
C
V
2022
,
C
ha
m
,
S
w
i
t
z
e
r
l
a
nd:
S
pr
i
nge
r
, 2022, pp. 1
–
21
, doi
:
10.1007/
978
-
3
-
031
-
20047
-
2_1.
[
9]
S
.
M
i
r
j
a
l
i
l
i
,
J
.
S
.
D
ong,
A
.
S
.
S
a
di
q,
a
nd
H
.
F
a
r
i
s
,
“
G
e
ne
t
i
c
a
l
gor
i
t
hm
:
t
h
e
or
y
,
l
i
t
e
r
a
t
ur
e
r
e
vi
e
w
,
a
nd
a
ppl
i
c
a
t
i
on
i
n
i
m
a
ge
r
e
c
ons
t
r
uc
t
i
on,”
i
n
St
udi
e
s
i
n C
om
put
at
i
onal
I
nt
e
l
l
i
ge
nc
e
, vol
. 811, 2020, pp. 69
–
85
, doi
:
10.1007/
978
-
3
-
030
-
12127
-
3_5.
[
10]
Y
.
B
.
‐
S
ha
l
om
,
X
.
‐
R
on.
L
i
,
a
nd
T
.
K
i
r
uba
r
a
j
a
n,
E
s
t
i
m
at
i
on
w
i
t
h
app
l
i
c
at
i
ons
t
o
t
r
ac
k
i
ng
and
nav
i
ga
t
i
on
,
3r
d
e
d.
N
e
w
Y
or
k:
J
ohn
W
i
l
e
y &
S
ons
, 2002
, doi
:
10.1002/
0471221279.
[
11]
S
.
Z
ha
ng
e
t
al
.
,
“
T
r
a
c
ki
ng
pe
r
s
ons
-
of
-
i
nt
e
r
e
s
t
vi
a
a
da
pt
i
ve
di
s
c
r
i
m
i
na
t
i
ve
f
e
a
t
ur
e
s
,”
i
n
C
om
put
e
r
V
i
s
i
on
–
E
C
C
V
2016
,
C
ha
m
,
S
w
i
t
z
e
r
l
a
nd:
S
pr
i
nge
r
, 2016, pp. 415
–
433
, doi
:
10.1007/
978
-
3
-
319
-
46454
-
1_26
.
[
12]
S
.
W
.
A
r
a
c
hc
hi
l
a
ge
a
nd
E
.
I
z
qui
e
r
do,
“
A
da
pt
i
ve
a
ggr
e
ga
t
e
d
t
r
a
c
kl
e
t
l
i
nki
ng
f
o
r
m
ul
t
i
-
f
a
c
e
t
r
a
c
ki
ng,”
i
n
2020
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on I
m
age
P
r
oc
e
s
s
i
ng (
I
C
I
P
)
, 2020, pp. 1366
–
1370
, doi
:
10.1109/
I
C
I
P
40778.2020.9190823.
[
13]
G
.
B
a
r
que
r
o,
C
.
F
e
r
na
nde
z
,
a
nd
I
.
H
upont
,
“
L
ong
-
t
e
r
m
f
a
c
e
t
r
a
c
ki
ng
f
or
c
r
ow
de
d
vi
de
o
-
s
ur
ve
i
l
l
a
nc
e
s
c
e
na
r
i
os
,”
i
n
I
J
C
B
2020
-
I
E
E
E
/
I
A
P
R
I
nt
e
r
nat
i
onal
J
oi
nt
C
onf
e
r
e
nc
e
on B
i
om
e
t
r
i
c
s
, 2020
, doi
:
10.1109/
I
J
C
B
48548.2020.9304892.
[
14]
S
.
Z
ha
ng
e
t
al
.
,
“
T
r
a
c
ki
ng
pe
r
s
ons
-
of
-
i
nt
e
r
e
s
t
vi
a
uns
upe
r
vi
s
e
d
r
e
pr
e
s
e
nt
a
t
i
on
a
da
pt
a
t
i
on,”
I
nt
e
r
nat
i
onal
J
our
nal
of
C
om
put
e
r
V
i
s
i
on
, vol
. 128, no. 1, pp. 96
–
120, 2020, doi
:
10.1007/
s
11263
-
019
-
01212
-
1.
[
15]
G
.
B
a
r
que
r
o,
I
.
H
upont
,
a
nd
C
.
F
.
T
e
na
,
“
R
a
nk
-
ba
s
e
d
v
e
r
i
f
i
c
a
t
i
on
f
or
l
on
g
-
t
e
r
m
f
a
c
e
t
r
a
c
ki
ng
i
n
c
r
ow
de
d
s
c
e
ne
s
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
B
i
om
e
t
r
i
c
s
,
B
e
hav
i
or
,
and
I
de
nt
i
t
y
Sc
i
e
nc
e
,
vol
.
3,
no.
4,
pp.
495
–
505,
2021,
doi
:
10.1109/
T
B
I
O
M
.2021.3099568.
[
16]
J
.
W
a
ng
a
nd
J
.
L
a
ng,
“
V
i
s
ua
l
m
ul
t
i
-
f
a
c
e
t
r
a
c
ki
ng
a
ppl
i
e
d
t
o
c
ounc
i
l
pr
oc
e
e
di
ngs
,”
I
E
E
E
I
ns
t
r
um
e
nt
at
i
on
&
M
e
as
ur
e
m
e
n
t
M
agaz
i
ne
, vol
. 24, no. 3, pp. 78
–
84, 2021, doi
:
10.1109/
M
I
M
.2021.9436089.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
209
-
218
218
[
17]
T
.
M
.
T
r
a
n
e
t
al
.
,
“
R
e
S
O
R
T
:
a
n
I
D
-
r
e
c
ove
r
y
m
ul
t
i
-
f
a
c
e
t
r
a
c
ki
ng
m
e
t
hod
f
or
s
ur
ve
i
l
l
a
nc
e
c
a
m
e
r
a
s
,”
i
n
2021
16t
h
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
on
A
ut
om
at
i
c
F
ac
e
and
G
e
s
t
ur
e
R
e
c
ogni
t
i
on
(
F
G
2021)
,
2021,
pp.
1
–
8
,
doi
:
10.1109/
F
G
52635.2021.9666941.
[
18]
G
.
R
e
n,
X
.
L
u,
a
nd
Y
.
L
i
,
“
A
c
r
os
s
-
c
a
m
e
r
a
m
ul
t
i
-
f
a
c
e
t
r
a
c
ki
ng
s
ys
t
e
m
ba
s
e
d
on
doubl
e
t
r
i
pl
e
t
ne
t
w
or
ks
,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
9,
pp. 43759
–
43774, 2021, doi
:
10.1109/
A
C
C
E
S
S
.2021.3061572.
[
19]
Z
.
W
e
ng,
H
.
Z
hua
ng,
H
.
L
i
,
B
.
R
a
m
a
l
i
nga
m
,
R
.
E
.
M
oha
n,
a
nd
Z
.
L
i
n,
“
O
nl
i
ne
m
ul
t
i
-
f
a
c
e
t
r
a
c
ki
ng
w
i
t
h
m
ul
t
i
-
m
oda
l
i
t
y
c
a
s
c
a
de
d
m
a
t
c
hi
ng,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
C
i
r
c
ui
t
s
and
S
y
s
t
e
m
s
f
or
V
i
de
o
T
e
c
h
nol
ogy
,
vol
.
33,
no.
6,
pp.
2738
–
2752,
2023,
doi
:
10.1109/
T
C
S
V
T
.2022.3224699.
[
20]
J
.
K
i
m
,
C
.
Y
.
J
u,
G
.
W
.
K
i
m
,
a
nd
D
.
H
.
L
e
e
,
“
B
oT
-
F
a
c
e
S
O
R
T
:
ba
g
-
of
-
t
r
i
c
ks
f
or
r
obus
t
m
ul
t
i
-
f
a
c
e
t
r
a
c
ki
ng
i
n
un
c
ons
t
r
a
i
ne
d
vi
de
os
,”
i
n
C
om
put
e
r
V
i
s
i
on
–
A
C
C
V
2024
, S
i
nga
por
e
:
S
pr
i
nge
r
, 2025, pp. 278
–
294
, doi
:
10.1007/
978
-
981
-
96
-
0901
-
7_17.
[
21]
R
. J
oc
hl
a
nd A
. U
hl
, “
F
a
c
e
Q
S
O
R
T
:
c
a
n a
c
om
bi
na
t
i
on of
t
w
o bi
om
e
t
r
i
c
f
e
a
t
ur
e
s
a
c
hi
e
ve
c
om
pe
t
i
t
i
ve
pe
r
f
or
m
a
nc
e
?
,”
i
n
2025 25t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on D
i
gi
t
al
Si
gnal
P
r
oc
e
s
s
i
ng (
D
SP
)
, 2025, pp. 1
–
5
, doi
:
10.1109/
D
S
P
65409.2025.11075151.
[
22]
L
.
L
i
a
nd
X
.
Z
ha
n,
“
A
nove
l
da
t
a
a
s
s
oc
i
a
t
i
on
a
l
gor
i
t
hm
ba
s
e
d
on
f
uz
z
y
l
ogi
c
f
or
vi
s
ua
l
obj
e
c
t
t
r
a
c
ki
ng,”
i
n
2019
12t
h
I
nt
e
r
nat
i
onal
C
ongr
e
s
s
on
I
m
age
and
Si
gnal
P
r
oc
e
s
s
i
ng,
B
i
oM
e
di
c
al
E
ngi
ne
e
r
i
ng
and
I
nf
or
m
at
i
c
s
,
2019,
pp.
1
–
6
,
doi
:
10.1109/
C
I
S
P
-
B
M
E
I
48845.2019.8965769.
[
23]
C
.
G
a
o,
Q
.
L
i
u,
Q
.
X
u,
L
.
W
a
ng,
J
.
L
i
u,
a
nd
C
.
Z
ou,
“
S
ke
t
C
hyC
oc
o:
i
m
a
ge
ge
ne
r
a
t
i
on
f
r
om
f
r
e
e
ha
nd
s
c
e
ne
s
ke
t
c
he
s
,”
i
n
2020
I
E
E
E
/
C
V
F
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
(
C
V
P
R
)
,
2020,
pp.
5173
–
5182
,
doi
:
10.1109/
C
V
P
R
42600.2020.00522.
[
24]
M
.
A
ya
z
ogl
u,
M
.
S
z
na
i
e
r
,
a
nd
O
.
I
.
C
a
m
p
s
,
“
F
a
s
t
a
l
gor
i
t
hm
s
f
or
s
t
r
uc
t
ur
e
d
r
obus
t
pr
i
nc
i
pa
l
c
om
pone
nt
a
na
l
ys
i
s
,”
i
n
2012
I
E
E
E
C
onf
e
r
e
nc
e
on C
om
put
e
r
V
i
s
i
on and P
at
t
e
r
n R
e
c
ogni
t
i
on
, 2012, pp. 1704
–
1711
, doi
:
10.1109/
C
V
P
R
.2012.6247865.
[
25]
C
.
D
i
c
l
e
,
O
.
I
.
C
a
m
ps
,
a
nd
M
.
S
z
na
i
e
r
,
“
T
he
w
a
y
t
he
y
m
ove
:
t
r
a
c
ki
ng
m
ul
t
i
p
l
e
t
a
r
ge
t
s
w
i
t
h
s
i
m
i
l
a
r
a
ppe
a
r
a
nc
e
,”
i
n
2013
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
, 2013, pp. 2304
–
2311
, doi
:
10.1
109/
I
C
C
V
.2013.286.
[
26]
C.
-
C
.
L
i
n
a
nd
Y
.
H
ung,
“
A
pr
i
or
-
l
e
s
s
m
e
t
hod
f
or
m
ul
t
i
-
f
a
c
e
t
r
a
c
ki
ng
i
n
unc
ons
t
r
a
i
ne
d
vi
de
os
,”
i
n
2018
I
E
E
E
/
C
V
F
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on and P
at
t
e
r
n R
e
c
ogni
t
i
on
, 2018, pp. 538
–
547
, doi
:
10.1109/
C
V
P
R
.2018.00063.
[
27]
S
.
H
.
A
bdul
na
bi
,
Y
.
S
.
M
udha
f
a
r
,
A
.
A
.
K
a
dhi
m
,
M
.
B
.
M
a
hdi
,
a
nd
H
.
H
.
S
oj
a
r
,
“
N
e
ur
a
l
ne
t
w
or
k
-
ba
s
e
d
s
ys
t
e
m
i
de
nt
i
f
i
c
a
t
i
on:
a
c
om
pr
e
he
ns
i
ve
F
P
G
A
de
s
i
gn
a
nd
i
m
pl
e
m
e
nt
a
t
i
on,”
i
n
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
and
M
e
c
hat
r
oni
c
s
Sy
s
t
e
m
, A
I
M
S 2024
, 2024, pp. 1
–
7
, doi
:
10.1109/
A
I
M
S
61812.2024.10512531.
[
28]
A
.
M
.
A
.
A
l
-
M
uqa
r
m
,
Y
.
M
udh
a
f
a
r
,
A
.
M
.
S
ha
ki
r
,
M
.
K
a
z
e
m
,
R
.
A
.
-
Y
a
hi
ya
,
a
nd
B
.
S
.
A
.
Z
a
hr
a
,
“
L
ow
-
c
o
s
t
s
m
a
r
t
l
e
a
r
ni
ng
w
i
t
h
m
oodl
e
-
ba
s
e
d
R
a
s
pbe
r
r
y
P
i
4
f
or
uni
ve
r
s
i
t
y
s
t
ude
nt
s
,”
i
n
6t
h
I
r
aqi
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
E
ngi
ne
e
r
i
ng
T
e
c
hnol
ogy
and
i
t
s
A
ppl
i
c
at
i
ons
, I
I
C
E
T
A
2023
, 2023, pp. 603
–
608
, doi
:
10.1109/
I
I
C
E
T
A
57613.20
23.10351266.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Adil
Abdulhur
Abushana
is
a
lecturer
at
the
Department
of
Co
mputer
Science,
University
of
Kufa.
He
received
his
B.Sc.
degree
in
Mathematics
f
ro
m
Salahaddin
University,
Iraq,
in
1989,
and
his
M.Sc.
degree
from
the
Department
of
I
nformation
Technology,
University
of
Utara
Malaysia,
in
2009,
respectively.
He
is
currently
pursuing
a
Ph.D.
in
the
Department
of
Information
Systems
at
Universiti
Teknologi
Malays
ia
(UTM).
His
research
interests
include
pattern
recognition
and
computer
vision,
with
ap
plications
to
biometrics.
He
was
one
of
the
participants
who
received
a
paper
awa
rd
at
the
I
EEE
conference
in
2015.
He can be contacted at email:
adelshana30
00@
gmail.co
m or adel
.alnasrawi
@
uokufa.edu
.iq
.
Yousif
Samer
Mudhafar
is
a
lecturer
at
the
D
epartment
of
Co
mputer
Science,
University of Kufa
. H
e earned hi
s
B.Sc.
in Com
puter
Techniqu
es Engi
neering fro
m the Is
lamic
University
in
Najaf
in
2018.
He
completed
his
M.Sc.
in
Computer
Sc
ience
Engineering
at
the
University
of
Debrecen
in
2022,
graduating
with
honors
and
receivin
g
the
outstanding
student
certificate.
He works
at
the Uni
versity
of Kufa,
Facult
y of
Educati
on,
Department
of Com
puter
Scienc
e. His
rese
arch
inter
ests
include co
mputer netw
orks
,
IoT, and
A
I. He can be
contacted
at
email:
yousif.mudhafar@
iunajaf.edu.iq.
Evaluation Warning : The document was created with Spire.PDF for Python.