I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
, pp.
4249
~
4262
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
13
.i
4
.pp
4249
-
4262
4249
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
D
r
i
ve
r
i
n
at
t
e
n
t
i
on
d
e
t
e
c
t
i
on
sys
t
e
m
u
si
n
g m
u
l
t
i
-
t
ask
c
asc
ad
e
d
c
on
vol
u
t
i
on
al
n
e
t
w
or
k
s
A
b
d
e
lf
e
t
t
ah
S
ou
lt
an
a,
F
aou
z
ia
B
e
n
ab
b
ou
, N
aw
al
S
ae
l,
S
ou
k
ai
n
a B
ou
h
s
is
s
in
L
a
bor
a
t
or
y of
I
nf
or
m
a
t
i
on
T
e
c
hnol
ogy a
nd M
ode
l
i
ng, F
a
c
ul
t
y of
S
c
i
e
nc
e
s
B
e
n
M
’
S
I
K
,
U
ni
ve
r
s
i
t
y H
a
s
s
a
n I
I
of
C
a
s
a
bl
a
n
c
a
,
C
a
s
a
bl
a
n
c
a
,
M
or
oc
c
o
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
O
c
t
31, 2023
R
e
vi
s
e
d
M
a
r
5, 2024
A
c
c
e
pt
e
d
M
a
r
21, 2024
Driver
inattention
has
emerged
as
a
critical
concern
impacting
road
safety,
resulting
in
an
alarming
surge
in
accidents
and
fatalities.
This
re
search
introduces
a
novel
system
for
detecting
inattention,
structure
d
across
six
levels:
percep
tion,
facial
feature
extractio
n,
tracking
driver
face,
and
driver
secondary
task
using
pre
-
trained
deep
learning
mod
els,
inattention
det
ection,
risk
estimation,
and
alert.
The
system
is
based
on
image
processing
captured
from
two
strategically
positioned
cameras
that
simultaneously
capt
ure
the
driver
’
s
activities
while
driving
and
their
facial
exp
ressions.
The
second
contribu
tion
concerns
the
driv
er
facial
features
extractio
n
using
mul
ti
-
task
cascaded
convolut
ional
networks
(MTCNN),
and
it
i
s
comparison
with
the
histogram
of
gradient
(HOG)
-
based
frontal
face
detector,
and
haar
feature
-
based
cascade
classifier.
The
algorithms
were
compared
based
o
n
their
runtime
efficiency,
robustness
in
handling
varying
lightin
g
condition
s,
and
various head movements. The MTCNN ac
hieves
high performance, re
aching
accuracy
levels
ranging
from
96.4%
to
99.5%
on
two
da
tasets
inc
luding
realistic
driving
scenarios:
the
DrivFace
dataset
and,
the
driver
dro
wsiness
dataset.
The
comparative
analysis
sheds
light
on
the
strength
s
and
weaknesses
of
each
algorithm,
providing
valuable
insights
for
selecti
ng
the
most suitable f
ace de
tection algo
rithm to use in
our system.
K
e
y
w
o
r
d
s
:
D
r
iv
e
r
di
s
tr
a
c
ti
on
D
r
iv
e
r
dr
ow
s
in
e
s
s
D
r
iv
e
r
e
m
ot
io
ns
D
r
iv
e
r
f
a
ti
gue
D
r
iv
e
r
i
na
tt
e
nt
io
n m
oni
to
r
in
g
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
:
A
bde
lf
e
tt
a
h S
oul
ta
na
L
a
bor
a
to
r
y of
I
n
f
or
m
a
ti
on T
e
c
hnol
ogy a
nd M
ode
li
ng, F
a
c
ul
ty
of
S
c
ie
nc
e
s
B
e
n
M
’
S
I
K
U
ni
ve
r
s
it
y H
a
s
s
a
n I
I
of
C
a
s
a
bl
a
nc
a
C
a
s
a
bl
a
n
c
a
,
M
or
oc
c
o
E
m
a
il
:
a
bde
lf
e
tt
a
h.s
oul
ta
na
-
e
tu
@
e
tu
.uni
vh2c
.m
a
1.
I
N
T
R
O
D
U
C
T
I
O
N
R
oa
d
s
a
f
e
ty
r
e
m
a
in
s
one
of
th
e
m
os
t
pr
e
s
s
in
g
c
on
c
e
r
ns
in
our
m
ode
r
n
s
oc
ie
ty
.
R
oa
d
a
c
c
id
e
nt
s
c
a
us
e
d by dr
iv
e
r
i
na
tt
e
nt
io
n c
ont
in
ue
t
o be
c
os
tl
y r
e
ga
r
di
ng huma
n l
iv
e
s
a
nd ma
te
r
ia
l
r
e
s
our
c
e
s
. A
c
c
or
di
ng t
o
th
e
W
or
ld
H
e
a
lt
h
O
r
ga
ni
z
a
ti
on
(
W
H
O
)
gl
oba
l
s
t
a
tu
s
r
e
por
t,
r
o
a
d
tr
a
f
f
ic
a
c
c
id
e
nt
s
c
a
us
e
1.35
m
il
li
on
de
a
th
s
ye
a
r
ly
.
T
hi
s
i
s
ne
a
r
ly
3
,
700
pe
opl
e
dyi
ng
on
th
e
w
or
ld
’
s
r
oa
ds
da
il
y
[
1]
.
I
n
th
is
c
ont
e
xt
[
2]
,
dr
iv
e
r
in
a
tt
e
nt
io
n
is
de
f
in
e
d
a
s
‘
in
s
uf
f
ic
ie
nt
,
or
no
a
tt
e
nt
io
n,
to
a
c
ti
vi
ti
e
s
c
r
it
ic
a
l
f
or
s
a
f
e
dr
iv
in
g
’
a
nd
c
a
n
be
br
ought
a
bout
th
r
ough
a
num
be
r
of
di
f
f
e
r
e
nt
m
e
c
ha
ni
s
m
s
s
uc
h
a
s
‘
dr
iv
e
r
-
r
e
s
t
r
ic
te
d
a
tt
e
nt
io
n’
(
e
.g.
due
to
bi
ol
ogi
c
a
l
s
ta
te
s
,
s
uc
h a
s
dr
ow
s
in
e
s
s
or
f
a
ti
gue
)
or
‘
dr
iv
e
r
m
is
pr
io
r
it
iz
e
d a
tt
e
nt
io
n’
(
e
.g. due
t
o f
oc
us
in
g a
tt
e
nt
io
n on one
a
s
pe
c
t
of
dr
iv
in
g
to
th
e
e
xc
lu
s
io
n
o
f
a
not
he
r
w
hi
c
h
is
m
or
e
c
r
it
ic
a
l
f
or
s
a
f
e
dr
iv
in
g)
.
T
hi
s
c
a
n
m
a
ni
f
e
s
t
in
va
r
io
us
f
or
m
s
,
s
uc
h
a
s
us
in
g
a
m
obi
le
phone
,
a
dj
us
ti
ng
th
e
r
a
di
o,
d
a
ydr
e
a
m
in
g,
or
e
nga
gi
ng
in
ot
he
r
di
s
tr
a
c
ti
ng
a
c
ti
vi
ti
e
s
.
T
h
e
c
on
s
e
que
nc
e
s
of
s
uc
h
in
a
tt
e
nt
io
n
a
r
e
f
a
r
-
r
e
a
c
hi
ng
a
nd
c
a
n
h
a
ve
de
v
a
s
ta
ti
ng
im
pa
c
t
s
on
r
oa
d
s
a
f
e
ty
.
I
t
s
ig
ni
f
ic
a
nt
ly
im
pa
ir
s
a
dr
iv
e
r
'
s
a
bi
li
ty
to
r
e
a
c
t
pr
om
pt
ly
to
s
udde
n
c
ha
nge
s
in
tr
a
f
f
ic
c
ondi
ti
ons
,
in
c
r
e
a
s
in
g
th
e
li
ke
li
hood
of
c
ol
l
is
io
ns
a
nd
a
c
c
id
e
nt
s
.
B
a
s
e
d
on
2017
pol
ic
e
a
nd
hos
pi
ta
l
r
e
por
ts
,
th
e
N
a
ti
ona
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4249
-
4262
4250
H
ig
hw
a
y
T
r
a
f
f
ic
S
a
f
e
ty
A
dm
in
is
tr
a
ti
on
(
N
H
T
S
A
)
id
e
nt
if
ie
d
9
1,000
c
a
r
a
c
c
id
e
nt
s
c
a
us
e
d
by
dr
ow
s
y
dr
iv
e
r
s
[
3]
.
A
s
tu
dy
by
th
e
A
m
e
r
ic
a
n
A
ut
om
obi
le
A
s
s
oc
ia
ti
on’
s
F
oun
da
ti
on
f
or
T
r
a
f
f
ic
S
a
f
e
ty
e
s
ti
m
a
te
d
th
a
t
m
or
e
th
a
n
320,000
dr
ow
s
y
dr
iv
in
g
a
c
c
id
e
nt
s
h
a
ppe
n
y
e
a
r
ly
,
in
c
lu
di
n
g
6
,
400
f
a
ta
l
c
r
a
s
he
s
[
4]
.
S
im
il
a
r
ly
,
th
e
r
e
por
t
of
th
e
N
H
T
S
A
in
th
e
U
S
A
c
on
c
lu
de
d
th
a
t
a
r
ound
64.4%
of
pe
opl
e
lo
s
e
li
f
e
due
to
th
e
di
v
e
r
s
io
n
of
a
tt
e
nt
io
n
f
r
om
dr
iv
in
g
[
5
]
.
M
or
e
ove
r
,
th
e
ir
r
e
por
t
a
ls
o
de
c
la
r
e
d
th
a
t
s
o
m
e
w
he
r
e
be
twe
e
n
94%
a
nd
96%
of
a
ll
m
ot
or
ve
hi
c
le
a
c
c
id
e
nt
s
a
r
e
c
a
u
s
e
d
by
s
om
e
hum
a
n
e
r
r
or
,
w
hi
le
m
a
ny
r
oa
d
a
c
c
id
e
nt
s
a
r
e
du
e
to
th
e
us
a
g
e
of
e
le
c
tr
oni
c
de
vi
c
e
s
s
uc
h
a
s
B
lu
e
to
ot
h
de
vi
c
e
s
a
nd
m
obi
le
phone
s
.
T
he
gr
a
vi
ty
of
th
e
s
e
s
ta
ti
s
ti
c
s
unde
r
s
c
or
e
s
th
e
ur
ge
nt
ne
e
d f
or
c
om
pr
e
he
ns
iv
e
m
e
a
s
ur
e
s
t
o c
om
ba
t
dr
iv
e
r
i
na
tt
e
nt
io
n a
nd f
or
ge
s
a
f
e
r
r
oa
ds
f
or
a
ll
.
W
hi
le
va
r
io
us
f
a
c
to
r
s
c
ont
r
ib
ut
e
to
dr
iv
e
r
in
a
tt
e
nt
io
n,
di
s
tr
a
c
ti
ons
,
f
a
ti
gue
,
dr
ow
s
in
e
s
s
,
a
nd
e
m
ot
io
na
l
s
ta
te
s
a
r
e
a
m
ong
th
e
m
os
t
pr
e
va
l
e
nt
.
D
is
tr
a
c
ti
on
b
y
a
s
e
c
onda
r
y
ta
s
k
i
s
one
of
th
e
m
a
in
f
a
c
to
r
s
im
pa
ir
in
g dr
iv
in
g;
c
om
m
on e
xa
m
pl
e
s
i
nc
lu
de
us
in
g a
m
obi
le
phone
, e
a
ti
ng or
dr
in
ki
ng, t
a
lk
in
g t
o pa
s
s
e
nge
r
s
,
gr
oom
in
g
li
ke
a
ppl
yi
ng
m
a
ke
up,
a
dj
us
ti
ng
c
ont
r
ol
s
,
a
nd
r
e
a
di
ng.
A
ddi
ti
ona
ll
y,
f
a
ti
gue
a
nd
dr
ow
s
in
e
s
s
a
r
e
a
ls
o
c
om
m
on
s
our
c
e
s
of
in
a
tt
e
nt
io
n.
C
e
c
c
a
c
c
i
e
t
al
.
[
6]
i
ndi
c
a
te
th
a
t
th
e
e
m
ot
io
na
l
s
ta
te
of
a
dr
iv
e
r
s
ig
ni
f
ic
a
nt
ly
i
nf
lu
e
nc
e
s
t
he
ir
l
e
ve
l
of
c
onc
e
nt
r
a
ti
on. F
or
e
xa
m
pl
e
,
S
te
r
ke
nbur
g a
nd J
e
on
[
7]
de
m
ons
tr
a
te
d t
ha
t
a
nge
r
de
gr
a
de
s
dr
iv
in
g
pe
r
f
or
m
a
nc
e
a
s
m
uc
h
a
s
or
m
or
e
th
a
n
ot
he
r
tr
a
di
ti
ona
l
di
s
tr
a
c
ti
on
ta
s
ks
.
D
r
iv
e
r
'
s
e
m
ot
io
n ha
s
be
c
om
e
a
f
a
c
to
r
t
ha
t
c
a
nnot
be
i
gnor
e
d i
n t
r
a
f
f
ic
s
a
f
e
ty
r
e
s
e
a
r
c
h
[
8]
, [
9]
.
I
n
r
e
s
po
ns
e
to
th
e
s
e
c
h
a
ll
e
n
ge
s
,
num
e
r
ou
s
r
e
s
e
a
r
c
h
e
f
f
or
t
s
h
a
ve
be
e
n
u
nde
r
t
a
ke
n
t
o
de
v
e
lo
p e
f
f
e
c
t
iv
e
s
ol
ut
io
ns
f
or
de
te
c
ti
ng
a
nd
m
it
ig
a
t
in
g
dr
i
ve
r
in
a
tt
e
nt
io
n.
T
he
s
e
e
f
f
or
t
s
of
t
e
n
f
o
c
us
on
i
ndi
vi
d
ua
l
a
s
pe
c
ts
of
in
a
tt
e
nt
io
n,
ut
il
i
z
in
g
v
a
r
io
u
s
t
e
c
hn
ol
ogi
e
s
r
a
ngi
ng
f
r
om
vi
de
o
im
a
ge
a
n
a
ly
s
is
to
ph
ys
io
lo
gi
c
a
l
s
ig
n
a
l
m
oni
to
r
in
g.
A
c
c
or
di
ng t
o
D
e
ng
a
nd
W
u
[
10]
,
c
a
ll
e
d
th
e
“
D
r
i
C
a
r
e
”
s
ys
t
e
m
,
pr
im
a
r
il
y
t
a
r
ge
t
s
d
e
te
c
ti
ng
dr
iv
e
r
s
’
f
a
ti
gue
s
ta
tu
s
t
hr
ough
vi
d
e
o
i
m
a
g
e
s
c
a
pt
ur
e
d
b
y
a
c
a
m
e
r
a
in
s
t
a
ll
e
d
in
th
e
v
e
hi
c
le
.
B
y
e
m
pl
oyi
n
g
c
onvol
ut
io
n
a
l
ne
ur
a
l
n
e
twor
k
s
(
C
N
N
)
,
th
i
s
a
ppr
oa
c
h
a
c
hi
e
ve
d
a
n
a
c
c
ur
a
c
y
of
92%
in
id
e
nt
if
yi
ng
dr
ow
s
in
e
s
s
a
nd
f
a
ti
g
ue
.
S
im
il
a
r
ly
,
C
hi
r
r
a
e
t
al
.
[
11]
d
e
ve
l
ope
d
a
nov
e
l
d
e
e
p
-
l
e
a
r
n
in
g
f
r
a
m
e
w
or
k
th
a
t
d
e
te
c
ts
dr
iv
e
r
dr
ow
s
in
e
s
s
ba
s
e
d
on
e
y
e
s
t
a
te
.
T
hi
s
a
p
pr
oa
c
h
a
c
hi
e
v
e
d
a
n
a
c
c
ur
a
c
y
of
96.4
2%
.
A
di
f
f
e
r
e
n
t
w
or
k
w
a
s
pr
opo
s
e
d
in
[
12]
to
d
e
te
c
t
dr
ow
s
i
ne
s
s
ba
s
e
d
on
e
l
e
c
tr
oe
n
c
e
ph
a
lo
gr
a
m
(
E
E
G
)
s
ig
n
a
l
s
. T
h
e
y
us
e
d
t
he
Q
-
f
a
c
to
r
w
a
v
e
le
t
tr
a
n
s
f
or
m
(
T
Q
W
T
)
c
oupl
e
d
w
it
h
th
e
e
xt
r
e
m
e
l
e
a
r
ni
n
g
m
a
c
hi
n
e
(
E
L
M
)
f
or
c
l
a
s
s
i
f
ic
a
ti
on
,
r
e
s
ul
ti
n
g
in
a
n
a
c
c
ur
a
c
y
of
91.84%
.
M
e
a
nw
hi
l
e
,
a
m
ul
ti
-
ta
s
ki
ng
C
N
N
m
od
e
l
i
s
pr
e
s
e
n
te
d
i
n
[
13]
,
c
om
bi
ni
ng
dr
ow
s
in
e
s
s
a
nd
f
a
ti
g
ue
d
e
te
c
ti
on
.
R
e
m
a
r
ka
bl
y,
th
e
ir
a
ppr
o
a
c
h
a
c
hi
e
v
e
d
a
hi
gh
a
c
c
ur
a
c
y
r
a
te
of
98.81%
w
hi
le
e
xpl
i
c
it
ly
e
va
lu
a
ti
ng
f
a
ti
gu
e
a
s
‘
ve
r
y
ti
r
e
d,
l
e
s
s
ti
r
e
d,
a
n
d
not
ti
r
e
d
.
’
B
a
khe
e
t
a
nd H
a
m
a
di
[
14]
pr
opos
e
d
a
f
r
a
m
e
w
or
k
f
or
in
s
t
a
nt
a
ne
ou
s
dr
iv
e
r
dr
ow
s
i
ne
s
s
d
e
te
c
ti
on.
T
h
e
y
e
m
pl
oye
d
a
n
a
d
a
pt
iv
e
va
r
i
a
nt
of
th
e
hi
s
to
gr
a
m
of
or
ie
nt
e
d
gr
a
di
e
nt
s
(
H
O
G
)
f
e
a
tu
r
e
s
to
r
e
pr
e
s
e
nt
th
e
e
ye
r
e
gi
on
a
nd
ut
il
iz
e
d
a
n
a
iv
e
B
a
ye
s
(
N
B
)
m
od
e
l
f
or
c
la
s
s
if
ic
a
ti
on.
T
h
e
ir
w
or
k
w
a
s
r
ig
or
ous
l
y
e
v
a
lu
a
te
d
us
in
g
th
e
publ
i
c
ly
a
va
il
a
bl
e
N
T
H
U
-
D
D
D
da
t
a
s
e
t,
d
e
m
on
s
tr
a
ti
ng
th
e
pot
e
nt
ia
l
of
th
e
ir
f
r
a
m
e
w
or
k
a
s
a
s
tr
on
g
c
o
nt
e
nd
e
r
a
ga
in
s
t
s
e
v
e
r
a
l
s
t
a
te
-
of
-
th
e
-
a
r
t
ba
s
e
li
n
e
s
.
N
ot
a
bl
y,
th
e
ir
f
r
a
m
e
w
or
k
a
c
hi
e
ve
d
a
c
om
pe
ti
ti
ve
d
e
te
c
ti
on
a
c
c
ur
a
c
y
of
85.6
2%
w
hi
l
e
m
a
in
ta
i
ni
ng
e
f
f
ic
ie
nc
y a
nd
s
ta
bi
l
it
y.
Z
ha
o
e
t
al
.
[
15]
a
ls
o
in
ve
s
ti
ga
te
dr
iv
e
r
di
s
tr
a
c
ti
on
de
te
c
ti
on,
f
oc
us
in
g
on
th
e
he
a
d
pos
e
.
T
he
y
e
m
pl
oye
d t
he
H
P
E
_R
e
s
ne
t5
0 a
lg
or
it
hm
t
o
a
c
hi
e
ve
a
n a
c
c
ur
a
c
y of
95%
in
i
de
nt
if
y
in
g i
ns
ta
nc
e
s
of
di
s
tr
a
c
ti
on.
O
n
th
e
ot
he
r
ha
nd,
J
a
m
s
he
e
d
e
t
al
.
[
16]
us
e
d
a
C
N
N
-
ba
s
e
d
m
e
th
od
f
or
de
ve
lo
pi
ng
dr
iv
e
r
a
c
ti
on
c
la
s
s
if
ie
r
s
.
T
he
ir
r
e
s
e
a
r
c
h
s
uc
c
e
s
s
f
ul
ly
c
la
s
s
if
ie
d
di
s
tr
a
c
te
d
dr
iv
e
r
s
in
to
te
n
c
a
te
gor
ie
s
w
it
h
a
n
a
c
c
ur
a
c
y
of
97%
,
ut
il
iz
in
g
th
e
S
ta
te
F
a
r
m
da
ta
s
e
t.
F
ur
th
e
r
m
or
e
,
P
a
nw
a
r
e
t
al
.
[
17]
pr
opos
e
d
a
de
e
p
l
e
a
r
ni
ng
c
onvolut
io
na
l
m
ode
l
to
ta
c
kl
e
di
s
tr
a
c
ti
on
a
nd
dr
ow
s
in
e
s
s
dur
in
g
dr
iv
in
g.
T
he
ir
c
o
m
pr
e
he
ns
iv
e
a
ppr
oa
c
h,
w
it
h
a
n
a
c
c
ur
a
c
y
of
99.95%
,
c
a
te
gor
iz
e
d
va
r
io
us
in
a
tt
e
nt
io
n
in
s
ta
n
c
e
s
,
s
uc
h
a
s
un
-
a
tt
e
nt
iv
e
dr
iv
in
g,
m
obi
le
phone
us
a
ge
,
f
r
e
que
nt
ya
w
ni
ng,
a
nd
s
le
e
pi
ng.
L
i
e
t
al
.
[
18]
pr
opos
e
a
nove
l
a
lg
or
it
hm
f
or
de
te
c
ti
ng
m
a
nua
l
di
s
tr
a
c
ti
ons
a
m
ong
dr
iv
e
r
s
.
T
hi
s
a
lg
or
it
hm
c
om
pr
is
e
s
two
m
odul
e
s
:
th
e
f
ir
s
t
pr
e
di
c
ts
bounding
boxe
s
f
or
th
e
dr
iv
e
r
’
s
r
ig
ht
ha
nd
a
nd
r
ig
ht
e
a
r
f
r
om
R
G
B
im
a
ge
s
,
w
hi
le
th
e
s
e
c
ond
m
odul
e
c
la
s
s
if
ie
s
th
e
ty
pe
of
di
s
tr
a
c
ti
on
ba
s
e
d
on
th
e
s
e
bounding
boxe
s
.
T
he
y
tr
a
in
e
d
a
nd
te
s
te
d
th
e
a
lg
or
it
hm
on
a
d
a
ta
s
e
t
c
ons
is
ti
ng
of
106,677
f
r
a
m
e
s
e
xt
r
a
c
te
d
f
r
om
vi
de
os
c
a
pt
ur
e
d
dur
in
g
s
im
ul
a
te
d
dr
iv
in
g
s
e
s
s
io
ns
w
it
h
twe
nt
y
pa
r
ti
c
ip
a
nt
s
.
N
ot
a
bl
y,
th
e
f
r
a
m
e
w
or
k
yi
e
ld
e
d
a
n
F
1
-
s
c
or
e
of
0.84,
0.69,
a
nd
0.82,
f
o
r
c
la
s
s
if
yi
ng
nor
m
a
l
dr
iv
in
g,
to
uc
hs
c
r
e
e
n
in
te
r
a
c
ti
on,
a
nd
phone
c
onve
r
s
a
ti
on
r
e
s
pe
c
ti
ve
ly
.
M
or
e
ove
r
,
in
th
e
c
a
s
e
of
[
19]
,
th
e
r
a
ndom
f
o
r
e
s
t
-
ba
s
e
d
a
ppr
oa
c
h
w
a
s
im
pl
e
m
e
nt
e
d
a
c
r
os
s
phys
io
lo
gi
c
a
l
f
unc
ti
ona
l
va
r
ia
bl
e
s
to
ta
ke
dr
iv
e
r
s
’
s
tr
e
s
s
le
ve
l
s
a
nd
a
c
c
or
di
ngl
y
c
a
te
gor
iz
e
th
e
m
.
T
he
a
na
ly
s
is
w
a
s
pe
r
f
or
m
e
d
on
e
xp
e
r
im
e
nt
a
l
da
ta
e
xt
r
a
c
te
d
f
r
om
th
e
dr
iv
e
db
op
e
n
da
t
a
ba
s
e
.
T
he
phys
io
lo
gi
c
a
l
m
e
a
s
ur
e
m
e
nt
s
of
in
te
r
e
s
t
a
r
e
e
le
c
tr
ode
r
m
a
l
a
c
ti
vi
ty
c
a
pt
ur
e
d
on
th
e
dr
iv
e
r
’
s
le
f
t
ha
nd
a
nd
f
oot
,
e
le
c
tr
om
yogr
a
m
(
E
M
G
)
,
r
e
s
pi
r
a
ti
on,
a
nd
he
a
r
t
r
a
te
;
th
e
y
a
c
hi
e
ve
d
a
n
a
c
c
ur
a
c
y
of
81%
.
A
c
c
or
di
ng
to
L
e
one
e
t
al
.
[
20]
a
s
ys
t
e
m
w
a
s
de
s
ig
ne
d
ba
s
e
d
on
a
lo
w
-
c
o
s
t
c
a
m
e
r
a
to
de
te
c
t
dr
iv
e
r
r
oa
d
r
a
ge
th
r
ough
a
m
e
ti
c
ul
ous
a
na
ly
s
i
s
of
th
e
dr
iv
e
r
'
s
f
a
c
i
a
l
e
xpr
e
s
s
io
ns
.
W
h
a
t
s
e
ts
th
i
s
a
ppr
oa
c
h
a
pa
r
t
is
it
s
s
ophi
s
ti
c
a
te
d
de
c
is
io
n
-
m
a
ki
ng
s
tr
a
te
gy,
w
hi
c
h
r
e
li
e
s
on
th
e
te
m
por
a
l
c
oh
e
r
e
nc
e
of
f
a
c
ia
l
e
xpr
e
s
s
io
ns
c
a
te
gor
iz
e
d
a
s
“
a
nge
r
”
a
nd
“
di
s
gus
t.
”
T
hi
s
m
e
th
odol
ogy
yi
e
ld
e
d
a
n
a
c
c
ur
a
c
y of
84.56%
w
he
n
e
m
pl
oyi
ng
th
e
s
uppor
t
ve
c
to
r
m
a
c
hi
n
e
(
S
V
M
)
a
lg
or
it
hm
.
V
a
r
io
us
s
e
ns
or
s
h
a
ve
be
e
n
e
m
pl
oye
d
to
de
te
c
t
di
f
f
e
r
e
nt
f
or
m
s
of
dr
iv
e
r
in
a
tt
e
nt
io
n,
in
c
lu
di
ng
di
s
tr
a
c
ti
on,
dr
ow
s
in
e
s
s
,
f
a
ti
gue
,
a
nd
e
m
ot
io
na
l
s
ta
te
.
P
hy
s
io
lo
gi
c
a
l
a
ppr
oa
c
he
s
in
vol
ve
di
s
c
r
e
te
ly
pl
a
c
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
r
iv
e
r
i
nat
te
nt
io
n
de
te
c
ti
on s
y
s
te
m
us
in
g m
ul
ti
-
ta
s
k
c
a
s
c
ade
d
c
onv
ol
ut
io
nal
... (
A
bde
lf
e
tt
ah Soult
ana)
4251
s
e
ns
or
s
on
th
e
dr
iv
e
r
’
s
body
to
e
xt
r
a
c
t
s
ig
na
ls
s
uc
h
a
s
E
E
G
,
e
le
c
tr
oc
a
r
di
ogr
a
m
(
E
C
G
)
,
or
E
M
G
.
A
not
he
r
a
ppr
oa
c
h
r
e
li
e
s
on
dr
iv
in
g
pe
r
f
or
m
a
nc
e
in
di
c
a
to
r
s
,
s
uc
h
a
s
m
oni
to
r
in
g
la
ne
de
pa
r
tu
r
e
,
pe
da
l
a
c
ti
vi
ty
,
a
nd
a
c
c
e
le
r
a
to
r
us
a
ge
.
H
ow
e
v
e
r
,
th
e
m
os
t
w
id
e
ly
a
dopt
e
d
a
ppr
oa
c
h
by
dr
iv
e
r
s
is
vi
s
io
n
-
ba
s
e
d,
w
hi
c
h
ut
il
iz
e
s
c
a
m
e
r
a
s
.
T
hi
s
non
-
in
tr
us
iv
e
na
tu
r
e
m
a
ke
s
it
a
pr
e
f
e
r
r
e
d
c
hoi
c
e
,
a
s
it
doe
s
not
r
e
qui
r
e
phys
i
c
a
l
c
ont
a
c
t
w
it
h
th
e
dr
iv
e
r
a
nd t
hus
e
ns
ur
e
s
a
hi
gh
e
r
l
e
ve
l
of
c
om
f
or
t
a
nd a
c
c
e
pt
a
nc
e
.
T
he
li
te
r
a
tu
r
e
c
ont
a
in
s
a
w
e
a
lt
h
of
pr
opos
a
ls
f
or
de
te
c
ti
ng
dr
iv
e
r
in
a
tt
e
nt
io
n,
a
s
e
xt
e
ns
iv
e
ly
de
s
c
r
ib
e
d
in
our
pr
e
vi
ou
s
w
or
k
[
21]
.
T
he
s
e
a
ppr
oa
c
h
e
s
h
a
ve
c
e
r
ta
in
ly
pr
ovi
de
d
v
a
lu
a
bl
e
in
s
ig
ht
s
in
to
unde
r
s
ta
ndi
ng
a
nd
a
ddr
e
s
s
in
g
pa
r
ti
c
ul
a
r
f
or
m
s
of
in
a
tt
e
nt
io
n.
H
ow
e
ve
r
,
th
e
s
e
s
tu
di
e
s
f
oc
us
in
di
vi
dua
ll
y
on
s
pe
c
if
ic
a
s
pe
c
t
s
of
dr
iv
e
r
i
na
tt
e
nt
io
n,
s
uc
h a
s
di
s
tr
a
c
ti
on, f
a
ti
gue
, dr
ow
s
in
e
s
s
, or
e
m
ot
io
na
l
s
ta
te
. R
e
c
ogni
z
in
g
th
e
s
e
f
a
c
to
r
s
in
di
vi
dua
ll
y
m
a
y
le
a
d
to
in
c
om
pl
e
te
or
le
s
s
a
c
c
ur
a
te
a
s
s
e
s
s
m
e
nt
s
of
dr
iv
e
r
in
a
tt
e
nt
io
n.
A
n
a
ppr
oa
c
h
th
a
t
ta
ke
s
in
to
a
c
c
ount
th
e
d
e
te
c
ti
on
of
di
f
f
e
r
e
nt
in
a
tt
e
nt
io
n
f
a
c
to
r
s
c
oul
d
pr
ovi
de
a
m
or
e
c
om
pl
e
te
a
nd nua
nc
e
d unde
r
s
ta
ndi
ng of
t
he
dr
iv
e
r
’
s
c
ondi
ti
on.
T
h
is
r
e
s
e
a
r
c
h
a
im
s
t
o
pr
o
po
s
e
a
n
ov
e
l
s
y
s
t
e
m
f
or
d
e
t
e
c
ti
ng
dr
iv
e
r
in
a
t
te
nt
i
on,
t
a
r
ge
ti
n
g
f
our
ke
y
a
s
p
e
c
t
s
:
di
s
tr
a
c
ti
on,
f
a
t
ig
u
e
,
dr
o
w
s
in
e
s
s
,
a
nd
e
m
ot
i
on
s
.
T
hi
s
s
y
s
te
m
i
s
s
tr
uc
tu
r
e
d
a
c
r
o
s
s
s
ix
l
e
v
e
l
s
:
pe
r
c
e
pt
io
n
,
f
a
c
ia
l
f
e
a
t
ur
e
s
de
t
e
c
ti
on
,
tr
a
c
ki
n
g
dr
i
ve
r
u
s
in
g
pr
e
-
tr
a
i
ne
d
d
e
e
p
le
a
r
n
in
g
m
o
de
ls
,
in
a
t
te
nt
i
on
d
e
t
e
c
ti
on,
r
i
s
k
e
s
ti
m
a
t
io
n
,
a
nd
a
l
e
r
t
.
E
a
c
h
ty
pe
of
d
is
tr
a
c
t
io
n
w
i
ll
b
e
a
d
dr
e
s
s
e
d
b
y
a
di
s
t
in
c
t
m
od
e
l
de
s
i
gn
e
d
to
c
a
pt
u
r
e
c
h
a
r
a
c
t
e
r
i
s
t
ic
s
s
ig
na
l
s
a
s
s
o
c
i
a
t
e
d
w
i
th
th
e
s
e
s
t
a
te
s
of
i
n
a
tt
e
n
ti
o
n.
B
y c
om
b
in
i
ng t
h
e
s
e
f
ou
r
m
od
e
l
s
,
th
e
pr
o
po
s
e
d
s
y
s
t
e
m
w
i
ll
a
c
c
ur
a
te
ly
a
s
s
e
s
s
th
e
l
e
v
e
l
of
dr
iv
e
r
in
a
tt
e
n
ti
o
n
i
n
r
e
a
l
-
ti
m
e
u
s
in
g
a
r
is
k
c
a
lc
ul
a
ti
on.
T
h
e
s
y
s
t
e
m
a
do
pt
s
a
n e
nt
ir
e
ly
non
-
in
t
r
u
s
iv
e
a
p
pr
o
a
c
h,
r
e
l
yi
n
g
o
n
im
a
g
e
pr
o
c
e
s
s
in
g f
r
om
t
w
o
c
a
m
e
r
a
s
i
n
s
t
a
l
le
d
in
t
h
e
c
a
r
t
o
m
oni
to
r
th
e
dr
iv
e
r
'
s
s
ta
te
. O
ne
of
ou
r
k
e
y
c
ont
r
ib
ut
i
on
s
l
ie
s
i
n
t
h
e
s
e
l
e
c
ti
o
n
of
a
n
opt
im
a
l
a
l
gor
i
th
m
f
or
l
e
v
e
l
2 of
th
e
s
y
s
t
e
m
,
w
h
ic
h
i
nvo
lv
e
s
f
a
c
i
a
l
f
e
a
t
ur
e
d
e
t
e
c
t
io
n
a
nd
tr
a
c
ki
ng
.
T
o
th
is
e
nd
,
w
e
c
o
ndu
c
t
e
d
a
r
ig
or
ou
s
c
om
pa
r
a
t
iv
e
a
na
ly
s
i
s
of
s
e
v
e
r
a
l
s
ta
te
-
of
-
t
he
-
a
r
t
a
l
gor
i
th
m
s
,
in
c
lu
d
in
g
m
ul
ti
-
t
a
s
k
c
a
s
c
a
d
e
d
c
on
vol
ut
i
on
a
l
ne
t
w
or
k
s
(
M
T
C
N
N
)
,
H
O
G
w
i
th
d
li
b
l
ib
r
a
r
y
,
a
n
d
H
a
a
r
f
e
a
tu
r
e
-
b
a
s
e
d
c
a
s
c
a
d
e
c
la
s
s
if
i
e
r
s
.
T
hi
s
s
te
p
i
s
c
r
uc
i
a
l
f
or
e
xt
r
a
c
ti
ng
t
he
m
out
h
r
e
gi
o
n
to
d
e
t
e
c
t
s
ig
ns
of
f
a
t
ig
u
e
,
w
h
il
e
s
i
gn
s
of
dr
ow
s
i
ne
s
s
a
r
e
i
de
nt
if
ie
d
by
a
n
a
ly
z
i
ng
t
he
e
y
e
r
e
gi
on.
A
ddi
ti
o
na
ll
y
, t
he
s
ys
te
m
e
xt
r
a
c
t
s
t
h
e
e
n
ti
r
e
f
a
c
e
r
e
gi
o
n t
o
de
te
c
t
a
nd i
nt
e
r
pr
e
t
f
a
c
i
a
l
e
x
pr
e
s
s
i
on
s
.
T
h
e
pr
op
o
s
e
d s
y
s
t
e
m
a
im
s
t
o
f
i
ll
t
h
e
g
a
p
l
e
f
t
by
pr
e
vi
ou
s
a
p
pr
o
a
c
he
s
by
d
e
t
e
c
ti
ng a
n
y
ty
pe
of
dr
iv
e
r
i
n
a
tt
e
nt
io
n
a
nd
to
i
m
pr
ov
e
r
o
a
d
s
a
f
e
t
y
by
pr
o
vi
d
in
g
r
e
l
e
v
a
nt
a
le
r
ts
c
or
r
e
s
po
ndi
n
g t
o
t
he
a
s
s
e
s
s
e
d
l
e
v
e
l
of
r
is
k.
T
h
e
p
a
p
e
r
i
s
or
g
a
ni
z
e
d a
s
f
ol
l
ow
s
:
s
e
c
ti
on
2 pr
e
s
e
nt
s
a
n
ov
e
l
s
y
s
te
m
f
or
dr
i
ve
r
in
a
tt
e
n
ti
o
n
m
o
ni
t
or
i
ng.
S
e
c
ti
o
n
3 i
s
de
vo
te
d t
o t
he
o
pt
im
a
l
a
lg
o
r
it
hm
pr
op
o
s
e
d f
or
f
a
c
ia
l
f
e
a
tu
r
e
s
d
e
t
e
c
ti
on
a
n
d f
e
a
tu
r
e
e
x
tr
a
c
t
io
n
,
w
hi
c
h
is
M
T
C
N
N
.
M
ovi
ng
o
n
to
s
e
c
t
io
n
4,
a
c
om
pr
e
h
e
n
s
i
ve
d
is
c
u
s
s
io
n
i
s
pr
e
s
e
n
te
d,
of
f
e
r
i
ng
i
ns
ig
ht
s
in
t
o
th
e
r
a
ti
on
a
l
e
b
e
h
in
d
s
e
le
c
ti
ng
th
e
M
T
C
N
N
a
lg
or
it
hm
.
A
ddi
t
io
na
ll
y
, t
hi
s
s
e
c
ti
on
ou
tl
i
ne
s
pot
e
nt
i
a
l
a
v
e
n
ue
s
f
or
f
ut
ur
e
r
e
f
in
e
m
e
n
t
a
n
d d
e
p
lo
ym
e
nt
of
th
e
s
y
s
te
m
. F
in
a
ll
y,
s
e
c
ti
on
5
e
n
c
a
ps
ul
a
t
e
s
t
h
e
p
a
p
e
r
w
it
h
a
c
on
c
l
udi
ng
s
um
m
a
r
y.
2.
N
O
V
E
L
S
Y
S
T
E
M
F
O
R
D
R
I
V
E
R
I
N
A
T
T
E
N
T
I
O
N
M
O
N
I
T
O
R
I
N
G
I
n
t
hi
s
r
e
s
e
a
r
c
h,
o
ur
pr
im
a
r
y
f
o
c
u
s
r
e
vol
ve
s
a
r
ou
nd
th
e
de
t
e
c
ti
o
n
o
f
dr
iv
e
r
in
a
tt
e
n
ti
v
e
n
e
s
s
,
th
e
ge
n
e
r
a
ti
on
of
a
le
r
t
m
e
s
s
a
g
e
s
,
a
n
d
pr
o
vi
d
in
g
pr
o
a
c
t
iv
e
a
s
s
i
s
t
a
n
c
e
to
th
e
dr
i
ve
r
.
T
o
a
c
hi
e
v
e
th
e
s
e
o
bj
e
c
ti
v
e
s
,
w
e
ha
v
e
i
nn
ov
a
ti
ve
ly
d
e
v
i
s
e
d
a
c
o
m
pr
e
h
e
n
s
i
ve
s
i
x
-
l
a
y
e
r
e
d
a
r
c
hi
t
e
c
tu
r
e
s
pe
c
if
ic
a
ll
y
t
a
il
or
e
d
f
or
dr
i
ve
r
in
a
tt
e
n
ti
o
n
de
t
e
c
ti
on,
c
om
pl
e
m
e
nt
e
d
by
t
he
im
pl
e
m
e
nt
a
ti
on
of
a
p
e
r
s
o
na
l
i
z
e
d
a
s
s
i
s
t
a
n
c
e
s
y
s
t
e
m
,
a
s
vi
s
ua
ll
y
d
e
p
ic
t
e
d
i
n
F
ig
ur
e
1.
O
u
r
s
ix
-
la
ye
r
e
d
a
r
c
hi
t
e
c
tu
r
e
i
s
d
e
s
ig
n
e
d
to
e
f
f
i
c
i
e
n
tl
y
de
te
c
t
dr
i
ve
r
i
na
tt
e
nt
i
on
.
E
a
c
h
l
a
y
e
r
s
e
r
v
e
s
a
di
s
ti
n
c
t
pur
po
s
e
:
s
e
n
s
or
s
c
ol
le
c
t
di
v
e
r
s
e
da
ta
, f
a
c
i
a
l
f
e
a
tu
r
e
e
x
tr
a
c
ti
on i
de
nt
if
ie
s
ke
y f
a
c
i
a
l
e
l
e
m
e
nt
s
, t
h
e
t
r
a
c
ki
ng
s
t
a
g
e
e
ns
ur
e
s
c
o
nt
i
nuo
u
s
tr
a
c
k
in
g
of
f
e
a
tu
r
e
s
,
i
na
tt
e
nt
io
n
de
t
e
c
ti
on
vi
a
a
t
e
m
por
a
l
w
in
dow
,
r
i
s
k
e
s
ti
m
a
ti
on
a
s
s
e
s
s
e
s
t
h
e
l
e
v
e
l
of
r
i
s
k,
a
nd
t
he
a
l
e
r
t
la
ye
r
i
s
s
u
e
s
w
a
r
ni
ng
s
i
n
c
a
s
e
of
c
r
it
i
c
a
l
r
i
s
k
.
T
hi
s
m
o
du
la
r
a
p
pr
o
a
c
h
a
ll
ow
s
f
or
in
d
e
p
e
nd
e
nt
e
nh
a
n
c
e
m
e
nt
s
a
n
d
a
d
va
nc
e
m
e
n
ts
,
s
tr
e
a
m
l
in
in
g
de
ve
lo
p
m
e
nt
a
n
d
m
a
i
nt
e
n
a
n
c
e
e
f
f
or
t
s
.
2.1. P
e
r
c
e
p
t
io
n
l
aye
r
I
n
th
e
in
it
ia
l
la
ye
r
,
w
e
s
tr
a
te
gi
c
a
ll
y
pos
it
io
n
two
c
a
m
e
r
a
s
:
one
on
th
e
dr
iv
e
r
’
s
s
id
e
to
c
om
pr
e
he
ns
iv
e
ly
c
a
pt
ur
e
s
e
c
onda
r
y t
a
s
ks
a
nd
a
not
he
r
di
r
e
c
te
d f
r
ont
a
ll
y t
o
f
oc
us
on f
a
c
ia
l
f
e
a
tu
r
e
s
. I
t
i
s
w
or
th
not
in
g
th
a
t
th
e
a
dopt
e
d
a
ppr
oa
c
h
is
vi
s
ua
ll
y
b
a
s
e
d,
r
e
ly
in
g
on
c
a
m
e
r
a
u
s
a
ge
.
H
ow
e
ve
r
,
th
is
a
r
c
hi
te
c
tu
r
e
r
e
m
a
in
s
ope
n
to
in
c
lu
de
ot
he
r
s
e
ns
or
s
w
he
n
it
is
ne
e
de
d.
F
or
e
xa
m
pl
e
,
e
na
bl
in
g
phys
io
lo
gi
c
a
l
m
e
a
s
ur
e
s
s
uc
h
a
s
E
E
G
,
E
C
G
, or
E
M
G
a
nd i
nt
e
gr
a
ti
ng othe
r
pr
e
di
c
ti
ve
m
ode
ls
t
o a
le
r
t
th
e
dr
iv
e
r
w
he
n a
pr
obl
e
m
a
r
is
e
s
.
2.2. F
ac
ia
l
f
e
at
u
r
e
s
e
xt
r
ac
t
io
n
l
aye
r
M
o
vi
n
g
on
to
th
e
s
e
c
on
d
l
a
ye
r
,
t
he
s
t
e
p
of
f
a
c
i
a
l
f
e
a
t
ur
e
d
e
t
e
c
ti
on
a
nd
e
x
tr
a
c
t
io
n
i
s
a
ppl
ie
d
t
o
th
e
s
tr
e
a
m
of
i
m
a
ge
s
f
r
om
c
a
m
e
r
a
t
w
o
,
s
im
u
lt
a
n
e
ou
s
ly
e
xt
r
a
c
ti
ng
v
i
ta
l
f
a
c
i
a
l
c
ha
r
a
c
t
e
r
i
s
t
ic
s
l
ik
e
th
e
f
a
c
e
, e
y
e
s
, a
n
d
m
out
h.
I
t
i
s
c
r
uc
ia
l
to
e
m
ph
a
s
i
z
e
t
ha
t
im
a
g
e
s
f
r
om
c
a
m
e
r
a
on
e
b
ypa
s
s
th
i
s
s
t
e
p
, a
s
th
e
y
a
r
e
u
s
e
d
i
n t
h
e
ir
e
nt
ir
e
t
y
f
or
id
e
nt
if
yi
ng
s
e
c
ond
a
r
y
ta
s
k
s
,
e
l
im
i
na
ti
n
g
th
e
n
e
e
d
f
or
im
a
g
e
s
e
gm
e
n
ta
t
io
n.
T
h
e
m
a
r
k
e
t
of
f
e
r
s
a
p
le
th
o
r
a
of
c
om
m
e
r
c
i
a
l
a
n
d
no
n
-
c
om
m
e
r
c
i
a
l
f
a
c
e
d
e
t
e
c
ti
o
n
a
nd
a
li
g
nm
e
nt
a
lg
or
it
hm
s
;
h
ow
e
v
e
r
,
du
e
to
t
im
e
c
on
s
tr
a
in
t
s
,
e
v
a
lu
a
t
in
g
th
e
m
i
s
im
pr
a
c
ti
c
a
l.
T
h
e
r
e
f
or
e
,
our
e
va
lu
a
ti
on
w
il
l
f
oc
u
s
on
t
he
m
o
s
t
e
m
pl
oy
e
d
a
lg
or
it
h
m
s
i
n
t
he
c
om
m
e
r
c
i
a
l
s
e
c
to
r
,
th
o
s
e
w
it
h
w
i
de
s
pr
e
a
d
pop
ul
a
r
it
y,
a
n
d
o
pe
n
-
s
our
c
e
im
p
le
m
e
nt
a
ti
on
s
f
r
e
e
ly
a
v
a
il
a
bl
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4249
-
4262
4252
C
o
ns
id
e
r
in
g
th
e
d
ua
l
ta
s
k
s
of
f
a
c
e
d
e
t
e
c
ti
on
a
nd
f
a
c
e
a
li
g
nm
e
nt
,
our
a
s
s
e
s
s
m
e
nt
w
il
l
c
e
nt
e
r
on
H
a
a
r
C
a
s
c
a
d
e
-
O
pe
nC
V
[
22]
,
H
oG
-
D
li
b
[
2
3]
,
a
nd
M
T
C
N
N
[
2
4]
.
T
h
e
s
e
a
l
gor
i
th
m
s
h
a
v
e
ga
r
n
e
r
e
d
e
x
te
ns
iv
e
a
d
op
ti
o
n
a
nd
h
a
v
e
pr
ov
e
n
to
b
e
e
f
f
e
c
t
iv
e
i
n
a
c
c
o
m
pl
i
s
hi
ng
bot
h
obj
e
c
ti
v
e
s
.
T
h
e
y
ha
ve
e
m
e
r
ge
d
a
s
h
ig
h
ly
vi
a
bl
e
opt
io
n
s
f
or
r
e
a
l
-
w
or
ld
a
pp
li
c
a
t
io
n
s
a
c
r
o
s
s
v
a
r
io
u
s
d
om
a
in
s
by
s
tr
i
ki
n
g
a
b
a
l
a
n
c
e
b
e
t
w
e
e
n
a
c
c
ur
a
c
y
a
n
d
e
f
f
i
c
i
e
n
c
y.
F
ig
ur
e
1. T
he
f
lo
w
c
ha
r
t
of
t
he
pr
opos
e
d dr
iv
e
r
i
na
tt
e
nt
io
n m
oni
to
r
in
g s
ys
te
m
2.3. T
r
ac
k
in
g u
s
in
g p
r
e
-
t
r
ai
n
e
d
m
od
e
ls
I
n
th
e
th
ir
d
la
ye
r
,
th
e
tr
a
c
ki
ng
s
ta
ge
ut
il
iz
e
s
de
e
p
le
a
r
ni
ng
m
ode
ls
tr
a
in
e
d
on
e
xt
e
ns
iv
e
da
ta
s
e
ts
of
f
li
ne
.
T
he
in
it
ia
l
m
ode
l
pr
e
di
c
ts
s
e
c
onda
r
y
ta
s
ks
us
in
g
im
a
ge
r
y
f
r
om
c
a
m
e
r
a
1.
T
he
s
e
s
e
c
ond
a
r
y
ta
s
ks
e
nc
om
pa
s
s
a
c
ti
vi
ti
e
s
li
ke
te
xt
in
g,
c
a
ll
in
g,
r
e
a
c
hi
ng_b
e
hi
nd,
ha
ir
_m
a
ke
up,
dr
in
ki
ng,
a
dj
us
ti
ng_r
a
di
o,
a
nd
ta
lk
in
g_t
o_pa
s
s
e
nge
r
.
A
not
he
r
de
e
p
-
le
a
r
ni
ng
m
ode
l
c
la
s
s
if
ie
s
f
a
c
ia
l
e
xpr
e
s
s
io
n
s
in
to
s
e
ve
n
di
s
ti
nc
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
r
iv
e
r
i
nat
te
nt
io
n
de
te
c
ti
on s
y
s
te
m
us
in
g m
ul
ti
-
ta
s
k
c
a
s
c
ade
d
c
onv
ol
ut
io
nal
... (
A
bde
lf
e
tt
ah Soult
ana)
4253
c
a
te
gor
ie
s
:
ha
ppi
ne
s
s
,
f
e
a
r
,
a
nge
r
,
s
a
dne
s
s
,
ne
ut
r
a
li
ty
,
s
ur
pr
is
e
,
a
nd
di
s
gus
t.
A
ddi
ti
ona
ll
y,
a
s
e
pa
r
a
te
de
e
p
le
a
r
ni
ng mode
l
m
oni
to
r
s
e
ye
s
ta
te
(
ope
n or
c
lo
s
e
d)
, w
hi
le
a
not
he
r
m
ode
l
tr
a
c
ks
m
out
h
m
ove
m
e
nt
t
o
a
s
c
e
r
ta
in
w
he
th
e
r
th
e
dr
iv
e
r
is
ya
w
ni
ng
or
not
.
D
e
e
p
le
a
r
ni
ng
m
ode
ls
f
or
de
te
c
ti
ng
dr
iv
e
r
di
s
tr
a
c
ti
on,
dr
ow
s
in
e
s
s
,
f
a
ti
gue
,
a
nd
e
m
ot
io
ns
gr
e
a
tl
y
be
ne
f
it
f
r
om
th
e
us
e
of
C
N
N
.
C
N
N
a
r
e
pow
e
r
f
ul
to
ol
s
f
or
th
e
s
e
ta
s
ks
a
s
th
e
y
a
r
e
s
pe
c
ia
ll
y
de
s
ig
ne
d
to
a
ut
om
a
ti
c
a
ll
y
e
xt
r
a
c
t
m
e
a
ni
ngf
ul
f
e
a
tu
r
e
s
f
r
om
vi
s
ua
l
da
ta
,
s
uc
h
a
s
im
a
ge
s
or
vi
de
os
.
T
h
e
y
a
r
e
c
a
pa
bl
e
of
id
e
nt
if
yi
ng
c
om
pl
e
x
pa
tt
e
r
ns
,
r
e
s
is
ti
ng
va
r
ia
ti
ons
in
obj
e
c
t
pos
it
io
n
a
nd
s
iz
e
w
it
hi
n
th
e
im
a
ge
,
a
nd
le
a
r
ni
ng
f
e
a
tu
r
e
s
a
t
di
f
f
e
r
e
nt
le
ve
ls
of
a
b
s
tr
a
c
ti
on.
T
hi
s
m
a
ke
s
th
e
m
na
tu
r
a
l
c
hoi
c
e
s
f
or
de
te
c
ti
ng
di
s
tr
a
c
ti
ng
be
ha
vi
or
s
,
s
ig
ns
of
dr
ow
s
in
e
s
s
or
f
a
ti
gue
,
a
s
w
e
ll
a
s
v
a
r
io
us
e
m
ot
io
na
l
e
xpr
e
s
s
io
ns
on
th
e
dr
iv
e
r
’
s
f
a
c
e
.
T
he
s
e
m
ode
ls
unde
r
go
pe
r
io
di
c
upda
te
s
b
a
s
e
d
on
dr
iv
e
r
im
a
ge
r
y,
a
ll
ow
in
g
f
or
c
ont
in
uous
im
pr
ove
m
e
nt
i
n pe
r
f
or
m
a
nc
e
a
nd pe
r
s
ona
li
z
e
d a
da
pt
a
ti
on t
o i
n
di
vi
dua
l
dr
iv
e
r
pr
of
il
e
s
.
2.4. I
n
at
t
e
n
t
io
n
d
e
t
e
c
t
io
n
l
aye
r
I
n
th
e
f
our
th
la
y
e
r
, dr
iv
e
r
in
a
t
te
nt
i
on
d
e
te
c
ti
on
ta
ke
s
pl
a
c
e
, e
m
pl
oyi
ng a
t
e
m
por
a
l
w
in
do
w
to
e
v
a
lu
a
te
va
r
io
u
s
f
or
m
s
of
in
a
tt
e
nt
io
n. R
e
l
yi
ng s
ol
e
ly
on a
s
in
gl
e
im
a
ge
(
f
r
a
m
e
)
d
oe
s
not
r
e
li
a
bl
y
a
ll
ow
f
or
th
e
de
t
e
c
ti
on
of
dr
iv
e
r
i
na
tt
e
nt
io
n
. I
ns
t
e
a
d
, i
t
is
m
or
e
e
f
f
e
c
t
iv
e
t
o b
a
s
e
t
he
a
na
ly
s
i
s
on t
he
n
um
be
r
of
f
r
a
m
e
s
p
e
r
s
e
c
ond
. T
h
is
m
e
a
n
s
th
a
t
by
a
n
a
ly
z
in
g
m
ul
ti
pl
e
im
a
g
e
s
p
e
r
s
e
c
o
nd,
a
m
or
e
c
o
m
pr
e
he
n
s
iv
e
a
n
d
dyn
a
m
ic
vi
e
w
of
th
e
dr
iv
e
r
’
s
be
ha
v
io
r
c
a
n
b
e
obt
a
i
ne
d.
H
ow
e
ve
r
,
th
i
s
r
e
qui
r
e
s
th
e
e
s
ta
bl
i
s
h
m
e
nt
of
a
t
hr
e
s
hol
d,
w
hi
c
h
i
s
a
pr
e
d
e
te
r
m
i
ne
d
va
lu
e
a
t
w
hi
c
h
in
a
tt
e
nt
io
n
is
pr
e
s
e
nt
.
T
hi
s
th
r
e
s
h
ol
d
is
c
r
uc
i
a
l
b
e
c
a
us
e
it
de
f
in
e
s
th
e
poi
nt
a
t
w
hi
c
h
a
s
p
e
c
if
i
c
s
ta
t
e
of
in
a
tt
e
nt
io
n
c
a
n
b
e
c
o
nf
ir
m
e
d.
F
or
in
s
ta
nc
e
,
if
th
e
nu
m
be
r
of
f
r
a
m
e
s
di
s
pl
a
yi
ng
s
i
gns
of
in
a
tt
e
nt
io
n
s
ur
pa
s
s
e
s
th
i
s
e
s
ta
bl
i
s
h
e
d
th
r
e
s
hol
d,
one
c
a
n
th
e
n c
on
c
lu
d
e
th
a
t
th
e
t
a
r
ge
t
e
d
f
or
m
of
in
a
tt
e
n
ti
on
i
s
pr
e
s
e
nt
. T
he
a
da
pt
i
ve
t
hr
e
s
hol
di
ng
pr
o
c
e
s
s
in
c
or
por
a
t
e
s
a
c
oun
te
r
to
m
oni
t
o
r
a
s
pe
c
if
ic
num
be
r
of
s
uc
c
e
s
s
i
ve
f
r
a
m
e
s
th
a
t
s
a
ti
s
f
y
th
e
c
r
it
e
r
ia
be
f
or
e
is
s
ui
n
g
a
w
a
r
ni
n
g.
F
or
e
xa
m
p
le
,
S
ha
ke
e
l
e
t
al
.
[
2
5]
in
te
gr
a
te
d
a
t
hr
e
s
hol
d
m
e
c
h
a
ni
s
m
:
if
th
e
c
l
a
s
s
if
ie
r
c
on
s
i
s
te
nt
l
y
id
e
nt
if
ie
s
te
n
c
o
ns
e
c
u
ti
ve
in
s
ta
nc
e
s
of
c
lo
s
e
d
e
ye
s
,
t
hi
s
ob
s
e
r
va
ti
on
s
ugg
e
s
t
s
th
a
t
th
e
in
di
vi
dua
l
i
s
e
xhi
bi
ti
ng
s
ig
n
s
of
dr
o
w
s
in
e
s
s
.
R
a
f
id
e
t
al
.
[
26]
e
s
t
a
bl
i
s
he
d
a
s
pe
c
if
ic
c
r
i
te
r
io
n
f
or
de
te
r
m
in
i
ng dr
ow
s
in
e
s
s
:
if
t
he
c
la
s
s
if
ie
r
c
o
ns
i
s
t
e
nt
ly
i
d
e
nt
if
i
e
s
30
c
on
s
e
c
ut
iv
e
i
ns
ta
n
c
e
s
of
c
lo
s
e
d e
y
e
s
, t
hi
s
obs
e
r
va
ti
o
n
in
di
c
a
t
e
s
dr
ow
s
in
e
s
s
.
F
a
s
a
nm
a
de
e
t
al
.
[
27]
pr
opo
s
e
s
t
he
u
s
e
of
a
s
e
qu
e
nc
e
of
f
r
a
m
e
s
to
ga
ug
e
th
e
dur
a
ti
on of
dr
iv
e
r
di
s
tr
a
c
t
io
n.
T
he
c
odi
n
g w
a
s
c
o
nf
ig
ur
e
d
s
u
c
h t
ha
t
onc
e
a
t
hr
e
s
h
ol
d of
125
c
on
s
e
c
ut
iv
e
f
r
a
m
e
s
is
r
e
a
c
he
d
(
e
qui
va
le
nt
to
5
s
e
c
on
ds
)
,
a
c
la
s
s
if
i
c
a
ti
on
de
c
is
i
on
i
s
tr
ig
g
e
r
e
d.
I
n
th
e
li
t
e
r
a
tu
r
e
,
d
iv
e
r
s
e
th
r
e
s
ho
ld
s
ha
ve
be
e
n
s
ugg
e
s
t
e
d
f
or
de
t
e
c
ti
n
g
dr
iv
e
r
in
a
tt
e
nt
i
on.
T
h
e
r
e
f
or
e
,
d
e
te
r
m
in
i
ng
th
e
a
ppr
opr
i
a
te
n
um
be
r
of
c
ons
e
c
ut
iv
e
f
r
a
m
e
s
n
e
c
e
s
s
it
a
t
e
s
r
ig
or
ous
te
s
ti
n
g
a
nd
va
li
da
ti
on
th
r
o
ugh
r
e
a
li
s
ti
c
dr
i
vi
ng
s
c
e
n
a
r
io
s
.
T
hi
s
e
ns
ur
e
s
t
ha
t
t
he
c
ho
s
e
n t
hr
e
s
h
ol
d
e
f
f
e
c
ti
ve
ly
c
a
pt
ur
e
s
i
n
s
ta
nc
e
s
of
i
na
tt
e
nt
io
n
i
n pr
a
c
ti
c
a
l
dr
iv
in
g
s
it
u
a
ti
o
ns
.
2.5. Ris
k
e
s
t
im
at
io
n
l
aye
r
T
he
f
if
th
l
a
ye
r
i
nvo
lv
e
s
r
is
k c
a
lc
ul
a
ti
on ba
s
e
d on the
pr
e
di
c
te
d
c
la
s
s
e
s
. T
he
s
c
or
in
g di
c
ti
ona
r
ie
s
pl
a
y
a
pi
vot
a
l
r
ol
e
in
qua
nt
if
yi
ng
th
e
pot
e
nt
ia
l
r
is
k
o
f
dr
iv
e
r
in
a
tt
e
nt
io
n.
E
a
c
h
di
c
ti
ona
r
y
c
or
r
e
s
ponds
to
a
s
pe
c
if
ic
c
a
te
gor
y:
di
s
tr
a
c
ti
on,
dr
ow
s
in
e
s
s
,
ya
w
ni
ng,
a
nd
e
m
ot
io
ns
.
T
he
s
ys
te
m
e
f
f
e
c
ti
ve
ly
c
a
pt
ur
e
s
th
e
de
gr
e
e
of
in
a
tt
e
nt
io
n
r
is
k
a
s
s
o
c
ia
te
d
w
it
h
th
e
dr
iv
e
r
’
s
a
c
ti
on
s
by
a
s
s
ig
ni
ng
pr
e
de
f
in
e
d
s
c
or
e
s
to
v
a
r
io
us
be
ha
vi
or
s
a
nd
e
xpr
e
s
s
io
ns
w
it
hi
n
e
a
c
h
c
a
te
gor
y.
F
or
in
s
ta
nc
e
,
hi
gh
-
r
is
k
a
c
ti
v
it
ie
s
li
ke
te
xt
in
g
r
e
c
e
iv
e
hi
ghe
r
s
c
or
e
s
,
w
hi
le
ne
ut
r
a
l
or
pos
it
iv
e
e
m
ot
io
ns
yi
e
ld
no
a
ddi
ti
ona
l
r
is
k.
T
he
in
di
vi
dua
l
s
c
or
e
s
f
r
om
e
a
c
h
c
a
te
gor
y
a
r
e
th
e
n
c
om
bi
ne
d
to
c
a
lc
ul
a
te
th
e
gl
oba
l
in
a
tt
e
nt
io
n
r
is
k
s
c
or
e
,
pr
ovi
di
ng
a
c
om
pr
e
he
ns
iv
e
a
s
s
e
s
s
m
e
nt
of
th
e
dr
iv
e
r
’
s
a
tt
e
nt
iv
e
ne
s
s
.
T
hi
s
m
e
th
od
e
na
bl
e
s
a
nu
a
nc
e
d
und
e
r
s
ta
ndi
ng
of
th
e
dr
iv
e
r
’
s
s
ta
te
.
I
t
f
a
c
il
it
a
te
s
th
e
c
la
s
s
if
ic
a
ti
on
of
r
is
k
le
ve
ls
,
th
e
r
e
by
c
ont
r
ib
ut
in
g
to
th
e
de
ve
lo
pm
e
nt
of
r
obus
t
dr
iv
e
r
in
a
tt
e
nt
io
n
m
on
it
or
in
g
s
ys
te
m
s
a
nd
e
nha
n
c
in
g
r
oa
d
s
a
f
e
ty
.
T
he
s
c
or
in
g
s
ys
te
m
ut
il
iz
e
d
in
th
e
s
e
di
c
ti
ona
r
ie
s
dr
a
w
s
it
s
f
ounda
ti
on
f
r
om
a
ut
hor
it
a
ti
ve
s
our
c
e
s
,
pa
r
ti
c
ul
a
r
ly
publ
ic
r
e
por
ts
s
uc
h
a
s
th
os
e
pr
ovi
de
d
by
th
e
N
H
T
S
A
.
T
h
e
s
e
r
e
por
ts
f
ur
ni
s
h
c
om
pr
e
he
ns
iv
e
s
ta
ti
s
ti
c
s
a
nd
in
s
ig
ht
s
in
to
va
r
io
us
f
or
m
s
of
da
nge
r
ous
in
a
tt
e
nt
io
n
th
a
t
d
r
iv
e
r
s
m
a
y
e
xhi
bi
t
w
hi
le
on
th
e
r
oa
d. B
y
a
li
gni
ng
our
s
c
or
in
g c
r
it
e
r
ia
w
it
h t
he
f
in
di
ngs
a
nd
a
s
s
e
s
s
m
e
nt
s
out
li
ne
d
in
th
e
s
e
r
e
por
ts
,
w
e
a
im
to
e
ns
ur
e
th
a
t
th
e
r
is
k
a
s
s
e
s
s
m
e
nt
s
a
r
e
r
oot
e
d
i
n
w
e
ll
-
doc
um
e
nt
e
d
a
nd
w
id
e
ly
a
c
knowle
dge
d
da
ta
, t
he
r
e
by e
nha
nc
in
g t
he
a
c
c
ur
a
c
y a
nd r
e
li
a
bi
li
ty
of
t
he
s
a
f
e
t
y e
va
lu
a
ti
ons
.
2.5.1. Dis
t
r
ac
t
io
n
s
c
or
e
s
M
a
ny
di
s
tr
a
c
ti
ons
a
ppe
a
r
to
in
c
r
e
a
s
e
th
e
r
e
la
ti
ve
r
is
k
of
c
r
a
s
h
e
s
a
nd
n
e
a
r
-
c
r
a
s
he
s
,
a
nd
di
s
tr
a
c
ti
on
s
th
a
t
r
e
qui
r
e
dr
iv
e
r
s
to
ta
ke
th
e
ir
e
y
e
s
of
f
th
e
r
oa
d
a
r
e
pot
e
nt
ia
l
ly
m
or
e
of
a
s
a
f
e
ty
pr
obl
e
m
th
a
n
di
s
tr
a
c
ti
ons
th
a
t
do
not
r
e
qui
r
e
dr
iv
e
r
s
to
ta
ke
th
e
i
r
e
ye
s
of
f
th
e
r
oa
d
[
2
8]
.
U
s
in
g
a
c
e
ll
phone
w
hi
le
dr
iv
in
g
c
r
e
a
te
s
e
nor
m
ous
pot
e
nt
ia
l
f
or
de
a
th
s
a
nd
in
ju
r
ie
s
on
r
oa
d
s
[
29]
.
T
a
b
le
1
pr
e
s
e
nt
s
th
e
di
s
tr
a
c
ti
on
s
c
or
e
s
,
pr
ovi
di
ng
in
s
ig
ht
s
i
nt
o t
he
ir
r
e
la
ti
ve
r
is
ks
.
2.5.2. Dr
ow
s
in
e
s
s
an
d
f
at
ig
u
e
s
c
or
e
s
A
s
ta
te
of
f
a
ti
gue
is
of
te
n
m
a
r
ke
d
by
th
e
oc
c
ur
r
e
nc
e
of
f
r
e
que
nt
ya
w
ni
ng.
Y
a
w
ni
ng,
a
lo
ng
w
it
h
th
e
s
e
ns
a
ti
on of
w
e
a
r
in
e
s
s
, s
e
r
ve
s
a
s
c
le
a
r
i
ndi
c
a
to
r
s
of
bot
h phys
ic
a
l
a
nd me
nt
a
l
e
xha
us
ti
on, pos
in
g a
s
ubs
ta
nt
ia
l
th
r
e
a
t
to
one
’
s
c
a
pa
c
it
y
to
dr
iv
e
s
a
f
e
ly
.
A
ddi
ti
ona
ll
y,
dr
ow
s
in
e
s
s
is
c
ha
r
a
c
t
e
r
iz
e
d
by
th
e
in
vol
unt
a
r
y
c
lo
s
ur
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4249
-
4262
4254
of
th
e
e
ye
s
,
f
ur
th
e
r
he
ig
ht
e
ni
ng
th
e
r
is
k.
R
e
c
ogni
z
in
g
th
e
s
e
v
e
r
it
y
of
th
is
da
ng
e
r
,
f
a
ti
gue
a
nd
dr
ow
s
in
e
s
s
ha
ve
be
e
n
a
s
s
ig
ne
d
a
hi
gh
s
c
or
e
of
3,
unde
r
s
c
or
in
g
th
e
s
ig
ni
f
ic
a
nt
p
e
r
il
a
s
s
oc
i
a
te
d
w
it
h
th
e
s
e
c
ondi
ti
ons
.
T
a
bl
e
2
pr
ovi
de
s
a
de
ta
il
e
d
br
e
a
kdown
of
th
e
dr
ow
s
in
e
s
s
a
nd
f
a
ti
gue
s
c
or
e
s
,
of
f
e
r
in
g
in
s
ig
ht
s
in
to
th
e
ir
po
te
nt
ia
l
im
pa
c
t
on dr
iv
in
g s
a
f
e
ty
.
T
a
bl
e
1. D
is
tr
a
c
ti
on
s
c
or
e
s
I
na
t
t
e
nt
i
on a
c
t
i
vi
t
y
(
s
e
ve
r
i
t
y s
c
or
e
)
D
e
s
c
r
i
pt
i
on
T
e
xt
i
ng (
3 poi
nt
s
)
T
e
xt
i
ng
w
hi
l
e
dr
i
vi
ng
i
s
c
ons
i
de
r
e
d
a
n
e
xt
r
e
m
e
l
y
da
nge
r
ous
f
or
m
of
di
s
t
r
a
c
t
i
on.
I
t
r
e
qui
r
e
s
s
i
gni
f
i
c
a
nt
vi
s
ua
l
,
m
a
nua
l
,
a
nd
c
ogni
t
i
ve
a
t
t
e
nt
i
on
a
w
a
y
f
r
om
t
he
r
oa
d,
m
a
ki
ng
i
t
a
hi
gh
-
r
i
s
k
a
c
t
i
vi
t
y.
C
a
l
l
i
ng (
2 poi
nt
s
)
W
hi
l
e
l
e
s
s
di
s
t
r
a
c
t
i
ng
t
ha
n
t
e
xt
i
ng,
m
a
ki
ng
a
c
a
l
l
s
t
i
l
l
di
ve
r
t
s
a
t
t
e
nt
i
on
f
r
om
dr
i
vi
ng.
I
t
i
nvol
ve
s
c
ogni
t
i
ve
a
nd
m
a
nua
l
di
s
t
r
a
c
t
i
on
a
s
t
he
dr
i
ve
r
m
us
t
hol
d
t
he
phone
a
nd
e
nga
ge
i
n
c
onve
r
s
a
t
i
on.
R
e
a
c
hi
ng be
hi
nd (
3 poi
nt
s
)
T
hi
s
a
c
t
i
on
i
nvol
ve
s
s
i
gni
f
i
c
a
nt
m
a
nua
l
a
nd
vi
s
ua
l
di
s
t
r
a
c
t
i
on,
a
s
t
he
dr
i
ve
r
'
s
a
t
t
e
nt
i
on
i
s
f
oc
us
e
d a
w
a
y f
r
om
t
he
r
oa
d w
hi
l
e
r
e
a
c
hi
ng f
or
a
n obj
e
c
t
.
H
a
i
r
a
nd m
a
ke
up (
1 poi
nt
)
T
hough
not
a
s
s
e
ve
r
e
a
s
t
e
xt
i
ng,
gr
oom
i
ng
a
c
t
i
vi
t
i
e
s
s
t
i
l
l
r
e
qui
r
e
m
a
nua
l
a
nd
v
i
s
ua
l
a
t
t
e
nt
i
on
a
w
a
y f
r
om
dr
i
vi
ng, m
a
ki
ng i
t
a
m
ode
r
a
t
e
-
r
i
s
k be
ha
vi
or
.
D
r
i
nki
ng or
e
a
t
i
ng (
1 poi
nt
)
T
a
ki
ng
a
dr
i
nk
or
e
a
t
i
ng
w
hi
l
e
dr
i
vi
ng
c
a
n
l
e
a
d
t
o
m
om
e
nt
a
r
y
di
s
t
r
a
c
t
i
on,
pa
r
t
i
c
ul
a
r
l
y
i
f
t
he
dr
i
ve
r
ha
s
t
o r
e
a
c
h f
or
a
c
ont
a
i
ne
r
.
A
dj
us
t
i
ng r
a
di
o (
1 poi
nt
)
A
dj
us
t
i
ng
t
he
r
a
di
o
c
a
n
l
e
a
d
t
o
br
i
e
f
m
a
nua
l
a
nd
vi
s
ua
l
di
s
t
r
a
c
t
i
on,
but
i
t
i
s
ge
ne
r
a
l
l
y
c
ons
i
de
r
e
d a
l
ow
e
r
-
r
i
s
k be
ha
vi
or
.
T
a
l
ki
ng t
o pa
s
s
e
nge
r
(
1 poi
nt
)
C
onve
r
s
a
t
i
ons
w
i
t
h
pa
s
s
e
nge
r
s
c
a
n
c
a
us
e
s
om
e
c
ogni
t
i
ve
di
s
t
r
a
c
t
i
on,
but
i
t
'
s
ge
ne
r
a
l
l
y
a
c
om
m
on a
nd r
e
l
a
t
i
ve
l
y l
ow
e
r
-
r
i
s
k be
ha
vi
or
.
-
T
a
bl
e
2.
D
r
ow
s
in
e
s
s
a
nd f
a
ti
gue
s
c
or
e
s
I
na
t
t
e
nt
i
on a
c
t
i
vi
t
y (
s
e
ve
r
i
t
y s
c
or
e
)
D
e
s
c
r
i
pt
i
on
D
r
ow
s
y (
3 poi
nt
s
)
D
r
ow
s
i
ne
s
s
s
i
gni
f
i
c
a
nt
l
y
i
m
pa
i
r
s
a
dr
i
ve
r
'
s
a
bi
l
i
t
y
t
o
r
e
a
c
t
qui
c
kl
y
a
nd
m
a
ke
s
ound
j
udgm
e
nt
s
.
I
t
'
s
a
hi
gh
-
r
i
s
k c
ondi
t
i
on a
s
i
t
i
nc
r
e
a
s
e
s
t
he
l
i
ke
l
i
hood of
a
c
c
i
de
nt
s
.
Y
a
w
ni
ng (
3 poi
nt
s
)
Y
a
w
ni
ng i
s
of
t
e
n i
ndi
c
a
t
i
ve
of
dr
ow
s
i
ne
s
s
or
f
a
t
i
gue
, w
hi
c
h c
a
n s
e
ve
r
e
l
y i
m
pa
i
r
a
dr
i
ve
r
'
s
a
bi
l
i
t
y
t
o f
oc
us
a
nd r
e
a
c
t
on t
i
m
e
.
N
o dr
ow
s
i
ne
s
s
(
0 poi
nt
s
)
W
he
n a
dr
i
ve
r
i
s
a
l
e
r
t
a
nd not
dr
ow
s
y, t
he
r
e
i
s
no a
ddi
t
i
ona
l
r
i
s
k a
s
s
oc
i
a
t
e
d w
i
t
h t
hi
s
f
a
c
t
or
.
N
o ya
w
ni
ng (
0 poi
nt
s
)
W
he
n a
dr
i
ve
r
i
s
not
ya
w
ni
ng, t
he
r
e
i
s
no a
ddi
t
i
ona
l
r
i
s
k a
s
s
oc
i
a
t
e
d w
i
t
h t
hi
s
f
a
c
t
or
.
2.5.3. E
m
ot
io
n
s
s
c
o
r
e
s
N
um
e
r
ous
s
tu
di
e
s
ha
v
e
hi
ghl
ig
ht
e
d
a
not
e
w
or
th
y
a
s
s
oc
ia
ti
o
n
be
twe
e
n
dr
iv
in
g
-
r
e
la
te
d
a
nge
r
a
nd
s
pe
c
if
ic
hi
gh
-
r
is
k
dr
iv
in
g
pr
a
c
ti
c
e
s
,
in
c
lu
di
ng
in
s
ta
n
c
e
s
of
s
p
e
e
di
ng,
a
ggr
e
s
s
iv
e
dr
iv
in
g,
a
nd
di
s
r
e
ga
r
di
ng
tr
a
f
f
ic
s
ig
na
ls
[
30]
.
F
ur
th
e
r
m
o
r
e
, s
c
hol
a
r
s
ha
ve
unde
r
s
c
or
e
d t
ha
t
f
e
e
li
ngs
of
a
nxi
e
ty
a
nd f
e
a
r
c
a
n a
ls
o s
e
r
ve
a
s
pr
e
di
c
to
r
s
f
or
e
nga
gi
ng
in
r
is
ky
dr
iv
in
g
be
ha
vi
or
s
[
31]
.
T
a
bl
e
3
pr
e
s
e
nt
s
th
e
e
m
ot
io
n
s
c
or
e
s
,
out
li
ni
ng
th
e
ir
pot
e
nt
ia
l
c
ont
r
ib
ut
io
n t
o r
is
ky dr
iv
in
g pr
a
c
ti
c
e
s
.
T
a
bl
e
3. E
m
ot
io
ns
s
c
or
e
s
I
na
t
t
e
nt
i
on a
c
t
i
vi
t
y (
s
e
ve
r
i
t
y s
c
or
e
)
D
e
s
c
r
i
pt
i
on
A
nge
r
(
3 poi
nt
s
)
A
nge
r
c
a
n
l
e
a
d
t
o
c
ogni
t
i
ve
di
s
t
r
a
c
t
i
on,
a
ggr
e
s
s
i
ve
dr
i
vi
ng
be
ha
vi
or
s
,
a
nd
r
oa
d
r
a
ge
,
pos
i
ng
s
i
gni
f
i
c
a
nt
r
i
s
ks
t
o r
oa
d s
a
f
e
t
y.
F
e
a
r
a
nd s
a
dne
s
s
(
2 poi
nt
s
e
a
c
h)
T
he
s
e
e
m
ot
i
ons
c
a
n
l
e
a
d
t
o
c
ogni
t
i
ve
di
s
t
r
a
c
t
i
on
a
nd,
i
n
s
om
e
c
a
s
e
s
,
phys
i
c
a
l
r
e
a
c
t
i
ons
t
ha
t
m
a
y
a
f
f
e
c
t
dr
i
vi
ng pe
r
f
o
r
m
a
nc
e
.
S
ur
pr
i
s
e
a
nd di
s
gus
t
(
1 poi
nt
e
a
c
h)
W
hi
l
e
t
he
s
e
e
m
ot
i
ons
m
a
y
m
om
e
nt
a
r
i
l
y
di
s
t
r
a
c
t
a
dr
i
ve
r
,
t
he
i
r
i
m
pa
c
t
i
s
ge
ne
r
a
l
l
y
c
ons
i
de
r
e
d
l
ow
e
r
c
om
pa
r
e
d t
o f
e
a
r
a
nd
s
a
dne
s
s
.
N
e
ut
r
a
l
a
nd ha
ppi
ne
s
s
(
0 poi
nt
s
)
W
he
n
a
dr
i
ve
r
i
s
e
m
ot
i
ona
l
l
y
ne
ut
r
a
l
or
e
xpe
r
i
e
nc
i
ng
ha
ppi
ne
s
s
,
t
he
r
e
i
s
n
o
a
ddi
t
i
ona
l
r
i
s
k
a
s
s
oc
i
a
t
e
d w
i
t
h t
he
s
e
f
a
c
t
or
s
. P
o
s
i
t
i
ve
e
m
ot
i
ons
m
a
y e
ve
n
c
ont
r
i
but
e
t
o a
m
or
e
a
l
e
r
t
a
nd f
oc
us
e
d
s
t
a
t
e
.
2.5.4. T
ot
al
r
is
k
T
he
r
is
k
c
a
t
e
gor
iz
a
ti
on
a
lg
or
it
hm
is
ba
s
e
d
on
a
s
im
pl
e
“
if
-
th
e
n
”
a
ppr
oa
c
h.
W
e
c
a
n
d
e
te
r
m
in
e
th
e
r
is
k
le
ve
l
a
s
s
oc
ia
te
d
w
it
h
e
a
c
h
c
om
bi
na
ti
on
by
us
in
g
th
e
s
um
o
f
poi
nt
s
obt
a
in
e
d
f
r
om
di
f
f
e
r
e
nt
c
om
bi
na
ti
ons
of
s
c
or
e
s
f
or
di
s
tr
a
c
ti
ng
a
c
ti
vi
ti
e
s
,
dr
ow
s
in
e
s
s
le
ve
l
s
,
ya
w
ni
ng
oc
c
ur
r
e
nc
e
s
,
a
nd
th
e
dr
iv
e
r
’
s
e
m
ot
io
ns
.
O
nc
e
w
e
c
a
lc
ul
a
te
th
e
to
ta
l
poi
nt
s
f
or
e
a
c
h
v
a
r
ie
ty
,
w
e
pa
s
s
it
to
th
e
c
a
te
gor
iz
a
ti
on
a
lg
or
it
hm
.
I
f
th
e
to
ta
l
poi
nt
s
a
r
e
le
s
s
th
a
n
or
e
qua
l
to
2,
th
e
r
is
k
c
a
te
gor
y
is
c
ons
id
e
r
e
d
“
lo
w
.
”
I
f
th
e
to
ta
l
poi
nt
s
f
a
ll
be
twe
e
n
3
a
nd
7
(
in
c
lu
s
iv
e
)
,
th
e
r
is
k
c
a
te
gor
y
is
d
e
f
in
e
d
a
s
“
m
e
di
um
.
”
L
a
s
tl
y,
if
th
e
to
ta
l
poi
nt
s
f
a
ll
be
twe
e
n
8
a
nd
10
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
r
iv
e
r
i
nat
te
nt
io
n
de
te
c
ti
on s
y
s
te
m
us
in
g m
ul
ti
-
ta
s
k
c
a
s
c
ade
d
c
onv
ol
ut
io
nal
... (
A
bde
lf
e
tt
ah Soult
ana)
4255
(
in
c
lu
s
iv
e
)
, t
he
r
is
k c
a
te
gor
y
is
l
a
be
le
d
“
hi
gh
.
”
I
f
t
he
t
ot
a
l
poi
nt
s
e
xc
e
e
d 10, the
r
is
k c
a
te
gor
y i
s
l
a
be
le
d
“
ve
r
y
hi
gh
.
”
F
ig
ur
e
2 pr
ovi
de
s
a
vi
s
ua
l
r
e
pr
e
s
e
nt
a
ti
on of
t
he
c
a
t
e
gor
iz
a
ti
on a
lg
or
it
hm
.
F
ig
ur
e
2
.
R
is
k f
unc
ti
on c
om
put
e
2.6. Ale
r
t
l
aye
r
I
n
th
e
c
ul
m
in
a
ti
on
of
th
e
s
ys
te
m
’
s
f
unc
ti
ona
li
ty
,
th
e
f
in
a
l
la
ye
r
a
s
s
um
e
s
th
e
vi
ta
l
r
ol
e
of
dr
iv
e
r
a
le
r
ti
ng.
I
t
e
m
pl
oys
a
nua
nc
e
d
r
is
k
a
s
s
e
s
s
m
e
nt
a
ppr
oa
c
h
to
de
te
r
m
in
e
th
e
a
ppr
opr
ia
te
le
ve
l
o
f
a
le
r
tn
e
s
s
r
e
qui
r
e
d.
T
he
a
le
r
t
s
ta
ge
r
e
pr
e
s
e
nt
s
a
pr
oa
c
ti
ve
a
ppr
oa
c
h
to
e
n
ha
nc
in
g
dr
iv
e
r
s
a
f
e
ty
by
pr
ovi
di
ng
ti
m
e
ly
a
nd
r
e
le
va
nt
f
e
e
dba
c
k
a
nd,
w
he
n
ne
c
e
s
s
a
r
y,
in
te
r
ve
ni
ng
to
pr
e
ve
nt
pot
e
nt
ia
l
ha
z
a
r
ds
a
s
s
oc
ia
te
d
w
it
h
in
a
tt
e
nt
io
n.
I
t
s
e
r
ve
s
a
s
a
va
lu
a
bl
e
t
ool
i
n pr
om
ot
in
g r
e
s
pons
ib
le
a
nd a
tt
e
nt
i
ve
dr
iv
in
g be
ha
vi
or
.
3.
D
R
I
V
E
R
F
A
C
I
A
L
F
E
A
T
U
R
E
S
D
E
T
E
C
T
I
O
N
P
R
O
P
O
S
I
T
I
O
N
A
pr
im
a
r
y
a
nd
s
ig
ni
f
ic
a
nt
c
ha
ll
e
nge
in
our
s
ys
te
m
li
e
s
in
e
xt
r
a
c
ti
ng
r
e
le
va
nt
f
e
a
tu
r
e
s
f
r
om
im
a
ge
s
(
dr
iv
e
r
f
a
c
ia
l
f
e
a
tu
r
e
s
e
xt
r
a
c
ti
on
l
a
ye
r
)
.
I
n
th
e
c
ont
e
xt
of
th
is
s
tu
dy,
w
e
w
il
l
pl
a
c
e
s
pe
c
ia
l
e
m
pha
s
is
on
th
is
c
r
uc
ia
l
c
om
pone
nt
.
F
ig
ur
e
3
il
lu
s
tr
a
te
s
our
a
ppr
oa
c
h
us
in
g
th
e
M
T
C
N
N
a
lg
or
it
hm
f
or
e
xt
r
a
c
ti
ng
f
a
c
ia
l
f
e
a
tu
r
e
s
s
uc
h a
s
e
ye
s
a
nd mout
h us
in
g t
he
c
a
m
e
r
a
i
n f
r
ont
of
t
h
e
dr
iv
e
r
.
T
hi
s
c
hoi
c
e
is
ju
s
ti
f
ie
d
by
two
e
xpe
r
im
e
nt
s
on
th
r
e
e
da
ta
s
e
ts
F
E
,
D
r
iv
F
a
c
e
,
a
nd
dr
iv
e
r
dr
ow
s
in
e
s
s
da
ta
s
e
t
.
T
he
f
ir
s
t
f
oc
us
e
s
on
lo
w
li
ght
in
g
a
nd
di
f
f
e
r
e
nt
he
a
d
m
ove
m
e
nt
s
(
a
c
r
os
s
a
180°
vi
e
w
)
,
w
hi
l
e
th
e
s
e
c
ond
e
xpe
r
im
e
nt
a
im
s
to
a
s
s
e
s
s
th
e
pe
r
f
or
m
a
nc
e
of
th
e
M
T
C
N
N
a
lg
or
it
hm
on
a
r
e
a
l
-
w
or
ld
da
ta
s
e
t
of
dr
iv
in
g
s
c
e
na
r
io
s
.
T
o
e
ns
ur
e
pr
e
c
is
e
a
n
a
ly
s
is
of
f
a
c
ia
l
f
e
a
tu
r
e
s
,
M
T
C
N
N
i
s
c
om
pa
r
e
d
to
a
hi
s
to
gr
a
m
of
gr
a
di
e
nt
(
H
O
G
)
-
ba
s
e
d
f
r
ont
a
l
f
a
c
e
de
te
c
to
r
,
a
nd
a
H
a
a
r
f
e
a
tu
r
e
-
ba
s
e
d
c
a
s
c
a
de
c
la
s
s
if
ie
r
.
T
he
s
e
a
lg
or
it
hm
s
w
e
r
e
e
xt
e
ns
iv
e
ly
c
om
pa
r
e
d f
or
t
he
ir
a
da
pt
a
bi
li
ty
t
o va
r
yi
ng l
ig
h
ti
ng c
ondi
ti
ons
, a
nd c
a
pa
c
it
y t
o a
c
c
om
m
oda
te
di
f
f
e
r
e
nt
he
a
d
m
ove
m
e
nt
s
in
e
xpe
r
im
e
nt
1.
F
ur
th
e
r
m
or
e
,
in
e
xpe
r
im
e
nt
2,
th
e
ir
a
de
pt
ne
s
s
in
a
c
c
ur
a
te
ly
de
te
c
ti
ng
e
s
s
e
nt
ia
l
f
a
c
ia
l
c
om
pone
nt
s
s
uc
h
a
s
th
e
f
a
c
e
,
e
ye
s
,
a
nd
m
out
h
w
a
s
m
e
ti
c
ul
ous
ly
a
s
s
e
s
s
e
d
us
in
g
e
xpa
ns
iv
e
da
ta
s
e
ts
obt
a
in
e
d
f
r
om
r
e
a
l
-
w
or
ld
dr
iv
in
g
s
c
e
na
r
io
s
a
nd
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
in
c
lu
di
ng
tr
ue
pos
it
iv
e
s
(
TP
)
,
tr
ue
ne
ga
ti
v
e
s
(
TN
)
,
f
a
ls
e
po
s
it
iv
e
s
(
FP
)
,
f
a
ls
e
pos
it
iv
e
s
(
FN
)
,
a
c
c
ur
a
c
y,
pr
e
c
is
io
n
,
F
1
-
s
c
or
e
,
a
nd
r
e
c
a
ll
.
F
ig
ur
e
3. P
r
opos
a
l
of
i
nc
or
por
a
ti
ng M
T
C
N
N
i
n
f
a
c
ia
l
f
e
a
tu
r
e
e
xt
r
a
c
ti
on l
a
ye
r
3.1. Dat
as
e
t
s
I
n
th
is
s
tu
dy,
w
e
ut
il
iz
e
d
th
r
e
e
da
ta
s
e
ts
. T
he
F
E
I
da
ta
s
e
t
w
a
s
e
m
pl
oye
d
to
a
s
s
e
s
s
th
e
pe
r
f
or
m
a
nc
e
of
te
c
hni
que
s
,
ta
ki
ng
in
to
a
c
c
ount
he
a
d
r
ot
a
ti
on
a
nd
li
ght
in
g
c
ondi
ti
ons
.
T
he
F
E
I
f
a
c
e
da
ta
s
e
t
is
a
c
ol
le
c
ti
on
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4249
-
4262
4256
B
r
a
z
il
ia
n
f
a
c
e
im
a
ge
s
c
a
pt
ur
e
d
be
twe
e
n
J
une
2005
a
nd
M
a
r
c
h
2006
a
t
th
e
A
r
ti
f
ic
ia
l
I
nt
e
ll
ig
e
nc
e
L
a
bor
a
to
r
y
of
F
E
I
in
S
ã
o
B
e
r
na
r
do
do
C
a
m
po,
S
ã
o
P
a
ul
o,
B
r
a
z
il
[
33]
.
T
hi
s
da
ta
s
e
t
c
on
s
is
ts
of
2
,
800
c
ol
or
f
ul
im
a
ge
s
,
w
it
h
14
im
a
ge
s
f
or
e
a
c
h
of
th
e
200
in
di
vi
dua
ls
.
T
h
e
phot
ogr
a
phs
w
e
r
e
ta
ke
n
a
ga
in
s
t
a
w
hi
te
hom
oge
ne
ou
s
ba
c
kgr
ound,
s
how
in
g
th
e
s
ubj
e
c
ts
in
a
n
upr
ig
ht
f
r
ont
a
l
pos
it
io
n
w
it
h
pr
of
il
e
r
ot
a
ti
on
of
up
to
a
ppr
oxi
m
a
te
ly
180
de
gr
e
e
s
.
T
he
im
a
ge
s
’
or
ig
in
a
l
s
iz
e
is
640
×
480
pi
xe
ls
,
w
it
h
a
pos
s
ib
le
v
a
r
ia
ti
on
in
s
c
a
le
of
a
bout
10%
.
T
he
f
a
c
e
s
f
e
a
tu
r
e
d
in
th
e
da
ta
s
e
t
be
lo
ng
to
in
di
vi
dua
ls
a
g
e
d
be
twe
e
n
19
a
nd
40
ye
a
r
s
,
m
a
in
ly
c
om
pr
is
in
g
s
tu
de
nt
s
a
nd
s
ta
f
f
a
t
F
E
I
,
di
s
pl
a
yi
ng
di
ve
r
s
e
a
ppe
a
r
a
nc
e
s
,
ha
i
r
s
ty
le
s
,
a
nd
a
dor
nm
e
nt
s
.
N
ot
a
bl
y,
th
e
da
ta
s
e
t
in
c
lu
de
s
a
n e
qua
l
num
be
r
of
m
a
le
a
nd f
e
m
a
l
e
s
ubj
e
c
t
s
, e
a
c
h t
ot
a
li
ng 100.
T
he
ot
he
r
two
da
ta
s
e
ts
w
e
r
e
ut
il
iz
e
d
to
e
va
lu
a
t
e
th
e
r
obu
s
tn
e
s
s
of
de
te
c
ti
on
s
in
r
e
a
l
dr
iv
in
g
s
c
e
na
r
io
s
.
T
he
publ
ic
D
r
iv
F
a
c
e
da
ta
s
e
t
[
32]
c
ont
a
in
s
im
a
ge
s
e
que
n
c
e
s
of
s
ubj
e
c
ts
w
hi
le
dr
iv
in
g
in
r
e
a
l
s
c
e
na
r
io
s
.
I
t
c
om
pr
is
e
s
606
s
a
m
pl
e
s
of
640
×
480
pi
xe
ls
e
a
c
h,
a
c
qui
r
e
d
ove
r
di
f
f
e
r
e
nt
da
ys
f
r
om
4
dr
iv
e
r
s
(
2
w
om
e
n
a
nd
2
m
e
n)
w
it
h
va
r
io
us
f
a
c
ia
l
f
e
a
tu
r
e
s
obs
tr
uc
ti
on
s
,
s
uc
h
a
s
gl
a
s
s
e
s
a
nd
f
a
c
ia
l
ha
ir
.
T
h
e
dr
iv
e
r
dr
ow
s
in
e
s
s
da
ta
s
e
t
[
33]
c
ont
a
in
s
1448
phot
o
s
,
di
vi
de
d
in
to
tw
o
gr
oups
:
723
im
a
g
e
s
l
a
be
le
d
“
ya
w
ni
ng
”
a
nd
725
im
a
ge
s
la
be
le
d
“
N
o
ya
w
ni
ng
”
.
I
t
is
im
po
r
ta
nt
to
no
te
th
a
t
e
a
c
h
of
th
e
‘
ya
w
ni
ng
’
a
nd
‘
N
o
ya
w
ni
ng
’
s
ubs
e
ts
w
il
l
be
us
e
d
s
e
pa
r
a
te
ly
f
or
e
va
lu
a
ti
on
pur
pos
e
s
.
F
ig
ur
e
4
s
how
c
a
s
e
s
e
xa
m
pl
e
s
of
im
a
ge
s
f
r
om
th
e
th
r
e
e
da
ta
s
e
ts
.
F
ig
ur
e
4. E
xa
m
pl
e
s
of
i
m
a
ge
s
f
r
om
t
he
t
hr
e
e
da
ta
s
e
ts
3.2. F
ac
ia
l
f
e
at
u
r
e
e
xt
r
ac
t
io
n
t
e
c
h
n
iq
u
e
s
3.2.1. Haar c
as
c
ad
e
-
op
e
n
c
v
I
n
2001,
V
io
la
a
nd
J
one
s
[
22]
pr
opos
e
d
a
n
e
f
f
e
c
ti
ve
obj
e
c
t
de
te
c
ti
on
te
c
hni
que
known
a
s
H
a
a
r
f
e
a
tu
r
e
-
ba
s
e
d
c
a
s
c
a
de
c
la
s
s
if
ie
r
s
.
I
t
is
a
m
a
c
hi
ne
le
a
r
ni
ng
-
ba
s
e
d
m
e
th
od
th
a
t
tr
a
in
s
th
e
c
la
s
s
if
ie
r
w
it
h
m
a
ny
pos
it
iv
e
phot
os
(
w
it
h
f
a
c
e
s
)
a
nd
ne
ga
ti
ve
im
a
ge
s
(
w
it
hout
f
a
c
e
s
)
.
I
n
th
e
ir
s
tu
dy,
va
r
io
us
e
xt
r
e
m
e
ly
ba
s
ic
or
w
e
a
k
f
a
c
ia
l
tr
a
it
s
a
r
e
le
a
r
ne
d
us
in
g
th
e
A
da
B
oos
t
m
ode
l
to
c
r
e
a
te
a
r
obus
t
c
la
s
s
if
ie
r
f
or
e
a
c
h
f
a
c
e
.
T
he
V
io
la
-
J
one
s
de
te
c
to
r
is
one
of
th
e
e
a
r
li
e
s
t
m
e
th
od
s
.
I
t
f
unc
ti
ons
on
gr
a
ys
c
a
le
im
a
ge
s
by
in
te
r
pr
e
ti
ng
th
e
im
a
ge
a
s
a
c
ol
le
c
ti
on
of
H
a
a
r
f
e
a
tu
r
e
s
(
li
ght
e
r
a
nd
da
r
ke
r
r
e
c
ta
ngl
e
s
)
.
T
he
r
e
a
r
e
num
e
r
ous
H
a
a
r
f
e
a
tu
r
e
ty
pe
s
w
it
h
va
r
io
us
pl
a
c
e
m
e
nt
s
of
th
e
r
e
c
ta
ngl
e
’
s
s
of
t
a
nd
da
r
k
a
r
e
a
s
.
T
he
y
c
a
n
be
c
om
put
e
d
ve
r
y
qui
c
kl
y
us
in
g
a
n
in
te
gr
a
l
im
a
ge
m
e
th
od.
T
he
in
te
gr
a
l
im
a
ge
is
a
c
om
put
in
g
t
e
c
hni
que
th
a
t
e
na
bl
e
s
th
e
r
a
pi
d
a
nd
e
f
f
ic
ie
nt
c
a
lc
ul
a
ti
on
of
th
e
s
um
of
pi
xe
l
v
a
lu
e
s
, a
c
hi
e
vi
ng
c
on
s
ta
nt
ti
m
e
c
om
pl
e
xi
ty
a
nd
li
tt
le
c
om
put
a
ti
ona
l
ove
r
he
a
d.
T
he
pr
oc
e
s
s
in
vol
ve
s
ge
ne
r
a
ti
ng
a
n
im
a
ge
of
e
qua
l
di
m
e
ns
io
ns
a
s
th
e
or
ig
in
a
l
im
a
ge
,
r
e
f
e
r
r
e
d
to
a
s
a
s
uppl
e
m
e
nt
a
l
a
r
e
a
ta
bl
e
.
T
he
s
um
m
a
ti
on
of
th
e
pi
xe
l
s
lo
c
a
te
d
t
o
th
e
le
f
t
a
nd
a
bove
e
a
c
h
gi
ve
n
pi
xe
l
(
x,
y)
in
th
e
or
ig
in
a
l
im
a
ge
i
s
c
a
lc
ul
a
te
d u
s
in
g
in
(
1)
.
(
,
)
=
∑
(
′
,
′
)
′
≤
,
′
≤
(
1)
T
he
f
unc
ti
on
ii
(
x,
y)
r
e
pr
e
s
e
nt
s
th
e
pi
xe
l
va
lu
e
s
of
th
e
in
te
gr
a
l
im
a
ge
,
w
hi
le
I
(
x,
y
)
r
e
pr
e
s
e
nt
s
th
e
pi
xe
l
va
lu
e
s
of
th
e
or
ig
in
a
l
im
a
ge
a
t
poi
nt
(
x,
y)
.
T
he
c
om
put
a
t
io
n
of
th
e
to
ta
l
pi
xe
l
va
lu
e
s
w
it
hi
n
a
r
e
c
ta
ngl
e
r
e
gi
on
c
a
n
be
s
im
pl
if
ie
d
by
ut
il
iz
in
g
onl
y
f
our
va
lu
e
s
f
r
om
th
e
in
te
gr
a
l
im
a
ge
,
r
a
th
e
r
th
a
n
s
um
m
in
g
th
e
va
lu
e
s
of
a
ll
in
di
vi
dua
l
pi
xe
ls
.
I
f
A
, B
,
C
,
a
nd D
r
e
pr
e
s
e
nt
th
e
va
lu
e
s
a
t
th
e
c
or
ne
r
s
of
th
e
ta
bl
e
b
e
in
g
to
ta
le
d,
th
e
t
ot
a
l
s
um
w
it
hi
n t
hi
s
r
e
c
ta
ngul
a
r
r
e
gi
on c
a
n be
c
a
l
c
ul
a
te
d us
in
g
in
(
2)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
r
iv
e
r
i
nat
te
nt
io
n
de
te
c
ti
on s
y
s
te
m
us
in
g m
ul
ti
-
ta
s
k
c
a
s
c
ade
d
c
onv
ol
ut
io
nal
... (
A
bde
lf
e
tt
ah Soult
ana)
4257
=
+
−
−
(
2)
C
a
s
c
a
di
ng
c
la
s
s
if
ie
r
s
a
r
e
e
m
pl
oye
d
to
e
f
f
ic
ie
nt
ly
e
li
m
in
a
te
non
-
f
a
c
e
in
s
ta
nc
e
s
a
nd
m
in
im
iz
e
s
upe
r
f
lu
ous
c
om
put
a
ti
ona
l
op
e
r
a
ti
ons
.
C
a
s
c
a
di
ng
in
vol
ve
s
th
e
c
ons
tr
uc
ti
on
of
a
c
a
s
c
a
d
e
c
l
a
s
s
if
ie
r
,
w
hi
c
h
is
c
om
pr
is
e
d
of
m
ul
ti
pl
e
s
ta
ge
s
,
e
a
c
h
hous
in
g
a
r
obus
t
tr
a
in
e
d
c
la
s
s
if
ie
r
.
I
n
th
e
e
ve
nt
of
a
s
ub
-
w
in
dow
e
xpe
r
ie
nc
in
g
f
a
il
ur
e
a
t
a
ny
poi
nt
,
it
w
il
l
pr
om
pt
ly
be
e
li
m
in
a
te
d.
T
he
A
da
boo
s
t
a
lg
or
it
hm
is
e
m
pl
oye
d
in
th
e
c
r
e
a
ti
on of
a
c
a
s
c
a
de
by me
ti
c
ul
ou
s
ly
c
ons
tr
uc
ti
ng e
a
c
h i
ndi
vi
d
ua
l
s
ta
ge
. T
h
e
pr
oc
e
s
s
i
nvol
ve
s
t
he
i
nt
e
gr
a
ti
on
of
m
ul
ti
pl
e
c
la
s
s
if
ie
r
s
t
o f
or
m
t
he
s
ta
ge
s
.
3.2.2.
H
oG
-
D
li
b
T
he
H
oG
-
D
li
b
p
r
ovi
de
s
a
n
a
ppr
oa
c
h
f
or
f
a
c
e
de
te
c
ti
on
ba
s
e
d
on
H
O
G
a
nd
L
in
e
a
r
S
V
M
[
23]
.
T
he
id
e
a
of
th
e
H
O
G
de
s
c
r
ip
to
r
is
to
c
r
e
a
te
a
ve
c
to
r
of
f
e
a
tu
r
e
s
s
o
th
a
t
th
e
ve
c
to
r
c
a
n
be
f
e
d
in
to
a
c
la
s
s
if
ic
a
ti
on
a
lg
or
it
hm
li
ke
S
V
M
to
pr
e
di
c
t
th
e
r
e
s
ul
t.
T
o
c
a
lc
ul
a
te
th
e
H
O
G
de
s
c
r
ip
to
r
,
w
e
ne
e
d
to
c
a
lc
ul
a
te
th
e
gr
a
di
e
nt
s
of
t
he
x
-
a
xi
s
-
a
nd y
-
a
xi
s
gr
a
di
e
nt
s
. T
he
c
a
lc
ul
a
ti
on of
t
he
gr
a
di
e
nt
ve
c
to
r
i
s
a
s
(
3)
a
nd (
4)
:
(
,
)
=
(
+
1
,
)
−
(
−
1
,
)
(
3)
(
,
)
=
(
,
+
1
)
−
(
,
−
1
)
(
4)
In
(
3)
a
nd
(
4)
r
e
pr
e
s
e
nt
th
e
ho
r
iz
on
ta
l
gr
a
di
e
nt
o
f
th
e
im
a
g
e
pi
x
e
l,
G
x(
x
,
y)
,
a
n
d
th
e
ve
r
ti
c
a
l
gr
a
di
e
nt
,
G
y(
x,
y)
,
r
e
s
p
e
c
ti
v
e
l
y.
In
(
5)
a
n
d (
6)
r
e
s
p
e
c
ti
v
e
l
y r
e
pr
e
s
e
nt
th
e
m
a
g
ni
t
ud
e
a
nd
d
ir
e
c
ti
on
of
t
h
e
gr
a
di
e
nt
a
t
pi
x
e
l
(
x,
y)
.
(
,
)
=
√
(
,
)
2
+
(
,
)
²
(
5)
(
,
)
=
−
1
(
(
,
)
(
,
)
)
(
6)
T
he
s
e
gr
a
di
e
nt
s
c
a
pt
ur
e
th
e
im
a
g
e
’
s
di
r
e
c
ti
on
a
nd
m
a
gni
tu
de
of
pi
xe
l
in
te
ns
it
y
c
h
a
nge
s
.
B
y
a
na
ly
z
in
g
th
e
s
e
gr
a
di
e
nt
s
,
D
li
b
c
a
n
c
ons
tr
uc
t
a
f
e
a
tu
r
e
ve
c
to
r
th
a
t
de
s
c
r
ib
e
s
th
e
c
ha
r
a
c
te
r
is
ti
c
pa
tt
e
r
ns
of
f
a
c
e
s
.
I
n
th
e
c
ont
e
xt
of
D
li
b,
f
iv
e
H
O
G
f
il
te
r
s
a
r
e
us
e
d
f
or
f
a
c
e
de
te
c
ti
on:
f
r
ont
-
lo
oki
ng,
le
f
t
-
lo
oki
ng,
r
ig
ht
-
lo
oki
ng,
f
r
on
t
-
lo
oki
ng
but
r
ot
a
te
d
le
f
t,
a
nd
f
r
ont
-
lo
oki
n
g
but
r
ot
a
te
d
r
ig
ht
.
T
he
s
e
f
il
te
r
s
he
lp
c
a
pt
ur
e
va
r
ia
ti
ons
in
f
a
c
ia
l
or
ie
nt
a
ti
on
a
nd
e
ns
ur
e
r
obus
t
f
a
c
e
de
te
c
ti
on
e
ve
n
w
he
n
f
a
c
e
s
a
r
e
r
ot
a
te
d
or
c
a
pt
ur
e
d
f
r
om
di
f
f
e
r
e
nt
a
ngl
e
s
.
3.2.3.
M
u
lt
i
-
t
as
k
c
as
c
ad
e
d
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
T
he
M
T
C
N
N
s
ta
nds
a
s
a
pr
om
in
e
nt
m
il
e
s
to
ne
in
th
e
r
e
a
lm
of
c
om
put
e
r
vi
s
io
n
a
nd
f
a
c
e
a
li
gnm
e
nt
.
U
nve
il
e
d
by
K
a
ip
e
ng
Z
ha
ng
a
nd
f
e
ll
ow
r
e
s
e
a
r
c
he
r
s
in
201
6
[
24]
.
T
he
M
T
C
N
N
is
w
e
ll
-
known
f
or
it
s
s
ta
te
-
of
-
th
e
-
a
r
t
pe
r
f
o
r
m
a
nc
e
on
a
va
r
ie
ty
of
be
nc
hm
a
r
k
da
ta
s
e
ts
a
s
w
e
ll
a
s
it
s
la
ndm
a
r
k
de
te
c
ti
on
a
bi
li
ty
,
w
hi
c
h
e
na
bl
e
s
it
to
id
e
nt
if
y
a
ddi
ti
ona
l
f
a
c
ia
l
f
e
a
tu
r
e
s
li
ke
th
e
e
ye
s
a
nd
m
out
h.
T
he
n
e
twor
k
e
m
pl
oys
a
c
a
s
c
a
d
e
s
tr
uc
tu
r
e
w
it
h
th
r
e
e
ne
twor
ks
.
T
he
im
a
ge
is
f
ir
s
t
r
e
s
i
z
e
d
to
va
r
io
us
s
iz
e
s
(
r
e
f
e
r
r
e
d
to
a
s
a
n
im
a
ge
pyr
a
m
id
)
.
T
he
f
ir
s
t
m
ode
l,
th
e
pr
opos
a
l
ne
twor
k
(P
-
N
e
t)
,
pr
opos
e
s
c
a
ndi
da
te
f
a
c
ia
l
r
e
gi
ons
;
th
e
s
e
c
ond
m
ode
l,
th
e
r
e
f
in
e
ne
twor
k
(R
-
N
e
t)
,
f
il
te
r
s
th
e
bounding
boxe
s
,
a
nd
th
e
th
ir
d
m
ode
l,
th
e
out
put
ne
twor
k
(O
-
N
e
t)
, pr
opos
e
s
f
a
c
ia
l
la
ndm
a
r
ks
.
U
pon
th
e
de
te
c
ti
on
of
a
f
a
c
e
,
th
e
P
-
N
e
t
a
lg
o
r
it
hm
p
r
ovi
de
s
th
e
c
oor
di
na
te
s
of
a
bounding
box.
T
he
ope
r
a
ti
on
w
il
l
be
r
e
pe
a
te
d
in
a
s
e
c
ti
on
-
w
is
e
m
a
nne
r
,
w
it
h
th
e
12
×
12
ke
r
ne
l
be
in
g
s
hi
f
te
d
2
pi
xe
ls
to
th
e
r
ig
ht
or
dow
n
a
t
e
a
c
h
it
e
r
a
ti
on.
T
he
di
s
pl
a
c
e
m
e
nt
of
2
pi
xe
ls
is
c
om
m
onl
y
r
e
f
e
r
r
e
d
to
a
s
th
e
s
tr
id
e
.
T
he
f
a
c
ia
l
f
e
a
tu
r
e
s
s
e
e
n
in
m
os
t
of
th
e
im
a
ge
s
e
xhi
bi
t
a
s
iz
e
e
xc
e
e
di
ng
2
pi
xe
ls
.
T
he
li
ke
li
hood
of
th
e
ke
r
ne
l
f
a
il
in
g
t
o
de
te
c
t
a
f
a
c
e
i
s
r
e
la
ti
ve
ly
l
ow
. T
he
R
-
N
e
t
in
c
or
por
a
te
s
t
he
pr
e
c
i
s
e
c
oor
di
na
te
s
of
t
he
upda
te
d bounding boxes
,
w
hi
c
h
a
r
e
m
or
e
a
c
c
ur
a
te
th
a
n
pr
e
vi
ou
s
ve
r
s
io
n
s
.
T
a
bl
e
4
pr
ovi
de
s
te
c
hni
c
a
l
in
f
or
m
a
ti
on
a
bout
th
e
th
r
e
e
a
lg
or
it
hm
s
,
pr
ovi
di
ng
in
s
ig
ht
s
in
to
th
e
ir
C
P
U
a
nd
G
P
U
us
a
ge
c
a
pa
bi
li
ti
e
s
,
s
uppor
t
f
or
c
ol
or
in
f
o
r
m
a
ti
on,
a
nd
r
e
c
om
m
e
nde
d i
m
a
ge
s
iz
e
s
.
T
he
ta
bl
e
pr
e
s
e
nt
s
te
c
hni
c
a
l
in
f
or
m
a
ti
on
on
th
r
e
e
w
id
e
ly
us
e
d
f
a
c
e
de
te
c
ti
on
a
nd
a
li
gnm
e
nt
m
e
th
ods
:
H
a
a
r
C
a
s
c
a
de
-
O
pe
nC
V
,
H
oG
-
D
li
b,
a
nd
M
T
C
N
N
.
A
ll
th
r
e
e
m
e
th
ods
s
uppor
t
C
P
U
us
a
ge
,
m
a
ki
ng
th
e
m
a
c
c
e
s
s
ib
le
f
or
s
ta
nd
a
r
d
c
om
put
in
g
de
vi
c
e
s
.
W
hi
le
H
a
a
r
C
a
s
c
a
de
a
nd
H
oG
-
D
li
b
a
ls
o
of
f
e
r
G
P
U
s
uppor
t,
M
T
C
N
N
s
ta
nd
s
out
by
ut
il
iz
in
g
c
ol
or
in
f
or
m
a
ti
on
f
o
r
e
nha
nc
e
d
a
c
c
ur
a
c
y,
w
hi
le
th
e
ot
he
r
two
m
e
th
ods
ope
r
a
te
on
gr
a
ys
c
a
le
im
a
g
e
s
.
R
e
c
om
m
e
nde
d
im
a
g
e
s
iz
e
s
v
a
r
y,
w
it
h
H
a
a
r
C
a
s
c
a
de
s
ugg
e
s
ti
ng
24×
24,
H
oG
-
D
li
b
>
80×
80,
a
nd
M
T
C
N
N
>
20×
20.
E
a
c
h
m
e
th
od
ha
s
uni
que
s
tr
e
ngt
hs
,
m
a
ki
ng
it
e
s
s
e
nt
ia
l
to
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4249
-
4262
4258
c
ons
id
e
r
s
pe
c
if
ic
a
ppl
ic
a
ti
on
r
e
qui
r
e
m
e
nt
s
a
nd
a
va
il
a
bl
e
c
om
put
a
ti
ona
l
r
e
s
our
c
e
s
w
he
n
s
e
le
c
ti
ng
th
e
m
o
s
t
s
ui
ta
bl
e
a
lg
or
it
hm
f
or
f
a
c
e
de
te
c
ti
on a
nd a
li
gnm
e
nt
t
a
s
k
s
.
T
a
bl
e
4.
T
e
c
hni
c
a
l
in
f
or
m
a
ti
on a
bout
t
he
t
hr
e
e
a
lg
or
it
hm
s
H
a
a
r
C
a
s
c
a
de
H
oG
-
dl
i
b
M
T
C
N
N
C
P
U
Y
e
s
, a
va
i
l
a
bl
e
Y
e
s
, a
va
i
l
a
bl
e
Y
e
s
, a
va
i
l
a
bl
e
G
P
U
C
e
r
t
a
i
n i
m
pl
e
m
e
nt
a
t
i
ons
(
N
ot
O
pe
nC
V
)
Y
e
s
, a
va
i
l
a
bl
e
Y
e
s
, a
va
i
l
a
bl
e
U
s
i
ng
C
ol
or
s
No
No
Y
e
s
I
m
a
ge
s
i
z
e
R
e
c
om
m
e
nds
24x24
>80x80
>20x20
3.3. E
xp
e
r
im
e
n
t
s
an
d
r
e
s
u
lt
s
W
e
unde
r
to
ok
two
di
s
ti
nc
t
e
xpe
r
im
e
nt
s
to
a
s
s
e
s
s
th
e
pe
r
f
or
m
a
nc
e
of
th
r
e
e
a
lg
or
it
hm
s
a
c
r
os
s
va
r
ie
d
c
ont
e
xt
s
.
I
n
th
e
f
ir
s
t
e
xpe
r
im
e
nt
,
w
e
ga
uge
d
pe
r
f
or
m
a
nc
e
b
y
qua
nt
if
yi
ng
th
e
s
uc
c
e
s
s
f
ul
f
a
c
e
de
te
c
ti
on
s
a
c
hi
e
ve
d.
F
or
th
e
s
e
c
ond
e
xpe
r
im
e
nt
,
w
e
e
m
pl
oy
e
d
w
id
e
ly
r
e
c
ogni
z
e
d
m
e
tr
ic
s
to
th
or
oughly
e
va
lu
a
te
th
e
pe
r
f
or
m
a
nc
e
of
th
e
s
e
th
r
e
e
a
lg
or
it
hm
s
.
T
he
s
e
e
xpe
r
im
e
nt
s
ut
il
iz
e
P
yt
hon
s
c
r
ip
ts
w
it
h
va
r
io
us
li
br
a
r
ie
s
f
or
f
a
c
e
de
te
c
ti
on
in
im
a
ge
s
.
T
he
f
ir
s
t
s
c
r
ip
t
e
m
pl
oys
O
p
e
nC
V
,
it
e
r
a
ti
ng
th
r
ough
a
di
r
e
c
to
r
y
of
im
a
ge
s
a
nd
ut
il
iz
in
g
th
e
H
a
a
r
C
a
s
c
a
de
c
la
s
s
if
ie
r
to
de
te
c
t
f
a
c
e
s
.
T
he
to
ta
l
c
ount
of
de
te
c
te
d
f
a
c
e
s
is
th
e
n
pr
in
te
d.
S
im
il
a
r
ly
,
th
e
s
e
c
ond
s
c
r
ip
t
us
e
s
th
e
D
li
b
li
br
a
r
y
a
lo
ngs
id
e
O
pe
nC
V
to
de
te
c
t
f
a
c
e
s
in
im
a
ge
s
.
I
t
it
e
r
a
te
s
th
r
ough
th
e
im
a
ge
di
r
e
c
to
r
y,
a
ppl
ie
s
f
a
c
e
de
te
c
ti
on
u
s
in
g
D
li
b
,
a
nd
pr
in
ts
th
e
to
ta
l
c
ount
of
de
te
c
te
d
f
a
c
e
s
.
F
in
a
ll
y,
th
e
th
i
r
d
s
c
r
ip
t
e
m
pl
oys
th
e
M
T
C
N
N
m
ode
l
f
or
f
a
c
e
de
te
c
ti
on,
a
ga
in
it
e
r
a
ti
ng
th
r
ough
th
e
i
m
a
ge
di
r
e
c
to
r
y,
a
ppl
yi
ng
f
a
c
e
de
te
c
ti
on
us
in
g
M
T
C
N
N
,
a
nd
pr
in
ti
ng
th
e
to
ta
l
c
ount
of
de
te
c
te
d
f
a
c
e
s
.
O
ve
r
a
ll
,
th
e
s
e
e
xpe
r
im
e
nt
s
d
e
m
ons
tr
a
te
di
f
f
e
r
e
nt
a
ppr
oa
c
he
s
t
o f
a
c
e
de
t
e
c
ti
on us
in
g va
r
io
us
l
ib
r
a
r
ie
s
i
n P
yt
hon.
T
o
c
om
pr
e
he
ns
iv
e
ly
e
va
lu
a
te
th
e
pe
r
f
or
m
a
nc
e
of
th
e
a
lg
or
it
hm
s
us
in
g
va
r
io
us
m
e
tr
ic
s
,
w
e
a
ugm
e
nt
e
d t
he
e
xi
s
ti
ng da
ta
s
e
ts
by i
nt
r
oduc
in
g a
ddi
ti
ona
l
im
a
g
e
s
t
ha
t
di
d not c
ont
a
in
t
he
dr
iv
e
r
. B
y i
nc
lu
di
ng
th
e
s
e
im
a
ge
s
w
it
hout
dr
iv
e
r
s
pr
e
s
e
nt
,
w
e
c
oul
d
c
a
lc
ul
a
te
T
N
a
nd,
F
P
,
a
nd
ot
he
r
e
va
lu
a
ti
on
m
e
tr
ic
s
.
T
hi
s
a
ppr
oa
c
h
a
ll
ow
e
d
us
to
ga
in
in
s
ig
ht
s
in
to
th
e
a
lg
o
r
it
hm
s
’
e
f
f
e
c
ti
ve
ne
s
s
in
de
te
c
ti
ng
f
a
c
e
s
a
nd
a
li
gni
ng
f
e
a
tu
r
e
s
pr
e
c
is
e
ly
in
dr
iv
in
g
-
r
e
la
te
d
s
c
e
na
r
io
s
,
th
us
pr
ovi
di
ng
a
m
or
e
r
obus
t
a
s
s
e
s
s
m
e
nt
of
th
e
ir
pe
r
f
or
m
a
nc
e
in
a
r
e
a
l
-
w
or
ld
c
ont
e
xt
.
H
e
r
e
i
s
a
n e
x
pl
a
n
a
ti
on
of
e
a
c
h m
e
tr
i
c
:
P
r
e
c
is
io
n
i
s
a
m
e
tr
i
c
t
h
a
t
qu
a
n
ti
f
i
e
s
t
he
a
c
c
ur
a
c
y
of
p
os
it
i
ve
pr
e
di
c
t
io
n
s
m
a
de
b
y
a
m
o
de
l:
pr
e
c
i
s
i
on
=
T
P
/
(
T
P
+
F
P
)
.
R
e
c
a
ll
,
a
ls
o
k
no
w
n
a
s
s
e
n
s
it
iv
i
ty
or
TP
r
a
te
,
m
e
a
s
ur
e
s
t
he
m
od
e
l
’
s
a
bi
li
ty
to
c
or
r
e
c
tl
y
i
de
nt
if
y
po
s
i
ti
v
e
in
s
t
a
n
c
e
s
f
r
o
m
a
ll
t
he
a
c
tu
a
l
po
s
i
ti
v
e
in
s
t
a
nc
e
s
in
th
e
da
t
a
s
e
t.
R
e
c
a
ll
=
T
P
/(
T
P
+
F
N
)
.
T
h
e
F
1
-
S
c
or
e
i
s
th
e
ha
r
m
oni
c
m
e
a
n
of
pr
e
c
i
s
io
n
a
n
d
r
e
c
a
l
l.
F1
-
S
c
or
e
=
2
×
(
P
r
e
c
is
io
n
×
R
e
c
a
l
l)
/(
P
r
e
c
i
s
io
n
+
R
e
c
a
ll
)
.
A
c
c
ur
a
c
y
i
s
a
m
e
tr
ic
th
a
t
m
e
a
s
ur
e
s
th
e
ov
e
r
a
ll
c
or
r
e
c
tn
e
s
s
of
t
he
m
o
de
l
’
s
pr
e
d
ic
ti
o
n
s
.
A
c
c
ur
a
c
y
=
(
T
P
+
T
N
)
/(
T
P
+
T
N
+
F
P
+
F
N
)
.
3.3.1. F
ac
e
d
e
t
e
c
t
io
n
on
f
e
i
d
at
as
e
t
T
he
obj
e
c
ti
ve
of
th
e
e
xpe
r
im
e
nt
is
to
de
m
on
s
tr
a
te
how
w
e
l
l
e
a
c
h
m
e
th
od
c
a
n
de
te
c
t
f
a
c
e
s
on
di
f
f
e
r
e
nt
he
a
d m
ove
m
e
nt
s
a
nd i
n l
ow
l
ig
ht
in
g c
ondi
t
io
ns
.
W
e
a
s
s
e
s
s
e
d f
a
c
e
de
te
c
ti
on i
n t
he
10 he
a
d pos
it
io
ns
a
nd
di
f
f
e
r
e
nt
li
ght
in
g
c
ondi
ti
ons
s
e
pa
r
a
te
ly
to
und
e
r
s
ta
nd
ho
w
e
a
c
h
of
th
e
a
lg
or
it
hm
s
pe
r
f
or
m
s
in
th
e
s
e
s
c
e
na
r
io
s
.
I
n
th
is
e
xp
e
r
im
e
nt
a
l
s
e
tu
p,
th
e
a
im
is
to
d
e
te
c
t
f
a
c
e
s
us
in
g
th
e
F
E
I
da
ta
s
e
t,
w
he
r
e
e
a
c
h
di
s
ti
nc
t
he
a
d
m
ove
m
e
nt
s
ta
te
c
on
s
is
ts
of
a
n
e
qu
a
l
num
be
r
of
200
im
a
ge
s
.
A
ddi
ti
ona
ll
y,
th
e
r
e
i
s
a
s
e
t
of
200
im
a
ge
s
s
pe
c
if
ic
a
ll
y
r
e
pr
e
s
e
nt
in
g
lo
w
-
li
ght
c
ondi
ti
ons
.
T
he
e
xp
e
r
im
e
nt
in
vol
ve
s
a
ppl
yi
ng
th
r
e
e
di
f
f
e
r
e
nt
f
a
c
e
de
te
c
ti
on
a
lg
or
it
hm
s
to
e
a
c
h
s
ta
te
w
it
hi
n
th
e
da
ta
s
e
t.
F
or
e
ve
r
y
s
ta
te
a
nd
e
ve
r
y
f
a
c
e
de
te
c
ti
on
c
la
s
s
if
ie
r
,
th
e
c
ode
c
ount
s
th
e
to
ta
l
num
be
r
of
f
a
c
e
s
de
te
c
te
d.
T
he
go
a
l
is
to
a
na
ly
z
e
a
nd
c
om
pa
r
e
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
th
e
s
e
a
lg
or
it
hm
s
in
va
r
io
us
s
c
e
na
r
io
s
w
it
hi
n
th
e
F
E
I
da
ta
s
e
t,
in
c
lu
di
ng
di
f
f
e
r
e
nt
he
a
d
m
ove
m
e
nt
s
a
nd
li
ght
in
g
c
ondi
ti
ons
.
T
a
bl
e
5
s
um
m
a
r
iz
e
s
th
e
c
om
pa
r
is
on
r
e
s
ul
ts
of
th
e
th
r
e
e
a
lg
or
it
hm
s
f
r
om
th
e
in
it
ia
l
e
va
lu
a
ti
on
e
xpe
r
im
e
nt
.
I
t
di
s
pl
a
ys
th
e
out
c
om
e
s
a
c
hi
e
ve
d
f
or
e
a
c
h
s
e
c
ti
o
n
of
th
e
da
ta
s
e
t.
S
ta
te
s
1
to
10
r
e
pr
e
s
e
nt
th
e
dr
iv
e
r
'
s
he
a
d
in
va
r
io
us
180°
he
a
d
m
ove
m
e
nt
s
c
e
na
r
io
s
,
w
h
il
e
th
e
f
in
a
l
s
ta
te
,
11,
de
pi
c
t
s
th
e
dr
iv
e
r
in
lo
w
-
li
ght
in
g c
ondi
ti
ons
.
T
a
bl
e
5. D
e
te
c
ti
on r
a
te
f
or
on F
E
I
A
l
gori
t
hm
s
1
2
3
4
5
6
7
8
9
10
11
H
a
a
r
Ca
s
c
a
de
40
182
197
200
200
200
198
197
164
5
154
H
oG
-
dl
i
b
159
200
200
2000
200
200
200
200
196
95
177
M
T
CN
N
154
188
197
197
198
197
197
196
187
86
155
Evaluation Warning : The document was created with Spire.PDF for Python.