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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
, pp.
561
~
569
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
1
4
.i
1
.pp
561
-
569
561
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
D
e
t
e
c
t
i
n
g h
u
m
an
f
al
l
u
si
n
g
i
n
t
e
r
n
e
t
of
t
h
i
n
gs
d
e
vi
c
e
s f
or
h
e
al
t
h
c
ar
e
ap
p
l
i
c
at
i
on
s
Z
ak
ar
ia
B
e
n
h
ai
li
, Y
ou
s
s
e
f
B
al
ou
k
i,
L
ah
c
e
n
M
ou
m
ou
n
L
a
bor
a
t
or
y
of
M
a
t
he
m
a
t
i
c
s
, C
om
put
e
r
S
c
i
e
nc
e
a
nd E
ngi
ne
e
r
i
ng S
c
i
e
nc
e
s
,
F
a
c
u
l
t
y of
S
c
i
e
nc
e
s
a
nd
T
e
c
hni
que
s
,
H
a
s
s
a
n F
i
r
s
t
U
ni
ve
r
s
i
t
y of
S
e
t
t
a
t
,
S
e
t
t
a
t
, M
or
o
cco
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
D
e
c
7
,
2023
R
e
vi
s
e
d
J
ul
4
,
2024
A
c
c
e
pt
e
d
J
ul
26
,
2024
Falls po
se a
signific
ant thr
eat to
uninte
ntional
injurie
s, par
ticular
ly
impacting
the
independence
of
older
individuals.
Existing
detection
methods
suffe
r
from
drawbacks,
including
inaccuracies,
wearer
discomfort,
complex
setup,
resource
-
intensive
computation,
and
limitations
in
detecting
falls
ou
tside
a
specific
setting.
In
r
esponse,
our
innovat
ive
fall
detection
system
int
egrates
with
a
pneumatic
solution,
analyzing
fundamental
human
activiti
es
like
running,
walking,
and
sitting
,
both
indoors
and
outdoors.
This
ap
proach
combines
wearable
sensors
with
a
vision
-
based
solution,
utilizing
a
sm
art
belt
with
embedded
accelerometer
and
gyroscope,
alongside
wall
-
in
stalled
cameras
in
a
smart
house.
The
system
triggers
an
airbag
and
se
nds
an
emergency
alarm
upon
fall
detection
.
To
achieve
this,
we
propose
FallMi
xer
a
lightweight
deep
le
arning
model,
combined
with
‘
y
ou
only
look
once’
version
8
(YOLO
v8
)
algorithm,
fine
-
tuned
on
a
collected
video
dat
aset
to
enable
real
-
time
detection.
We
found
that
the
models
result
in
comp
etitive
performance,
as
demonstrated
on
SisFall,
UCI
human
activit
y
recognition
(
HAR
)
,
and
mobile
health
(
MHEALTH
)
datasets
with
a
remarkable
mean
a
verage
precision.
Subsequently,
we
assess
the
hardware
performance
of
our
solution on edge devic
es.
K
e
y
w
o
r
d
s
:
A
c
ti
vi
ty
r
e
c
ogni
ti
on
D
e
e
p l
e
a
r
ni
ng
F
a
ll
de
te
c
ti
on
I
nt
e
r
ne
t
of
t
hi
ngs
S
e
ns
or
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
:
Z
a
ka
r
ia
B
e
nha
il
i
L
a
bor
a
to
r
y
of
M
a
th
e
m
a
ti
c
s
, C
om
put
e
r
S
c
ie
nc
e
a
nd E
ngi
ne
e
r
in
g S
c
ie
nc
e
s
F
a
c
ul
ty
of
S
c
ie
nc
e
s
a
nd
T
e
c
hni
que
s
, H
a
s
s
a
n F
ir
s
t
U
ni
ve
r
s
it
y of
S
e
tt
a
t
S
e
tt
a
t
26000
, M
or
oc
c
o
E
m
a
il
:
z
.be
nha
il
i@uhp.a
c
.m
a
1.
I
N
T
R
O
D
U
C
T
I
O
N
H
um
a
n
a
c
ti
vi
ty
r
e
c
ogni
ti
on
(
H
A
R
)
ha
s
e
m
e
r
ge
d
a
s
a
c
r
uc
ia
l
a
r
e
a
of
r
e
s
e
a
r
c
h
w
it
hi
n
a
m
bi
e
nt
a
nd
c
ont
e
xt
-
a
w
a
r
e
c
om
put
in
g
[
1]
,
[
2]
.
I
t
ha
s
a
s
ubs
ta
nt
ia
l
im
pa
c
t
in
di
f
f
e
r
e
nt
f
ie
ld
s
,
s
uc
h
a
s
in
te
ll
ig
e
nt
s
ur
ve
il
la
nc
e
s
ys
te
m
s
[
3]
, he
a
lt
hc
a
r
e
[
4]
,
[
5]
,
hum
a
n
-
c
om
put
e
r
i
nt
e
r
a
c
ti
on
[
6
]
, a
nd e
ld
e
r
c
a
r
e
a
s
s
is
ta
n
c
e
[
7]
. W
hi
le
H
A
R
h
a
s
br
oa
d
a
ppl
ic
a
ti
ons
,
one
c
r
uc
ia
l
a
r
e
a
w
he
r
e
it
pl
a
ys
a
vi
ta
l
r
ol
e
is
in
f
a
ll
d
e
te
c
ti
on
f
or
e
ld
e
r
ly
pe
r
s
on
s
.
T
h
e
de
te
c
ti
on
of
f
a
ll
s
hol
ds
pa
r
a
m
ount
im
por
ta
nc
e
,
pa
r
ti
c
ul
a
r
ly
f
o
r
th
os
e
a
ge
d
65
a
nd
a
bove
,
a
s
it
r
e
pr
e
s
e
nt
s
a
s
ig
ni
f
ic
a
nt
publ
ic
he
a
lt
h
c
onc
e
r
n
a
nd
s
ta
nds
a
s
th
e
le
a
di
ng
c
a
us
e
of
in
ju
r
y
-
r
e
la
te
d
f
a
ta
li
ti
e
s
w
it
hi
n
th
is
de
m
ogr
a
phi
c
.
T
he
im
pl
e
m
e
nt
a
ti
on
of
r
e
a
l
-
ti
m
e
f
a
ll
de
te
c
ti
on
d
e
vi
c
e
s
hol
d
s
im
m
e
ns
e
va
lu
e
,
of
f
e
r
in
g
a
m
e
a
n
s
to
pr
om
pt
ly
s
um
m
on
a
s
s
is
ta
n
c
e
a
nd
th
e
r
e
by
pr
e
ve
nt
in
g
in
ju
r
ie
s
.
A
ddi
ti
ona
ll
y,
s
uc
h
d
e
vi
c
e
s
c
ont
r
ib
ut
e
to
a
n
e
nha
nc
e
d
qua
li
ty
of
li
f
e
f
o
r
s
e
ni
or
s
a
nd
pr
ovi
de
c
a
r
e
gi
ve
r
s
w
i
th
gr
e
a
te
r
he
lp
.
H
A
R
f
r
a
m
e
w
or
ks
e
na
bl
e
th
e
de
te
c
ti
on,
c
la
s
s
if
ic
a
ti
on,
a
nd
unde
r
s
ta
ndi
ng
of
s
pe
c
if
ic
body
m
o
ve
m
e
nt
or
a
c
ti
vi
ty
th
r
ough
da
ta
c
ol
le
c
te
d
f
r
om
di
ve
r
s
e
s
e
ns
or
s
.
I
n
li
te
r
a
tu
r
e
r
e
s
e
a
r
c
he
r
s
de
te
c
t
th
e
a
c
ti
vi
ti
e
s
th
r
ough
two
m
a
in
m
e
th
ods
:
vi
s
io
n
-
ba
s
e
d,
w
hi
c
h
a
na
ly
z
e
s
vi
de
o
or
im
a
ge
da
ta
c
a
pt
ur
e
d
by
c
a
m
e
r
a
s
,
a
nd
w
e
a
r
a
bl
e
s
e
ns
or
-
ba
s
e
d,
w
hi
c
h
le
ve
r
a
ge
s
da
ta
f
r
om
non
-
in
tr
us
iv
e
, e
m
be
dde
d s
e
ns
or
s
i
n s
m
a
r
t
de
vi
c
e
s
l
ik
e
s
m
a
r
tp
h
one
s
a
nd s
m
a
r
twa
t
c
he
s
.
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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
:
561
-
569
562
H
ow
e
ve
r
,
bui
ld
in
g
m
ode
ls
th
a
t
ge
ne
r
a
li
z
e
w
e
ll
a
c
r
os
s
va
r
io
us
a
c
ti
vi
ti
e
s
a
nd
s
e
ns
or
s
r
e
m
a
in
s
a
c
ha
ll
e
ngi
ng
ta
s
k.
H
um
a
n
a
c
ti
vi
ty
s
ig
na
ls
a
nd
im
a
ge
s
c
a
n
s
ig
ni
f
ic
a
nt
ly
va
r
y
a
m
ong
in
di
vi
dua
ls
a
nd
e
ve
n
f
or
th
e
s
a
m
e
i
ndi
vi
dua
l
pe
r
f
or
m
in
g a
n a
c
ti
vi
ty
a
t
s
e
pa
r
a
te
t
im
e
s
. A
ddi
ti
ona
ll
y, di
f
f
e
r
e
nt
a
c
ti
vi
ti
e
s
m
a
y e
xhi
bi
t
s
im
il
a
r
s
ig
na
l
pa
tt
e
r
ns
,
f
ur
th
e
r
c
om
pl
ic
a
ti
ng
a
c
ti
vi
ty
c
la
s
s
if
ic
a
ti
on.
T
r
a
di
ti
ona
ll
y,
r
e
s
e
a
r
c
he
r
s
ha
ve
u
s
e
d
ha
nd
c
r
a
f
te
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
ods
c
om
bi
ne
d
w
it
h
s
upe
r
vi
s
e
d
m
a
c
hi
ne
le
a
r
ni
ng
te
c
hni
que
s
li
k
e
k
-
ne
a
r
e
s
t
n
e
ig
hbor
s
(
K
N
N
)
[
8]
,
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
s
(
S
V
M
)
[
9]
,
de
c
is
io
n
tr
e
e
s
(
D
T
)
[
10]
,
a
nd
e
ns
e
m
bl
e
a
ppr
oa
c
he
s
f
or
c
la
s
s
if
ic
a
ti
on
[
11]
.
N
e
ve
r
th
e
le
s
s
,
th
e
s
e
a
ppr
oa
c
h
e
s
r
e
qui
r
e
dom
a
in
e
xpe
r
ti
s
e
,
r
ig
or
ous
da
ta
pr
e
-
pr
oc
e
s
s
in
g,
a
nd
m
ig
ht
la
c
k
f
le
xi
bi
li
ty
due
to
th
e
di
f
f
ic
ul
ty
of
e
s
ta
bl
is
hi
ng
s
pa
ti
a
l
a
nd
te
m
por
a
l
r
e
la
ti
ons
hi
ps
a
m
ong
ha
ndc
r
a
f
te
d
f
e
a
tu
r
e
s
.
R
e
c
e
nt
de
e
p
le
a
r
ni
ng
te
c
hni
que
s
ha
ve
ga
in
e
d
popula
r
it
y,
e
s
pe
c
ia
ll
y
in
a
r
e
a
s
li
ke
na
tu
r
a
l
la
ngua
ge
pr
oc
e
s
s
in
g
[
12]
,
im
a
ge
r
e
c
ogni
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on
[
13]
.
P
r
e
c
is
e
ly
m
ul
ti
la
ye
r
pe
r
c
e
pt
r
on
(
M
L
P
)
m
ix
e
r
s
[
14]
,
a
nove
l
a
r
c
hi
te
c
tu
r
e
th
a
t
ha
s
a
tt
r
a
c
te
d
c
ons
id
e
r
a
bl
e
in
te
r
e
s
t
in
na
t
ur
a
l
la
ngua
ge
pr
oc
e
s
s
in
g
a
nd
c
om
put
e
r
vi
s
io
n.
I
n
c
ont
r
a
s
t
to
th
e
popula
r
t
r
a
ns
f
or
m
e
r
[
15
]
,
th
a
t
r
e
li
e
s
on
s
e
lf
-
a
tt
e
nt
io
n
m
e
c
ha
ni
s
m
s
,
M
L
P
-
m
ix
e
r
s
us
e
M
L
P
s
f
or
pr
oc
e
s
s
in
g
in
put
da
ta
.
T
hi
s
a
lt
e
r
na
ti
ve
a
ppr
oa
c
h
pr
ovi
de
s
s
e
v
e
r
a
l
a
dva
nt
a
ge
s
,
s
u
c
h
a
s
s
im
pl
ic
it
y
a
nd
c
om
put
a
ti
ona
l
e
f
f
ic
ie
nc
y, l
e
a
di
ng t
o f
a
s
te
r
t
r
a
in
in
g t
im
e
s
a
nd
r
e
duc
e
d m
e
m
or
y c
ons
um
pt
i
on, whic
h
i
s
s
ui
ta
bl
e
f
or
in
te
r
ne
t
of
th
in
gs
(
I
oT
)
a
nd
e
dge
de
vi
c
e
s
.
G
iv
e
n
th
e
s
e
pr
om
is
in
g
c
ha
r
a
c
te
r
is
ti
c
s
,
w
e
in
ve
s
ti
ga
te
th
e
de
pl
oym
e
nt
of
th
is
c
la
s
s
of
ne
twor
k
f
or
ti
m
e
s
e
r
ie
s
d
a
ta
,
a
nd
t
he
ir
s
ui
ta
bi
li
ty
f
or
ta
s
ks
in
vol
vi
ng
lo
ng
-
r
a
nge
de
pe
nde
nc
ie
s
.
W
e
pr
e
s
e
nt
a
s
y
s
te
m
de
s
ig
n
a
im
e
d
a
t
e
ns
ur
in
g
th
e
s
a
f
e
ty
of
e
ld
e
r
ly
in
di
vi
dua
ls
w
it
hi
n
th
e
ir
hom
e
s
a
nd
m
in
im
iz
in
g
th
e
r
is
k
of
in
ju
r
y
in
th
e
e
ve
nt
of
a
f
a
ll
out
door
s
.
T
o
de
te
c
t
m
ove
m
e
nt
s
,
w
e
e
m
pl
oy
two
ty
pe
s
o
f
s
e
ns
or
s
:
a
c
c
e
le
r
om
e
te
r
s
a
nd
gyr
os
c
ope
s
.
T
he
s
e
s
e
ns
or
s
he
lp
in
te
r
pr
e
t
body
m
ot
io
ns
a
nd
id
e
nt
if
y
pot
e
nt
ia
l
f
a
ll
s
.
I
n
a
ddi
ti
on,
w
e
us
e
c
a
m
e
r
a
s
s
tr
a
te
gi
c
a
ll
y
in
s
ta
ll
e
d
on
th
e
w
a
ll
s
.
T
o
a
c
c
om
pl
is
h
th
is
obj
e
c
ti
ve
,
w
e
in
tr
oduc
e
F
a
ll
M
ix
e
r
,
a
de
e
p
n
e
ur
a
l
ne
twor
k
de
s
ig
ne
d
to
e
xpl
oi
t
s
e
n
s
or
y
da
ta
a
nd
e
xt
r
a
c
t
te
m
p
or
a
l
f
e
a
tu
r
e
s
f
r
om
it
.
F
or
r
e
a
l
-
ti
m
e
f
a
ll
de
te
c
ti
on,
w
e
ut
il
iz
e
d
Y
O
L
O
v8,
a
s
ta
te
-
of
-
th
e
-
a
r
t
obj
e
c
t
de
te
c
ti
on
s
ys
t
e
m
.
W
he
r
e
f
or
tr
a
in
in
g
a
nd
te
s
ti
ng
w
e
e
m
pl
oy
e
d
th
r
e
e
da
t
a
s
e
t
s
f
r
om
w
e
a
r
a
bl
e
s
e
ns
or
s
,
a
lo
ngs
id
e
a
c
ol
le
c
te
d
vi
d
e
o
da
ta
s
e
t
f
e
a
tu
r
in
g
in
s
ta
nc
e
s
of
f
a
ll
s
.
T
o
im
pl
e
m
e
nt
th
is
s
ol
ut
io
n,
w
e
pr
opos
e
e
m
be
ddi
ng
th
e
s
ys
te
m
in
to
a
s
m
a
r
t
be
lt
or
s
m
a
r
t
ja
c
ke
t.
T
hi
s
w
e
a
r
a
bl
e
de
vi
c
e
w
il
l
he
lp
to
de
te
c
t
pr
e
-
f
a
ll
a
c
ti
ons
us
in
g
th
e
c
a
pa
bi
li
ti
e
s
of
F
a
ll
M
ix
e
r
.
I
n
th
e
e
ve
nt
of
a
pot
e
nt
ia
l
f
a
ll
, t
he
s
ys
te
m
c
a
n t
r
ig
ge
r
t
he
pr
ot
e
c
ti
on
m
e
c
ha
ni
s
m
, s
uc
h a
s
de
pl
oyi
ng a
n a
ir
ba
g, t
o s
a
f
e
gua
r
d
s
e
ns
it
iv
e
bone
s
a
nd
m
in
im
iz
e
th
e
r
is
k
of
in
ju
r
y.
T
he
in
te
gr
a
t
io
n
of
vi
s
ua
l
de
te
c
ti
on
f
ur
th
e
r
c
ont
r
ib
ut
e
s
to
r
e
duc
in
g f
a
ls
e
pos
it
iv
e
s
a
nd ne
ga
ti
ve
s
,
e
ns
ur
in
g t
he
a
r
r
iv
a
l
of
t
he
a
le
r
t
to
t
he
m
e
di
c
a
l
pe
r
s
onne
l.
I
n
th
is
r
e
s
e
a
r
c
h,
our
c
ont
r
ib
ut
io
n
li
e
s
in
pr
e
s
e
nt
in
g
a
r
e
s
il
ie
nt
a
ppr
oa
c
h
f
or
f
a
ll
de
te
c
ti
on
f
or
e
ld
e
r
ly
pe
opl
e
w
it
hi
n
a
hous
e
hol
d,
le
ve
r
a
gi
ng
de
e
p
le
a
r
ni
ng
a
lg
or
it
hm
s
.
I
n
c
ont
r
a
s
t
to
pr
io
r
s
tu
di
e
s
th
a
t
c
onc
e
nt
r
a
te
on
s
in
gl
e
-
m
oda
li
ty
de
te
c
ti
on, our
nove
l
a
ppr
oa
c
h a
dopt
s
a
dua
l
-
s
e
ns
or
s
tr
a
te
gy t
o a
ugm
e
nt
de
te
c
ti
on c
a
p
a
bi
li
ti
e
s
.
O
ur
pr
opos
e
d
a
lg
or
it
hm
e
xhi
bi
ts
th
e
a
bi
li
ty
to
id
e
nt
if
y
a
di
ve
r
s
e
r
a
nge
of
a
c
ti
on
s
,
s
ur
pa
s
s
in
g
th
e
p
e
r
f
or
m
a
nc
e
of
e
xi
s
ti
ng
m
e
th
odol
ogi
e
s
,
pa
r
ti
c
ul
a
r
ly
in
f
a
ll
de
te
c
ti
on.
A
ddi
ti
ona
ll
y
,
w
e
e
xpl
or
e
th
e
pr
a
c
ti
c
a
li
ty
of
im
pl
e
m
e
nt
in
g our
a
ppr
oa
c
h on e
dge
de
vi
c
e
s
, hi
ghl
ig
ht
in
g i
ts
po
te
nt
ia
l
f
or
r
e
a
l
-
w
or
ld
a
ppl
ic
a
ti
ons
.
T
he
r
e
m
a
in
de
r
of
th
e
pa
pe
r
unf
ol
ds
in
a
s
tr
uc
tu
r
e
d
m
a
nne
r
.
S
e
c
ti
on
2
p
r
ovi
de
s
a
n
ove
r
vi
e
w
o
f
m
e
th
odol
ogi
e
s
pr
opos
e
d
by
ot
he
r
r
e
s
e
a
r
c
he
r
s
in
th
e
f
ie
ld
.
T
he
s
pe
c
if
ic
s
of
our
pr
opo
s
e
d
m
e
th
od
a
r
e
out
li
ne
d
in
s
e
c
ti
on
3.
S
e
c
ti
on
4
pr
e
s
e
nt
s
th
e
e
xp
e
r
im
e
nt
a
l
s
e
tu
p,
de
ta
il
i
ng
th
e
c
onf
ig
ur
a
ti
on
a
nd
pa
r
a
m
e
te
r
s
us
e
d
f
or
e
va
lu
a
ti
on.
R
e
s
ul
ts
a
nd
di
s
c
us
s
io
ns
a
r
e
c
om
pr
e
he
ns
iv
e
ly
c
ove
r
e
d
in
s
e
c
ti
on
5,
a
nd
la
s
t,
w
e
c
onc
lu
de
th
e
pa
pe
r
.
2.
R
E
L
A
T
E
D
WORKS
S
e
ve
r
a
l
not
e
w
or
th
y
s
tu
di
e
s
ha
ve
c
ont
r
ib
ut
e
d
to
a
dva
nc
in
g
th
e
a
c
c
ur
a
c
y
a
nd
e
f
f
ic
ie
nc
y
of
de
te
c
ti
ng
f
a
ll
s
a
nd
r
e
la
te
d
a
c
ti
vi
ti
e
s
.
S
e
ngül
e
t
al
.
[
16]
e
xpl
or
e
d
de
e
p
le
a
r
ni
ng
-
ba
s
e
d
f
a
ll
de
te
c
ti
on
us
in
g
s
m
a
r
twa
tc
he
s
,
w
it
h
a
pr
im
a
r
y
e
m
pha
s
is
on
he
a
lt
hc
a
r
e
a
ppl
ic
a
ti
ons
.
T
h
e
ir
m
e
th
od
in
vol
ve
d
c
ol
l
e
c
ti
ng
gyr
os
c
op
e
a
nd
a
c
c
e
le
r
om
e
te
r
us
in
g
a
s
m
a
r
twa
tc
h,
a
ugm
e
nt
e
d
by
a
n
in
te
r
pol
a
ti
on
te
c
hni
que
.
T
hi
s
a
ppr
oa
c
h
hi
gh
li
ght
e
d
e
f
f
ic
a
c
y i
n a
c
c
ur
a
te
ly
i
de
nt
if
yi
ng i
ns
t
a
nc
e
s
of
f
a
ll
in
g.
T
hi
s
a
ug
m
e
nt
a
ti
on t
e
c
hni
que
h
a
s
a
dva
nc
e
d t
he
a
c
c
ur
a
c
y
of
bi
di
r
e
c
ti
ona
l
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
Bi
-
L
S
T
M
)
a
lg
or
it
hm
to
r
e
a
c
h
99%
,
c
ol
le
c
t
e
d
da
ta
h
a
ve
s
ubm
it
te
d
th
e
c
r
e
a
ti
on of
38
f
e
a
tu
r
e
s
.
I
n a
not
he
r
w
or
k
[
17]
,
a
f
a
ll
d
e
te
c
ti
on
s
ys
te
m
,
us
in
g
opt
im
iz
e
d c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
(
C
N
N
)
a
nd
w
e
a
r
a
bl
e
I
oT
s
e
n
s
or
da
ta
, w
a
s
de
ve
lo
pe
d.
D
a
ta
f
r
om
14 i
ndi
vi
dua
ls
a
nd
s
ix
s
e
ns
or
s
w
e
r
e
pr
oc
e
s
s
e
d
to
e
xt
r
a
c
t
di
s
ti
nc
ti
ve
f
e
a
tu
r
e
s
,
w
hi
c
h
w
e
r
e
th
e
n
r
e
duc
e
d
us
in
g
m
ul
ti
li
ne
a
r
pr
in
c
ip
a
l
c
om
pone
nt
a
na
ly
s
is
.
A
n
8
-
la
ye
r
,
A
le
xN
e
t
-
ba
s
e
d
C
N
N
s
ig
ni
f
ic
a
nt
ly
a
dva
n
c
e
d
f
a
ll
de
te
c
ti
on.
A
not
he
r
a
ppr
oa
c
h,
pr
e
s
e
nt
e
d
by
L
e
e
e
t
al
.
[
18]
,
us
e
s
f
oot
pl
a
nt
a
r
pr
e
s
s
ur
e
a
nd
a
c
c
e
le
r
a
ti
on
da
ta
.
T
he
ir
in
nova
ti
ve
s
ys
te
m
e
f
f
e
c
ti
ve
ly
di
s
ti
ngui
s
he
d
be
twe
e
n
va
r
io
us
a
c
ti
vi
ti
e
s
of
da
il
y
li
vi
ng
(
A
D
L
ta
s
ks
)
a
nd
di
ve
r
s
e
ty
pe
s
of
f
a
ll
s
.
B
y
ut
il
iz
in
g
D
T
s
a
nd
a
th
r
e
s
hol
d
te
c
hni
que
,
th
e
m
e
th
od
a
c
hi
e
ve
d
a
95%
a
c
c
ur
a
c
y
r
a
te
in
r
e
c
ogni
z
in
g
f
a
ll
a
c
ti
vi
ti
e
s
a
t
a
n
a
ve
r
a
ge
s
pe
e
d
of
317
m
il
li
s
e
c
onds
.
L
uo
e
t
al
.
[
19]
in
tr
oduc
e
d
a
bi
na
r
y
c
onvolut
io
na
l
ne
twor
k
ta
il
or
e
d
f
or
r
e
a
l
-
ti
m
e
H
A
R
. T
he
ir
f
oc
us
on i
m
pr
ovi
ng f
a
ll
de
te
c
ti
on us
in
g w
e
a
r
a
bl
e
s
e
ns
or
da
ta
, pa
r
ti
c
ul
a
r
ly
f
or
m
obi
le
de
vi
c
e
s
, l
e
d
th
e
m
t
o i
nc
or
por
a
te
di
la
te
d c
onvolut
io
ns
t
o e
f
f
e
c
ti
ve
ly
c
a
pt
ur
e
f
e
a
tu
r
e
s
f
r
om
t
i
m
e
s
e
r
ie
s
da
ta
.
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
e
te
c
ti
ng human fall
us
in
g i
nt
e
r
ne
t
of
t
hi
ngs
d
e
v
ic
e
s
f
or
he
al
th
c
ar
e
appli
c
at
io
ns
(
Z
ak
ar
ia
B
e
nhai
li
)
563
W
e
a
r
a
bl
e
s
,
be
in
g
b
a
tt
e
r
y
-
ope
r
a
te
d
w
it
h
li
m
it
e
d
noni
nva
s
iv
e
a
p
pl
ic
a
bi
li
ty
,
ha
ve
le
d
ot
he
r
r
e
s
e
a
r
c
he
r
s
to
a
dopt
a
vi
s
ua
l
a
ppr
oa
c
h
f
or
f
a
ll
de
te
c
ti
on.
T
he
y
in
ve
s
ti
ga
te
hum
a
n
pos
tu
r
e
e
s
ti
m
a
ti
on
us
in
g
va
r
io
us
ty
pe
s
of
c
a
m
e
r
a
s
.
I
nt
e
r
e
s
ti
ngl
y,
m
o
s
t
of
th
e
s
e
s
tu
di
e
s
u
s
e
v
a
r
io
us
ty
pe
s
of
C
N
N
or
Y
O
L
O
a
lg
or
it
hm
s
f
or
r
e
a
l
-
ti
m
e
de
te
c
ti
on.
H
a
s
a
n
e
t
al
.
[
20]
u
s
e
a
two
-
la
ye
r
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
L
S
T
M
)
on
vi
de
o
d
a
ta
,
ut
il
iz
in
g
th
e
O
pe
nP
os
e
a
lg
or
it
hm
f
or
2D
body
po
s
e
e
s
ti
m
a
ti
on,
a
nd
e
m
pl
oyi
ng
s
e
que
nt
ia
l
f
r
a
m
e
s
f
or
f
a
ll
id
e
nt
if
ic
a
ti
on.
F
e
ng
e
t
al
.
[
21]
a
ddr
e
s
s
e
d
c
ha
ll
e
nge
s
in
c
om
pl
e
x
e
nvi
r
onm
e
nt
s
f
or
pe
de
s
tr
ia
n
de
te
c
ti
on,
by
in
te
gr
a
ti
ng
Y
O
L
O
v3,
D
e
e
p
-
S
or
t
f
or
obj
e
c
t
tr
a
c
ki
ng,
a
nd
V
G
G
-
16
w
it
h
a
n
a
tt
e
nt
io
n
-
L
S
T
M
ne
twor
k
f
or
f
a
ll
de
te
c
ti
on.
T
hi
s
s
how
c
a
s
e
d
a
da
pt
a
bi
li
ty
in
in
tr
ic
a
te
s
c
e
na
r
io
s
.
F
e
i
e
t
al
.
[
22]
a
ppr
oa
c
he
d
f
a
ll
de
te
c
ti
on
by
us
in
g
th
e
F
lo
w
-
pos
it
io
n
N
e
t
m
ode
l,
c
om
bi
ni
ng
opt
ic
a
l
f
lo
w
a
nd
hum
a
n
pos
e
da
ta
.
N
ot
a
bl
y,
th
e
ir
s
y
s
te
m
de
m
ons
tr
a
te
d
good
a
c
c
ur
a
c
y
on
two
op
e
n
da
ta
s
e
ts
U
R
F
D
a
nd
L
e
2i
,
hi
ghl
ig
ht
in
g
it
i
s
pot
e
nt
ia
l
f
or
r
e
li
a
bl
e
f
a
ll
de
te
c
ti
on.
R
e
s
e
a
r
c
h
by
K
a
n
e
t
al
.
[
23]
,
th
e
Y
O
L
O
n
e
twor
k
unde
r
goe
s
m
odi
f
ic
a
ti
ons
a
nd
e
nha
n
c
e
m
e
nt
s
in
c
onj
unc
ti
on
w
it
h
ot
he
r
m
odul
e
s
to
a
tt
a
in
a
c
om
m
e
nd
a
bl
e
m
e
a
n
a
ve
r
a
ge
pr
e
c
is
io
n
(
m
A
P
)
a
nd
gui
de
s
th
e
de
ve
lo
pm
e
nt
of
a
li
ght
w
e
ig
ht
f
a
ll
de
te
c
ti
on ne
twor
k.
T
he
pr
e
c
e
di
ng
m
e
th
odol
ogi
e
s
ha
ve
pr
im
a
r
il
y
c
onc
e
nt
r
a
te
d
on
f
a
ll
de
te
c
ti
on
th
r
ough
va
r
io
us
s
e
ns
or
m
oda
li
ti
e
s
but
ha
ve
not
pr
io
r
it
iz
e
d
th
e
a
nt
ic
ip
a
ti
on
of
a
f
a
ll
e
ve
nt
,
in
it
ia
ti
ng
a
w
a
r
ni
ng
s
ig
na
l
in
a
dva
nc
e
of
a
f
a
ll
oc
c
ur
r
e
nc
e
hol
ds
th
e
pot
e
nt
ia
l
f
or
li
f
e
s
a
vi
ng
in
te
r
ve
nt
io
ns
.
F
ur
th
e
r
m
or
e
,
a
r
e
f
in
e
d
a
ppr
oa
c
h
e
nt
a
il
s
di
s
ti
ngui
s
hi
ng
be
twe
e
n
in
door
a
nd
out
door
f
a
ll
de
te
c
ti
on,
e
nha
nc
in
g
th
e
s
ys
te
m
’
s
r
e
s
il
ie
nc
e
by
c
ons
id
e
r
in
g
th
e
di
s
ti
nc
ti
ve
e
nvi
r
onm
e
nt
a
l
a
tt
r
ib
ut
e
s
of
e
a
c
h
s
e
tt
in
g.
C
on
s
id
e
r
in
g
th
e
s
e
c
ons
id
e
r
a
ti
o
ns
,
w
e
put
f
or
th
our
a
ppr
oa
c
h,
w
hi
c
h
c
e
nt
e
r
s
on
th
e
de
te
c
ti
on
of
f
a
ll
s
th
r
ough
th
e
in
te
gr
a
te
d
us
e
of
bot
h
w
e
a
r
a
bl
e
s
e
n
s
or
s
a
nd
s
ur
ve
il
la
nc
e
c
a
m
e
r
a
s
.
3.
M
E
T
H
O
D
I
n
th
is
s
e
c
ti
on,
w
e
pr
e
s
e
nt
our
a
ppr
oa
c
h
f
or
a
c
ti
vi
ty
de
te
c
ti
on,
w
it
h
a
s
pe
c
if
ic
f
oc
us
on
id
e
nt
if
yi
ng
f
a
ll
s
.
L
e
ve
r
a
gi
ng
w
e
a
r
a
bl
e
s
e
ns
or
s
s
e
a
m
le
s
s
ly
in
te
gr
a
te
d
in
to
s
m
a
r
tp
hone
s
,
our
s
ys
te
m
pr
om
pt
ly
r
e
c
ogni
z
e
s
f
a
ll
pa
tt
e
r
ns
,
tr
ig
ge
r
in
g
a
n
a
le
r
t
to
in
it
ia
te
a
pr
ot
e
c
ti
ve
m
e
c
ha
n
is
m
.
T
hi
s
m
e
c
ha
ni
s
m
ta
ke
s
th
e
f
or
m
of
a
b
e
lt
a
ir
ba
g
de
s
ig
ne
d
to
s
a
f
e
gua
r
d
vul
ne
r
a
bl
e
bone
s
.
I
nt
r
oduc
in
g
F
a
ll
M
ix
e
r
,
a
ne
ur
a
l
ne
twor
k,
w
e
e
va
lu
a
te
it
s
pe
r
f
or
m
a
nc
e
th
r
ough
im
pl
e
m
e
nt
a
ti
on
on
a
s
m
a
r
tp
hone
.
T
he
a
n
a
ly
s
is
c
oul
d
a
ls
o
be
don
e
on
c
l
oud
s
e
r
ve
r
s
to
f
ur
th
e
r
e
nha
nc
e
de
te
c
ti
on
c
a
pa
bi
li
ti
e
s
.
W
e
in
c
or
por
a
te
s
ur
ve
il
la
nc
e
c
a
m
e
r
a
s
ta
il
or
e
d
f
or
s
m
a
r
t
hom
e
s
c
e
na
r
io
s
.
T
he
n e
m
pl
oyi
ng Y
O
L
O
[
24]
,
w
e
t
r
a
in
t
he
m
ode
l
us
in
g a
c
ol
le
c
te
d da
ta
s
e
t,
e
na
bl
in
g i
t
to
e
f
f
e
c
ti
ve
ly
t
r
a
c
k a
nd
de
te
c
t
f
a
ll
s
w
it
h
pr
e
c
is
io
n.
T
hi
s
c
om
pr
e
h
e
ns
iv
e
s
ys
te
m
not
onl
y
le
ve
r
a
ge
s
w
e
a
r
a
bl
e
t
e
c
hnol
ogy
but
a
l
s
o
in
te
gr
a
te
s
s
ophi
s
ti
c
a
te
d
c
a
m
e
r
a
-
ba
s
e
d
s
ur
ve
il
la
nc
e
,
e
ns
ur
in
g
a
r
obus
t
a
nd
r
e
li
a
bl
e
a
ppr
oa
c
h
to
f
a
ll
de
te
c
ti
on
a
nd pr
ot
e
c
ti
on.
F
ig
ur
e
1
pr
e
s
e
nt
s
t
he
de
ta
il
s
of
t
he
s
ol
ut
io
n.
F
ig
ur
e
1
. F
a
ll
de
te
c
ti
on s
ys
te
m
i
n s
m
a
r
t
hom
e
us
in
g c
a
m
e
r
a
a
nd
w
e
a
r
a
bl
e
s
e
n
s
or
s
3.1. F
al
lM
ix
e
r
ar
c
h
it
e
c
t
u
r
e
T
he
pr
opos
e
d
a
r
c
hi
te
c
tu
r
e
pr
e
s
e
nt
e
d
in
F
ig
ur
e
2
i
s
in
s
pi
r
e
d
f
r
om
th
e
pa
pe
r
[
25]
w
e
na
m
e
d
it
“
F
a
ll
M
ix
e
r
”
it
is
de
s
ig
ne
d
to
a
ddr
e
s
s
th
e
c
ha
ll
e
nge
of
e
f
f
ic
ie
nt
ly
pr
oc
e
s
s
in
g
ti
m
e
s
e
r
ie
s
da
ta
,
us
in
g
li
ght
w
e
ig
ht
ope
r
a
ti
ons
.
T
he
m
ode
l
ut
il
iz
e
s
a
c
om
bi
na
ti
on
of
c
onvolut
io
na
l
a
nd
m
ix
e
r
la
ye
r
s
to
e
xt
r
a
c
t
r
e
le
va
nt
f
e
a
tu
r
e
s
a
nd
f
a
c
il
it
a
te
i
nf
or
m
a
ti
on
f
lo
w
.
A
r
c
hi
te
c
tu
r
e
’
s
i
nput
:
it
i
s
a
one
-
di
m
e
ns
io
na
l
ti
m
e
s
e
r
ie
s
da
ta
, w
hi
c
h i
s
r
e
s
h
a
pe
d t
o
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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
:
561
-
569
564
m
a
ke
it
c
om
pa
ti
bl
e
w
it
h
c
onvolut
io
na
l
la
ye
r
s
.
L
ik
e
t
he
or
ig
in
a
l
a
r
c
hi
te
c
tu
r
e
w
e
a
ls
o
us
e
d
th
e
pa
tc
h
e
m
be
ddi
ng
la
ye
r
, s
pe
c
if
ic
a
ll
y w
e
a
ppl
ie
d a
2D
c
onvolut
io
n ope
r
a
ti
on w
it
h
a
pa
tc
h s
iz
e
e
qua
l
to
t
he
ke
r
ne
l
s
iz
e
, t
o c
a
pt
ur
e
lo
c
a
l
in
f
or
m
a
ti
on
f
r
om
th
e
in
put
s
e
que
nc
e
.
T
he
out
put
is
th
e
n
pa
s
s
e
d
th
r
ough
a
G
a
us
s
ia
n
e
r
r
or
li
ne
a
r
uni
t
(
G
E
L
U
)
a
c
ti
va
ti
on
f
unc
ti
on
f
ol
lo
w
e
d
by
ba
tc
h
nor
m
a
li
z
a
ti
on
to
in
tr
oduc
e
non
-
li
ne
a
r
it
y
a
nd
s
ta
bi
li
z
e
tr
a
in
in
g.
T
he
c
or
e
of
th
e
a
r
c
hi
te
c
tu
r
e
is
th
e
F
a
ll
M
ix
e
r
la
ye
r
c
om
pos
e
d
of
a
s
e
r
ie
s
of
de
pt
hw
is
e
c
onvolut
io
na
l
la
ye
r
s
.
E
a
c
h F
a
ll
M
ix
e
r
c
ons
is
ts
of
t
hr
e
e
m
a
in
c
om
pon
e
nt
s
.
C
onvolut
io
n
la
ye
r
th
a
t
us
e
s
a
de
pt
hw
is
e
c
onvolut
io
n
w
it
h
a
ke
r
ne
l
s
iz
e
of
two
to
id
e
nt
if
y
s
pa
ti
a
l
r
e
la
ti
ons
hi
ps
be
twe
e
n
n
e
ig
hbor
in
g
pa
tc
he
s
a
nd
to
c
a
pt
ur
e
de
pe
nde
nc
ie
s
w
it
hi
n
lo
c
a
l
pa
tc
he
s
of
th
e
in
put
s
e
que
nc
e
.
S
ki
p
c
onne
c
ti
on
th
a
t
ke
e
ps
da
ta
f
r
om
e
a
r
li
e
r
la
ye
r
s
i
n
th
e
m
ode
l.
T
hi
s
pr
om
ot
e
s
in
f
or
m
a
ti
on
f
lo
w
a
nd
r
e
duc
e
s
th
e
va
ni
s
hi
ng
gr
a
di
e
nt
,
w
hi
c
h
f
a
c
il
it
a
te
s
th
e
m
od
e
l
’
s
a
bi
li
ty
to
pi
c
k
up
pe
r
ti
ne
nt
r
e
pr
e
s
e
nt
a
ti
ons
.
P
oi
nt
w
is
e
c
onvolut
io
n
th
a
t
ta
ke
s
th
e
out
put
of
th
e
s
ki
p
c
onne
c
t
io
n
a
nd
pa
s
s
e
s
it
th
r
ough
a
1
×
1
c
onvolut
io
na
l
la
ye
r
to
a
ggr
e
ga
te
in
f
o
r
m
a
ti
on
a
nd
in
c
r
e
a
s
e
s
th
e
m
ode
l’
s
c
a
pa
c
it
y
to
c
a
pt
ur
e
m
o
r
e
c
om
pl
e
x
pa
tt
e
r
ns
in
th
e
da
ta
.
A
gl
oba
l
a
ve
r
a
ge
pooli
ng l
a
ye
r
a
nd a
de
n
s
e
l
a
ye
r
a
r
e
u
s
e
d i
n t
h
e
a
r
c
hi
te
c
tu
r
e
’
s
out
put
.
F
ig
ur
e
2
. T
he
a
r
c
hi
te
c
tu
r
e
of
F
a
ll
M
ix
e
r
us
e
d f
or
a
c
ti
vi
ty
r
e
c
ogn
it
io
n
3.2. Re
al
t
im
e
f
al
l
d
e
t
e
c
t
io
n
al
gor
it
h
m
I
n t
hi
s
s
tu
dy, we
ut
il
iz
e
t
he
Y
O
L
O
v8 ne
twor
k, t
he
l
a
te
s
t
it
e
r
a
ti
on i
n t
he
Y
O
L
O
f
a
m
il
y of
a
lg
o
r
it
hm
s
.
T
he
Y
O
L
O
v8 mod
e
l
c
om
pr
is
e
s
f
our
m
a
in
ne
twor
k
c
om
pone
nt
s
:
th
e
n
eck
-
e
nd
f
or
f
e
a
tu
r
e
f
us
io
n, t
he
b
a
c
kbone
ne
twor
k
f
or
e
xt
r
a
c
ti
ng
im
a
ge
f
e
a
tu
r
e
s
,
th
e
in
put
f
or
da
ta
e
nh
a
nc
e
m
e
nt
,
a
nd
th
e
de
c
oupl
e
d
he
a
d
e
r
out
put
,
w
hi
c
h
s
e
pa
r
a
te
s
th
e
c
la
s
s
if
ic
a
ti
on
f
r
om
th
e
de
te
c
ti
on
he
a
de
r
.
W
it
h
n
um
e
r
ous
im
pr
ove
m
e
nt
s
bui
lt
on
th
e
Y
O
L
O
v5
f
r
a
m
e
w
or
k,
Y
O
L
O
v8
s
ur
pa
s
s
e
s
Y
O
L
O
v5
in
t
e
r
m
s
of
s
pe
e
d
a
n
d
a
c
c
ur
a
c
y.
A
ddi
ti
ona
ll
y,
it
pr
ovi
de
s
a
uni
f
ie
d
f
r
a
m
e
w
or
k
f
or
t
r
a
in
in
g
m
ode
ls
th
a
t
m
a
na
ge
ta
s
ks
s
u
c
h
a
s
im
a
g
e
c
la
s
s
if
ic
a
ti
on,
obj
e
c
t
r
e
c
ogni
ti
on,
a
nd
in
s
ta
n
c
e
s
e
gm
e
nt
a
ti
on.
4.
E
X
P
E
R
I
M
E
N
T
S
E
T
U
P
T
hi
s
s
e
c
ti
on
of
f
e
r
s
a
c
om
pr
e
he
n
s
iv
e
ove
r
vi
e
w
of
th
e
da
ta
s
e
t
s
a
nd
c
onf
ig
ur
a
ti
ons
ut
il
iz
e
d
in
th
e
e
xpe
r
im
e
nt
s
c
onduc
te
d
in
th
is
s
tu
dy.
T
o
e
va
lu
a
te
our
m
ode
ls
a
c
r
os
s
va
r
io
us
s
e
tt
in
gs
,
w
e
s
e
l
e
c
te
d
f
our
da
ta
s
e
ts
,
th
r
e
e
publ
ic
ly
a
va
il
a
bl
e
a
nd
one
c
ol
le
c
te
d
by
our
te
a
m
.
E
a
c
h
da
ta
s
e
t
is
di
s
ti
ngui
s
he
d
by
uni
que
tr
a
it
s
a
nd
e
nvi
r
onm
e
nt
s
.
W
e
pr
ovi
de
de
ta
il
e
d
in
s
ig
ht
s
in
to
th
e
da
ta
s
e
t
pa
r
ti
ti
oni
ng
t
r
a
in
in
g
c
onf
ig
ur
a
ti
ons
e
m
pl
oye
d
in
our
s
tu
dy, a
im
in
g f
or
r
e
p
r
oduc
ib
il
it
y a
c
r
os
s
a
ll
a
na
ly
s
e
s
.
4.1.
D
at
as
e
t
s
d
e
s
c
r
ip
t
io
n
U
C
I
H
A
R
[
26]
da
ta
s
e
t
w
a
s
ga
th
e
r
e
d u
s
in
g a
n i
P
hone
6s
pl
a
c
e
d
in
t
he
f
r
ont
poc
ke
t
of
t
he
pa
r
ti
c
ip
a
nt
s
.
A
to
ta
l
of
24
pa
r
ti
c
ip
a
nt
s
,
in
c
lu
di
ng
10 w
om
e
n
a
nd
14
m
e
n,
w
i
th
di
ve
r
s
e
a
ge
s
,
w
e
ig
ht
s
,
a
nd
he
ig
ht
s
,
e
nga
g
e
d
in
s
ix
a
c
ti
vi
ti
e
s
a
c
r
os
s
15
tr
ia
ls
unde
r
c
on
s
is
te
nt
e
nvi
r
on
m
e
nt
a
l
c
ondi
ti
ons
.
T
h
e
a
c
ti
vi
ti
e
s
pe
r
f
or
m
e
d
w
e
r
e
w
a
lk
in
g,
jo
ggi
ng,
s
it
ti
ng,
s
ta
ndi
ng,
s
ta
ir
s
dow
n,
a
nd
s
ta
ir
s
up.
E
a
c
h
a
c
c
e
le
r
om
e
te
r
a
nd
gyr
os
c
ope
a
xi
s
in
th
e
da
ta
s
e
t
c
ont
a
in
s
1,304,950
s
a
m
pl
e
s
.
T
he
d
a
ta
w
a
s
c
ol
le
c
te
d
a
t
a
s
a
m
pl
in
g
r
a
te
of
50
H
z
,
r
e
s
ul
ti
ng
in
a
c
um
ul
a
ti
ve
r
e
c
or
di
ng
ti
m
e
o
f
435
m
in
ut
e
s
(
a
bout
7
a
nd
a
ha
lf
h
our
s
)
.
M
obi
le
he
a
lt
h
(
M
H
E
A
L
T
H
)
[
27]
da
ta
s
e
t
10
pa
r
ti
c
ip
a
nt
s
pe
r
f
or
m
s
ta
ti
c
a
nd
dyna
m
ic
a
c
ti
vi
ti
e
s
,
c
a
pt
ur
e
d
us
in
g
va
r
io
us
s
e
ns
or
s
,
in
c
lu
di
ng
th
r
e
e
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
e
te
c
ti
ng human fall
us
in
g i
nt
e
r
ne
t
of
t
hi
ngs
d
e
v
ic
e
s
f
or
he
al
th
c
ar
e
appli
c
at
io
ns
(
Z
ak
ar
ia
B
e
nhai
li
)
565
a
c
c
e
le
r
om
e
te
r
s
e
ns
or
s
, a
n
e
le
c
tr
oc
a
r
di
ogr
a
m
s
e
ns
or
,
two
gyr
os
c
ope
s
e
n
s
or
s
,
a
nd
th
r
e
e
m
a
gne
to
m
e
t
e
r
s
e
n
s
or
s
.
T
he
po
s
it
io
n
of
a
ll
th
e
s
e
s
e
ns
or
s
is
on
th
e
le
f
t
a
nkl
e
,
th
e
c
h
e
s
t,
a
nd
th
e
r
ig
ht
w
r
is
t.
M
H
E
A
L
T
H
e
nc
om
p
a
s
s
e
s
twe
lv
e
va
r
ie
d a
c
ti
ons
s
uc
h a
s
s
it
ti
ng, c
yc
li
ng, a
nd c
li
m
bi
ng s
ta
ir
s
.
S
is
F
a
ll
[
28]
s
e
le
c
te
d
f
or
te
s
ti
ng
our
m
ode
l,
th
e
S
is
F
a
ll
da
ta
s
e
t
s
e
ts
it
s
e
lf
a
pa
r
t
f
r
om
ot
he
r
publ
ic
ly
a
va
il
a
bl
e
da
ta
s
e
ts
.
I
t
in
vol
ve
s
da
ta
f
r
om
f
if
te
e
n
he
a
lt
hy,
in
de
pe
nde
nt
e
ld
e
r
ly
in
di
vi
dua
l
s
a
nd
out
s
hi
ne
s
c
om
pe
ti
to
r
s
w
it
h
it
s
in
c
lu
s
io
n
of
m
or
e
pa
r
ti
c
ip
a
nt
s
,
di
ve
r
s
e
a
c
ti
vi
ty
ty
pe
s
,
a
nd
a
la
r
ge
r
num
be
r
of
r
e
c
or
di
ngs
.
C
om
pr
is
in
g
2,706
a
c
ti
vi
ti
e
s
of
da
il
y
li
vi
ng
a
nd
1,798
f
a
ll
s
,
t
hi
s
da
ta
s
e
t
a
ddr
e
s
s
e
s
th
e
s
c
a
r
c
it
y
of
da
ta
s
e
ts
r
e
a
li
s
ti
c
a
ll
y
s
im
ul
a
ti
ng
a
c
ti
vi
ti
e
s
a
nd
f
a
ll
s
by
e
ld
e
r
ly
in
di
vi
dua
ls
in
th
e
dom
a
in
of
e
ld
e
r
ly
f
a
ll
d
e
te
c
ti
on
r
e
s
e
a
r
c
h.
S
pe
c
if
ic
a
ll
y,
it
in
tr
oduc
e
s
up
to
th
ir
ty
-
f
our
a
c
ti
vi
ti
e
s
(
f
a
ll
s
a
nd
a
c
ti
vi
ti
e
s
of
da
il
y
li
vi
ng)
pe
r
f
or
m
e
d
by
th
ir
ty
-
e
ig
ht
pa
r
ti
c
ip
a
nt
s
e
qui
ppe
d
w
it
h
a
w
e
a
r
a
bl
e
de
vi
c
e
f
ix
e
d
to
th
e
ir
w
a
is
t.
T
he
in
f
or
m
a
ti
on
f
or
th
os
e
th
r
e
e
da
ta
s
e
ts
is
or
ga
ni
z
e
d i
n T
a
bl
e
1 t
o f
a
c
il
it
a
te
a
s
tr
a
ig
ht
f
or
w
a
r
d c
om
pa
r
is
on.
V
id
e
o
da
ta
s
e
t
in
our
vi
s
ua
l
a
ppr
oa
c
h,
w
e
c
ol
le
c
te
d
a
da
ta
s
e
t
s
our
c
e
d
f
r
om
Y
ouT
ube
a
nd
K
a
ggl
e
,
c
om
pr
is
in
g
va
r
io
us
s
c
e
na
r
io
s
de
pi
c
ti
ng
in
di
vi
dua
ls
e
xpe
r
ie
nc
i
ng
f
a
ll
s
.
I
t
is
im
po
r
ta
nt
to
h
ig
hl
ig
ht
th
a
t
th
is
da
ta
s
e
t
i
s
l
im
it
e
d.
T
o
m
it
ig
a
te
t
hi
s
l
im
it
a
ti
on, we
a
ppl
ie
d da
t
a
a
ugm
e
nt
a
ti
on t
e
c
hni
que
s
t
o
e
xpa
nd t
he
da
ta
s
e
t’
s
di
ve
r
s
it
y. T
he
da
ta
s
e
t
w
a
s
c
a
te
gor
iz
e
d i
nt
o t
w
o di
s
ti
nc
t
c
l
a
s
s
e
s
:
f
a
ll
a
nd not f
a
ll
.
T
a
bl
e
1. D
a
ta
s
e
ts
r
e
c
or
di
ng de
ta
il
s
V
a
r
i
a
bl
e
S
e
ns
or
s
A
c
t
i
vi
t
i
e
s
F
r
e
que
nc
y (
H
z
)
S
a
m
pl
e
s
U
C
I
H
A
R
2
6
50
10,299
M
H
E
A
L
T
H
3
12
50
120
S
i
s
F
a
l
l
2
2
200
4,505 us
e
d
4.2.
D
at
a p
ar
t
it
io
n
i
n
g
W
e
a
ppr
oa
c
he
d
d
a
ta
s
e
t
p
a
r
ti
ti
oni
ng
uni
f
or
m
ly
.
O
nl
y
th
e
U
C
I
da
ta
s
e
t
ha
d
a
pr
e
-
de
f
in
e
d
s
pl
it
in
to
tr
a
in
in
g a
nd t
e
s
t
s
e
t
s
, a
nd
w
e
m
a
in
ta
in
e
d
c
ons
is
t
e
nt
t
e
s
t
pa
r
ti
ti
ons
f
or
c
om
pa
r
a
ti
ve
a
na
ly
s
i
s
. O
ur
pr
e
pr
oc
e
s
s
in
g
m
e
th
odol
ogy
e
xc
lu
s
iv
e
ly
in
vol
ve
d
th
e
us
e
of
r
a
w
s
ig
na
l
s
pos
t
-
dow
n
s
a
m
pl
in
g,
di
s
ti
ngui
s
hi
ng
our
a
ppr
oa
c
h
f
r
om
th
e
r
e
s
t
of
s
tu
di
e
s
on
U
C
I
H
A
R
.
T
he
s
e
s
tu
di
e
s
of
te
n
e
it
he
r
a
ppl
ie
d
th
e
ir
ow
n
pr
e
pr
oc
e
s
s
in
g
t
e
c
hni
que
s
to
r
a
w
da
ta
or
ut
il
iz
e
d
pr
ovi
de
d
h
a
nd
-
c
r
a
f
te
d
f
e
a
tu
r
e
s
.
H
a
nd
-
c
r
a
f
te
d
f
e
a
tu
r
e
s
w
e
r
e
e
m
pl
oye
d
f
or
th
e
S
is
F
a
ll
da
ta
s
e
t,
a
nd
w
e
a
dopt
e
d
th
e
s
a
m
e
s
e
tu
p
f
or
th
e
M
H
E
A
L
T
H
da
ta
s
e
t.
M
ul
ti
pl
e
da
ta
tr
a
ns
f
or
m
a
ti
ons
w
e
r
e
a
ppl
ie
d
to
e
nha
nc
e
th
e
qu
a
li
ty
of
th
e
i
m
a
g
e
da
t
a
s
e
t.
A
c
r
os
s
a
ll
d
a
ta
s
e
t
s
,
w
e
a
dhe
r
e
d
to
a
s
ta
nda
r
di
z
e
d s
pl
it
of
30%
f
or
te
s
ti
ng a
nd 70%
f
or
t
r
a
in
in
g.
5.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
5.1.
E
val
u
at
io
n
m
e
as
u
r
e
s
W
e
pr
ovi
de
a
c
onc
is
e
ove
r
vi
e
w
of
th
e
e
va
lu
a
ti
on
c
r
it
e
r
ia
e
m
pl
oye
d
in
th
is
s
tu
dy.
P
r
e
c
is
io
n:
is
pr
e
s
e
nt
e
d
a
s
th
e
pr
opor
ti
on
of
c
or
r
e
c
tl
y
pr
e
di
c
te
d
pos
it
iv
e
obs
e
r
va
ti
ons
r
e
la
ti
ve
to
th
e
to
ta
l
pr
e
di
c
te
d
pos
it
iv
e
obs
e
r
va
ti
ons
. I
t
is
c
a
lc
ul
a
t
e
d us
in
g t
he
t
e
r
m
s
t
r
ue
pos
it
iv
e
(
T
P
)
a
nd f
a
ls
e
pos
it
iv
e
(
F
P
)
.
=
+
(
1)
R
e
c
a
ll
(
s
e
n
s
it
iv
it
y
)
:
r
e
c
a
ll
,
a
ls
o
r
e
f
e
r
r
e
d
to
a
s
s
e
ns
it
iv
it
y,
is
th
e
r
a
ti
o
of
c
or
r
e
c
tl
y
id
e
nt
if
ie
d
pos
it
iv
e
obs
e
r
va
ti
ons
to
t
he
t
ot
a
l
num
be
r
of
obs
e
r
va
ti
ons
i
n t
he
c
or
r
e
s
ponding a
c
tu
a
l
c
la
s
s
.
=
+
(
2)
F
1
s
c
or
e
:
it
s
e
r
ve
s
a
s
a
ha
r
m
oni
c
m
e
a
n
of
bot
h
’
pr
e
c
is
io
n
’
a
nd
’
r
e
c
a
ll
.’
C
ons
e
que
nt
ly
,
it
in
c
or
por
a
te
s
in
f
or
m
a
ti
on a
bout
f
a
ls
e
pos
it
iv
e
s
a
nd f
a
ls
e
ne
ga
ti
ve
s
t
o yi
e
ld
a
m
or
e
c
om
pr
e
he
ns
iv
e
a
s
s
e
s
s
m
e
nt
. W
hi
le
not
a
s
s
tr
a
ig
ht
f
or
w
a
r
d a
s
a
c
c
ur
a
c
y, F
1 i
s
of
te
n m
or
e
i
nf
or
m
a
ti
ve
, e
s
pe
c
ia
ll
y i
n s
c
e
na
r
io
s
w
h
e
r
e
t
he
r
e
i
s
a
n i
m
ba
l
a
nc
e
in
c
la
s
s
di
s
tr
ib
ut
io
n.
1
=
2
∗
∗
+
(
3)
W
e
a
ls
o
e
m
pl
oy
m
A
P
,
a
m
e
tr
ic
th
a
t
of
f
e
r
s
a
m
or
e
c
om
pr
e
he
ns
iv
e
a
s
s
e
s
s
m
e
nt
th
a
n
a
c
c
ur
a
c
y
a
lo
ne
f
or
e
va
lu
a
ti
ng
a
m
ode
l’
s
pr
e
c
is
io
n a
nd r
e
c
a
ll
t
r
a
de
-
of
f
i
n obje
c
t
de
te
c
ti
on t
a
s
k
s
.
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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
:
561
-
569
566
5
.2.
C
la
s
s
if
i
c
at
io
n
r
e
s
u
lt
s
T
r
a
in
in
g
c
onf
ig
ur
a
ti
on
in
our
le
a
r
ni
ng
e
nvi
r
onm
e
nt
,
w
e
s
e
t
up
t
he
tr
a
in
in
g
pr
oc
e
s
s
f
or
th
e
F
a
ll
M
ix
e
r
m
ode
l,
w
hi
c
h
is
ta
il
or
e
d
f
or
a
c
l
a
s
s
if
ic
a
ti
on
t
a
s
k.
W
e
tr
a
in
e
d
th
e
m
ode
l
us
in
g
th
e
A
da
m
opt
im
iz
e
r
,
s
e
tt
in
g
th
e
le
a
r
ni
ng
r
a
te
to
0.001,
a
nd
e
m
pl
oye
d
th
e
c
a
te
gor
ic
a
l
c
r
os
s
-
e
nt
r
opy
lo
s
s
f
unc
ti
on.
T
o
m
it
ig
a
te
ove
r
f
it
ti
ng,
w
e
in
c
or
por
a
te
d
e
a
r
ly
s
to
ppi
ng
w
it
h
a
pa
ti
e
nc
e
of
500
e
poc
h
s
.
A
d
di
ti
ona
ll
y,
a
c
a
ll
ba
c
k
w
a
s
im
pl
e
m
e
nt
e
d
to
s
a
v
e
th
e
opt
im
a
l
m
ode
l
ba
s
e
d
on
va
li
da
ti
on
a
c
c
ur
a
c
y.
T
h
e
tr
a
in
in
g
pr
oc
e
s
s
e
xt
e
nd
s
up
to
a
m
a
xi
m
um
of
1
,
000
e
poc
hs
,
ut
il
iz
in
g
a
ba
tc
h
s
iz
e
of
512.
T
hi
s
s
e
tu
p
e
n
s
ur
e
s
th
e
m
ode
l’
s
e
f
f
ic
ie
nt
le
a
r
ni
ng
a
nd
h
e
lp
s
u
s
a
c
hi
e
ve
a
c
c
ur
a
te
c
la
s
s
if
ic
a
ti
on
r
e
s
ul
t
s
,
m
a
ki
ng
it
s
ui
t
a
bl
e
f
or
ha
ndl
in
g
s
e
que
nt
ia
l
da
t
a
.
D
ur
in
g
th
e
e
xpe
r
im
e
nt
s
,
w
e
th
or
oughly
e
va
lu
a
te
d
th
e
F
a
ll
M
ix
e
r
m
ode
l
w
it
h
va
r
yi
ng
c
onf
ig
ur
a
ti
ons
,
e
xpl
or
in
g
di
f
f
e
r
e
nt
num
be
r
s
of
f
il
te
r
s
a
nd
de
pt
hs
w
hi
le
ke
e
pi
ng
th
e
ke
r
ne
l
s
i
z
e
a
nd
pa
t
c
h
s
iz
e
c
ons
ta
nt
.
T
hi
s
s
y
s
te
m
a
ti
c
a
na
ly
s
is
e
na
bl
e
d
u
s
to
id
e
nt
if
y
th
e
opt
im
a
l
s
e
tt
in
gs
f
or
t
he
numbe
r
of
f
il
te
r
s
a
nd
de
pt
h, m
a
xi
m
i
z
in
g t
he
m
ode
l’
s
f
e
a
tu
r
e
c
a
pt
ur
in
g c
a
pa
bi
li
ti
e
s
a
nd
a
c
hi
e
vi
ng
a
c
c
ur
a
te
c
la
s
s
if
ic
a
ti
ons
.
B
y
m
a
in
ta
in
in
g
c
ons
t
a
nt
ke
r
ne
l
a
nd
pa
tc
h
s
iz
e
,
w
e
e
ns
ur
e
d
a
f
a
ir
c
om
pa
r
is
on a
nd f
oc
us
e
d on unde
r
s
ta
ndi
ng t
he
i
m
pa
c
t
of
ot
he
r
hype
r
pa
r
a
m
e
te
r
s
on pe
r
f
or
m
a
nc
e
. I
n s
ubj
e
c
ti
ng
our
w
or
k
to
a
c
om
pa
r
a
ti
ve
a
na
ly
s
i
s
w
it
h
va
r
io
us
a
ppr
oa
c
h
e
s
,
a
ll
a
dhe
r
in
g
to
th
e
s
a
m
e
d
a
ta
s
pl
it
ti
ng
c
onf
ig
ur
a
ti
on,
a
not
a
bl
e
obs
e
r
va
ti
on
be
c
om
e
s
known,
F
a
ll
M
ix
e
r
e
m
e
r
ge
s
a
s
pa
r
ti
c
ul
a
r
ly
not
e
w
or
th
y
f
or
a
tt
a
in
in
g
s
upe
r
io
r
a
c
c
ur
a
c
y.
T
a
bl
e
2
hi
ghl
ig
ht
s
th
e
pe
r
f
or
m
a
nc
e
of
di
f
f
e
r
e
nt
a
r
c
hi
te
c
tu
r
e
s
,
a
nd
a
m
id
s
t
th
e
m
.
T
he
c
om
pa
r
e
d
m
e
th
odol
ogi
e
s
in
th
e
ta
bl
e
e
m
pl
oye
d
m
e
th
ods
li
ke
f
us
io
n
of
L
S
T
M
a
nd
C
N
N
a
r
c
hi
te
c
tu
r
e
s
or
e
ns
e
m
bl
e
m
e
th
ods
.
H
ow
e
ve
r
,
c
ons
i
s
te
nt
ly
,
F
a
ll
M
ix
e
r
e
xhi
bi
ts
e
nha
nc
e
d
pe
r
f
or
m
a
nc
e
i
n
te
r
m
s
of
a
c
c
ur
a
c
y.
T
hi
s
out
c
om
e
not
onl
y
unde
r
s
c
or
e
s
th
e
e
f
f
ic
a
c
y
of
our
a
ppr
oa
c
h
.
T
hi
s
s
ig
ni
f
ic
a
nc
e
is
f
ur
th
e
r
e
xe
m
pl
if
ie
d
in
th
e
c
ont
e
xt
of
th
e
M
H
E
A
L
T
H
da
ta
s
e
t,
a
s
il
lu
s
tr
a
te
d
in
T
a
bl
e
3
.
H
e
r
e
,
th
e
ut
il
iz
a
ti
on
of
onl
y
a
n
a
c
c
e
le
r
om
e
te
r
w
it
h
F
a
ll
M
ix
e
r
de
m
ons
tr
a
te
s
c
om
m
e
nda
bl
e
a
c
ti
vi
ty
de
te
c
ti
on
c
a
p
a
bi
li
ti
e
s
,
a
nd
unde
r
s
c
or
in
g
th
e
ve
r
s
a
ti
li
ty
a
nd
e
f
f
e
c
ti
ve
ne
s
s
of
F
a
ll
M
ix
e
r
, pa
r
ti
c
ul
a
r
ly
i
n s
c
e
n
a
r
io
s
i
nvol
vi
ng dive
r
s
e
c
om
bi
na
ti
ons
of
w
e
a
r
a
bl
e
s
e
ns
or
s
.
T
a
bl
e
2. P
r
e
vi
ous
w
or
ks
on U
C
I
H
A
R
A
r
c
hi
t
e
c
t
ur
e
A
c
c
ur
a
c
y
M
obi
l
e
H
A
R
T
[
29]
97.67%
F
a
l
l
M
i
xe
r
97.56%
V
i
T
[
29]
93.66%
C
N
N
L
S
T
M
[
30]
92.79%
±
0.34
T
a
bl
e
3. P
r
e
vi
ous
w
or
ks
on
M
H
E
A
L
T
H
da
ta
s
e
t
A
r
c
hi
t
e
c
t
ur
e
F
a
l
l
M
i
xe
r
E
kV
N
[
31]
M
E
M
M
[
32]
A
c
c
ur
a
c
y
97.71%
67%
90.91%
F
a
ll
de
te
c
ti
on
r
e
s
ul
ts
:
F
ol
lo
w
in
g
th
e
c
om
pr
e
he
ns
iv
e
te
s
ti
ng
of
F
a
ll
M
ix
e
r
a
c
r
os
s
va
r
io
us
a
c
ti
vi
ti
e
s
,
our
f
oc
us
s
hi
f
te
d
to
e
va
lu
a
ti
ng
it
s
pe
r
f
or
m
a
nc
e
u
s
in
g
th
e
S
is
F
a
ll
da
ta
s
e
t
w
it
h a
s
pe
c
if
ic
e
m
ph
a
s
is
on
th
e
de
te
c
ti
on
of
a
s
in
gul
a
r
a
c
ti
vi
ty
,
na
m
e
ly
f
a
ll
s
i
n
da
il
y l
iv
in
g. T
he
r
e
s
ul
ts
s
how
t
ha
t
it
c
a
n de
te
c
t
th
e
f
a
ll
w
i
th
a
n a
c
c
ur
a
c
y
c
lo
s
e
t
o 99%
. T
he
c
onf
u
s
io
n m
a
tr
ix
i
s
pr
e
s
e
nt
e
d i
n
T
a
bl
e
4.
T
a
bl
e
4
. C
onf
us
io
n m
a
tr
ix
of
S
is
F
a
ll
da
ta
s
e
t
P
r
e
di
c
t
e
d f
a
l
l
P
r
e
di
c
t
e
d
ADL
T
r
ue
f
a
l
l
556
0
T
r
ue
ADL
0
345
5.3.
H
ar
d
w
ar
e
p
e
r
f
o
r
m
an
c
e
W
e
c
ho
s
e
th
e
T
e
ns
or
f
lo
w
L
it
e
m
ode
l
m
a
ke
r
li
br
a
r
y
f
or
it
s
u
s
e
r
-
f
r
ie
ndl
y
in
te
r
f
a
c
e
in
m
obi
le
m
ode
l
de
ve
lo
pm
e
nt
,
s
e
a
m
le
s
s
in
te
gr
a
ti
on
w
it
h
T
e
ns
or
F
lo
w
te
m
pl
a
te
s
,
a
nd
th
e
a
bi
li
ty
to
c
r
e
a
te
e
f
f
ic
ie
nt
,
li
ght
w
e
ig
ht
T
F
L
it
e
m
ode
ls
s
ui
ta
bl
e
f
or
m
obi
le
de
pl
oym
e
nt
. W
e
m
ove
d our
e
xpe
r
im
e
nt
a
l
m
ode
ls
t
o a
n i
P
hone
11 P
r
o M
a
x
w
it
h
a
H
e
xa
-
c
or
e
C
P
U
,
4
G
B
R
A
M
,
a
nd
a
n
A
ppl
e
A
13
B
io
ni
c
C
hi
ps
e
t
(
7
nm
+
)
w
it
h
a
4
-
c
or
e
gr
a
phi
c
s
A
ppl
e
G
P
U
.
O
ve
r
1
,
000
in
f
e
r
e
nc
e
s
w
e
r
e
pe
r
f
or
m
e
d
us
in
g
f
our
C
P
U
th
r
e
a
ds
,
th
e
in
f
e
r
e
nc
e
s
w
e
r
e
e
xc
lu
s
iv
e
ly
c
onduc
te
d on the
de
vi
c
e
’
s
C
P
U
, a
nd mul
ti
pl
e
m
e
a
s
ur
e
s
w
e
r
e
t
a
ke
n dur
in
g t
he
pr
oc
e
s
s
.
A
na
ly
s
is
:
th
e
out
c
om
e
s
of
t
he
e
xpe
r
im
e
nt
s
a
r
e
de
ta
il
e
d
i
n
T
a
bl
e
5.
I
t
hi
ghl
ig
ht
s
a
s
ub
s
ta
nt
ia
l
de
c
r
e
a
s
e
in
th
e
a
ve
r
a
g
e
in
f
e
r
e
nc
e
ti
m
e
f
or
F
a
ll
M
ix
e
r
in
c
om
pa
r
is
on
to
V
i
T
a
nd
H
A
R
T
,
s
ugge
s
ti
ng
it
s
s
upe
r
io
r
e
f
f
ic
ie
nc
y
in
m
a
ki
ng
pr
e
di
c
ti
ons
on
th
e
de
vi
c
e
.
W
hi
le
our
M
ode
l
e
xhi
bi
ts
a
hi
ghe
r
m
e
m
or
y
f
oot
pr
in
t
th
a
n
H
A
R
T
but
a
lo
w
e
r
one
th
a
n
V
iT
,
it
f
e
a
tu
r
e
s
th
e
s
m
a
ll
e
s
t
m
ode
l
s
iz
e
,
in
di
c
a
ti
n
g
a
r
e
duc
e
d
ne
e
d
f
or
s
to
r
a
ge
s
pa
c
e
.
T
hi
s
pr
ove
s
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
e
te
c
ti
ng human fall
us
in
g i
nt
e
r
ne
t
of
t
hi
ngs
d
e
v
ic
e
s
f
or
he
al
th
c
ar
e
appli
c
at
io
ns
(
Z
ak
ar
ia
B
e
nhai
li
)
567
a
dva
nt
a
ge
ous
,
pa
r
ti
c
ul
a
r
ly
f
or
de
vi
c
e
s
w
it
h l
im
it
e
d
s
to
r
a
ge
c
a
pa
c
it
y. H
ow
e
ve
r
, t
he
opt
im
a
l
a
r
c
hi
te
c
tu
r
e
c
hoi
c
e
de
pe
nds
on
s
p
e
c
if
ic
a
ppl
ic
a
ti
on
r
e
qui
r
e
m
e
nt
s
,
c
ons
id
e
r
in
g
f
a
c
to
r
s
li
ke
a
v
a
il
a
bl
e
m
e
m
or
y
a
nd
th
e
de
s
ir
e
d
ba
la
nc
e
be
twe
e
n s
pe
e
d a
nd mode
l
c
om
pl
e
xi
t
y.
T
a
bl
e
5
. I
nf
e
r
e
nt
ia
l
dur
a
ti
on a
nd me
m
or
y us
a
ge
t
hr
oughout 1000 i
nf
e
r
e
nc
e
s
on t
he
de
vi
c
e
N
e
t
w
or
k
A
ve
r
a
ge
i
nf
e
r
e
nc
e
t
i
m
e
(
m
s
)
M
e
m
or
y (
M
B
)
M
ode
l
s
i
z
e
(
M
B
)
V
i
T
[
29]
8.213±
1.518
32.07
15.22
H
A
R
T
[
29]
5.376±
1.104
12.74
5.9
O
ur
m
ode
l
2.46 (
m
e
a
n)
25.5
3
5.4.
F
al
l
d
e
t
e
c
t
io
n
u
s
in
g c
am
e
r
as
B
y
ut
il
iz
in
g
Y
O
L
O
’
s
de
te
c
ti
on
c
a
pa
bi
li
ti
e
s
a
nd
f
in
e
-
tu
ni
ng
it
on
our
da
ta
s
e
t,
th
e
obt
a
in
e
d
p
r
e
c
is
io
n
is
e
qua
l
to
97.2%
, t
he
r
e
c
a
ll
is
82.6%
a
nd t
he
m
A
P
i
s
e
qua
l
to
89.
6%
. H
um
a
n f
a
ll
s
a
r
e
s
how
n i
n
F
ig
ur
e
3.
I
n t
he
im
a
ge
,
w
e
obs
e
r
ve
s
uc
c
e
s
s
f
ul
f
a
ll
de
te
c
ti
on,
w
it
h
th
e
bounding
box
e
f
f
e
c
ti
ve
ly
e
nc
om
pa
s
s
in
g
th
e
in
di
vi
dua
l.
T
hi
s
out
c
om
e
i
s
e
nc
our
a
gi
ng, pa
r
ti
c
ul
a
r
ly
gi
ve
n t
he
us
e
of
a
m
o
de
s
t
da
ta
s
e
t,
a
nd a
l
im
it
e
d t
r
a
in
in
g t
im
e
.
F
ig
ur
e
2
. I
ndoor
a
nd outdoor
f
a
ll
de
te
c
ti
on us
in
g s
ur
ve
il
la
nc
e
c
a
m
e
r
a
s
a
nd Y
O
L
O
a
lg
or
it
hm
5.5
.
D
is
c
u
s
s
io
n
T
hi
s
s
tu
dy
in
tr
oduc
e
s
a
nov
e
l
a
ppr
oa
c
h
to
f
a
ll
de
te
c
ti
on
by
in
te
gr
a
ti
ng
w
e
a
r
a
bl
e
s
e
n
s
or
s
w
it
h
a
c
c
e
le
r
om
e
te
r
s
a
nd
gyr
os
c
ope
s
,
a
lo
ng
w
it
h
vi
s
u
a
l
s
e
n
s
or
s
ut
il
iz
in
g
c
a
m
e
r
a
s
.
B
y
c
om
bi
ni
ng
th
e
s
e
two
m
oda
li
ti
e
s
,
th
e
pr
opos
e
d
a
lg
or
it
hm
s
a
c
hi
e
ve
d
r
e
m
a
r
ka
bl
e
a
c
c
ur
a
c
y
r
a
te
s
on
va
r
io
us
da
ta
s
e
t
s
,
s
ur
pa
s
s
in
g
pr
e
vi
ous
m
e
th
ods
.
N
ot
a
bl
y,
F
a
ll
M
ix
e
r
a
tt
a
in
e
d
97.56%
a
c
c
ur
a
c
y
on
th
e
U
C
I
H
A
R
da
t
a
s
e
t,
97.71%
on
M
H
E
A
L
T
H
,
a
nd
99%
on
S
is
f
a
ll
da
ta
s
e
t,
w
hi
le
Y
ol
ov8
de
m
ons
tr
a
te
d
97.2%
pr
e
c
is
io
n
on
a
c
us
to
m
f
a
ll
vi
de
o
da
ta
s
e
t.
T
hi
s
a
dv
a
nc
e
m
e
nt
out
pe
r
f
or
m
s
tr
a
di
ti
ona
l
te
c
hni
qu
e
s
li
ke
L
S
T
M
a
nd
c
onvolut
io
n,
a
s
w
e
ll
a
s
tr
a
ns
f
or
m
e
r
-
ba
s
e
d
a
lg
or
it
hm
s
,
s
how
c
a
s
in
g
th
e
pot
e
nt
ia
l
f
or
a
tt
e
nt
io
n
-
f
r
e
e
a
r
c
hi
te
c
tu
r
e
s
in
e
xt
r
a
c
ti
ng
te
m
por
a
l
pr
ope
r
ti
e
s
f
r
om
s
e
ns
or
da
ta
.
D
e
s
pi
te
th
e
s
e
pr
om
is
in
g
r
e
s
ul
t
s
,
f
u
r
th
e
r
r
e
s
e
a
r
c
h
is
e
s
s
e
nt
ia
l
to
opt
im
iz
e
th
e
in
te
gr
a
ti
on
of
vi
s
ua
l
a
nd
w
e
a
r
a
bl
e
s
e
n
s
or
da
ta
,
a
ddr
e
s
s
im
pl
e
m
e
nt
a
ti
on
c
ha
ll
e
nge
s
s
uc
h
a
s
in
c
or
por
a
ti
ng
a
ir
ba
g
pr
ot
e
c
ti
on
f
or
bone
s
a
nd
e
xpl
or
e
de
pl
oym
e
nt
opt
io
ns
in
bot
h
d
e
vi
c
e
a
nd
c
lo
ud
e
nvi
r
onm
e
nt
s
.
T
he
s
e
f
in
d
in
gs
s
ugge
s
t
e
xc
it
in
g
pr
os
pe
c
ts
f
or
e
nh
a
nc
in
g
de
te
c
ti
on
a
c
c
ur
a
c
y
a
n
d
r
e
li
a
bi
li
ty
us
in
g
hybr
id
s
e
ns
in
g,
p
a
r
ti
c
ul
a
r
ly
in
s
m
a
r
t
hom
e
s
a
nd da
il
y l
iv
in
g a
c
ti
vi
ty
m
oni
to
r
in
g f
or
ol
de
r
a
d
ul
ts
.
6.
C
O
N
C
L
U
S
I
O
N
T
hi
s
in
ve
s
ti
ga
ti
on
ha
s
f
oc
us
e
d
on
th
e
de
v
e
lo
pm
e
nt
of
a
f
a
ll
de
t
e
c
ti
on
s
y
s
te
m
ta
il
or
e
d
to
th
e
s
p
e
c
if
ic
ne
e
ds
of
s
e
ni
or
c
it
iz
e
ns
,
c
a
pa
bl
e
of
be
in
g
de
pl
oye
d
in
bot
h
in
do
or
a
nd
out
door
s
e
tt
in
gs
.
T
he
pr
opos
e
d
ne
twor
k
c
a
n de
te
c
t
m
ove
m
e
nt
s
, pa
r
ti
c
ul
a
r
ly
f
a
ll
s
, by
ut
il
iz
in
g s
e
ns
or
s
bui
lt
i
nt
o s
m
a
r
tp
hone
s
or
ot
he
r
I
oT
de
vi
c
e
s
. T
he
us
e
of
a
vi
de
o
-
ba
s
e
d
a
ppr
oa
c
h
e
nha
n
c
e
d
th
e
de
te
c
ti
on
of
th
e
f
a
ll
,
w
hi
le
a
ls
o
pr
ovi
di
ng
th
e
a
bi
li
ty
to
m
oni
to
r
a
nd
a
na
ly
z
e
va
r
io
us
ot
he
r
m
ove
m
e
nt
s
,
pr
om
ot
in
g
a
m
or
e
a
c
t
iv
e
li
f
e
s
ty
le
.
T
he
pr
opos
e
d
F
a
ll
M
ix
e
r
m
ode
l
s
how
c
a
s
e
s
a
s
ig
ni
f
ic
a
nt
le
ve
l
of
r
e
li
a
bi
li
ty
a
nd
a
c
c
ur
a
c
y
in
r
e
c
ogni
z
in
g
ba
s
ic
m
ove
m
e
nt
s
li
ke
w
a
lk
in
g
a
nd
s
it
ti
ng.
I
t
a
c
hi
e
ve
s
a
n
a
c
c
ur
a
c
y
of
97.71%
on
M
H
E
A
L
T
H
a
nd
97.67%
on
U
C
I
H
A
R
.
M
or
e
ov
e
r
,
it
r
e
a
c
he
s
a
99%
a
c
c
ur
a
c
y i
n de
te
c
ti
ng f
a
ll
s
i
n
t
he
S
is
f
a
ll
da
ta
s
e
t,
ut
il
iz
in
g i
nf
or
m
a
ti
on f
r
om
a
c
c
e
le
r
om
e
te
r
a
nd gy
r
os
c
ope
s
e
ns
or
s
.
T
he
in
c
or
por
a
ti
on
of
Y
O
L
O
v8
f
in
e
tu
ne
d
a
id
s
in
a
c
hi
e
vi
ng
a
pr
e
c
is
io
n
of
97.2%
f
or
de
te
c
ti
ng
two
c
la
s
s
e
s
,
na
m
e
ly
F
a
ll
a
nd
N
oF
a
ll
,
on
a
c
ol
le
c
te
d
vi
de
o
da
ta
s
e
t.
I
n
th
e
f
ut
ur
e
,
w
e
a
r
e
th
in
ki
ng
a
bout
im
pr
ovi
n
g
th
e
s
ys
te
m
by
a
ddi
ng
a
be
lt
w
it
h
a
n
a
ir
ba
g
th
a
t
is
in
te
nde
d
to
s
hi
e
ld
im
por
ta
nt
bone
s
in
th
e
e
ve
nt
of
a
f
a
ll
.
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
.
1
4
, N
o.
1
,
F
e
br
ua
r
y
202
5
:
561
-
569
568
R
e
a
l
-
ti
m
e
de
te
c
ti
on
c
oul
d
b
e
pe
r
f
or
m
e
d
us
in
g
a
R
a
s
pb
e
r
r
y
P
i
w
it
h
a
c
a
m
e
r
a
,
a
nd
th
e
s
e
ns
or
s
c
oul
d
be
in
te
gr
a
te
d
in
to
a
s
m
a
r
twa
tc
h.
T
hi
s
a
ll
ow
s
th
e
s
ugge
s
te
d
s
ys
te
m
to
be
r
e
li
a
bl
y
de
pl
oye
d
in
a
s
m
a
r
t
hom
e
c
ont
e
xt
,
r
e
duc
in
g
w
or
r
ie
s
a
m
ong olde
r
r
e
s
id
e
nt
s
w
hi
le
a
ls
o i
m
pr
ovi
ng t
he
ir
s
a
f
e
ty
a
nd w
e
ll
-
be
in
g.
R
E
F
E
R
E
N
C
E
S
[
1]
Y
.
A
s
i
m
,
M
.
A
.
A
z
a
m
,
M
.
E
.
-
U
l
-
H
a
q,
U
.
N
a
e
e
m
,
a
nd
A
.
K
ha
l
i
d,
“
C
ont
e
xt
-
a
w
a
r
e
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on
(
C
A
H
A
R
)
i
n
-
t
he
-
w
i
l
d
us
i
ng
s
m
a
r
t
phone
a
c
c
e
l
e
r
om
e
t
e
r
,”
I
E
E
E
Se
ns
or
s
J
our
nal
,
vol
.
20,
no.
8,
pp.
4361
–
4371,
2020
,
doi
:
10.1109/
J
S
E
N
.2020.2964278
.
[
2]
L
.
C
a
o,
Y
.
W
a
ng,
B
.
Z
ha
ng,
Q
.
J
i
n,
a
nd
A
.
V
.
V
a
s
i
l
a
kos
,
“
G
C
H
A
R
:
A
n
e
f
f
i
c
i
e
nt
gr
oup
-
ba
s
e
d
c
ont
e
xt
—
a
w
a
r
e
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on
on
s
m
a
r
t
phone
,”
J
our
nal
of
P
ar
al
l
e
l
and
D
i
s
t
r
i
but
e
d
C
om
put
i
ng
,
vol
.
118,
pp.
67
–
80,
2018,
doi
:
doi
:
10.1016/
j
.j
pdc
.2017.05.007.
[
3]
A
.
H
us
s
a
i
n,
S
.
U
.
K
ha
n, N
.
K
ha
n, M
.
S
ha
ba
z
, a
nd
S
.
W
. B
a
i
k,
“
A
I
-
dr
i
ve
n
be
ha
v
i
or
bi
om
e
t
r
i
c
s
f
r
a
m
e
w
or
k
f
or
r
obus
t
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on
i
n
s
ur
ve
i
l
l
a
nc
e
s
y
s
t
e
m
s
,”
E
ngi
ne
e
r
i
ng
A
ppl
i
c
at
i
ons
of
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
,
vol
.
127,
2024,
doi
:
doi
:
10.1016/
j
.e
nga
ppa
i
.2023.107218.
[
4]
A
. S
uba
s
i
, K
.
K
ha
t
e
e
b,
T
. B
r
a
hi
m
i
, a
nd
A
. S
a
r
i
r
e
t
e
,
“
C
ha
pt
e
r
5
-
H
um
a
n
a
c
t
i
vi
t
y r
e
c
ogni
t
i
on us
i
ng m
a
c
hi
ne
l
e
a
r
ni
ng m
e
t
hods
i
n
a
s
m
a
r
t
he
a
l
t
hc
a
r
e
e
nvi
r
onm
e
nt
,”
i
n
I
nnov
at
i
on i
n
H
e
al
t
h
I
nf
or
m
at
i
c
s
, A
c
a
de
m
i
c
P
r
e
s
s
, 2020, pp. 123
–
144
, doi
:
10.1016/
B
978
-
0
-
12
-
819043
-
2.00005
-
8.
[
5]
R
.
L
i
u,
A
.
A
.
R
a
m
l
i
,
H
.
Z
ha
ng,
E
.
H
e
nr
i
c
s
on,
a
nd
X
.
L
i
u,
“
A
n
ove
r
vi
e
w
of
h
um
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on
u
s
i
ng
w
e
a
r
a
bl
e
s
e
n
s
or
s
:
he
a
l
t
hc
a
r
e
a
nd a
r
t
i
f
i
c
i
a
l
i
nt
e
l
l
i
ge
nc
e
,”
i
n
I
nt
e
r
ne
t
of
T
hi
ngs
–
I
C
I
O
T
2021
, C
ha
m
:
S
pr
i
nge
r
I
nt
e
r
na
t
i
ona
l
P
ubl
i
s
hi
ng, 2022, pp.
1
–
14
,
doi
:
10.1007/
978
-
3
-
030
-
96068
-
1_1
.
[
6]
R
.
Z
ha
ng,
C
.
J
i
a
ng,
S
.
W
u,
Q
.
Z
hou,
X
.
J
i
ng,
a
nd
J
.
M
u,
“
W
i
-
Fi
s
e
n
s
i
ng
f
or
j
oi
n
t
ge
s
t
ur
e
r
e
c
ogni
t
i
on a
nd
hum
a
n
i
d
e
nt
i
f
i
c
a
t
i
on
f
r
om
f
e
w
s
a
m
pl
e
s
i
n hum
a
n
-
c
om
put
e
r
i
nt
e
r
a
c
t
i
on,”
I
E
E
E
J
our
nal
on Se
l
e
c
t
e
d A
r
e
a
s
i
n C
om
m
uni
c
at
i
ons
, vol
.
40, no. 7,
pp. 2193
–
2205
,
2022, doi
:
10.1109/
J
S
A
C
.2022.3155526.
[
7]
L
.
S
c
hr
a
de
r
e
t
al
.
,
“
A
dva
nc
e
d
s
e
ns
i
ng
a
nd
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on
i
n
e
a
r
l
y
i
nt
e
r
ve
nt
i
on
a
nd
r
e
ha
bi
l
i
t
a
t
i
on
of
e
l
de
r
l
y
pe
op
l
e
,”
J
our
nal
of
P
opul
at
i
on A
ge
i
ng
, vol
. 13, no. 2, pp. 139
–
165, J
un. 2020, doi
:
10.1
007/
s
12062
-
020
-
09260
-
z.
[
8]
G
.
G
uo,
H
.
W
a
ng,
D
.
B
e
l
l
,
Y
.
B
i
,
a
nd
K
.
G
r
e
e
r
,
“
K
N
N
m
ode
l
-
ba
s
e
d
a
ppr
oa
c
h
i
n
c
l
a
s
s
i
f
i
c
a
t
i
on,”
O
n
T
he
M
ov
e
t
o
M
e
ani
ngf
ul
I
nt
e
r
ne
t
Sy
s
t
e
m
s
2003:
C
oopI
S, D
O
A
, and O
D
B
A
SE
(
O
T
M
2003
)
, C
a
t
a
ni
a
, I
t
a
l
y, 2003, pp. 986
–
996
, doi
:
10.1007/
978
-
3
-
540
-
39964
-
3_62
.
[
9]
C
. J
. B
ur
ge
s
,
“
A
t
ut
or
i
a
l
on s
uppor
t
v
e
c
t
or
m
a
c
hi
ne
s
f
or
pa
t
t
e
r
n r
e
c
ogni
t
i
on,”
D
at
a m
i
ni
ng and k
now
l
e
dg
e
di
s
c
o
v
e
r
y
, vol
. 2, no.
2,
pp. 121
–
167, 1998
, doi
:
10.1023/
A
:
1009715923555
.
[
10]
R
.
A
.
B
e
r
k,
“
C
l
a
s
s
i
f
i
c
a
t
i
on
a
nd
r
e
gr
e
s
s
i
on
t
r
e
e
s
(
C
A
R
T
)
,”
i
n
St
at
i
s
t
i
c
al
L
e
ar
ni
ng
f
r
om
a
R
e
gr
e
s
s
i
on
P
e
r
s
p
e
c
t
i
v
e
,
N
e
w
Y
or
k:
S
pr
i
nge
r
, 2008, pp. 1
–
65
, doi
:
10.1007/
978
-
0
-
387
-
77501
-
2_3.
[
11]
L
.
R
oka
c
h,
“
E
ns
e
m
bl
e
m
e
t
hods
f
or
c
l
a
s
s
i
f
i
e
r
s
,”
i
n
D
at
a
M
i
ni
ng
and
K
now
l
e
dge
D
i
s
c
ov
e
r
y
H
andbook
,
B
os
t
on,
M
A
:
S
pr
i
nge
r
U
S
,
2005, pp. 957
–
980
, doi
:
10.1007/
0
-
387
-
25465
-
X
_45.
[
12]
I
.
L
a
ur
i
ol
a
,
A
.
L
a
v
e
l
l
i
,
a
nd
F
.
A
i
ol
l
i
,
“
A
n
i
nt
r
oduc
t
i
on
t
o
de
e
p
l
e
a
r
ni
ng
i
n
na
t
ur
a
l
l
a
ngua
ge
pr
oc
e
s
s
i
ng:
M
ode
l
s
,
t
e
c
hni
que
s
,
a
nd
t
ool
s
,”
N
e
ur
o
c
om
put
i
ng
, vol
. 470, pp. 443
–
456, 2022
, doi
:
10.1016/
j
.ne
uc
om
.2021.05.103
.
[
13]
A
.
B
oc
hkovs
ki
y,
C
.
-
Y
.
W
a
ng,
a
nd
H
.
-
Y
.
M
.
L
i
a
o,
“
Y
ol
ov4:
O
pt
i
m
a
l
s
pe
e
d
a
nd
a
c
c
ur
a
c
y
of
obj
e
c
t
de
t
e
c
t
i
on,”
ar
X
i
v
-
C
om
put
e
r
Sc
i
e
nc
e
,
pp. 1
-
17,
2020.
[
14]
I
.
O
.
T
ol
s
t
i
khi
n
e
t
al
.
,
“
M
L
P
-
m
i
xe
r
:
A
n
a
l
l
-
m
l
p
a
r
c
hi
t
e
c
t
ur
e
f
or
vi
s
i
on,”
A
dv
anc
e
s
i
n
ne
u
r
al
i
nf
or
m
at
i
on
p
r
oc
e
s
s
i
ng
s
y
s
t
e
m
s
,
vol
.
34, pp. 24261
–
24272, 2021.
[
15]
A
. V
a
s
w
a
ni
e
t
al
.
, “
A
t
t
e
nt
i
on i
s
A
l
l
you N
e
e
d,”
i
n
A
dv
anc
e
s
i
n N
e
ur
al
I
nf
or
m
at
i
on P
r
oc
e
s
s
i
ng Sy
s
t
e
m
s
,
2017.
[
16]
G
.
Ş
e
ngül
,
M
.
K
a
r
a
ka
ya
,
S
.
M
i
s
r
a
,
O
.
O
.
A
ba
yom
i
-
A
l
l
i
,
a
nd
R
.
D
a
m
a
š
e
vi
č
i
us
,
“
D
e
e
p
l
e
a
r
ni
ng
ba
s
e
d
f
a
l
l
de
t
e
c
t
i
on
us
i
ng
s
m
a
r
t
w
a
t
c
he
s
f
or
he
a
l
t
hc
a
r
e
a
ppl
i
c
a
t
i
ons
,”
B
i
om
e
di
c
al
Si
gnal
P
r
oc
e
s
s
i
ng
and
C
ont
r
ol
,
vol
.
71,
2022,
doi
:
10.1016/
j
.bs
pc
.2021.103242.
[
17]
A
.
A
l
a
r
i
f
i
a
nd
A
.
A
l
w
a
da
i
n,
“
K
i
l
l
e
r
he
ur
i
s
t
i
c
opt
i
m
i
z
e
d
c
onvol
ut
i
on
ne
u
r
a
l
ne
t
w
or
k
-
ba
s
e
d
f
a
l
l
de
t
e
c
t
i
on
w
i
t
h
w
e
a
r
a
bl
e
I
oT
s
e
ns
or
de
vi
c
e
s
,”
M
e
as
u
r
e
m
e
nt
, vol
. 167, 2021, doi
:
10.1016/
j
.m
e
a
s
ur
e
m
e
nt
.2020.108258.
[
18]
C
.
M
.
L
e
e
,
J
.
P
a
r
k,
S
.
P
a
r
k,
a
nd
C
.
H
.
K
i
m
,
“
F
a
l
l
-
de
t
e
c
t
i
on
a
l
gor
i
t
hm
us
i
ng
pl
a
nt
a
r
pr
e
s
s
ur
e
a
nd
a
c
c
e
l
e
r
a
t
i
on
d
a
t
a
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
P
r
e
c
i
s
i
on E
ngi
ne
e
r
i
ng and M
anuf
ac
t
ur
i
ng
, vol
. 21, no. 4, pp. 725
–
7
37, A
pr
. 2020, doi
:
10.1007/
s
12541
-
019
-
00268
-
w.
[
19]
F
.
L
uo,
S
.
K
ha
n,
Y
.
H
ua
ng,
a
nd
K
.
W
u,
“
B
i
na
r
i
z
e
d
n
e
ur
a
l
ne
t
w
or
k
f
or
e
dge
i
nt
e
l
l
i
ge
nc
e
of
s
e
ns
or
-
ba
s
e
d
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
o
n,”
I
E
E
E
T
r
ans
ac
t
i
ons
on M
obi
l
e
C
om
put
i
ng
, vol
. 22, no. 3, pp. 1356
–
1368, 2023, doi
:
10.1109/
T
M
C
.2021.3109940.
[
20]
M
.
M
.
H
a
s
a
n,
M
.
S
.
I
s
l
a
m
,
a
nd
S
.
A
bdul
l
a
h,
“
R
obus
t
pos
e
-
ba
s
e
d
hum
a
n
f
a
l
l
de
t
e
c
t
i
on
us
i
ng
r
e
c
ur
r
e
nt
ne
ur
a
l
n
e
t
w
or
k,”
i
n
201
9
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on R
obot
i
c
s
, A
ut
om
at
i
on, A
r
t
i
f
i
c
i
al
-
i
n
t
e
l
l
i
ge
nc
e
and I
nt
e
r
ne
t
-
of
-
T
hi
ngs
(
R
A
A
I
C
O
N
)
, 2019, pp. 48
–
51
, doi
:
10.1109/
R
A
A
I
C
O
N
48939.2019.23.
[
21]
Q
.
F
e
ng,
C
.
G
a
o,
L
.
W
a
ng,
Y
.
Z
ha
o,
T
.
S
ong,
a
nd
Q
.
L
i
,
“
S
pa
t
i
o
-
t
e
m
por
a
l
f
a
l
l
e
ve
nt
de
t
e
c
t
i
on
i
n
c
om
pl
e
x
s
c
e
n
e
s
us
i
ng
a
t
t
e
nt
i
o
n
gui
de
d L
S
T
M
,”
P
at
t
e
r
n R
e
c
ogni
t
i
on L
e
t
t
e
r
s
, vol
. 130, pp. 242
–
249, 2020, doi
:
10.1016/
j
.pa
t
r
e
c
.2018.08.031.
[
22]
K
.
F
e
i
,
C
.
W
a
ng,
J
.
Z
ha
ng,
Y
.
L
i
u,
X
.
X
i
e
,
a
nd
Z
.
T
u,
“
F
l
ow
-
pos
e
N
e
t
:
a
n
e
f
f
e
c
t
i
ve
t
w
o
-
s
t
r
e
a
m
ne
t
w
or
k
f
or
f
a
l
l
de
t
e
c
t
i
on,”
T
he
V
i
s
ual
C
om
put
e
r
, vol
. 39, no. 6, pp. 2305
–
2320, F
e
b. 2022, doi
:
10.1007/
s
0037
1
-
022
-
02416
-
2.
[
23]
X
.
K
a
n,
S
.
Z
hu,
Y
.
Z
ha
ng,
a
nd
C
.
Q
i
a
n,
“
A
l
i
ght
w
e
i
ght
hum
a
n
f
a
l
l
de
t
e
c
t
i
on
ne
t
w
or
k,”
Se
ns
or
s
,
vol
.
23,
no.
22,
2023,
doi
:
10.3390/
s
23229069.
[
24]
J
.
T
e
r
ve
n,
D
.
-
M
.
C
ór
dova
-
E
s
pa
r
z
a
,
a
nd
J
.
-
A
.
R
om
e
r
o
-
G
onz
á
l
e
z
,
“
A
c
om
pr
e
he
ns
i
ve
r
e
vi
e
w
of
Y
O
L
O
a
r
c
hi
t
e
c
t
ur
e
s
i
n
c
om
put
e
r
vi
s
i
on:
f
r
o
m
Y
O
L
O
v1 t
o Y
O
L
O
v8 a
nd Y
O
L
O
-
N
A
S
,”
M
ac
hi
ne
L
e
ar
ni
ng and
K
now
l
e
dge
E
x
t
r
ac
t
i
on
, vol
. 5, no. 4, pp. 1680
–
1716,
2023, doi
:
10.3390/
m
a
ke
5040083.
[
25]
A
.
T
r
oc
km
a
n
a
nd
J
.
Z
.
K
ol
t
e
r
,
“
P
a
t
c
he
s
a
r
e
a
l
l
you
n
e
e
d
?
,”
ar
X
i
v
-
C
om
put
e
r
Sc
i
e
nc
e
,
pp.
1
-
16,
2022
,
doi
:
10.48550/
a
r
X
i
v.2201.09792
.
[
26]
D
.
A
ngui
t
a
,
A
.
G
hi
o,
L
.
O
ne
t
o,
X
.
P
a
r
r
a
,
a
nd
J
.
L
.
R
e
ye
s
-
O
r
t
i
z
,
“
A
publ
i
c
do
m
a
i
n
da
t
a
s
e
t
f
or
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on
us
i
ng
s
m
a
r
t
phone
s
,”
i
n
E
SA
N
N
2013 p
r
oc
e
e
di
ng
s
, E
ur
ope
an
Sy
m
pos
i
um
on A
r
t
i
f
i
c
i
al
N
e
ur
al
N
e
t
w
o
r
k
s
, C
om
put
at
i
onal
I
nt
e
l
l
i
ge
nc
e
and
M
ac
hi
ne
L
e
ar
ni
ng. B
r
uge
, B
r
uge
s
,
B
e
l
gi
um
,
pp. 437
-
442,
A
pr
. 2013.
[
27]
O
.
B
a
nos
e
t
al
.
,
“
m
H
e
a
l
t
hD
r
oi
d:
A
nove
l
f
r
a
m
e
w
or
k
f
or
a
gi
l
e
de
ve
l
opm
e
nt
o
f
m
obi
l
e
he
a
l
t
h
a
ppl
i
c
a
t
i
o
ns
,”
i
n
A
m
bi
e
nt
A
s
s
i
s
t
e
d
L
i
v
i
ng and D
ai
l
y
A
c
t
i
v
i
t
i
e
s
, C
ha
m
:
S
pr
i
nge
r
I
nt
e
r
na
t
i
ona
l
P
ubl
i
s
hi
ng, 2014, pp.
91
–
98
, doi
:
10.1007/
978
-
3
-
319
-
13105
-
4_14
.
[
28]
A
.
S
uc
e
r
qui
a
,
J
.
D
.
L
ópe
z
,
a
nd
J
.
F
.
V
a
r
ga
s
-
B
oni
l
l
a
,
“
S
i
s
F
a
l
l
:
A
f
a
l
l
a
nd
m
ov
e
m
e
nt
da
t
a
s
e
t
,”
Se
n
s
or
s
,
vol
.
17,
no.
1,
2017,
doi
:
10.3390/
s
17010198.
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
e
te
c
ti
ng human fall
us
in
g i
nt
e
r
ne
t
of
t
hi
ngs
d
e
v
ic
e
s
f
or
he
al
th
c
ar
e
appli
c
at
io
ns
(
Z
ak
ar
ia
B
e
nhai
li
)
569
[
29]
E
.
K
.
S
a
nna
r
a
,
F
.
P
or
t
e
t
,
a
nd
P
.
L
a
l
a
nda
,
“
L
i
ght
w
e
i
ght
t
r
a
n
s
f
or
m
e
r
s
f
or
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on
on
m
obi
l
e
de
vi
c
e
s
,”
a
r
X
i
v
-
C
om
put
e
r
Sc
i
e
n
c
e
,
pp. 1
-
20,
2022
, doi
:
10.48550/
a
r
X
i
v.2209.11750
.
[
30]
R
.
M
ut
e
ge
ki
a
nd
D
.
S
.
H
a
n,
“
A
C
N
N
-
L
S
T
M
a
ppr
oa
c
h
t
o
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogn
i
t
i
on,”
i
n
2020
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
i
n I
nf
or
m
at
i
on and C
om
m
uni
c
at
i
on (
I
C
A
I
I
C
)
, 2020, pp. 362
–
366
,
doi
:
10.1109/
I
C
A
I
I
C
48513.2020.9065078.
[
31]
K
.
D
.
G
a
r
c
i
a
e
t
al
.
,
“
A
n
e
ns
e
m
bl
e
of
a
ut
onom
ous
a
ut
o
-
e
nc
ode
r
s
f
or
hum
a
n
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on,”
N
e
ur
oc
om
put
i
ng
,
vol
.
439,
pp.
271
–
280, 2021, doi
:
10.1016/
j
.ne
uc
om
.2020.01.125.
[
32]
Y.
-
J
.
K
i
m
,
B
.
-
N
.
K
a
ng,
a
nd
D
.
K
i
m
,
“
H
i
dde
n
M
a
r
kov
m
ode
l
e
ns
e
m
bl
e
f
o
r
a
c
t
i
vi
t
y
r
e
c
ogni
t
i
on
us
i
ng
t
r
i
-
a
xi
s
a
c
c
e
l
e
r
om
e
t
e
r
,”
i
n
2015 I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on Sy
s
t
e
m
s
, M
an, and C
y
be
r
ne
t
i
c
s
,
O
c
t
. 2
015, pp. 3036
–
3041
, doi
:
10.1109/
S
M
C
.2015.528.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Zakaria
Benhaili
received
the
engineering
degree
from
the
Nati
onal
School
of
Applied
Sciences
at
Sultan
Moulay
Slimane
University,
Khouribga,
M
orocco,
currently
a
Ph
.
D
.
student
in
Faculty
of
Sciences
and
Technologies
at
Hassan
1
st
Unive
rsity.
His
main
research
area
includes
internet
of
things,
deep
learning
,
and
it
i
s
applic
at
ions
in
human
activity
recognition
,
pattern
recognition
,
and
smart
homes.
He
can
b
e
contacted
at
email:
z.benhail
i@
uhp.ac.ma.
Youssef
Balouki
Laboratory
of
Mathematics,
Computer
and
Engine
ering
Sciences,
Faculty
of
Scienc
es
and
Techn
iques,
Hassa
n
I
Univer
sity,
Settat,
Mo
rocc
o.
He
is
currently
a
professor
of
computer
science
with
the
Faculty
of
Science
and
Technologies,
Hassan
I
University,
Settat,
Morocco.
He
has
conducted
many
Ph.D.
theses
and
h
as
written
fifty
scientific
articles
in
the
domain
of
artificial
intell
igence,
softwar
e
eng
ineering,
model
-
driven
development,
data
mining,
and
formal
methods.
He
can
be
contacted
at
email:
balouki.youssef@
gmail.com
.
Lahcen
Moumoun
received
the
advanced
graduate
diploma
in
structural
calculati
on
from
the
National
School
of
Electrici
ty
and
Mechanics
of Casabl
anca
from
Hassan
Second
University,
Casablanca,
Moro
cco,
in
1995,
and
the
Ph.D.
deg
ree
in
image
proc
essing
from
in
Faculty
of
S
ciences
and
Technolog
ies
at
Hassan
1st
Universi
t
y,
Morocco,
in
2011.
He
has
been
a
professor
of
computer
science
engineering
with
Hassan
1
st
University,
since
2014.
He has authored or co
-
authored
more th
an 100
refereed jou
rnal and
co
nference papers
, 10 bo
ok
chapters,
and
three
edited
books
with
Elsevier
and
Springer
.
His
research
interest
s
include
pattern recognition, computer vision, big data
,
and deep learning. He
c
an be contacted
at
email:
lahcen.mou
moun@
uhp.ac.ma
.
Evaluation Warning : The document was created with Spire.PDF for Python.