Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
4
,
A
ugus
t
2020
,
pp.
4093
~
41
00
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
4
.
pp
4093
-
41
00
4093
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Analytic
al fram
ewo
rk for optim
ized f
ea
tu
re ext
raction f
or
up
gr
adi
ng occup
ancy sen
sing perf
or
m
ance
Preethi K
. M
ane, K
. Nar
as
i
mha
R
ao
Depa
rtment
o
f
E
le
c
troni
cs
and
In
strum
ent
at
ion
E
ngine
er
ing, BMS
Coll
eg
e
of
En
gine
er
ing, Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Oct
9
, 2
019
Re
vised
N
ov
2
5
,
2019
Accepte
d
Dec
1,
2019
The
adop
ti
on
of
the
o
cc
up
an
c
y
sensors
has
bec
om
e
an
i
nevi
t
abl
e
in
comm
erc
ia
l
and
non
-
comm
erc
ia
l
sec
urity
d
evi
c
es,
owing
to
th
ei
r
profic
i
e
n
c
y
in
the
ene
rg
y
m
a
nage
m
ent
.
I
t
has
bee
n
found
tha
t
the
usage
s
of
conve
nti
on
a
l
sensors
is
shrouded
with
oper
a
tional
pr
ob
le
m
s,
h
enc
e
the
use
of
t
he
Doppler
rad
ar
offe
rs
be
tt
er
m
it
igation
of
such
prob
le
m
s.
How
eve
r,
the
usage
of
Doppler
rad
ar
towar
ds oc
cupanc
y
sensing
in
e
xisti
ng
s
y
st
em i
s found
to
b
e
ver
y
m
uch
in
inf
ancy
st
age.
Moreove
r,
th
e
per
for
m
anc
e
of
m
onit
oring
using
Doppler
rad
ar
is
y
et
to
be
impro
ved
m
ore
.
The
re
fore
,
thi
s
pape
r
int
roduc
es
a
sim
pli
fie
d
fr
amework
for
enr
ic
hing
th
e
e
vent
sensing
per
form
ance
b
y
eff
i
cient
sel
ec
t
ion
of
m
ini
m
al
robust
at
tributes
using
Do
pple
r
rad
ar
.
A
d
o
p
t
i
o
n
o
f
a
n
a
l
y
t
i
c
a
l
m
e
t
h
o
d
o
l
o
g
y
h
a
s
b
e
e
n
c
a
r
r
i
e
d
o
u
t
t
o
f
i
n
d
t
h
a
t
d
i
f
f
e
r
e
n
t
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s
c
o
u
l
d
b
e
f
u
r
t
h
e
r
u
s
e
d
f
o
r
i
m
p
r
o
v
i
n
g
t
h
e
a
c
c
u
r
a
c
y
p
e
r
f
o
r
m
a
n
c
e
f
o
r
t
h
e
f
e
a
t
u
r
e
t
h
a
t
h
a
s
b
e
e
n
e
x
t
r
a
c
t
e
d
i
n
t
h
e
p
r
o
p
o
s
e
d
s
y
s
t
e
m
o
f
o
c
c
u
a
n
c
y
s
y
s
t
e
m
.
Ke
yw
or
d
s
:
Dop
pler rada
r
Ma
chine
le
a
rn
i
ng
Moti
on sensi
ng
Occupa
ncy se
nsi
ng
Op
ti
m
iz
ation
Su
r
veill
ance
Copyright
©
202
0
Instit
ute of
Ad
v
ance
d
Engi
ne
eri
ng
and
Sc
ie
n
ce
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Pr
eet
hi
K. M
an
e
,
Dep
a
rt
m
ent
of
Ele
ct
ro
nics
an
d In
st
ru
m
entat
i
on E
ng
i
neer
i
ng,
BM
S
Coll
ege
of Enginee
rin
g
,
Ba
ng
al
or
e,
In
di
a
.
E
-
Ma
il
: pk
m
ph2008@
gm
ai
l.c
om
1.
INTROD
U
CTION
The
im
ple
m
e
ntati
on
of
t
he
var
i
ou
s
occ
upancy
se
nsors
is
widely
f
ound
in
com
m
ercial
and
non
-
com
m
erci
al
app
li
cat
ion
s
[1
]
.
The
se
se
ns
ors
are
of
l
ow
c
os
t
an
d
e
ff
ect
ive
in
ide
ntifyi
ng
obj
ect
based
distur
a
nces
i
n
the
surveil
la
nc
e
area
[
2].
At
pr
ese
nt,
occup
ancy
sensi
ng
op
e
rati
on
is
carried
out
by
va
rio
us
sens
or
s
e.
g.
pa
ssive
in
fr
a
red
sens
or
,
ultra
s
onic
sen
sor
(a
ct
ive
an
d
passi
ve
),
m
ic
ro
wa
ve
-
base
d
i
den
ti
fi
cat
ion
,
et
c
[3
]
.
Usag
e
of
s
uc
h
c
onve
ntion
al
se
nso
rs
is
re
ported
to
offe
r
issue
s
associat
ed
wi
th
the
acc
ur
ac
y
of
the
sense
d
inf
or
m
at
ion
[4
]
.
The
pr
ob
le
m
with
su
c
h
se
nsors
is
that
th
ey
of
fe
r
m
axi
m
iz
ed
occu
r
re
nces
of
ou
tl
ie
rs.
As
a
resu
lt
,
the
pro
po
e
r
e
xec
ution
of
the
occupa
ncy
se
ns
or
t
o
en
s
ur
e
reli
able
operati
on
bec
om
es
a
chall
en
ging
ta
sk
.
In
this
c
ase,
the
a
doption
of
ra
dar
ba
sed
m
echan
is
m
can
be
e
xpect
ed
to
offe
r
bette
r
so
luti
on
w
hic
h
w
orks
on
e
qu
i
valent
pri
nc
iples
of
r
efle
ct
ion
of
sou
nd.
A
ny
ra
dar
syst
e
m
m
akes
us
e
of
el
ect
ro
m
agn
et
ic
wav
e
s,
w
hic
h
are
tra
ns
m
it
t
ed
an
d
the
n
st
ud
ie
d
ba
sed
on
their
be
hav
i
our
of
or
ie
ntati
on
with
resp
ect
to
s
pee
d,
distance,
a
nd
dire
ct
ion.
Th
e
transm
itted
sign
al
tra
vels
th
r
ou
gh
ai
r
at
th
e
sp
eed
of
li
ght
and
it
al
ways
travels
in
strai
ght
li
ne
s
unti
l
and
un
l
ess
it
encou
nters
im
ped
i
m
ents
li
ke
a
surfac
e
that
is
el
ect
rical
ly
disruptive
.
The
syst
e
m
can
com
pu
te
the
distance
w
hich
basi
cal
ly
rep
resen
t
s
tim
e
delay
for
the
two
oppo
sit
e
ly
scat
te
red
sig
na
ls.
Dif
fer
e
nt
r
esearche
s
ha
ve
ind
ic
at
ed
th
e
sign
i
ficant
i
m
ple
m
entat
ion
of
D
oppler
r
adar
t
o
perform
the
identific
at
ion
of
hu
m
an
m
otion
relat
ed
bio
si
gnal
s
[
5
,
6].
T
he
prop
e
r
usa
ge
of
these
bio
-
s
ign
al
helps
to know th
e
pr
e
ci
se
det
ect
ion
of
t
he
hum
an
body
m
o
ti
on
. A
lo
ng
w
it
h
these researc
hes,
the
re
e
xist
m
any
c
o
s
t
e
f
f
e
c
t
i
v
e
ha
r
d
w
a
r
e
d
e
s
i
g
n
w
h
i
c
h
s
u
p
p
o
r
t
s
t
h
e
d
i
g
i
t
a
l
s
i
gn
a
l
p
r
o
c
e
s
s
i
n
g
(
D
S
P
)
i
n
f
r
a
m
i
n
g
o
f
t
h
e
o
c
c
u
p
a
n
c
y
s
e
n
s
i
n
g
s
y
s
t
e
m
[
7]
.
T
h
e
i
m
p
l
e
m
e
n
t
a
t
i
o
n
o
f
t
h
e
s
e
s
e
n
s
o
r
s
i
s
c
o
n
s
t
r
u
c
t
e
d
b
a
s
e
d
o
n
t
h
e
c
a
r
d
i
o
pu
l
m
o
n
a
r
y
a
c
t
i
o
n
[
8
,
9
]
.
T
h
e
l
i
m
i
t
a
t
i
o
n
o
f
t
h
e
c
o
n
v
e
n
t
i
o
n
a
l
o
c
c
u
p
a
n
c
y
s
e
n
s
o
r
i
s
t
h
a
t
i
t
u
s
e
s
h
i
g
h
l
y
non
-
i
n
v
a
s
i
v
e
m
ec
han
ism
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
4093
-
4100
4094
consi
der
e
d
for
the
li
ne
of
s
igh
t
ob
j
ect
s
s
ensin
g
an
d
a
r
e
us
ed
i
n
the
diag
no
sis
of
cl
inica
l
app
li
cat
ion
s.
This
m
anu
scrip
t
aim
s
to
intro
duce
cost
ef
fe
ct
ive
m
achine
le
arn
in
g
m
echan
ism
fo
r
en
ha
ncin
g
the
accu
racy
of
occupa
ns
y
se
nsi
ng
us
in
g
D
opple
r
ra
dar.
T
he
m
anu
scri
pt
is
orga
nized
w
it
h
existi
ng
res
earches
on
detect
ion
schem
es
m
ean
t
with
power
transm
issi
on
li
nes
,
hi
gh
li
gh
te
d
th
e
resear
ch
pr
oble
m
con
sidere
d
for
the
desi
gn
and
the
s
olu
ti
on
int
rod
uced
fo
r
th
e
pro
bl
e
m
is
br
ie
fe
d
in
sect
ion
1
.
T
he
syst
em
i
m
ple
m
enta
ti
on
of
the
pro
posed
s
yst
e
m
is
descr
ibed
i
n
sect
io
n
2
th
rou
gh
im
ple
m
entat
ion
str
at
egy
an
d
al
go
rithm
i
m
p
lem
e
ntati
on
in se
ct
io
n 3. T
he
c
on
cl
us
ive
po
i
nts
of
t
he
m
anuscript is
d
is
cusse
d
in
secti
on 4.
The
var
i
ou
s
te
chn
i
qu
e
s
w
hic
h
are
ass
ociat
ed
with
the
occ
up
a
ncy
se
ns
in
g
a
nd
t
he
rese
arch
t
rend
i
s
highli
gh
te
d
th
e
wor
k
of
Ma
ne
a
nd
Ra
o
[
10]
.
I
n
Am
ay
ri
et
al
.
[11]
ha
ve
intr
oduce
d
a
Ba
ye
sia
n
ne
twor
k
base
d
m
od
el
wh
ic
h
helps
to
est
im
at
e
the
sensi
ng
occ
up
a
ncy.
The
furter
i
nv
e
sti
ga
ti
on
on
occupan
cy
sensing
pe
rform
ance
is
de
s
cribe
d
in
Elt
abie
et
al
.
[
12]
ov
e
r
t
he
c
onditi
on
of
un
der
us
e
d
sp
ec
trum
.
Ca
rr
yi
ng
the
s
pectr
um
us
age
in
the
occ
up
ancy
sen
sin
g,
the
w
ork
of
Hlophe
et
al
.
[13]
ha
ve
disc
us
se
d
the
dee
p
le
a
rn
i
ng
ba
sed
stoc
ha
sti
c
appro
a
ch
.
The
inc
orpora
ti
on
of
I
nter
ne
t
of
Thi
ngs
(
I
oT
)
with
occup
ancy
sensing
is
f
ound
i
n
Ya
ng
et
al
.
[14]
wh
e
r
e
channel
sta
te
inform
at
ion
is
con
si
der
e
d.
The
m
achine
le
arn
i
ng
appr
oach
with
the
m
otion
se
ns
in
g
m
echan
ism
is
intro
duc
ed
f
or
the
occ
up
a
ncy
sen
sin
g
in
Wu
et
al
.
[15].
In
t
he
w
ork
of
Ca
o
et
al
.
[
16]
,
the
sm
art
se
ns
io
n
a
ppro
ac
h
is
pr
ese
nted
w
hich
a
da
pts
the
sign
ific
a
nt
bui
ldin
g
m
echan
ism
with
bette
r
po
wer
m
anag
em
ent.
T
he
hum
ans
c
ountin
g
base
d
on
the
bi
o
-
si
gn
al
s
thr
ough
occupa
ncy
se
nsi
ng
is
descr
i
be
d
in
the
w
ork
of
Sa
ntra
et
al
.
[
17]
.
T
he
se
nsory
data
base
d
obj
ect
detect
ion
is
pro
vid
e
d
in
L
uppe
a
nd
Sha
ban
i
[18].
T
he
sense
d
data
f
ro
m
the
occup
ancy
sensi
ng
is
analy
sed
by
us
in
g
the
evide
ncy
theo
ry
in
Nesa
and
Ba
ne
rj
ee
[19].
The
w
or
k
of
Ji
n
et
al
.
[20]
s
m
art
m
e
te
r
based
occupan
cy
sensing
m
echan
ism
is
pr
esen
te
d
to
perf
or
m
the
power
m
a
nag
em
ent.
The
blu
et
hoot
h
ge
ner
at
e
d
sig
nals
ar
e
us
e
d
f
or
the
oc
cup
a
ncy
detect
ion
by
Shen
an
d
Ne
ws
ha
m
[2
1].
The
s
pecific
f
orm
s
of
sens
ors
are
us
e
d
i
n
the
w
ork
of
Liu
et
al
.
[
22
]
for
occupa
nc
y
sensing
by
us
i
ng
the
hi
dd
e
n
m
arkov
m
od
el
.
The
idea
of
env
i
ronm
ental
factor
s
har
i
ng
i
s
discuss
e
d
in
Jin
et
al
.
[2
3]
f
or
the
oc
cu
pa
nc
y
sensing.
Th
e
hybr
i
d
m
ech
anis
m
for
e
nh
a
ncem
ent
of
acc
ur
a
cy
in
detect
ion
process
is
pres
ented
i
n
Kh
a
n
et
al
.
[
24]
.
T
he
s
pectr
um
sensing
thr
ough
the
de
ep
le
ar
ning
ba
sed
stoc
hastic
appr
oach
is
f
ound
in
Li
u
et
al
.
[25]
an
d
s
pectr
um
sension
i
n
Ma
riani
et
al
.
[26]
inc
orp
ora
ti
ng
t
he
c
ogniti
ve
ra
dio.
The
w
ork
of
Albesa
an
d
G
asulla
[
27
]
ha
s
us
e
d
the inducti
ve
li
nk
s
f
or
the
occ
up
a
ncy fact
or
detect
ion
by ada
pting
t
he
idea
o
f
m
agn
et
ic
co
upli
ng. A cap
aci
ti
ve
m
od
el
li
ng
is
per
f
or
m
ed
in
th
e
work
of
C
oole
y
et
al
.
[2
8]
and
t
he
w
ork
of
Ho
ss
ai
n
[
29]
adap
te
d
the
P
ost
eriori
m
echan
ism
.
T
he
in
du
ct
uce
pri
nciple
inc
orp
or
at
e
d
in
the
work
of
Ge
orge
et
al
.
[30]
wh
il
e
dis
pu
te
s
of
te
xt
cl
us
te
r
in
the
work
of
Nikha
rth
an
d
S
ubra
hm
anya
m
[3
1]
f
or
occupa
ncy
sensing.
T
he
pro
blem
s
identi
fied
i
n
the ex
ist
i
ng
res
earches
are
h
i
gl
igh
te
d bel
ow.
The
re
view
of
li
te
ratur
e
cl
earl
y
ind
ic
at
es
t
hat,
the
existi
ng
li
te
ratur
e
la
cks
s
ign
ific
a
nt
w
or
k
realt
ed
t
o
the
eval
ution
of
i
den
ti
ficat
io
n
of
the
occ
upancy
se
nsors
and
m
any
of
the
ap
proac
he
s
inclu
des
th
e
us
e
of
the
D
oppe
r
Ra
dar
base
d
com
pu
ta
ti
on
usi
ng
wav
e
si
gnal
in
te
rp
retat
io
ns
usi
ng
t
he
intrin
s
ic
featur
es
real
te
d
to
the
ph
ysi
ol
og
i
cal
at
tribu
te
s.
On
e
of
the
cr
it
ic
al
req
uirem
ent
in
the
appl
ic
at
ion
s
based
on
the
occ
upancy
require
a
n
ac
c
ur
at
e
cal
ssif
cat
ion
s
of
the
act
i
on
s
of
the
oc
cup
a
nts,
w
her
e
as
the
tra
diti
onal
ap
proac
hes
la
ck
s
the
capaci
ty
to
cl
assify
the
oc
cup
a
nts
act
io
ns
or
t
he
ty
pical
act
ivit
ie
s
in
m
or
e
com
pu
ta
t
ion
al
y
ef
fcienc
t
an
d
accurate
way.
The
chall
en
ge
s
li
es
in
or
der
to
dev
el
op
s
uc
h
cl
assifi
cat
ion
m
od
el
s
a
re
it
s
dep
e
nd
e
ncie
s
on
the
hi
gh
e
r
dim
entions
of
the
featu
re
set
f
r
om
a
la
rg
e
set
of
e
xpli
ci
t
sensors
s
o
that
s
yst
e
m
m
od
el
desig
n
beco
m
es
capa
ble
of
ha
ndli
ng
ei
the
r
si
ng
le
cl
ass
of
ect
ivit
y
decti
on
or
t
he
m
ulti
ple
cl
asses
of
the
ac
ti
on
or
act
ivit
y
detectio
n
in
m
or
e
com
pu
ta
ti
on
al
eff
ci
ue
nt
m
a
nn
e
r
to
m
inize
the
com
pu
ta
ti
on
al
cost.
Hen
ce
,
the
desi
gn
ap
proac
h
of
s
uc
h
cl
assifi
cat
io
n
m
od
el
s
re
quires
ha
ndli
ng
t
he
huge
ver
ac
it
y
of
the
occ
up
a
nt’
s
act
ion
s
a
nd
act
ivit
ie
s
in
the
r
egio
n
of
occ
up
ant
ap
plica
ti
ons
of
inte
rest.
T
her
e
fore,
the
i
niti
al
obj
ect
ive
is
to
handle
the
iss
ue
is
desig
n
a
c
om
pu
ta
ti
on
al
c
os
t
ef
fecti
ve
le
arn
i
ng
m
od
el
wh
ic
h
cl
assifi
e
s
the
occ
upant
act
ion
s
cat
egorical
ly
t
hen
the
ai
m
is
to
op
ti
m
iz
e
the
m
od
el
with
the
ind
entifcat
io
n
of
m
or
e
signi
ficant
feat
ur
e
cl
asses
that
i
m
pact
w
el
l
on
t
he
act
i
on
s
s
ub
-
s
pce
cl
assifi
cat
ion
i
n
case
of
am
biguit
y
so
t
ha
t
the
se
pecifit
y
and
ensiti
vity
o
f
th
e m
od
el
s ar
e a
ccepta
ble.
The
propose
d
s
yst
e
m
is
design
ed
a
nd
de
velo
ped
f
or
im
pr
ov
ing
the
pri
or
m
od
el
[32]
w
here
the
pri
m
e
fo
c
us
is
offe
re
d
to
wards
c
on
s
tructi
ng
a
r
obust
m
at
he
m
atica
l
fr
am
ewo
r
k
in
order
t
o
opti
m
iz
e
the
perfor
m
ance
of
ra
dar
-
based
sensi
ng.
The
stud
y
i
ntr
oduc
es
a
novel
f
orm
of
desi
gn
of
the
Dop
pler
rad
a
r
dif
fer
e
nt
from
the
tra
diti
on
al
m
od
el
in
ord
er
to
car
ry
ou
t
an
ef
fecti
ve
e
ven
t
cl
assifi
cat
ion
of
t
he
s
ub
j
ect
s.
The
pr
opose
d
arch
it
ect
ural
sc
hem
e fo
r
t
his s
e
ns
in
g o
per
at
io
n
is s
how
n
i
n
F
igure
1
.
The
c
om
plete
desig
n
of
the
pro
po
se
d
syst
e
m
is
carried
out
us
in
g
a
naly
ti
cal
m
od
el
ing
that
perform
dev
el
op
i
ng
cer
ta
in
at
tribu
t
es
for
assist
in
g
i
n
cl
assify
ing
di
f
fer
e
nt
f
orm
s
of
sense
d
eve
nt
i
n
order
to
facil
it
ate
pr
eci
se
se
ns
i
ng
of
s
ubj
ect
oc
cup
a
ncy.
T
he
pro
po
se
d
st
udy
al
so
form
ulate
s
a
novel
extracti
on
of
e
ssentia
l
featur
e
s in o
rder to
im
pr
ovise
the
perform
ance o
f
classi
fica
ti
on
on the
basi
s of signal
s e
xtracted
from
D
opple
r
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
An
alyti
cal fra
mework
for
opt
imize
d
fe
atu
re
ext
ra
ct
ion f
or…
(
Preet
hi K.
Ma
ne
)
4095
rad
a
r.
T
he
stu
dy
m
a
inly
tracks
do
wn
t
he
m
ob
il
it
y
-
base
d
m
ov
e
m
ent
as
t
he
pri
m
e
so
ur
c
e
of
e
ven
t
generati
on
wh
e
re
Dop
pler
rad
a
r
is
us
e
d
trakin
g
with
r
espect
to
t
rack
points.
The
s
yst
e
m
carries
ou
t
i
den
ti
ficat
ion
of
the
s
ubj
ect
s
on
the
basis
of
th
is
track
points
.
The
syst
em
i
m
ple
m
entat
ion
be
gin
s
with
th
e
in
vestigat
io
n
over
the
com
pu
ta
ti
on
of
s
uch
at
ti
bute
s
in
order
t
o
plan
ou
t
ef
fe
ct
ive
log
ic
al
conditi
on
for
in
corp
or
at
in
g
accuracy
wh
il
e
pe
rfo
rm
i
ng
est
im
at
ion
op
e
ra
ti
on.
F
or
this
pur
po
e
,
th
e
stud
y
f
ram
es
up
a
n
al
gorit
hm
that
is
resp
onsible
for
ch
oo
si
ng
an
el
it
e
fo
rm
of
an
at
tribu
te
in
or
de
r
to
m
ake
the
syst
e
m
ind
epe
ndent
f
rom
pr
ocessi
ng
higher
nu
m
ber
of
sy
stem
at
tribu
te
s.
This
has
a
direct
in
flue
nce
to
wards
r
edu
ci
ng
com
pu
ta
ti
on
al
c
omplexit
y.
The
st
ud
y
will
al
so
ha
ve
a
n
inclusio
n
of
m
ulti
ple
var
ia
nts
of
i
den
ti
ficat
i
on
at
tri
bu
te
s
usi
ng
D
oppler
r
adar
wh
il
e
it
can
per
f
or
m
diff
ere
nt
angular
ori
ent
at
ion
f
or
trac
kin
g
the
diff
e
rent
even
ts
with
de
fine
d
an
d
unkn
ow
n
or
ie
ntati
on
.
T
he
con
sec
utive
par
t
of
the
sec
ti
on
hi
gh
li
ghts
about
the
im
plem
entat
ion
appr
oach
a
dopt
ed
in
the pr
opos
e
d
st
ud
y.
A
n
a
l
y
t
i
c
a
l
M
o
d
e
l
F
e
a
t
u
r
e
C
o
n
s
t
r
u
c
t
i
o
n
F
e
a
t
u
r
e
C
o
m
p
u
t
a
t
i
o
n
A
c
c
u
r
a
c
y
o
f
F
e
a
t
u
r
e
E
s
t
i
m
a
t
e
s
F
e
a
t
u
r
e
S
e
l
e
c
t
i
o
n
A
n
a
l
y
s
i
s
S
t
a
t
i
s
t
i
c
s
O
r
i
e
n
t
a
t
i
o
n
A
n
g
l
e
T
e
s
t
i
f
y
a
c
c
u
r
a
c
y
I
d
e
n
t
i
f
i
c
a
t
i
o
n
o
f
I
m
p
o
r
t
a
n
t
F
e
a
t
u
r
e
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
Figure
1.
Pro
pose
d
arc
hitec
tu
re
of
sensi
ng a
nd classi
ficat
io
n
2.
PR
OPOSE
D
IMPLEME
N
TATION
A
PP
ROAC
H
This
sect
io
n
di
scusses
a
bout
the
i
m
ple
m
entat
ion
ap
proac
h
to
wards
f
ra
m
ei
ng
up
a
m
od
el
that
ca
n
enh
a
nce
t
he
pe
rfor
m
ance
of
sensi
ng
of
propose
d
syst
em
us
i
ng
it
s
opti
m
iz
at
ion
ap
proach.
T
he
pro
po
s
e
d
syst
e
m
ta
rg
et
s
to
i
m
pr
ov
e
the
per
f
or
m
ance
of
the
syst
em
as
a
m
echan
ism
to
carry
ou
t
op
ti
m
iz
at
ion
without
any
form
of
de
pende
ncies
of
so
phist
ic
at
ed
r
eso
ur
ces
.
The
discuss
i
on
of
the
pro
pose
d
syst
e
m
i
m
ple
m
e
ntati
on
is
carried
out
with
res
pect
to
the
strat
egy
a
dopted
f
or
t
he
i
m
ple
m
entat
io
n,
c
or
e
desi
gn
of
syst
em
,
and
flo
w
of
execu
ti
on
f
or
il
lustrati
ng the
a
lgorit
hm
steps.
2.1.
S
tr
at
e
gy
of impl
ement
ati
on
The
im
po
rtant
strat
egy
of
t
he
i
m
ple
m
entation
is
a
bout
c
arr
yi
ng
ou
t
t
he
optim
iz
at
ion
process
f
or
extracti
on
of
f
eat
ur
e
wh
e
re
the
sig
nals
ass
ociat
ed
with
t
he
m
ov
em
ent
is
c
ollec
te
d
by
the
D
oppler
r
adar.
It
shou
l
d
be
no
te
d
that
the
se
ns
in
g
is
car
ried
out
over
dif
f
eren
t
set
of
e
ve
nts.
T
he
pr
opos
e
d
syst
em
m
ake
use
of
t
he
sta
ti
c
m
echan
ism
for
perform
ing
fea
ture
e
xtracti
on
in
pr
ese
nce
of
m
ulti
ple
num
ber
of
fluctu
at
ion
of
env
i
ron
m
ent
var
ia
bles
e.g.
ti
m
e
fo
r
set
tl
ing,
ang
le
of
ori
entat
ion
,
qu
al
it
y
of
sign
al
.
T
her
e
fore,
the
s
tud
y
is
m
or
e
e
m
ph
asi
zed
ov
e
r
in
for
m
at
ion
extract
ion
i
n
preci
se
m
ann
er
t
hat
is
co
nn
ect
e
d
w
it
h
the
at
trib
utes
for
even
t.
This
op
erati
on
posit
iv
el
y
influ
e
nces
the
cl
assifi
cat
ion
m
echan
ism
an
d
com
pu
ta
ti
on
al
perf
or
m
ance
of
pro
po
se
d
syst
e
m
u
sing
m
achine lear
ning.
T
he
n
e
xt secti
on i
ll
us
ta
trat
es ab
out co
re
desig
n of sy
ste
m
.
2.2.
Co
re
des
ign
of sy
s
tem
The
pro
posed
stud
y
is
init
ia
te
d
on
the
ba
sis
of
t
he
c
onsid
erati
on
th
a
t
a
Dop
pler
ra
dar
is
ge
ner
at
in
g
a
sp
eci
fic
f
orm
of
signa
l
in
the
form
of
m
od
ulate
d
f
re
qu
e
ncy
of
li
ne
arized
f
or
m
.
The
pro
posed
stud
ie
s
consi
der
the
c
on
ce
pt
of
eve
nt
as
set
of
s
pecific
e
ven
ts
or
act
io
ns
c
r
eat
ed
by
t
he
hu
m
an
m
ov
em
ent
as
the
pr
im
e
ind
ic
at
or
of
the
oc
cup
a
ncy
sens
ing
.
Howe
ver,
it
is
qu
it
e
a
c
om
plex
ta
sk
in
orde
r
to
unde
rstan
d
the
or
ie
ntati
on
of
the
act
io
ns
exh
i
bited
by
the
hum
an
as
they
are
hi
gh
ly
unpr
e
dicta
ble.
Howe
ver
,
t
he
re
are
al
ways
so
m
e
s
et
of
act
ion
s
t
hat
sign
ifie
s
te
ntati
ve
eve
nts.
The
D
oppler
rad
a
r
colle
ct
s
the
traject
or
y
that
is
ref
le
ct
ed
f
r
om
the
par
ts
of
th
e
body
in
the
f
or
m
of
the
m
od
ulate
d
si
gn
al
s
.
At
pr
e
sent,
th
ere
are
tw
o
dis
crete
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
4093
-
4100
4096
te
chn
iq
ues
i
n
orde
r
to
be
us
e
d i
n
occ
up
a
ncy
sensing
wh
e
re
on
e
of
t
he
a
ppr
oach
e
s is conne
ct
ed
with
k
ine
m
at
ic
syst
e
m
wh
il
e
ano
t
her
ap
proa
ch
is
as
so
ci
at
e
d
with
t
he
m
otion
-
base
d
at
tr
ibu
te
s.
I
n
t
he
existi
ng
ap
pro
aches,
m
ajo
rity
of
t
he
us
age
is
ass
ociat
ed
with
t
he
co
nventio
na
l
kin
em
atic
m
od
el
ing
ap
proach.
S
uch
m
od
el
is
com
plete
ly
de
sign
e
d
on
t
he
basis
of
the
inv
est
igati
on
t
ow
a
r
ds
sp
eci
fic
m
ov
em
ent
of
t
he
hum
an
body
consi
sti
ng
of
joint
a
nd
point
s
m
ov
em
ent
c
onnected
with
the
hum
an
bo
dy.
As
per
the
co
nv
e
ntio
nal
syst
e
m
,
it
is
belie
ve
d
that
the
D
oppl
er
si
gn
al
f
or
the
hum
an
is
accum
ulate
d
from
the
m
ulti
p
le
dif
fer
e
nt
pa
rts
of
the
ta
r
get
po
i
nt
of
the
hum
an
body
that
is
be
li
eved
t
o
be
c
har
act
erize
d
by
trajecto
ry
with
ti
m
e
var
yi
ng
trai
t
s
as
per
the
kin
e
m
at
ic
m
od
el
.
Howe
ver,
this
m
od
el
is
restrict
ed
on
ly
to
th
e
scenario
wh
e
n
the
subj
ect
is
fo
un
d
m
ob
il
e
ov
er
th
ei
r
feet.
T
he
de
vices
associat
e
d
with
the
m
oti
on
ca
pture
are
connecte
d
with
fu
rt
her
accu
ra
cy
in
the
con
te
xt
of
the
extracte
d
s
pectr
ogram
of
a
hu
m
an.
Wh
il
e
com
par
ing
w
it
h
the
kin
em
atic
m
od
el
ing
(t
hat
is
fou
nd
to
ha
ve
m
or
e
dep
e
nde
ncies
on
the
co
m
pu
ta
ti
on
of
t
he
te
m
po
ra
l
fa
ct
or
a
sso
ia
te
d
with
t
he
hu
m
an
body)
,
su
c
h
a
ppro
ac
h
on
the
basis
of
the
m
otion
.
T
his
a
ppro
ac
h
is
capa
ble
of
t
ra
ns
f
or
m
ing
t
he
m
ov
e
m
ent
radar
data
colle
ct
ed
f
r
om
certai
n
sens
or
s
in
t
he
form
of
the
sim
ulate
d
data.
T
his
s
i
m
ulate
d
versi
on
of
t
he
real
sign
al
s
us
es
the
sta
nda
rd
ra
da
r
co
nce
pt
in
orde
r
to
c
om
pu
te
the
extracte
d
sig
nal.
I
t
will
m
ean
that
the
gen
e
rated
data
is
abso
l
utely
no
t
in
the
ge
ne
r
ic
fo
rm
and
it
ref
e
rs
to
certai
n
ex
plici
t
fo
rm
of
sig
nal
an
d
hen
ce
t
he
pro
pose
d
syst
e
m
co
ns
ide
rs
the
d
at
a t
o b
e g
e
ner
at
e
d on
the b
a
sis o
f
m
otion
c
on
t
ro
ls.
Fr
om
the
c
onve
ntion
al
the
or
y
of
Dop
pler
ra
dar
syst
em
,
it
i
s
belie
ve
d
that
aggre
gation
of
interfe
ren
c
e
and
re
pons
e
of
t
he
ta
r
get
resu
lt
s
in
cu
m
ula
ti
ve
return
sig
nal
that
is
extracte
d
by
the
D
oppler
rada
r
.
The
pro
po
sed
stud
y
as
s
umes
that
it
im
ple
m
ents
sup
pr
es
sion
of
cl
utter
an
d
t
her
e
fore,
the
m
od
el
pe
rfor
m
s
represe
ntati
on
of
the
rem
nan
t
interfer
e
nce
in
the
form
of
no
rm
al
distribu
ti
ve
no
ise
.
The
r
efore,
the
sig
na
l
that
is
obta
ined
f
r
om
the
so
urce
t
arg
et
point
ca
n
be
basical
ly
co
ns
ide
re
d
as
the
re
plica
te
d
co
py
of
the
s
hifted
fr
e
qu
e
ncy
c
ha
racteri
zed
by
delay
ed
ti
m
e
associat
ed
with
the
f
orward
ed
sig
nal.
Th
e
ne
xt
pa
rt
of
stud
y
i
m
ple
m
entat
io
n
is
m
or
e
fo
c
use
d
on
extracti
ng
the
feat
ur
es
fo
r
bette
r
acc
ur
acy
of
perf
orm
ing
occupa
nt
act
ion
sensing
.
At
pr
esent,
t
her
e
ar
e
va
rio
us
f
orm
s
of
inf
orm
a
ti
on
t
hat
can
be
obta
ined
f
r
om
the
eve
nt
feed
in
the
form
of
diff
ere
nt
act
ion
s.
Hen
ce
,
an
ex
pl
ic
it
app
ro
ac
h
is
pr
ese
nted
in
pro
po
se
d
syst
em
wh
ere
the
syst
e
m
consi
ders three
ty
pes
of
data i
.e.
trai
ni
ng
dat
a,
te
sti
ng
data,
and
featu
res da
ta
. A
ll
these d
a
ta
acts as an
in
pu
t t
o
the
pr
opos
e
d
s
yst
e
m
.
The
pr
opose
d
stu
dy
co
ns
tr
ucts
the
dat
a
struct
ur
e
d
of
po
te
ntial
featu
r
es
w
hich
is
f
orm
ed
us
in
g
both
te
st
data
and
trai
ne
d
data.
The
da
ta
base
act
ually
rep
resen
ts
c
ertai
n
sp
eci
fic
set
of
m
ov
em
e
nt
of
su
bject
s
a
nd
therefo
re
the
pro
po
se
d
syst
e
m
ob
ta
ins
t
he
in
form
at
io
n
ass
ociat
ed
with
the
ide
nt
it
y
of
the
sub
j
ect
s
in b
ot
h
trai
n
a
nd
t
est
data.
All
th
e
ob
ta
i
ned
d
at
a
is
now
norm
alized
to o
btain
t
he
m
axi
m
u
m
value.
The
pro
posed
st
ud
y
pe
rfor
m
s
an
assessm
e
nt
to
check
if
there
is
a
pu
re
st
fo
rm
of
the
data
with
res
pect
to
the in
div
id
ual i
den
ti
ty
of the
a
ct
ion
s a
nd s
ubje
ct
. F
inall
y, th
e p
r
opose
d st
udy o
btains t
he norm
al
iz
ed
data with
resp
ect
t
o
th
e a
ct
ion
s c
orres
pond
i
ng to
t
he
e
ve
nts.
The
ne
xt
pa
rt
of
the
im
ple
m
entat
ion
is
ass
ociat
ed
with
t
he
im
plica
ti
on
of
the
m
achi
ne
le
ar
ning
.
The
c
or
e
m
otive
of
this
pa
r
t
of
th
e
im
ple
m
entat
ion
is
to
te
sti
fy
us
a
ge
of
t
hr
ee
dif
f
eren
t
m
echan
i
sm
of
m
achine lear
nin
g ap
proac
h
i.e
. d
ee
p
le
ar
ning
appr
oach,
K
-
ne
arest nei
ghbor
ing
al
gorithm
(
KNN), a
nd Suppo
rt
Vecto
r
Ma
chi
ne
.
For
this
pur
po
s
e,
al
l
the
e
xpli
ci
ts
featur
es
are
loa
de
d
an
d
sta
ti
sti
cal
info
rm
ation
is
ext
r
act
ed.
The
pr
opos
e
d
stud
y
ext
racts
sta
nd
a
rd
de
viati
on
from
t
he
feat
ur
es
a
nd
chec
ks
if
t
he
obta
ine
d
f
eat
ur
es
(or
sta
ndar
d
de
viati
on)
are
f
ound
m
or
e
than
ideal
value
of
sta
ti
sti
cs.
The
stud
y
obta
in
s
al
l
the
sa
m
pl
es
with
resp
ect
to
the
sta
ti
sti
cal
featur
e
wh
ic
h
is
fu
r
ther
ra
ndom
iz
e
d.
The
nex
t
pa
rt
of
the
im
ple
m
entat
ion
is
about
perform
ing
tra
ining
of
the
da
ta
con
si
der
i
ng
prese
nce
of
error.
Th
e
stu
dy
sel
ect
s
the
cum
ulati
ve
num
ber
of
trai
ning
data
as
well
as
co
ns
ide
rs
the
a
r
bitrary
fla
g
of
the
trai
ne
d
da
ta
.
This
op
e
r
at
ion
is
f
ollo
wed
by
extracti
ng
the
i
nput
of
the
trai
ning
f
ollow
e
d
by
extracti
ng
the
co
rr
es
po
ndi
ng
ta
r
get
of
th
e
trai
ning.
O
wi
ng
t
o
the
ad
opti
on
of
la
ye
r
-
ba
sed
m
achine
le
a
rn
i
ng
syst
em
,
the
pro
posed
stu
dy
consi
ders
tw
o
equ
i
valent
la
ye
rs
a
nd
sp
eci
fic
value
of
the
i
nput
si
de
with
discret
e
de
finiti
on
of
the
num
ber
of
cl
asses
.
T
he
pr
im
ary
op
e
rati
on
of
the
pro
pose
d
m
achine
le
ar
ni
ng
syst
em
ini
ti
at
es
with
the
declarat
io
n
of
the
a
ppr
opri
at
e
at
tribu
te
s
wh
e
re
the
input
siz
e
corres
ponds
w
it
h
the
siz
e
of
the
trai
ned
data.
This
operati
on
is
al
so
f
ollow
e
d
by
init
ia
li
zi
ng
the
antic
ipate
d
m
ean
act
ivatio
n
functi
o
n
f
or
the
hidde
n
uni
ts,
weig
ht
re
duct
ion
at
trib
ute
and
pe
nalty
at
t
rib
ute
.
The
seco
nd
ste
p
of
operati
on
of
th
e
pro
pose
d
syst
e
m
is
m
eant
fo
r
pe
rfor
m
ing
the
encodin
g
m
echan
ism
.
The
im
ple
m
ention
is
then
f
oc
us
e
d
on
perf
orm
ing
op
ti
m
izati
on
us
i
ng
sta
nd
a
r
ds
of
qu
as
i
-
Ne
wton
m
ec
han
ism
wh
ic
h
interest
ing
ly
offer
s
a
bette
r
form
of
m
e
m
or
y
op
tim
iz
at
ion
.
The
presente
d
op
ti
m
iz
ation
appr
oach
perform
s
a
co
m
pu
ta
ti
on
of
the
hessia
n
m
a
t
rix
in
orde
r
to
enh
a
nce
the
se
arch
s
pace.
T
hi
s
m
echan
is
m
of
de
e
p
le
arn
in
g
a
ppr
oa
ch
offers
bet
te
r
perform
ance
ap
plica
ble
ove
r
diff
e
re
nt
do
m
ai
ns
inclusive
of
la
ngua
ge,
s
pe
ech,
visio
n,
et
c.
A
nothe
r
sig
nificant
advanta
geous
featu
re
of
us
in
g
dee
p
ne
ur
al
net
wor
k
is
the
m
ini
m
izati
on
of
the
de
pende
nc
ie
s
fo
r
a
ny
s
or
t
of
e
xp
e
ns
i
ve
en
gin
ee
rin
g
proc
ess
that
is
con
si
der
e
d
to
save
m
axi
m
u
m
pro
du
ct
io
n
tim
e
wh
il
e
pr
act
ic
ing
m
achine
le
arn
i
ng
a
ppro
a
c
h.
At
the
sam
e
tim
e,
the
adap
ta
bili
ty
char
ect
erist
ic
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
An
alyti
cal fra
mework
for
opt
imize
d
fe
atu
re
ext
ra
ct
ion f
or…
(
Preet
hi K.
Ma
ne
)
4097
of
the
dee
p
le
arn
i
ng
a
ppr
oa
ch
is
ano
t
her
su
it
able
featu
r
e
that
increase
s
the
app
li
ca
bi
li
t
y
of
the
m
achine
le
arn
in
g
a
ppr
oa
ch.
In
sim
pler
sens
e,
it
will
m
ean
that
over
a
n
inc
reas
ing
set
of
th
e
data,
de
ep
le
arn
i
ng
appr
oach cl
aims t
o offer
enha
nced pe
rfo
rm
a
nce in com
par
i
so
n t
o exist
in
g appr
oach
es
.
The
m
achine
le
arn
i
ng
a
ppr
oa
ch
is
to
perfor
m
i
m
pl
ic
at
ion
s
of
t
he
K
N
N
al
gorith
m
that
us
es
cl
ass
m
e
m
ber
sh
ip
as
the
rep
rese
nta
ti
on
of
the
out
pu
t.
The
cl
ass
ific
at
ion
of
a
n
obj
ect
is
carried
out
with
an
ai
d
of
m
ul
ti
ple
sel
ect
ion
s
of
it
s
neighb
or
i
ng
no
de
s
w
her
e
the
cl
ass
is
assi
gn
e
d
a
n
obj
ect
w
it
h
highest
c
om
m
on
a
m
on
g
the
s
pe
ci
fic
integ
er
nu
m
ber
of
near
es
t
neig
hborh
ood.
O
ne
inte
resti
ng
fact
a
bout
the
K
N
N
al
go
rithm
is
that
it
per
form
s
the
approxim
at
ion
of
the
functi
on
local
ly
and
the
syst
em
def
er
s
al
l
the
com
pu
ta
ti
on
unti
l
an
d
un
le
ss
t
he
cl
as
sific
at
ion
is
no
t
carried
out.
S
om
e
of
the
ess
e
ntial
char
ect
e
risti
cs
for
w
hic
h
the
KNN
al
gorithm
has been
sel
ect
ed
f
or
pe
rfor
m
i
ng
t
rainin
g
a
nd test
ing
are
as
f
ollows:
Th
is al
gorithm
is q
uite sim
ple an
d high
ly
sp
onta
neous
w
her
e
the
sim
pli
ci
ty
of
the
com
pu
ta
ti
on
is
c
arr
ie
d
ou
t
by
r
eadin
g
al
l
the
values
i
n
the
da
ta
set
.
Ther
e
is
no
s
pe
ci
fic
assum
ption
co
nnect
ed
to
the
us
age
of
the
K
N
N
al
gor
it
h
m
un
li
ke
va
rio
us
e
xisti
ng
l
inear
regressio
n
al
gorithm
as
well
as
there
is
no
inclusio
n
of
s
te
ps
of
the
tra
ining.
Th
ere
is
no
m
od
el
buil
din
g
process
in
KNN
but
only
the
ind
e
xing
of
the
no
vel
entri
es
of
data
is
carried
out
in
it
.
Anothe
r
sig
ni
ficant
char
ect
e
risti
cs o
f
K
N
N
is t
hat it
sp
on
ta
ne
ous
ly
ev
olv
es. Ca
pab
il
it
y t
o
so
lv
e the
m
ulti
-
cl
a
ss p
r
ob
le
m
is a
no
t
her
interest
ing cha
recteri
sti
cs of the
KNN
al
go
rithm
w
it
h
enh
a
nce c
a
pab
il
it
y t
o
a
dopt
diff
e
re
nt d
ist
an
ce m
od
el
s i
n
it
viz.
i)
Mi
nkowski
distanc
e,
ii
)
Ma
nhat
ta
n
distance,
ii
i)
H
a
m
m
ing
distan
ce,
iv
)
E
uclide
an
distance.
T
he
ne
xt
par
t
of
the
im
p
lem
entat
ion
for
the
m
achine
l
earn
i
ng
w
il
l
be
Su
pp
or
t
Vect
or
Ma
chine
(
SVM
).
Ba
sic
al
ly
,
SVM
is
a
t
ype
of
the
su
pe
rv
ise
d
le
arn
i
ng
m
od
el
that
is
associat
ed
with
enh
a
nce
d
cl
assifi
cat
ion
perform
ance.
Along
with
perf
or
m
i
ng
li
near
cl
as
sific
at
ion
oper
at
ion
,
the
SVM
ap
proac
h
i
s
capa
ble
of
perform
ing
no
n
-
li
nea
r
cl
assifi
cat
ion
operati
on
that
is
al
s
o
te
rm
ed
as
kernel
-
base
d
op
e
rati
on.
T
he
i
m
plica
ti
on
of
the
SV
M
a
ppr
oach
of
the
m
achine
le
arn
in
g
perform
s
app
ly
ing
a
hyperplan
e
ov
er
a
hi
gh
dim
ension
al
sp
ace
that
is
no
rm
al
ly
util
iz
ed
f
or
pe
r
form
ing
re
gr
es
sion
or
cl
assifi
cat
ion
operati
on.
The
co
nvent
ion
al
u
sa
ge
of
SV
M
is
re
port
edly
fou
nd
to
be
use
d
in
cl
assifi
cat
ion
of
sign
al
,
cat
egorizat
ion
of
hyper
te
xt
or
te
xt,
recogn
iz
at
ion
of
ha
ndw
ritt
en
char
act
e
rs.
The
pri
m
e
reaso
ns
f
or
wide
ra
ng
e
of
a
vaila
bili
ty
are
that
-
It
is
the
m
os
t
su
it
ab
le
for
t
he
a
pp
li
cat
ion
wh
e
r
e
the
in
put
data
is
ver
y
vag
ue
or
do
es
not
hav
e
m
uch
cl
arit
y.
Its
app
li
cabil
it
y
is
qu
it
e
hig
h,
especia
ll
y
in
case i
f
the
dat
a is sem
i
-
sru
ct
ured
or
unstruct
ur
e
d, w
hich
is
the m
ajo
rity
of
the cases
in p
r
esent ti
m
es.
2.3
.
E
xecut
i
on fl
ow
The
al
gorithm
of
the
pro
ose
d
syst
em
pe
rfor
m
s
op
ti
m
i
zat
ion
of
the
sensing
perf
or
m
ance
of
the
D
oppler
ra
dar
t
ow
a
r
ds
tr
ackin
g
dow
n
t
he
dynam
ic
even
ts.
Diff
e
re
nt
recor
d
of
a
n
e
ven
t
is
c
onside
red
a
s
an
input
for
th
e
pr
op
os
e
d
syst
e
m
that
is
par
ti
cular
to
the
a
ct
ion
s
of
hum
a
n.
Th
e
stud
y
offe
rs
an
outc
om
e
as
featur
e
t
hat
is
extracte
d
from
the
eve
nts.
T
h
e
a
l
g
o
r
i
t
h
m
c
a
r
r
y
o
u
t
n
o
n
-
r
e
c
u
r
s
i
v
e
o
p
e
r
a
t
i
o
n
f
o
r
e
x
t
r
a
c
t
i
o
n
f
e
a
t
u
r
e
s
f
r
o
m
t
h
e
g
i
v
e
n
s
c
e
n
a
t
i
o
o
f
a
n
e
v
e
n
t
a
n
d
e
x
t
r
a
c
t
t
h
e
f
i
n
a
l
f
e
a
t
u
r
e
.
T
h
e
s
t
e
p
s
o
f
t
h
e
a
l
g
o
r
i
t
h
m
a
r
e
a
s
f
o
l
l
o
w
s
:
Algorithm for Extracting Feature
Input:
d
(data)
Output: d
feat
(extracted feature)
Start
1. init d
2. Extract R
op
=[d
test
d
train
d
feat
];
3. val
max
= arg
max
(R
op
-
arg
min
(d1))
4. For act
id
=1:max(y)
5. For s
id
=1:max(sub
id
)
6. x
norm
x
q
/m
at
(v
al
max
)
7. d
feat
=(d
feat
, x
norm
)
8. End
9. End
End
The
el
ab
or
at
io
ns
of
the
esse
ntial
al
go
rithm
ic
ste
ps
are:
The
first
li
ne
of
act
io
n
of
the
pro
posed
al
gorithm
is
to
con
si
der
a
n
ev
ent
data
d
as
it
s
input.
Using
this
input
valu
e,
the
sys
tem
ob
ta
ins
va
rio
us
form
s
of
featur
e
s
in
the
f
or
m
of
dif
f
eren
t
disc
rete
act
ion
that
are
store
d
in
R
op
m
at
rix.
This
m
at
rix
has
i
nclu
sion
of
d
train
trai
ned
da
ta
,
d
test
te
st
data,
d
f
eat
data
with
featu
re.
T
his
op
e
rati
on
is
f
ollow
e
d
by
ob
t
ai
nin
g
val
m
ax
hig
hest
po
s
sible
value
ob
ta
ine
d
f
ro
m
the
sense
d
dat
a
of
an
eve
nt
wh
ic
h
is
su
pport
ed
by
dual
at
tribu
te
s.
T
he
pri
m
ary
at
tribu
te
is
R
op
wh
il
e
the
seconda
ry
at
tribu
te
will
be
d
1
th
at
are
essenti
ally
a
m
at
rix
to
store
trai
ne
d
da
ta
an
d
te
st
data.
The
possibil
it
y
of
the
highest
val
ue
from
the
d
1
m
at
rix
is
val
m
ax
whil
e
the
c
on
se
cutive
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
4093
-
4100
4098
i
m
ple
m
entat
io
n
of
the
al
gor
it
h
m
ta
kes
thi
s
hi
gh
est
val
ue
with
res
pect
to
y
par
am
et
er
that
is
esse
ntial
ly
a
m
a
trix
with
al
l
trai
nin
g
an
d
te
st
data.
T
he
stu
dy
ta
kes
t
he
ide
ntit
y
of
the
s
ubj
ect
i
n
orde
r
to
f
ram
eup
x
norm
norm
al
iz
ed
m
at
rix
wh
ic
h
t
w
o
s
upporti
ng
a
tt
ribu
te
s
i.e
.
x
q
an
d
m
at
(v
al
m
a
x
).
The
m
a
trix
upon
subj
e
ct
ed
to
resizi
ng
operat
ion
is
sto
red
ba
ck
in
x
q
m
at
ri
x
that
has
poss
ession
of
al
l
th
e
resp
ect
ive
d
1
data
associat
e
d
with
the
s
pecific
e
ve
nt
of
act
io
n
a
s
well
as
it
s
li
nked
ide
ntit
y.
A
ll
cum
ulati
ve
featur
es
a
re
c
om
pu
te
d
with
a
n
ai
d
of
the
ob
ta
ine
d
x
nor
m
vector
.
Th
e
stud
y
al
so
con
st
ru
ct
s
d
f
eat
ano
t
her
m
at
rix
that
stores
va
lues
of
cum
ulati
ve
act
ion
s
li
nke
d
with
al
l
the
f
eat
ur
e
i
n
or
de
r
to
re
pr
ese
nt
a
par
ti
c
ular
ca
se
of
a
n
oc
cu
r
red
e
ve
nt.
Ma
chine
le
arn
in
g
ap
pro
ach
is
app
li
ed
in
nex
t
ste
p
wh
ic
h
be
gin
s
with
sel
ect
ing
trai
ning
par
am
et
ers
e.g
.
nu
m
ber
of
la
ye
rs,
decay
at
tribu
te
weig
ht,
sp
a
rsity
weigh
t,
et
c.
A
n
un
s
up
e
r
vised
le
arn
in
g
al
gorithm
is
design
ed
f
or
reducin
g
t
he
dim
ension
of
the
data
in
pro
pos
ed
m
achine
le
arn
i
ng
ap
proac
h.
Th
e
stu
dy
perform
s
op
ti
m
iz
a
ti
on
op
e
rati
on
us
i
ng
Broy
den
-
Fle
tc
her
-
G
oldfar
b
-
Shan
no
m
e
tho
d
in
orde
r
to
hav
e
a
sig
nific
ant
con
t
ro
l
ov
er
cost
.
The
stu
dy
ap
plies
feedf
orw
ard
neural
net
work
i
n
orde
r
to
perform
tr
ai
nin
g
rou
nds
us
in
g
uns
up
e
rv
ise
d
m
et
ho
dolo
gy.
The
cl
assifi
cat
ion
is
ca
rr
ie
d
ou
t
ne
xt
on
de
pendin
g
upon
the
at
trb
utes
th
at
are
obta
ine
d
from
the
le
arn
in
g
m
et
hod.
T
he
ass
essm
ent
of
the
stud
y
was
car
ried
out
us
i
ng
oth
e
r
relat
ed
tr
ai
nin
g
m
echani
s
m
as
exh
i
bited i
n
Fi
gure
2.
E
x
t
r
a
c
t
e
d
F
e
a
t
u
r
e
D
e
e
p
N
e
u
r
a
l
N
e
t
w
o
r
k
K
N
N
A
l
g
o
r
i
t
h
m
S
V
M
A
l
g
o
r
i
t
h
m
A
c
c
u
r
a
c
y
E
f
f
i
c
i
e
n
c
y
Figure
2
.
Feat
ure e
xtracti
on a
nd traini
ng proc
ess
3.
RESU
LT
A
N
ALYSIS
The
outc
om
es
of
the
pr
opos
e
d
m
od
el
are
a
naly
zed
by
sta
nd
a
r
d
dataset
[
33
]
that
exh
i
bi
ts
diff
ere
nt
act
ivit
ie
s
capt
ur
e
d
f
or
resea
r
ch
w
hich
incl
ud
e
s
10,29
9
instances
of
da
ta
set
s
and
561
at
tribu
te
s.
T
he
dat
a
dataset
is
cha
r
act
erized
with
m
ul
ti
-
var
ia
te
a
nd
tim
e
series
analy
sis.
T
he
pe
roform
ance
analy
sis
of
the
s
yst
e
m
is
done
with
c
om
par
isi
on
of
rece
nt
m
achi
ne
le
ar
ning
ap
proac
hes
co
ns
i
der
i
ng
t
he
sca
le
of
er
r
or
rate
an
d
eff
ic
ie
ncy.
Th
e
f
ollow
i
ng
Fi
gures
3
an
d
4
giv
es
the
c
omparati
ve
analy
s
is
of
er
r
or
rate
an
d
e
ff
ic
ie
nc
y
wit
h
diff
e
re
nt tech
niq
ue
s.
Figure
3 St
udy o
f
E
rror Ra
te
Figure
4.
Com
par
at
ive
an
al
ys
is of ef
fici
ency
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
An
alyti
cal fra
mework
for
opt
imize
d
fe
atu
re
ext
ra
ct
ion f
or…
(
Preet
hi K.
Ma
ne
)
4099
T
he
s
t
ud
y
o
ut
c
om
e
s
ho
w
s
t
ha
t
pr
o
po
s
e
d
s
y
s
t
em
of
f
e
r
s
be
t
t
e
r
e
f
f
i
c
i
e
nc
y
w
i
t
h
S
V
M
a
n
d
ne
xt
be
t
t
e
r
pe
r
f
or
m
a
nc
e
obs
e
r
ve
d
i
s
f
or
K
N
N
w
hi
l
e
t
he
e
f
f
i
c
i
e
nc
y
of
D
N
N
i
s
a
f
t
e
r
t
ha
t
.
T
he
pr
im
e
r
e
a
s
on
be
hi
n
d
t
hi
s
ou
t
c
om
e
i
s
-
S
V
M
of
f
e
r
s
be
t
t
e
r
f
e
a
t
ur
e
f
or
L
2
r
e
gu
l
a
r
i
z
a
t
i
on
t
ha
t
s
uc
c
e
s
s
f
ul
l
y
r
e
s
i
s
t
s
f
r
om
va
r
i
ou
s
ov
e
r
-
f
i
t
t
i
ng
pr
o
bl
em
s
a
nd
he
nc
e
a
c
c
u
r
a
c
y
f
a
c
t
or
o
f
S
V
M
i
s
f
ou
nd
t
o
be
hi
gh
e
r
t
ha
n
e
xi
s
t
i
ng
K
N
N
a
s
w
e
l
l
a
s
D
N
N
a
s
s
ho
w
n
i
n
F
i
gu
r
e
4
.
A
t
t
he
s
am
e
t
im
e
,
S
V
M
al
s
o
of
f
e
r
s
be
t
t
e
r
s
t
a
bi
li
ty
pe
r
f
or
m
a
nc
e
e
ve
n
c
om
pa
r
e
d
t
o
K
N
N
a
nd
D
N
N
r
e
s
ul
t
i
ng
i
n
be
t
t
e
r
e
f
f
i
ci
e
nc
y
e
ve
n
i
n
in
c
r
e
a
s
i
ng
t
r
a
i
ni
ng
r
a
t
i
o.
F
i
gu
r
e
4
e
xh
i
bi
t
s
t
ha
t
S
V
M
ou
t
pe
r
f
or
m
s
K
N
N
a
nd
D
N
N
w
i
t
h
t
he
i
nc
r
e
a
s
e
d
t
r
a
i
ni
ng
r
a
t
i
o
w
he
n
bo
t
h
e
f
f
i
c
i
e
nc
y
(
%
)
a
nd
e
r
r
or
r
a
t
e
(
%
)
a
r
e
c
o
nc
e
r
n
e
d.
T
he
i
nv
e
s
t
i
ga
t
i
on
r
e
s
ul
t
s
ho
w
s
t
ha
t
p
r
o
po
s
e
d
f
r
a
m
e
w
or
k
o
f
f
e
r
s
be
t
t
e
r
e
f
f
e
c
t
i
ve
ne
s
s
w
i
t
h
S
V
M
.
T
he
pe
r
f
or
m
a
nc
e
a
na
l
y
si
s
i
s
obs
e
r
ve
d
f
o
r
K
N
N
a
n
d
D
N
N
e
f
f
i
c
i
e
nc
y
la
t
e
r
i
s
a
f
t
e
r
t
ha
t
.
I
t
i
s
ob
s
e
r
ve
d
t
ha
t
S
V
M
of
f
e
r
s
be
t
t
e
r
f
e
a
t
ur
e
s
f
or
L
2
r
e
gu
l
a
r
i
z
a
t
i
on
t
ha
t
e
f
f
e
c
t
i
ve
ly
op
po
s
e
s
f
r
o
m
di
f
f
e
r
e
nt
ov
e
r
-
f
i
t
t
i
ng
i
s
s
ue
s
a
nd
he
nc
e
f
o
r
t
h
e
xa
c
t
ne
s
s
f
a
c
t
or
of
S
V
M
i
s
s
e
e
n
a
s
hi
gh
e
r
t
ha
n
e
xi
s
t
i
ng
K
N
N
j
us
t
a
s
D
N
N
s
e
e
F
i
g
ur
e
4
.
S
im
ul
t
a
ne
ou
s
l
y
,
S
V
M
l
i
ke
w
i
s
e
of
f
e
r
s
be
t
t
e
r
s
t
a
bi
l
it
y
w
hi
c
h
i
s
a
na
l
y
z
ed
w
i
t
h
K
N
N
a
nd
D
N
N
l
e
a
di
ng
t
o
s
i
gn
i
f
i
c
a
nt
e
f
f
i
c
i
e
nc
y
a
s
w
e
l
l
a
s
im
pr
ov
e
d
t
r
a
i
ni
ng
r
a
t
i
on
.
F
i
gu
r
e
4
di
s
pl
a
y
s
t
h
a
t
S
V
M
be
a
t
s
K
N
N
a
n
d
D
N
N
w
i
t
h
t
he
e
xp
a
n
de
d
pr
e
pa
r
i
n
g
pr
o
po
r
t
i
on
w
he
n
b
ot
h
e
r
r
or
r
a
t
e
(
%
)
a
n
d
e
f
f
e
c
t
i
ve
ne
s
s
(
%
)
a
r
e
c
on
c
e
r
ne
d.
4.
CONCL
US
I
O
N
As
the
a
co
oupa
ncy
detect
io
n
is
no
t
li
m
i
te
d
on
ly
to
t
he
fi
nd
the
num
ber
of
acc
oupa
nts,
wh
e
re
m
os
t
popula
r
m
od
e
ls
of
c
om
pu
ta
ti
on
li
ke
Dopp
le
r
m
et
ho
ds
are
us
ed
as
an
in
vasi
ve
ap
proach,
w
her
eas
the
ide
ntific
at
ion
of
the
oc
cup
a
nts’
act
io
ns
play
s
an
i
m
po
ryant
r
ole
to
bu
il
d
s
om
e
con
te
xt
ori
ented
app
li
cat
io
ns
.
T
he
dataset
incl
ud
e
s
the
se
nso
r
at
trib
utes
we
ables
of
the
de
vices
of
the
use
rs
as
a
n
non
-
i
nv
a
sive
appr
oach
t
o
br
ing
int
rinsic
in
te
rp
re
seat
ion
s
as
oppose
t
o
the
bi
o
-
si
gnal
s
ta
ken
from
the
inv
asi
ve
ses
nors
a
nd
app
ly
in
g
com
pu
ta
ion
al
y
com
plex
al
gorithm
.
The
pr
opos
e
d
le
arn
i
ng
m
od
el
s
ta
ke
the
s
ta
ti
sti
ca
l
cor
al
ti
on
as
a
featur
e
vect
or
in
or
der
to
desig
n
the
analy
ti
cal
fr
am
ewo
r
k.
T
he
fr
am
ewo
r
k
inclu
des
eval
ua
ti
on
of
the
di
ff
e
ren
t
l
earn
i
ng
m
od
le
s
to
validat
e
it
s
beh
a
vi
or
i
n
te
rm
s
of
the
accu
racy,
ti
m
e
com
plexiti
es
an
d
com
pu
ta
ti
on
al
com
plexiti
es.
The
ap
proac
h
of
the
fra
m
ewo
r
k
desi
gn
pro
vid
es
a
sim
ple
and
highly
non
-
rec
ru
si
ve
appr
oach
of
t
he
feat
ur
e
de
scripto
rs
a
nd
thei
r
co
relat
uo
n
as
com
par
e
d
to
the
t
rad
i
ti
on
al
appr
oach
es
of
the
occ
up
a
nt
c
la
ssific
at
ion
.
T
he
m
od
el
i
m
pr
ov
ise
d
a
la
ye
rs
approac
h
of
c
onvoluti
onal
ne
ur
al
netw
ork
(L
A
-
CNN
)
by
m
eans
of
optim
iz
at
i
on
a
nd
it
s
be
nc
hm
ark
ing
with
r
epects
to
t
he
SV
M
as
w
el
l
as
K
N
N
ou
t
p
er
form
s it
s
accu
racy.
R
E
F
E
R
E
N
C
E
S
[1]
Yogana
ndini,
A.
P.,
and
G.
S.
A
nit
ha
,
"A
cost
ef
fec
t
ive
computational
d
esign
of
m
axi
m
um
powe
r
point
tr
ac
k
ing
for
photo
-
voltai
c
cell,
"
Int
ernat
ional
Journal
o
f
El
e
ct
ri
cal
an
d
Computer
Enginee
ring
(
IJE
CE
)
,
vol.
9,
no.
2
,
pp.
851
-
860
,
20
19.
[2]
Angadi,
Parv
at
h
i,
M.
Nag
endr
a
,
and
M.
Hanum
a
ntha
ppa
,
"A
novel
fra
m
ework
fo
r
eff
icien
t
ide
n
tification
of
br
a
in
ca
nc
er
reg
ion
fr
om
bra
in
MRI,
"
Inte
rnational
J
ournal
of
El
e
ct
rical
and
Computer
Engi
ne
ering
(
IJE
CE)
,
vol.
9,
no.
2
,
pp
.
1410
-
1417
,
2019
.
[3]
Dipoal
a
,
W
.
S,
"M
ic
rowave
i
ntrusion
det
e
ct
o
r
with
thre
shol
d
adj
ustm
ent
i
n
response
to
per
iodic
signal
s
,
"
U.S.
Pat
ent
5
,
58
1,
237,
no.
3,
199
6.
[4]
W
eng,
T
,
and
Y.
Agarwal
,
"F
r
om
buil
dings
to
sm
art
buil
d
ings
-
sen
sing
and
actua
t
i
o
n
to
improve
energ
y
e
fficie
n
c
y
,
"
IEE
E
Design
&
Test
of
Comput
e
rs
,
vol. 29, no. 4, pp. 36
-
44,
2012
.
[5]
Yava
ri,
E.,
C.
Song,
V.
Lubeck
e
and
O.
Boric
-
Lube
ck
e,
"Is
there
an
y
bod
y
in
th
ere?
:
Inte
ll
ig
ent
rad
ar
oc
cupa
n
c
y
sensors
,
"
IEE
E
Mic
rowave
Mag
azine
,
vo
l. 15, n
o.
2
,
pp
.
57
-
64
,
2014.
[6]
Y
a
v
a
r
i
,
E
.
,
A
.
L
e
e
,
K
.
P
a
n
g
,
N
.
A
n
d
r
e
w
M
c
C
a
b
e
a
n
d
O
.
B
o
r
i
c
-
L
u
b
e
c
k
e
,
"
R
a
d
a
r
a
n
d
c
o
n
v
e
n
t
i
o
n
a
l
o
c
c
u
p
a
n
c
y
s
e
n
s
o
r
s
p
e
r
f
o
r
m
a
n
c
e
c
o
m
p
a
r
i
s
o
n
,
"
A
s
i
a
-
P
a
c
i
f
i
c
M
i
c
r
o
w
a
v
e
C
o
n
f
e
r
e
n
c
e
,
p
p
.
4
4
4
-
4
4
6
,
2
0
1
4
.
[7]
S
ark
ar,
Abhi
ji
t
,
Mark
Fairc
h
il
d
a
nd
Carl
Salva
gg
i
o,
"Inte
g
rated
d
a
y
l
ight
h
arv
est
ing
and
o
cc
up
ancy
det
e
ct
ion
using
digi
tal
imaging,
"
Sensors
,
Camer
as,
and
S
yste
ms
for I
ndustrial/
S
c
ie
ntific A
pp
li
ca
tions
IX,
vol. 681
6,
2008
.
[8]
Yava
ri,
E.,
C
.
Song,
V.
Lub
ec
k
e,
and
O.
Bori
c
-
Lube
ck
e.
"Is
th
e
re
an
y
bod
y
in
th
ere?
:
Int
e
lligent
rad
ar
o
cc
upan
c
y
sensors
,
"
IEE
E
Mic
rowave
M
ag
azine,
vo
l. 15, n
o.
2
,
pp
.
57
-
64
,
2014.
[9]
Santra
,
A.
,
R.
V
aga
rap
p
an
Ulag
a
nat
han
,
and
T
.
Finke,
"S
hort
-
ran
ge
m
il
li
m
et
ri
c
-
w
ave
rad
ar
s
y
st
e
m
fo
r
occ
upancy
sensing
applicati
on,
"
IE
EE
s
ensors
le
tters
,
vol
.
2
,
no.
3
,
pp
.
1
-
4
,
2
018.
[10]
M
a
n
e
,
P
r
e
e
t
h
i
K
.
,
K
.
N
a
r
a
s
i
m
h
a
R
a
o
,
"
G
r
a
n
u
l
a
r
m
o
b
i
l
i
t
y
-
f
a
c
t
o
r
a
n
a
l
y
s
i
s
f
r
a
m
e
w
o
r
k
f
o
r
e
n
r
i
c
h
i
n
g
o
c
c
u
p
a
n
c
y
s
e
n
s
i
n
g
w
i
t
h
d
o
p
p
l
e
r
r
a
d
a
r
,
"
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
E
l
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
(
IJE
CE)
,
v
o
l
.
8
,
n
o
.
2
,
p
p
.
9
7
9
-
988
,
2
0
1
8
.
[11]
Am
ay
r
i,
Man
ar,
Stepha
ne
Ploix
,
Hus
sain
Kaz
m
i,
Quoc
-
Dung
Ngo
and
E.
L.
Saf
a
di,
"Estimat
ing
occ
upan
c
y
from
m
ea
surem
ent
s
a
nd
knowledge
u
sing
the
ba
y
esi
a
n
net
work
fo
r
en
erg
y
m
ana
g
ement
,
"
Journal
of
S
ensors
,
vol.
201
9,
pp.
1
-
12
,
2019
.
[12]
El
ta
b
ie,
Om
ar
M.,
Moham
ed,
F.
Abdelka
der
,
and
A.
M.
Ghuniem,
"Incorporat
ing
Prim
ar
y
oc
c
upancy
patter
ns
i
n
compress
ive
spe
ct
rum
sensing
,
"
IEE
E
Acce
ss
,
vo
l.
7
,
pp
.
29096
-
2
9106,
2019
.
[13]
Hlophe,
Mduduzi
C
and
Sunil
BT
Maha
raj,
"S
pec
trum
occ
upa
nc
y
re
constructi
on
in
distri
buted
cogni
ti
ve
rad
i
o
net
works
using
dee
p
learni
ng,
"
I
EE
E
Acce
ss
,
vol
.
7
,
pp
.
14294
-
1
4307,
2019
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
4093
-
4100
4100
[14]
Yang,
Jianf
ei
,
Ha
n
Zou,
Hao
Jia
ng,
and
Li
hua
X
ie
,
"D
ev
ic
e
-
fre
e
occ
upan
t
a
ct
iv
ity
sensing
using
wifi
-
ena
b
le
d
io
t
devi
c
es
for
sm
ar
t
hom
es,
"
IE
EE
I
nte
rnet
of
Thing
s J
ournal,
vol. 5, no. 5, pp. 3991
-
4002,
2018
.
[15]
W
u,
Li
bo,
Ya
W
ang,
and
Hai
li
L
iu,
"O
cc
up
ancy
det
e
ct
ion
and
lo
ca
l
iz
a
ti
on
b
y
m
o
nit
oring
non
li
ne
ar
en
erg
y
flow
o
f
a
sh
uttere
d
passi
ve
infr
are
d
sensor,"
IE
EE Se
nso
rs
Journal,
vol
.
18,
no
.
21
,
pp
.
8
656
-
8666,
2018
.
[16]
Cao,
Ning
y
u
an,
Jus
ti
n
T.
,
Shre
y
a
s
Sen,
and
Ariji
t
R
.
,
"S
m
art
sensi
ng
for
HVAC
co
ntrol
:
Coll
abor
ative
int
e
ll
i
g
ence
in
optica
l and
IR
ca
m
er
as,
"
IE
EE Tr
ansac
ti
ons on
Industrial Elect
ronics,
vol
.
65
,
n
o.
12
,
pp
.
9785
-
9794,
2018
.
[17]
Santra
,
A.,
R
.
V
.
Ulag
ana
th
an,
a
nd
T.
Fink
e,
"S
hort
-
ran
ge
m
il
lim
et
ric
-
wave
rad
ar
s
y
st
em
fo
r
oc
cupa
nc
y
s
ensing
appl
i
ca
t
ion,
"
I
E
EE
sensor
s lett
e
rs
,
vol. 2, no. 3,
pp.
1
-
4
,
2018
.
[18]
Luppe
,
Christop
her
and
Am
ir
Shabani
,
"Towa
rds
rel
i
abl
e
in
telli
gen
t
o
cc
up
an
c
y
de
tecti
on
for
sm
a
rt
buil
d
ing
appl
i
ca
t
ions,"
IE
EE
30
th
Canadi
an
Confe
ren
ce o
n
Elec
tri
cal an
d
Computer
Engi
n
ee
ring,
pp.
1
-
4,
2017.
[19]
Nesa,
Nas
hre
en,
and
Indra
ji
t
Ban
erj
e
e,
"IoT
-
base
d
sensor
data
fus
i
on
for
o
cc
up
ancy
s
ensing
using
d
emps
te
r
–
Shafe
r
evi
d
en
ce t
heor
y
for
sm
art
bui
ldi
n
gs,
"
IEEE
Int
ern
et
o
f
Things
Journal,
vo
l. 4, no. 5, pp. 1563
-
1570,
2017.
[20]
Jin,
Ming,
Ruo
xi
Jia
and
Cost
as
J.
Spanos,
"
Virtua
l
o
cc
up
an
c
y
sensing:
Us
i
ng
sm
art
m
et
ers
to
ind
ic
a
te
y
ou
r
pre
senc
e
,
"
I
EE
E
T
rans
act
ions o
n
Mobil
e
Computi
ng,
vol
.
16
,
no
.
1
1,
p
p
.
3264
-
327
7,
2017
.
[21]
Shen,
W
ei
m
ing
and
Gu
y
News
ham,
"S
m
art
ph
one
base
d
occ
u
panc
y
detec
ti
on
in
offic
e
bui
ldings
,
"
IEE
E
20t
h
Inte
rnational
Co
nfe
renc
e
on
Co
mputer
Suppo
rte
d
Cooperative
W
ork
in
Design
,
p
p.
632
-
636
,
201
6.
[22]
Li
u,
Pengch
eng,
Sing
-
Kiong
N
guang
and
As
hton
Partri
dge
,
"O
cc
upancy
inferenc
e
using
p
y
r
oel
e
ct
ri
c
infra
r
e
d
senso
rs t
hrough hidden
Markov
m
odel
,
"
IE
E
E
Se
nsor
s J
ournal,
v
ol.
16
,
no
.
4
,
pp
.
1062
-
1068,
201
5.
[23]
Jin,
M,
N.
Bekiaris
-
Li
b
eri
s,
K.
W
ee
kl
y
,
C.
J.
Spanos
and
A.
M
.
Ba
y
en,
"O
cc
up
ancy
dete
ction
via
envi
ronm
enta
l
sensing,
"
IE
EE
Tr
ansacti
ons on Aut
omation
Scie
nce
and
Eng
ine
e
ring,
vol
.
15
,
no
.
2,
pp.
443
-
455,
2016.
[24]
Khan,
M
.
D
Abdul
la
h
Al
Hafi
z,
HM
Saj
ja
d
Hos
sain
and
Ni
rm
aly
a
Ro
y
,
"S
ense
pre
senc
e:
I
nfra
struct
ur
e
-
l
ess
occ
upan
c
y
detec
ti
on
for
opp
ortu
nisti
c
sensing
ap
pli
c
at
ions,
"
16th
IEE
E
In
te
rnatio
nal
Confe
ren
ce
on
Mobil
e
Dat
a
Manage
ment,
vo
l.
2
,
pp
.
56
-
61
,
2
015.
[25]
Li,
B
in,
S.
Li,
A.
Nallana
th
an,
Y.
Nan,
C
.
Zh
ao
a
nd
Z.
Zhou
,
"D
e
ep
sensing
for
n
ext
-
gen
era
t
ion
d
y
nami
c
spec
tru
m
sharing:
More
t
han
det
e
ct
ing
th
e
occ
u
pan
c
y
state
of
primar
y
sp
ec
trum,
"
IEEE
Tr
an
sacti
ons
on
Comm
unic
ati
on
s,
vol.
63
,
no
.
7
,
pp
.
2442
-
2457
,
20
15.
[26]
Ma
ria
ni
,
Andrea
,
S.
K
ande
ep
an,
and
A.
Giorge
t
ti
,
"P
eri
odi
c
spe
ct
rum
sensing
with
non
-
contin
uo
us
primar
y
user
tra
nsm
issions
,
"
IEE
E
Tr
ansact
io
ns on
Wireless
C
omm
unic
ati
ons,
vol.
14
,
no
.
3
,
pp
.
1636
-
1649
,
20
14.
[27]
A
l
b
e
s
a
,
J
.
a
n
d
M
.
G
a
s
u
l
l
a
,
"
M
o
n
i
t
o
r
i
n
g
s
w
i
t
c
h
-
t
y
p
e
s
e
n
s
o
r
s
v
i
a
i
n
d
u
c
t
i
v
e
c
o
u
p
l
i
n
g
:
A
p
p
l
i
c
a
t
i
o
n
t
o
o
c
c
u
p
a
n
c
y
a
n
d
b
e
l
t
d
e
t
e
c
t
i
o
n
i
n
r
e
m
o
v
a
b
l
e
v
e
h
i
c
l
e
s
e
a
t
s
,
"
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
P
o
w
e
r
E
l
e
c
t
r
o
n
i
c
s
,
v
o
l
.
2
7
,
n
o
.
1
1
p
p
.
4
4
7
2
-
4
4
8
0
,
2
0
1
2
.
[28]
Cooley
,
John J.,
Al
-
Tha
ddeus
Av
estruz
and
St
even B.
Leeb,
"A
re
trofi
t
ca
p
aciti
v
e s
ensing
occ
up
an
c
y
d
e
te
c
tor
usin
g
fluore
sce
n
t
l
amps
,
"
IEEE
Tr
ansac
ti
ons on
Industrial Elect
ronics
,
vol.
59
,
no
.
4
,
pp
.
1898
-
1911
,
20
11.
[29]
Hos
sain,
K.
an
d
B.
Champag
ne,
"W
ide
b
and
spec
trum
sensing
for
cognitive
rad
ios
wi
th
co
rre
lated
subban
d
occ
upan
c
y
,
"
IEEE
Signa
l
Proc
es
sing Le
tters
,
vol
.
18,
no.
1,
pp.
35
-
38,
2010
.
[30]
G
e
o
r
g
e
,
B
,
H
.
Z
a
n
g
l
,
T
.
B
r
e
t
t
e
r
k
l
i
e
b
e
r
a
n
d
G
.
B
r
a
s
s
e
u
r
,
"
A
c
o
m
b
i
n
e
d
i
n
d
u
c
t
i
v
e
–
c
a
p
a
c
i
t
i
v
e
p
r
o
x
i
m
i
t
y
s
e
n
s
o
r
f
o
r
s
e
a
t
o
c
c
u
p
a
n
c
y
d
e
t
e
c
t
i
o
n
,
"
I
E
E
E
t
r
a
n
s
a
c
t
i
o
n
s
o
n
i
n
s
t
r
u
m
e
n
t
a
t
i
o
n
a
n
d
m
e
a
s
u
r
e
m
e
n
t
,
v
o
l
.
5
9
,
n
o
.
5
,
p
p
.
1
4
6
3
-
1
4
7
0
,
2
0
1
0
.
[31]
Nikhat
h,
A.
Ko
usar
and
K.
Subrahm
an
y
am,
"F
ea
ture
se
le
c
ti
o
n,
opti
m
izati
on
and
cl
ust
eri
ng
strategie
s
of
t
ex
t
documents,
"
Int
ernati
onal
Journ
al
of
El
e
c
tric
a
l
and
Computer
Engi
nee
ring
(
IJECE)
,
vol.
9,
no.
2,
pp.
1313
-
132
0
,
2019.
[32]
Mane
,
P
reethi
K
.
,
and
K.
Nar
asi
m
ha
Rao,
"Revi
ew
of
R
ese
arc
h
progre
ss
,
tre
nds
and
gap
in
oc
cu
pancy
sensing
fo
r
sophistic
ated
sen
sor
y
op
era
t
ion,
"
Computer
Scien
ce
On
-
l
ine Conf
ere
nce,
pp
.
212
-
222,
2018
.
[33]
"H
um
an
Acti
vity
R
ec
ogn
it
ion
Us
ing
Sm
art
phone
s Dat
a
Se
t",
htt
ps://
arc
hi
ve.ics
.
uci
.
edu/
m
l
/d
ataset
s/hum
an+
a
ctivity
+r
ec
ogn
it
io
n+using+
sm
art
p
hones,
r
et
ri
eve
d
on
20
-
09
-
2019
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Pr
e
ethi
K
Mane
re
ce
iv
ed
her
BE
degr
e
e
in
In
strum
ent
at
ion
T
ec
hnolog
y
,
in
th
e
y
e
ar
1996
fro
m
Banga
lor
e
univers
ity
and
ME
in
El
ectroni
cs
from
Banga
lore
un
ive
rsit
y
in
the
yea
r
2000.
She
is
pursuing
PhD
in
El
e
ct
roni
cs
an
d
comm
unic
at
io
n
Engi
ne
eri
ng
f
rom
Visvesvara
y
a
Technol
og
i
cal
Univer
sit
y
.
She
has
work
expe
ri
enc
e
of
21
y
e
ars
in
teac
hing
and
is
cur
ren
t
l
y
serv
ing
as
As
socia
te
Profess
or
in
th
e depa
rtment
of El
ec
tron
ic
s
and
Ins
trumenta
t
ion
T
e
chnol
og
y
.
Dr
K.
Narasimha
Rao
has
obta
ine
d
his
B.
Te
ch
,
M.T
ec
h
and
Ph
D
degr
ee
s
in
the
y
e
ar
1983,
1985
and
2011
respe
ct
iv
ely
.
He
h
as
26
y
e
ars
of
t
ea
ch
ing
expe
r
ience
.
Presen
tly
,
he
is
Profess
or
Depa
rtmentt
Of
Instrum
ent
ation
Technol
og
y
,
B
MS
CE,
and
B
an
gal
ore
-
19
.
His
r
ese
arc
h
intere
sts
are
in
the
ar
ea
s
of
ene
rg
y
eff
icient
DC
D
rive
s,
Digit
al
Signal
Proce
ss
ing,
Proce
ss
Control
and
Biom
edi
cal
Sign
al
Proc
essing
.
Evaluation Warning : The document was created with Spire.PDF for Python.