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
-
A
I
)
V
ol
. 13, No.
4
, D
e
c
e
m
be
r
2024, pp.
4121
~
4128
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v13.i4.pp
4121
-
4128
4121
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
E
n
h
a
n
c
i
n
g vi
d
e
o an
o
m
al
y
d
e
t
e
c
t
i
on
f
or
h
u
m
an
su
sp
i
c
i
ou
s
b
e
h
avi
or
t
h
r
ou
gh
d
e
e
p
h
y
b
r
i
d
t
e
m
p
o
r
al
sp
at
i
al
n
e
t
w
or
k
K
u
s
u
m
a S
r
ir
am
1
,2
, K
ir
an
P
u
r
u
s
h
ot
h
am
3
1
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
on
S
c
i
e
nc
e
a
nd
E
ngi
ne
e
r
i
ng
,
M
.
S
.
R
a
m
a
i
a
h I
ns
t
i
t
ut
e
of
T
e
c
hnol
ogy
, B
e
nga
l
ur
u, I
ndi
a
2
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd
E
ngi
ne
e
r
i
ng,
V
i
s
ve
s
va
r
a
ya
T
e
c
hnol
ogi
c
a
l
U
ni
ve
r
s
i
t
y, B
e
nga
l
ur
u, I
ndi
a
3
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd
E
ngi
ne
e
r
i
ng
,
R
N
S
I
ns
t
i
t
ut
e
of
T
e
c
hnol
o
gy
, B
e
nga
l
ur
u, I
ndi
a
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
15
,
2023
R
e
vi
s
e
d
A
pr
2
,
2024
A
c
c
e
pt
e
d
A
pr
17
,
2024
Abnormal
behavior
exhibited
by
individuals
with
pa
rticular
intenti
ons
is
common,
and
when
such
behavior
occurs
in
public
places,
it
can
cause
physical
and
mental
harm
to
others.
Considering
the
rise
in
th
e
aut
omated
approach
for
anomaly
detection
in
videos
,
accuracy
becomes
essential.
Most
existin
g
models
follow
a
deep
learning
architectu
re,
which
faces
chal
lenges
due
to
variations
in
motion.
This
research
work
develops
a
deep
le
arning
-
based
hybrid
architecture
with
temporal
and
spatial
feature
s
.
The
hybrid
temporal spa
tial networ
k (HTSNet)
consists of tw
o customize
d archite
ctures:
a
graph
neural
network
(GNN)
and
a
convolutional
neural
network
(
CNN).
HTSNet
combined
with
a
novel
classifier
to
extract
features
and
c
lassify
normal
and
abnormal
behavior.
The
p
erformance
of
HTSNet
is
rigo
rously
evaluated
using t
he
University
of California,
San Diego
-
Pedestr
ian
1
(
UCSD
Ped1
)
dataset,
a
benchmark
in
computer
vision
research
for
anomaly
de
tection
in
video
surveillance.
The
effectiveness
of
HTSNet
is
demonstrated
th
r
ough
a
comparativ
e
analysis
with
current
stat
e
-
of
-
the
-
art
methods,
using
th
e
area
under
the
curve
(AUC)
metric
as
a
standard
measure
of
performanc
e.
This
paper
contributes
to
the
advanc
ement
of
video
surveillance
techn
ology,
providing
a
robust
framework
f
or
enhancing
public
safety
and
securit
y
in
an
increasin
gly digital wor
ld.
K
e
y
w
o
r
ds
:
A
nom
a
ly
de
te
c
ti
on
C
r
ow
d a
na
ly
s
is
D
e
e
p l
e
a
r
ni
ng
G
r
a
ph ne
ur
a
l
ne
twor
k
H
ybr
id
t
e
m
por
a
l
s
pa
ti
a
l
ne
twor
k
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
:
K
us
um
a
S
r
ir
a
m
D
e
pa
r
tm
e
nt
of
I
nf
or
m
a
ti
on S
c
ie
nc
e
a
nd
E
ngi
ne
e
r
in
g,
M
. S
. R
a
m
a
ia
h I
ns
ti
tu
te
of
T
e
c
hnol
ogy
B
e
nga
lu
r
u, I
ndi
a
E
m
a
il
:
kus
um
a
s
_12@
r
e
di
f
f
m
a
il
.c
om
1.
I
N
T
R
O
D
U
C
T
I
O
N
T
he
w
id
e
s
pr
e
a
d
a
dopt
io
n
of
vi
de
o
c
a
pt
ur
e
e
qui
pm
e
nt
h
a
s
l
e
d
to
th
e
ge
ne
r
a
ti
on
of
a
s
ig
ni
f
ic
a
nt
vol
um
e
of
vi
de
o
da
ta
. T
he
s
e
c
ur
it
y
of
th
e
r
e
s
id
e
nt
s
a
nd
th
e
ir
pos
s
e
s
s
io
ns
is
c
ont
in
ge
nt
upon
th
e
in
te
gr
it
y
of
th
is
da
ta
.
T
he
id
e
nt
if
ic
a
ti
on
of
a
ty
pi
c
a
l
c
r
ow
d
be
ha
vi
or
is
of
ut
m
os
t
im
por
ta
nc
e
,
pa
r
ti
c
ul
a
r
ly
in
s
c
e
na
r
io
s
in
vol
vi
ng
la
r
g
e
c
ongr
e
ga
ti
ons
of
pe
opl
e
.
A
nom
a
ly
de
te
c
ti
on
is
a
c
r
uc
ia
l
c
om
pone
nt
f
or
e
nha
nc
in
g
s
oc
ia
l
s
e
c
ur
it
y
a
nd
s
a
f
e
gua
r
di
ng i
ndi
vi
dua
ls
.
T
he
ut
il
iz
a
ti
on of
vi
de
o s
ur
ve
il
la
nc
e
s
ys
te
m
s
f
or
publi
c
s
a
f
e
t
y i
s
pr
e
va
le
nt
[
1]
.
T
he
s
tu
dy
pr
e
s
e
nt
s
two
di
s
ti
nc
t
a
ppr
oa
c
he
s
,
na
m
e
ly
a
di
r
e
c
t
t
e
c
hni
que
a
nd
a
n
in
di
r
e
c
t
m
e
th
od,
f
or
c
r
ow
d
de
te
c
ti
on.
T
he
di
r
e
c
t
a
ppr
oa
c
h
is
a
de
te
c
ti
on
te
c
hni
que
th
a
t
e
m
pl
oys
e
it
he
r
s
e
gm
e
nt
a
ti
on
or
hum
a
n
de
te
c
ti
on
to
a
c
c
ur
a
te
ly
id
e
nt
if
y
a
nd
di
f
f
e
r
e
nt
ia
te
be
twe
e
n
in
di
vi
dua
ls
w
it
hi
n
a
gi
ve
n
s
c
e
ne
.
T
he
m
a
p
-
ba
s
e
d
in
di
r
e
c
t
a
ppr
oa
c
h
be
gi
ns
by
id
e
nt
if
yi
ng
vi
s
u
a
l
f
e
a
tu
r
e
s
a
nd
s
ubs
e
que
nt
ly
e
m
pl
oys
m
a
ppi
ng
te
c
hni
que
s
to
e
s
ta
bl
is
h
a
s
s
o
c
ia
ti
ons
be
tw
e
e
n
th
e
s
e
a
tt
r
ib
ut
e
s
a
nd
th
e
po
pul
a
ti
on
[
2]
.
T
he
im
pl
e
m
e
nt
a
ti
on
of
th
e
s
e
m
e
th
odol
ogi
e
s
w
il
l
r
e
s
ul
t
in
a
n
e
nha
nc
e
m
e
nt
of
pe
de
s
tr
ia
n
s
a
f
e
ty
a
nd
a
de
e
pe
r
c
om
pr
e
he
n
s
io
n
of
c
r
ow
d
be
ha
vi
or
.
T
he
m
a
c
hi
ne
le
a
r
ni
ng
s
y
s
te
m
ha
s
b
e
e
n
s
pe
c
if
ic
a
ll
y
d
e
s
ig
ne
d
to
e
va
lu
a
te
th
e
nor
m
a
li
ty
or
a
bnor
m
a
li
ty
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
. 13, No. 4, D
e
c
e
m
be
r
2024:
4121
-
4128
4122
of
va
r
io
us
c
r
ow
d s
c
e
na
r
io
s
, s
uc
h a
s
a
nxi
e
ty
, c
la
s
he
s
, a
nd s
ta
m
pe
de
s
.
T
he
l
a
be
l
di
s
tr
ib
ut
io
n
a
ppr
oa
c
h ha
s
be
e
n
e
m
pl
oye
d
by
th
e
a
ut
hor
s
to
a
ddr
e
s
s
th
e
i
s
s
ue
of
m
ix
e
d
be
h
a
vi
or
.
T
he
te
r
m
"
m
ix
e
d
be
ha
vi
or
"
is
us
e
d
to
de
s
c
r
ib
e
th
e
s
im
ul
ta
ne
ous
m
a
ni
f
e
s
ta
ti
on of
m
u
lt
ip
le
be
ha
vi
or
s
.
T
hi
s
phe
nom
e
non r
e
s
ul
ts
i
n a
s
in
gul
a
r
c
onc
e
nt
r
a
ti
on on
a
s
pe
c
if
ic
ta
s
k,
a
c
c
om
pa
ni
e
d
by
a
la
c
k
of
in
te
r
e
s
t
in
e
nga
gi
ng
in
ot
he
r
a
c
ti
vi
ti
e
s
.
T
he
c
onve
r
ge
nc
e
of
be
ha
vi
or
a
l
pa
tt
e
r
ns
c
a
n
b
e
obs
e
r
ve
d
w
he
n
th
e
r
e
is
a
s
im
ul
ta
ne
ous
pr
e
s
e
n
c
e
of
f
e
a
r
f
ul
or
c
onf
us
e
d
b
e
ha
vi
or
a
lo
ngs
id
e
a
ggr
e
s
s
iv
e
be
h
a
vi
or
[
3]
, [
4]
.
T
he
pr
opos
e
d
a
ppr
oa
c
h
in
vol
ve
s
th
e
in
te
gr
a
ti
on
of
m
ot
i
on
in
f
or
m
a
ti
on
im
a
ge
s
(
M
I
I
)
w
it
h
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
(
C
N
N
)
.
T
he
v
a
r
ia
ti
ona
l
a
be
r
r
a
nt
be
ha
vi
or
de
te
c
ti
on
(
V
A
B
D
)
is
a
pr
oba
bi
li
s
ti
c
m
e
th
od.
T
he
obj
e
c
ti
ve
of
th
is
a
ppr
oa
c
h
is
to
pr
om
p
tl
y
id
e
nt
if
y
a
bnor
m
a
l
c
r
ow
d
be
ha
v
io
r
in
vi
de
o
c
li
ps
.
T
he
a
nom
a
ly
a
na
ly
s
i
s
m
e
th
od
d
e
s
c
r
ib
e
d
in
[
5]
is
ba
s
e
d
on
a
n
im
pr
ove
d
ve
r
s
io
n
of
th
e
k
-
m
e
a
n
s
a
lg
or
it
hm
.
A
pr
e
-
tr
a
in
e
d
2D
-
C
N
N
w
a
s
de
ve
lo
p
e
d
f
or
m
ot
io
n
in
put
,
a
lo
ng
w
i
th
a
s
im
pl
if
ie
d
2D
-
C
N
N
. T
he
obj
e
c
ti
ve
w
a
s
to
a
c
hi
e
ve
th
e
hi
ghe
s
t
r
e
c
ogni
ti
on
a
c
c
ur
a
c
y
w
hi
le
m
in
im
iz
in
g
th
e
c
om
put
a
ti
ona
l
r
e
qui
r
e
m
e
nt
s
.
M
e
hm
ood
[
6]
in
c
lu
de
d
a
c
om
pr
e
he
ns
iv
e
e
va
lu
a
ti
on
of
th
e
pr
ogr
e
s
s
m
a
de
i
n
th
e
f
ie
ld
of
c
r
ow
d
a
na
ly
s
is
u
s
in
g
phys
ic
a
l
a
ppr
oa
c
he
s
.
T
he
id
e
nt
if
ie
d
a
ppr
oa
c
he
s
c
a
n
be
c
a
te
gor
iz
e
d
in
to
t
hr
e
e
di
s
ti
nc
t
c
a
te
gor
ie
s
:
c
om
pl
e
x
c
r
ow
d
m
ot
io
n
s
ys
te
m
s
,
f
lu
id
dyna
m
ic
s
,
a
nd
in
te
r
a
c
ti
on
f
or
c
e
s
.
H
u
e
t
al
.
[
7]
f
oc
us
e
d
on
e
va
lu
a
ti
ng
a
f
r
a
m
e
w
or
k
th
a
t
ha
d
li
m
it
e
d
m
oni
to
r
in
g
c
a
pa
bi
li
ti
e
s
f
or
de
te
c
ti
ng
a
nd
id
e
nt
if
yi
ng
a
bnor
m
a
l
be
ha
vi
or
de
te
c
ti
on
a
nd
lo
c
a
li
z
a
ti
on
(
A
B
D
L
)
i
n c
r
ow
de
d e
nvi
r
onm
e
nt
s
.
Z
ha
ng
e
t
al
.
[
8]
in
tr
oduc
e
d
a
va
lu
a
bl
e
e
m
be
ddi
ng
t
e
c
hni
que
kn
ow
n
a
s
b
a
g
-
of
-
e
ve
nt
-
m
ode
ls
(
B
oE
M
)
to
c
ha
r
a
c
te
r
iz
e
vi
de
o
c
li
ps
th
a
t
di
s
pl
a
y
bot
h
nor
m
a
l
a
nd
a
bn
or
m
a
l
be
ha
vi
or
.
T
he
r
e
s
e
a
r
c
he
r
s
e
m
pl
oye
d
a
m
e
th
odol
ogy
to
ge
ne
r
a
te
s
ynt
he
ti
c
a
nom
a
lo
us
e
ve
nt
s
th
a
t
f
a
it
hf
ul
ly
r
e
pl
ic
a
te
s
pe
c
if
ic
in
s
ta
nc
e
s
of
a
nom
a
lo
us
in
c
id
e
nt
s
.
C
ha
ndr
a
ka
la
e
t
al
.
[
9]
pr
opos
e
d
a
n
a
ppr
oa
c
h
f
or
s
c
e
ne
pe
r
c
e
pt
io
n
th
a
t
in
te
gr
a
te
s
pr
in
c
ip
le
s
f
r
om
ps
yc
hol
ogy
th
e
or
y
w
it
h
th
e
r
e
pr
e
s
e
nt
a
ti
on
of
f
lu
id
dyna
m
ic
s
.
T
hi
s
s
tu
dy
in
ve
s
ti
ga
te
s
a
te
c
hni
que
f
or
e
xt
r
a
c
ti
ng
a
c
ti
ons
f
r
om
c
ont
in
uous
unc
ons
tr
a
in
e
d
vi
de
os
.
T
he
pr
opos
e
d
a
ppr
oa
c
h
c
ons
is
ts
of
th
r
e
e
ke
y
c
om
pone
nt
s
:
te
m
por
a
l
a
c
ti
on
r
out
e
s
e
a
r
c
hi
ng,
s
pa
ti
a
l
-
te
m
por
a
l
a
c
ti
on
c
om
pe
ns
a
ti
on,
a
nd
s
p
a
ti
a
l
lo
c
a
ti
on
e
s
ti
m
a
ti
on.
D
e
e
p
le
a
r
ni
ng
ha
s
d
e
m
ons
tr
a
te
d
not
a
bl
e
a
dva
n
c
e
m
e
nt
s
,
s
pe
c
if
ic
a
ll
y
i
n
th
e
dom
a
in
s
of
f
a
c
ia
l
r
e
c
ogni
ti
on
a
nd
ta
r
ge
t
tr
a
c
k
in
g,
a
m
ong
va
r
io
us
ot
he
r
f
ie
ld
s
[
10]
.
C
N
N
s
a
nd
lo
ng
s
h
or
t
-
te
r
m
m
e
m
or
y
ne
twor
ks
(
L
S
T
M
s
)
a
r
e
two
pr
om
in
e
nt
ty
pe
s
of
ne
ur
a
l
ne
twor
k
s
ut
il
iz
e
d
in
th
e
f
ie
ld
of
de
e
p
le
a
r
ni
ng.
C
N
N
e
m
pl
oys
a
c
om
bi
na
ti
on
of
f
or
w
a
r
d
a
nd
r
e
ve
r
s
e
pr
opa
ga
ti
on
a
lg
or
it
hm
s
to
it
e
r
a
ti
ve
ly
m
od
if
y
th
e
th
r
e
s
hol
ds
a
nd
w
e
ig
ht
s
i
n
th
e
t
r
a
in
in
g
pr
oc
e
s
s
.
T
he
a
c
hi
e
ve
m
e
nt
of
th
is
ta
s
k
is
a
c
c
om
pl
is
he
d
by
s
uppl
yi
ng
th
e
m
ode
l
w
it
h
in
put
la
be
ls
a
nd
out
put
ti
ng
vi
de
o i
m
a
ge
s
.
L
i
e
t
al
.
[
1
1]
ut
il
iz
e
d
c
a
s
c
a
de
d
c
la
s
s
if
ie
r
s
to
gr
a
dua
ll
y
di
s
ti
n
gui
s
h
be
twe
e
n
ty
pi
c
a
l
a
nd
a
bnor
m
a
l
pe
de
s
tr
ia
n
be
ha
vi
or
.
T
he
id
e
nt
if
ic
a
ti
on
of
a
bnor
m
a
l
r
e
gi
ons
w
a
s
s
uc
c
e
s
s
f
ul
ly
a
c
hi
e
ve
d
by
e
m
pl
oyi
ng
c
a
s
c
a
de
d
C
N
N
s
a
nd c
a
s
c
a
de
d a
ut
o
e
nc
ode
r
s
.
T
he
r
e
f
or
e
, t
o
e
xt
r
a
c
t
a
s
pe
c
t
s
o
f
pe
de
s
tr
ia
n a
c
ti
vi
ty
, t
he
ut
il
iz
a
ti
on of
opt
ic
a
l
f
lo
w
da
ta
f
r
om
a
n
in
put
im
a
ge
a
nd
dua
l
-
s
tr
e
a
m
C
N
N
s
w
a
s
im
pl
e
m
e
nt
e
d.
T
he
c
ha
ll
e
nge
of
ha
ndl
in
g
in
f
or
m
a
ti
on
tr
a
ns
f
e
r
ove
r
e
xt
e
nde
d
i
nput
s
e
qu
e
nc
e
s
pos
e
s
a
s
ig
ni
f
ic
a
nt
obs
ta
c
le
f
or
c
onve
nt
io
na
l
r
e
c
ur
r
e
nt
n
e
ur
a
l
ne
twor
ks
[
12]
.
T
he
L
S
T
M
ne
twor
k
w
a
s
s
pe
c
if
ic
a
ll
y
de
ve
lo
pe
d
to
a
dd
r
e
s
s
th
is
pa
r
ti
c
ul
a
r
pr
obl
e
m
.
T
he
m
ode
l
f
or
de
te
c
ti
ng
a
bnor
m
a
l
be
h
a
vi
or
w
a
s
de
ve
lo
pe
d
us
in
g
L
S
T
M
,
in
c
or
por
a
ti
ng
ti
m
e
-
dom
a
in
a
nd
ge
ogr
a
phi
c
a
l
da
ta
a
c
qui
r
e
d
th
r
ough
a
ut
oe
nc
ode
r
s
[
13]
.
A
de
e
p
le
a
r
ni
ng
ne
twor
k
w
a
s
c
ons
tr
uc
te
d
by
c
om
bi
ni
ng
a
n
L
S
T
M
a
nd
s
pa
ti
a
l
-
te
m
por
a
l
C
N
N
to
e
f
f
e
c
ti
ve
ly
id
e
nt
if
y
a
nd
c
la
s
s
if
y
pe
de
s
tr
ia
n
be
ha
vi
or
s
.
T
he
ne
twor
k
w
a
s
s
ubs
e
que
nt
ly
e
m
pl
oye
d t
o de
te
c
t
a
bnor
m
a
l
pe
de
s
tr
ia
n be
ha
vi
or
.
D
ir
e
kogl
u
[
14]
pr
e
s
e
nt
s
a
f
r
a
m
e
w
or
k
f
or
a
nom
a
ly
de
te
c
ti
on
ba
s
e
d
on
de
e
p
ne
ur
a
l
ne
twor
ks
.
T
hi
s
f
r
a
m
e
w
or
k
ut
il
iz
e
s
w
e
a
k
s
upe
r
vi
s
io
n
a
nd
is
tr
a
in
e
d
us
in
g
onl
y
vi
de
o
-
le
ve
l
la
be
ls
.
T
he
s
e
lf
-
r
e
a
s
oni
ng
-
ba
s
e
d
tr
a
in
in
g
te
c
hni
que
in
vol
ve
s
ut
il
iz
in
g
bi
na
r
y
c
lu
s
t
e
r
in
g
of
s
pa
ti
o
-
te
m
por
a
l
vi
de
o
da
ta
to
c
on
s
tr
uc
t
ps
e
udo
la
b
e
ls
.
T
hi
s
f
e
a
tu
r
e
a
id
s
in
m
it
ig
a
ti
ng
th
e
noi
s
e
pr
e
s
e
nt
in
th
e
la
be
ls
of
f
il
m
s
th
a
t
e
xhi
bi
t
a
nom
a
lo
us
c
ha
r
a
c
te
r
is
ti
c
s
.
T
o
e
nha
nc
e
th
e
a
c
c
ur
a
c
y
of
a
nom
a
ly
de
te
c
ti
on,
our
pr
opos
e
d
f
o
r
m
u
la
ti
on
a
dvoc
a
te
s
f
or
th
e
in
te
gr
a
ti
on
of
th
e
c
or
e
ne
twor
k a
nd c
lu
s
te
r
in
g, e
na
bl
in
g t
he
m
t
o w
or
k c
ol
la
bor
a
ti
ve
ly
.
T
he
dua
l
-
s
tr
e
a
m
va
r
ia
ti
ona
l
a
ut
o
-
e
nc
ode
r
(
D
S
V
A
E
)
[
15
]
is
a
pr
opos
e
d
m
ode
l
f
or
voi
c
e
a
c
ti
vi
ty
de
te
c
ti
on
(
V
A
D
)
,
c
ons
i
s
ti
ng
of
two
s
ta
c
ke
d
va
r
ia
ti
ona
l
a
ut
o
-
e
n
c
ode
r
s
(
V
A
E
)
m
ode
ls
. T
he
f
ir
s
t
m
ode
l
c
ons
i
s
ts
of
two
s
ha
ll
ow
ge
ne
r
a
ti
ve
m
od
e
ls
:
th
e
f
ul
ly
c
onne
c
t
e
d
va
r
i
a
ti
ona
l
a
ut
o
-
e
nc
ode
r
(
F
C
V
A
E
)
a
nd
th
e
s
ki
p
c
onne
c
te
d
va
r
ia
ti
ona
l
a
ut
o
-
e
nc
od
e
r
s
(
S
C
V
A
E
)
.
T
he
F
C
V
A
E
m
ode
l
a
im
s
to
le
a
r
n
th
e
ove
r
a
ll
f
e
a
tu
r
e
s
of
th
e
m
ode
l
w
hi
le
e
xc
lu
di
ng s
pe
c
if
ic
und
e
s
ir
a
bl
e
a
s
pe
c
ts
. T
he
s
pa
ti
a
l
a
nd
te
m
por
a
l
pr
ope
r
ti
e
s
of
th
e
pi
c
tu
r
e
f
r
a
m
e
s
a
r
e
a
c
c
ur
a
te
ly
e
xt
r
a
c
te
d
by
th
e
S
C
V
A
E
.
T
o
m
in
im
iz
e
in
f
or
m
a
ti
on
lo
s
s
,
th
e
s
ki
p
-
c
onne
c
te
d
va
r
ia
ti
ona
l
a
ut
oe
nc
ode
r
(
S
C
V
A
E
)
e
s
ta
bl
is
he
s
a
c
onne
c
ti
on
be
twe
e
n
th
e
e
nc
ode
r
a
nd
de
c
ode
r
f
e
a
tu
r
e
s
.
T
hi
s
c
onn
e
c
ti
on
he
lp
s
m
a
in
ta
in
t
he
f
lo
w
of
i
nf
or
m
a
ti
on t
hr
oughout t
he
m
ode
l.
T
he
m
ot
iv
a
ti
on
f
o
r
th
is
pa
pe
r
is
gr
ounde
d
in
r
e
li
a
bl
e
vi
de
o
s
ur
ve
il
la
nc
e
s
ys
te
m
s
in
a
n
e
r
a
w
he
r
e
publ
ic
s
a
f
e
ty
a
nd s
e
c
ur
it
y
a
r
e
pa
r
a
m
ount
. W
it
h
th
e
e
xpone
nt
ia
l
in
c
r
e
a
s
e
in
th
e
a
m
ount
of
vi
de
o
da
ta
ge
n
e
r
a
te
d
da
il
y
,
tr
a
di
ti
ona
l
m
a
nua
l
m
oni
to
r
in
g
m
e
th
ods
a
r
e
no
lo
nge
r
f
e
a
s
ib
le
,
ne
c
e
s
s
it
a
ti
ng
th
e
de
ve
lo
pm
e
nt
of
a
ut
om
a
te
d
s
ys
te
m
s
c
a
pa
bl
e
of
e
f
f
e
c
ti
ve
ly
id
e
nt
if
yi
ng
a
nom
a
lo
us
e
ve
nt
s
.
T
he
c
ha
ll
e
nge
li
e
s
in
th
e
in
he
r
e
nt
c
om
pl
e
xi
ty
of
vi
de
o
da
ta
,
w
hi
c
h
in
c
lu
de
s
di
ve
r
s
e
a
nd
of
te
n
s
ubt
le
va
r
ia
ti
ons
in
be
ha
vi
or
a
nd
e
nvi
r
onm
e
nt
.
T
he
a
bi
li
ty
to
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
E
nhanc
in
g v
id
e
o anomaly
de
te
c
ti
on f
or
hum
an
s
us
pi
c
io
us
b
e
ha
v
io
r
t
hr
ough de
e
p h
y
br
id
…
(
K
u
s
um
a Sr
i
r
am
)
4123
a
c
c
ur
a
te
ly
di
s
ti
ngui
s
h
be
twe
e
n
nor
m
a
l
va
r
ia
ti
ons
a
nd
ge
nui
ne
a
nom
a
li
e
s
is
c
r
uc
ia
l
in
a
r
a
nge
of
a
ppl
ic
a
ti
ons
,
f
r
om
ur
ba
n
s
ur
ve
il
la
nc
e
to
tr
a
f
f
ic
c
ont
r
ol
a
nd
f
r
om
r
e
ta
il
e
nvi
r
onm
e
nt
s
to
hom
e
s
e
c
ur
it
y
.
B
y
e
nh
a
nc
in
g
a
nom
a
ly
de
te
c
ti
on
a
lg
or
it
hm
s
,
th
is
r
e
s
e
a
r
c
h
a
im
s
to
c
ont
r
ib
ut
e
to
s
a
f
e
r
a
nd
m
or
e
s
e
c
ur
e
e
nvi
r
onm
e
nt
s
,
w
hi
le
a
l
s
o
a
ddr
e
s
s
in
g
th
e
s
ig
ni
f
ic
a
nt
c
om
put
a
ti
ona
l
a
nd
a
c
c
ur
a
c
y
c
ha
ll
e
n
ge
s
pos
e
d
by
th
e
v
a
s
t
a
nd
gr
ow
in
g
vol
um
e
s
of
vi
de
o
da
ta
. T
hi
s
a
dva
nc
e
m
e
nt
is
not
ju
s
t
a
te
c
hnol
ogi
c
a
l
pur
s
ui
t
;
it
is
a
s
te
p
t
ow
a
r
ds
c
r
e
a
ti
ng
m
or
e
r
e
s
pons
iv
e
a
nd
in
te
ll
ig
e
nt
m
oni
to
r
in
g
s
ys
te
m
s
th
a
t
c
a
n
pl
a
y
a
c
r
uc
ia
l
r
ol
e
in
e
ns
ur
in
g
publ
ic
s
a
f
e
ty
a
nd
s
e
c
ur
it
y
in
a
n
in
c
r
e
a
s
in
gl
y di
gi
ta
l
w
or
ld
.
−
H
ybr
id
te
m
por
a
l
s
pa
ti
a
l
n
e
twor
k
(
H
T
S
N
e
t)
a
r
c
hi
te
c
tu
r
e
:
ba
s
e
d
h
ybr
id
a
r
c
hi
te
c
tu
r
e
c
om
bi
ni
ng
te
m
por
a
l
a
nd
s
pa
ti
a
l
f
e
a
tu
r
e
a
na
ly
s
i
s
(
T
S
F
A
)
,
le
ve
r
a
gi
ng
th
e
s
tr
e
ngt
hs
of
gr
a
ph
ne
ur
a
l
ne
twor
k
s
(
G
N
N
)
a
nd
C
N
N
f
or
e
nha
nc
e
d m
ot
io
n va
r
ia
ti
on a
na
ly
s
i
s
.
−
C
us
to
m
iz
e
d
G
N
N
a
nd
C
N
N
ut
il
iz
a
ti
on:
th
e
ta
il
or
e
d
G
N
N
e
f
f
e
c
ti
ve
ly
is
ol
a
te
s
a
nom
a
lo
u
s
pa
tt
e
r
ns
,
w
hi
le
th
e
c
us
to
m
iz
e
d
C
N
N
i
m
pr
ove
s
s
pa
ti
a
l
-
te
m
por
a
l
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
le
a
di
ng t
o m
or
e
pr
e
c
i
s
e
a
nom
a
ly
de
te
c
ti
on.
I
nc
or
por
a
ti
on
of
a
nove
l
c
la
s
s
if
ie
r
w
it
hi
n
H
T
S
N
e
t
s
ig
ni
f
ic
a
nt
ly
b
oos
ts
th
e
a
c
c
ur
a
c
y
in
di
f
f
e
r
e
nt
i
a
ti
ng
nor
m
a
l
f
r
om
a
bnor
m
a
l
be
ha
vi
or
s
.
−
B
e
nc
hm
a
r
ki
ng
a
nd
e
va
lu
a
ti
on:
th
e
m
ode
l’
s
e
f
f
e
c
ti
ve
ne
s
s
is
va
li
da
te
d
th
r
ough
r
ig
or
ous
te
s
ti
ng
on
th
e
U
ni
ve
r
s
it
y
of
C
a
li
f
or
ni
a
,
S
a
n
D
ie
go
–
P
e
de
s
tr
ia
n
1
(
U
C
S
D
P
e
d1
)
da
ta
s
e
t,
o
f
f
e
r
in
g
c
om
pa
r
a
ti
ve
in
s
ig
ht
s
a
ga
in
s
t
e
xi
s
ti
ng
s
ta
te
-
of
-
th
e
-
a
r
t
m
e
th
ods
.
T
he
s
tu
dy
c
ont
r
ib
ut
e
s
t
o
publ
ic
s
a
f
e
ty
a
dva
n
c
e
m
e
nt
s
by
e
nha
nc
in
g
th
e
a
c
c
ur
a
c
y a
nd
r
e
li
a
bi
li
ty
of
vi
de
o
s
ur
ve
il
la
nc
e
s
ys
te
m
s
in
de
te
c
ti
ng
a
bnor
m
a
l
be
ha
vi
or
s
. T
he
us
e
of
th
e
a
r
e
a
unde
r
th
e
c
ur
ve
(
A
U
C
)
m
e
tr
ic
f
or
pe
r
f
or
m
a
nc
e
e
va
lu
a
ti
o
n
s
e
ts
ne
w
b
e
nc
hm
a
r
ks
f
or
f
ut
ur
e
a
no
m
a
ly
de
te
c
ti
on s
ys
te
m
s
.
2.
P
R
O
P
O
S
E
D
M
E
T
H
O
D
O
L
O
G
Y
T
he
m
e
th
odol
ogy
pr
opos
e
d
is
d
e
e
p
H
T
S
N
e
t
f
or
a
nom
a
ly
de
te
c
ti
on
is
di
s
c
us
s
e
d
in
de
ta
il
in
th
is
s
e
c
ti
on
of
th
e
s
tu
dy
,
w
he
r
e
a
nom
a
lo
us
pa
tt
e
r
n
de
te
c
ti
on
is
de
a
lt
w
it
h
a
s
im
pl
e
pr
obl
e
m
of
c
la
s
s
if
ic
a
ti
on,
a
vi
de
o
is
s
pl
it
in
to
va
r
io
us
le
ngt
hs
of
vi
de
o
s
e
gm
e
nt
s
ha
vi
ng
c
ons
ta
nt
le
ngt
h
.
F
ur
th
e
r
,
a
c
la
s
s
if
ie
r
is
tr
a
in
e
d
f
or
th
e
us
ua
l
c
la
s
s
a
s
w
e
ll
a
s
e
xt
r
a
c
ti
on
of
th
e
c
ha
r
a
c
te
r
is
ti
c
ve
c
to
r
f
r
o
m
e
ve
r
y
vi
de
o
s
e
gm
e
nt
a
nd
e
ve
r
y
vi
de
o
s
e
gm
e
nt
r
e
c
e
iv
e
s
a
n
a
nom
a
ly
va
lu
e
.
T
he
c
ha
r
a
c
te
r
is
ti
c
ve
c
to
r
is
d
e
n
ot
e
d
by
of
th
e
vi
de
o
s
e
gm
e
nt
,
w
he
r
e
th
e
−
ℎ
vi
de
o
s
e
gm
e
nt
ha
s
th
e
c
ha
r
a
c
te
r
is
ti
c
ve
c
to
r
.
H
ow
e
ve
r
,
th
e
a
tt
r
ib
ut
e
s
a
r
e
r
e
pr
e
s
e
nt
e
d
gr
a
phi
c
a
ll
y
=
(
,
,
)
f
or
s
im
il
a
r
f
e
a
tu
r
e
s
a
nd
=
(
,
,
)
f
or
c
ons
is
te
n
c
y
be
in
g
te
m
por
a
l.
H
e
r
e
,
th
e
ve
r
te
x
s
e
t
is
de
not
e
d
a
s
,
is
us
e
d
to
de
not
e
th
e
f
r
ont
ie
r
s
e
t,
a
nd
th
e
ve
r
te
x
a
tt
r
ib
ut
e
is
r
e
pr
e
s
e
nt
e
d
a
s
.
T
he
vi
de
o
s
e
gm
e
nt
is
de
not
e
d a
s
, t
he
f
e
a
tu
r
e
r
e
s
e
m
bl
a
nc
e
a
s
w
e
ll
a
s
t
he
c
ons
i
s
te
nc
y t
e
m
por
a
ll
y i
s
de
not
e
d a
s
,
a
nd t
he
f
e
a
tu
r
e
ha
vi
ng d
-
di
m
e
ns
io
n f
or
t
he
vi
de
o s
e
gm
e
nt
s
i
s
de
not
e
d
a
s
×
.
T
he
e
xpone
nt
ia
l
f
unc
ti
on t
ha
t
is
us
e
d
he
r
e
i
s
nor
m
a
li
z
e
d
be
c
a
u
s
e
th
e
a
dj
a
c
e
nc
y
i
s
s
how
n
a
s
non
-
ne
ga
ti
ve
f
or
th
e
c
onf
in
e
m
e
nt
of
s
im
il
a
r
it
y
in
s
id
e
r
a
nge
s
0
a
nd
1.
S
im
u
lt
a
ne
ous
ly
,
th
e
a
dj
a
c
e
nt
m
a
tr
ix
f
or
f
e
a
tu
r
e
s
is
s
how
n
by
th
a
t
de
pi
c
ts
s
im
il
a
r
f
e
a
tu
r
e
s
qua
nt
it
a
ti
ve
ly
f
or
vi
de
o
s
e
gm
e
nt
s
.
H
ow
e
ve
r
,
th
e
ti
m
e
du
r
a
ti
on
of
th
e
vi
de
o
is
e
xpr
e
s
s
e
d
by
th
e
a
dj
a
c
e
nt
m
a
tr
ix
ha
vi
ng
a
c
ons
is
te
nc
y
th
a
t
is
te
m
por
a
l
w
hi
c
h
i
s
s
how
n
a
s
.
T
he
r
e
f
or
e
,
a
nd
a
r
e
bot
h
de
s
c
r
ib
e
d
a
s
gi
ve
n i
n (
1)
a
nd (
2)
.
(
,
)
=
e
x
p
(
.
−
(
.
)
)
(
1)
(
,
)
=
e
x
p
(
−
‖
−
‖
)
(
2)
I
n
F
ig
ur
e
1,
w
e
s
e
e
th
a
t
th
e
vi
de
o
s
e
gm
e
nt
s
de
pi
c
t
s
im
il
a
r
f
e
a
tu
r
e
s
.
T
he
y
a
r
e
li
nke
d
th
r
ough
th
e
s
m
a
ll
e
r
f
r
ont
ie
r
,
w
he
r
e
a
s
th
e
de
e
p
-
c
ol
or
e
d
ve
r
te
xe
s
de
pi
c
t
th
e
vi
de
o
s
e
gm
e
nt
s
th
a
t
ha
ve
in
c
r
e
a
s
e
d
a
nom
a
ly
va
lu
e
s
.
L
a
s
tl
y
,
th
e
ve
r
te
xe
s
th
a
t
a
r
e
c
lo
s
e
a
r
e
a
ls
o
m
a
r
ke
d
us
in
g
th
e
s
a
m
e
la
be
l
of
a
nom
a
ly
th
r
ough
th
e
L
a
pl
a
c
ia
n
gr
a
phi
c
a
l
ope
r
a
to
r
.
S
pe
c
if
ic
a
ll
y
,
th
e
uni
t
m
a
tr
ix
is
e
xpr
e
s
s
e
d
us
in
g
×
.
W
hi
le
th
e
m
a
tr
ix
a
dj
a
c
e
nc
y a
lo
ng
w
it
h e
dge
s
i
s
s
how
n
a
s
̂
.
T
hi
s
i
s
f
or
m
ul
a
te
d a
s
gi
ve
n i
n (
3)
. H
e
r
e
, t
he
de
gr
e
e
of
t
he
m
a
tr
ix
i
s
e
xpr
e
s
s
e
d
a
s
̃
.
T
hi
s
is
de
f
in
e
d
a
s
gi
ve
n
(
5)
.
T
he
r
e
f
or
e
,
th
e
gr
a
phi
c
a
l
r
e
pr
e
s
e
nt
a
ti
on
of
th
e
s
im
il
a
r
it
y
in
f
e
a
tu
r
e
s
r
e
s
ul
ts
i
n t
he
f
ol
lo
w
in
g f
or
m
ul
a
ti
on a
t
th
e
m
ode
l
le
ve
l
a
s
s
how
n
i
n (
6)
.
̂
=
̃
−
1
2
⁄
̃
̃
−
1
2
⁄
(
3)
̃
=
+
(
4)
̃
(
,
)
=
∑
̃
(
,
)
(
5)
=
(
̂
)
(
6)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
. 13, No. 4, D
e
c
e
m
be
r
2024:
4121
-
4128
4124
F
ig
ur
e
1
.
P
r
opos
e
d f
r
a
m
e
w
or
k f
or
th
e
huma
n a
nom
a
ly
pa
tt
e
r
n de
te
c
ti
on
C
ons
id
e
r
in
g
in
(
6)
,
th
e
a
c
ti
va
ti
on
f
unc
ti
on
is
gi
ve
n
a
s
,
th
e
tr
a
i
na
bl
e
m
a
tr
ix
va
r
ia
bl
e
is
gi
ve
n
a
s
,
a
nd t
he
c
ha
r
a
c
te
r
is
ti
c
s
a
t
th
e
i
nput
l
e
v
e
l
a
r
e
gi
ve
n a
s
. L
a
s
tl
y
, t
he
m
ode
l
of
s
im
il
a
r
c
ha
r
a
c
te
r
is
ti
c
s
r
e
s
ul
ti
ng i
n
th
e
out
put
is
c
om
bi
ne
d
w
it
h
th
e
te
m
po
r
a
l
c
ons
is
te
nt
s
ys
te
m
a
t
th
e
pool
in
g
le
ve
l
a
nd
ut
il
iz
e
s
th
e
a
c
ti
va
ti
on
f
unc
ti
on
s
ig
m
oi
d.
T
he
r
e
f
or
e
,
th
e
p
r
oba
bi
li
ty
o
f
pr
e
di
c
ti
on
f
or
e
ve
r
y
gr
a
phi
c
a
l
ve
r
te
x
m
a
tc
he
s
w
it
h
th
e
pr
oba
bi
li
ty
of
a
nom
a
ly
of
om
it
ti
ng
noi
s
e
f
or
th
e
ℎ
vi
de
o
s
e
gm
e
nt
a
s
w
e
ll
a
s
th
e
la
s
t
lo
s
s
f
unc
ti
on
th
a
t
r
e
s
ul
te
d i
n t
he
s
um
m
a
ti
on of
t
w
o t
ype
s
of
l
os
s
t
ha
t
a
r
e
gi
ve
n
in
(
7)
.
=
+
(
7)
T
he
va
lu
e
s
of
a
nd
a
r
e
obt
a
in
e
d
by
e
va
lu
a
ti
on
of
di
r
e
c
t
a
s
w
e
ll
a
s
in
di
r
e
c
t
s
upe
r
vi
s
io
n
r
e
s
pe
c
ti
ve
ly
.
B
e
f
or
e
noi
s
e
e
li
m
in
a
ti
on,
a
c
la
s
s
if
ie
r
is
tr
a
in
e
d
f
r
om
w
he
r
e
a
n
a
ppr
oxi
m
a
te
va
lu
e
f
o
r
a
nom
a
ly
pr
oba
bi
li
ty
c
oul
d
be
obt
a
in
e
d
f
or
e
v
e
r
y
vi
de
o
s
e
gm
e
nt
ha
vi
ng
a
c
on
s
ta
nt
le
ngt
h.
W
e
a
r
e
gi
ve
n
th
e
va
lu
e
of
a
nom
a
ly
pr
oba
bi
li
ty
a
s
̃
=
{
̃
}
=
1
,
th
e
a
lg
or
it
hm
s
th
a
t
a
r
e
di
r
e
c
tl
y
s
upe
r
vi
s
e
d
a
r
e
us
e
d
f
or
e
xpr
e
s
s
in
g
lo
s
s
of
c
r
os
s
-
e
nt
r
opy a
s
gi
ve
n i
n (
8)
.
=
−
(
‖
‖
)
−
1
∑
[
̃
ln
+
(
1
−
̃
)
ln
(
1
−
)
]
(
8)
I
n
th
is
e
qua
ti
on,
th
e
vi
de
o
s
e
gm
e
nt
s
th
a
t
ha
v
e
hi
gh
c
onf
id
e
nc
e
a
r
e
r
e
pr
e
s
e
nt
e
d
a
s
.
T
e
n
im
a
ge
s
a
r
e
c
r
oppe
d
f
r
om
e
ve
r
y
f
r
a
m
e
o
f
th
e
vi
de
o
th
a
t
is
us
e
d
in
th
e
a
ugm
e
nt
a
ti
on
of
da
ta
a
s
w
e
ll
a
s
th
e
c
om
put
a
ti
on
of
th
e
m
e
a
n
a
nom
a
ly
pr
oba
bi
li
ty
̃
a
nd
th
e
va
r
ia
nc
e
is
pr
e
di
c
te
d
in
th
e
c
la
s
s
if
ie
r
. T
he
unc
e
r
ta
in
pr
e
di
c
ti
ons
a
r
e
m
e
a
s
ur
e
d us
in
g va
r
ia
nc
e
.
T
he
l
e
s
s
e
r
t
he
va
lu
e
of
va
r
ia
nc
e
, t
he
hi
ghe
r
t
he
c
onf
id
e
nc
e
. I
ndi
r
e
c
t
s
upe
r
vi
s
io
n i
s
a
ty
pe
of
te
m
por
a
l
a
ppr
oa
c
h
th
a
t
is
ut
il
iz
e
d
f
or
f
ut
u
r
e
im
pl
e
m
e
nt
a
ti
on
of
s
om
e
la
be
le
d
da
ta
in
s
ta
n
c
e
s
,
a
s
onl
y
a
por
ti
on
of
th
e
e
nt
ir
e
vi
de
o
ha
s
a
pr
e
di
c
ti
on
of
hi
gh
c
onf
id
e
nc
e
.
T
he
goa
l
is
to
e
a
s
e
th
e
pr
e
di
c
ti
on
of
th
e
ne
twor
k
f
or
a
ll
th
e
vi
de
o
s
e
gm
e
nt
s
f
or
va
r
io
us
tr
a
in
in
g
pha
s
e
s
a
s
gi
ve
n
in
(
9)
.
I
n
th
e
(
9)
,
th
e
m
e
a
n
o
f
w
e
ig
ht
e
d
pr
e
di
c
ti
on
f
or
noi
s
e
om
it
ti
ng i
s
gi
ve
n a
s
̅
in
e
ve
r
y t
r
a
in
in
g pha
s
e
.
=
(
)
−
1
∑
|
−
̅
|
=
1
(
9)
2.1.
A
n
om
al
y p
at
t
e
r
n
c
la
s
s
if
ic
at
io
n
I
n
pr
e
vi
ous
s
tu
di
e
s
,
a
nom
a
ly
pa
tt
e
r
n
c
l
a
s
s
if
ic
a
ti
on
w
a
s
p
e
r
f
or
m
e
d
di
r
e
c
tl
y
ba
s
e
d
on
tr
a
in
in
g
a
s
w
e
ll
a
s
a
n
e
nt
ir
e
vi
de
o ha
vi
ng
a
nom
a
li
e
s
be
in
g c
l
a
s
s
if
ie
d.
A
lt
hough,
pr
a
c
ti
c
a
ll
y
, i
f
a
huge
c
ount
of
nor
m
a
l
be
ha
vi
or
vi
de
o
s
e
gm
e
nt
s
c
a
n
be
s
e
pa
r
a
te
d
f
r
om
th
e
vi
de
os
ha
vi
ng
a
n
om
a
li
e
s
,
th
e
pe
r
f
or
m
a
nc
e
of
a
nom
a
ly
pa
tt
e
r
n
c
la
s
s
if
ic
a
ti
on
w
oul
d
be
e
nha
n
c
e
d. I
n t
he
pr
opos
e
d m
e
th
od, we
s
pl
it
e
ve
r
y vi
de
o i
nt
o
vi
de
o s
e
gm
e
nt
s
ha
vi
ng
a
pa
r
ti
c
ul
a
r
c
ons
ta
nt
l
e
ngt
h. F
or
e
ve
r
y vi
de
o s
e
gm
e
nt
, a
c
or
r
e
s
ponding a
nom
a
ly
va
lu
e
i
s
obt
a
in
e
d a
t
th
e
pha
s
e
of
a
nom
a
ly
pa
tt
e
r
n
de
te
c
ti
on.
H
e
nc
e
,
a
li
br
a
r
y
of
vi
de
o
s
e
gm
e
nt
s
is
obt
a
in
e
d
th
a
t
h
a
s
e
le
m
e
nt
s
of
a
nom
a
ly
va
lu
e
s
a
s
gi
v
e
n i
n (
10)
.
=
{
|
{
1
,
2
,
…
…
,
}
}
(
10)
T
o
om
it
th
e
nor
m
a
l
vi
de
o
s
e
gm
e
nt
s
,
a
n
a
ppr
oa
c
h
is
pr
opos
e
d
th
a
t
ha
s
a
f
ix
e
d
th
r
e
s
hol
d
.
T
he
vi
de
o
s
e
gm
e
nt
s
th
a
t
a
r
e
hi
ghe
r
in
c
om
pa
r
is
on
to
th
e
th
r
e
s
hol
d
a
r
e
la
b
e
le
d
a
nd
th
e
one
s
lo
w
e
r
th
a
n
th
e
th
r
e
s
hol
d
a
r
e
c
ons
id
e
r
e
d
a
s
nor
m
a
l,
th
e
nor
m
a
l
one
s
a
r
e
de
le
te
d.
T
he
r
e
f
or
e
,
th
e
vi
de
o
pos
t
in
te
r
c
e
pt
io
n
is
e
xpr
e
s
s
e
d
a
s
gi
ve
n
in
(
1
1)
.
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
E
nhanc
in
g v
id
e
o anomaly
de
te
c
ti
on f
or
hum
an
s
us
pi
c
io
us
b
e
ha
v
io
r
t
hr
ough de
e
p h
y
br
id
…
(
K
u
s
um
a Sr
i
r
am
)
4125
=
{
|
ℎ
,
}
(
11)
In
th
e
(
1
1)
,
th
e
a
nom
a
ly
va
lu
e
pr
e
di
c
ti
on
is
gi
ve
n
a
s
by
th
e
m
ode
l
f
or
a
nom
a
ly
de
te
c
ti
on
f
or
th
e
vi
de
o
s
e
gm
e
nt
.
C
ons
id
e
r
in
g
th
e
(
1
1)
,
w
e
ob
s
e
r
ve
th
a
t
dur
in
g
t
he
in
c
r
e
a
s
e
in
th
r
e
s
hol
d,
th
e
c
ount
of
vi
de
o
s
e
gm
e
nt
s
ha
vi
ng
a
nom
a
li
e
s
th
a
t
m
e
e
t
th
is
r
e
qui
r
e
m
e
nt
is
le
s
s
f
or
a
n
in
di
vi
dua
l
v
id
e
o
w
it
h
a
nom
a
li
e
s
,
w
hi
c
h
c
oul
d
r
e
s
ul
t
in
in
a
de
qua
te
in
f
or
m
a
ti
on
f
or
tr
a
in
in
g
th
e
m
ode
l
f
or
c
la
s
s
if
ic
a
ti
on
a
nd
la
s
tl
y
le
a
ds
to
ove
r
f
it
ti
ng
o
f
th
e
m
ode
l.
O
n t
he
c
ont
r
a
r
y
,
th
e
l
e
s
s
e
r
t
he
t
hr
e
s
hol
d, t
he
hi
ghe
r
t
he
vi
de
o s
e
gm
e
nt
s
w
it
h a
nom
a
li
e
s
t
ha
t
s
a
ti
s
f
y
th
is
r
e
qui
r
e
m
e
nt
th
a
t
e
xi
s
t
in
th
e
s
a
m
e
a
nom
a
ly
vi
de
o,
a
nd
th
e
hi
ghe
r
th
e
pos
s
ib
il
it
y
f
or
nor
m
a
l
vi
de
o
s
e
gm
e
nt
s
not
be
in
g
c
om
pl
e
te
ly
om
it
te
d.
T
he
r
e
f
or
e
,
s
e
le
c
ti
ng
a
n
a
ppr
opr
ia
te
th
r
e
s
hol
d
is
e
s
s
e
nt
ia
l.
T
he
or
e
ti
c
a
ll
y
,
w
e
c
oul
d
pr
e
di
c
t,
a
s
s
um
in
g
th
e
r
e
i
s
no
ove
r
f
it
ti
ng,
in
c
r
e
a
s
e
d
th
r
e
s
hol
d
le
a
ds
to
le
s
s
e
r
noi
s
e
in
in
put
in
f
or
m
a
ti
on
f
or
th
e
c
la
s
s
if
ic
a
ti
on s
y
s
te
m
a
nd t
he
r
e
i
s
a
n i
nc
r
e
a
s
e
i
n t
he
r
a
te
of
a
c
c
u
r
a
c
y
.
C
onve
r
s
e
ly
,
ba
s
e
d
on
(
1
1)
if
th
e
a
nom
a
ly
pa
tt
e
r
n
lo
c
a
li
z
a
ti
on
i
s
m
a
de
,
m
a
ny
vi
de
o
s
e
gm
e
nt
s
w
oul
d
m
e
e
t
th
e
th
r
e
s
hol
d
f
or
s
e
le
c
ti
on.
T
he
s
e
s
e
pa
r
a
te
vi
de
o
s
e
gm
e
nt
s
a
r
e
c
om
bi
ne
d
a
s
on
e
vi
de
o
a
nd
f
ur
th
e
r
c
la
s
s
if
ie
d. T
hi
s
c
oul
d
r
e
s
ul
t
in
te
m
por
a
l
s
e
que
nc
e
da
ta
l
os
s
of
th
e
vi
de
o
a
t
th
e
ti
m
e
of
e
xt
r
a
c
ti
ng
f
e
a
tu
r
e
s
,
th
e
lo
c
a
li
z
a
ti
on a
ppr
oa
c
h i
s
e
nh
a
nc
e
d a
s
gi
ve
n
in
(
12)
.
=
{
|
∀
{
,
+
1
,
…
.
,
+
}
,
{
1
,
2
,
…
.
,
−
}
}
(
12)
I
n
(
12)
s
ig
ni
f
ie
s
th
a
t
th
e
c
ont
in
ui
ty
li
m
it
a
ti
on
in
c
lu
de
s
w
hi
le
lo
c
a
li
z
a
ti
on
of
a
nom
a
ly
s
e
gm
e
nt
s
. T
hi
s
is
tr
ue
if
,
a
t
le
a
s
t
c
ont
in
uous
vi
de
o
s
e
gm
e
nt
s
s
a
ti
s
f
y
th
e
(
12)
,
th
e
s
e
c
on
s
e
c
ut
iv
e
vi
de
o
s
e
gm
e
nt
s
a
r
e
put
in
th
e
vi
de
o
s
e
gm
e
nt
s
e
t
w
it
h
a
nom
a
li
e
s
a
nd
in
put
in
to
th
e
a
nom
a
ly
pa
tt
e
r
n
c
la
s
s
if
ic
a
ti
on
m
ode
l
f
or
te
s
ti
ng
or
tr
a
in
in
g.
L
a
s
tl
y
,
th
is
s
e
c
ti
on
f
or
s
e
le
c
ti
on
in
th
e
pr
opos
e
d
m
e
th
odol
ogy
is
s
um
m
a
r
iz
e
d
a
s
gi
ve
n
in
(
13)
.
U
s
in
g
th
is
a
ppr
oa
c
h, w
e
om
it
t
he
noi
s
e
i
n t
he
vi
de
o
s
e
gm
e
nt
s
a
s
w
e
ll
a
s
t
he
t
e
m
por
a
l
f
e
a
tu
r
e
s
a
r
e
c
on
s
e
r
ve
d.
′
=
{
|
ℎ
,
,
}
(
13)
3.
P
E
R
F
O
R
M
A
N
C
E
E
V
A
L
U
A
T
I
O
N
A
th
or
ough
a
s
s
e
s
s
m
e
nt
of
th
e
pr
opos
e
d
d
e
e
p
H
T
S
N
e
t
a
ppr
o
a
c
h
on
th
e
U
C
S
D
P
ed
1
da
t
a
s
e
t
s
f
or
a
nom
a
ly
id
e
nt
if
ic
a
ti
on
is
c
a
r
r
ie
d
out
in
th
is
s
e
c
ti
on.
A
c
om
pr
e
he
ns
iv
e
s
tu
dy
is
c
onduc
te
d
to
s
how
th
e
r
obus
tn
e
s
s
a
nd e
f
f
ic
a
c
y of
t
he
t
e
c
hni
que
by c
om
pa
r
in
g e
s
ti
m
a
te
d a
bnor
m
a
l
f
r
a
m
e
s
w
it
h gr
ound tr
ut
h l
a
be
ls
.
T
he
s
ys
te
m
'
s
pe
r
f
or
m
a
nc
e
is
e
va
lu
a
te
d
a
ga
in
s
t
c
ur
r
e
nt
s
ta
te
-
of
-
th
e
-
a
r
t
m
e
th
ods
by
c
a
lc
ul
a
ti
ng
th
e
A
U
C
by
s
how
c
a
s
in
g
th
e
a
nom
a
ly
s
c
or
e
s
.
3.1. Dat
as
e
t
d
e
t
ai
ls
T
he
U
C
S
D
P
e
d1 da
ta
s
e
t,
c
r
e
a
te
d by the
U
ni
ve
r
s
it
y of
C
a
li
f
or
ni
a
, S
a
n D
ie
go, i
s
a
pr
om
in
e
nt
r
e
s
our
c
e
in
c
om
put
e
r
vi
s
io
n
r
e
s
e
a
r
c
h,
pa
r
ti
c
ul
a
r
ly
f
or
a
nom
a
ly
de
t
e
c
ti
on
in
vi
de
o
s
ur
ve
il
la
nc
e
.
I
t
c
om
pr
is
e
s
lo
w
-
r
e
s
ol
ut
io
n,
gr
a
ys
c
a
le
vi
de
os
f
oc
us
in
g
on
pe
de
s
tr
ia
n
w
a
lk
w
a
ys
,
w
he
r
e
a
nom
a
li
e
s
a
r
e
de
f
in
e
d
a
s
non
-
ty
pi
c
a
l
pe
de
s
tr
ia
n
be
ha
vi
or
s
li
ke
s
ka
te
bo
a
r
di
ng,
bi
ki
ng,
or
de
vi
a
ti
ng
f
r
om
w
a
lk
w
a
ys
.
T
he
s
e
vi
de
os
a
r
e
a
nnot
a
te
d
to
m
a
r
k a
nom
a
lo
us
e
ve
nt
s
,
a
id
in
g i
n t
he
de
ve
lo
pm
e
nt
a
nd t
e
s
ti
ng of
s
ur
ve
il
la
nc
e
a
lg
or
it
hm
s
.
3.2. M
e
t
r
ic
u
s
e
d
f
or
c
om
p
ar
is
on
T
he
A
U
C
m
e
tr
ic
,
pa
r
ti
c
ul
a
r
ly
a
s
pa
r
t
of
th
e
r
e
c
e
iv
e
r
ope
r
a
ti
ng
c
ha
r
a
c
te
r
is
ti
c
(
R
O
C
)
a
n
a
ly
s
is
,
s
e
r
ve
s
a
s
a
c
r
it
ic
a
l
to
ol
f
o
r
e
va
lu
a
ti
ng
th
e
pe
r
f
or
m
a
nc
e
of
a
nom
a
ly
de
te
c
ti
on
a
lg
or
it
hm
s
.
A
U
C
in
vi
de
o
a
nom
a
ly
de
te
c
ti
on
qua
nt
if
ie
s
how
w
e
ll
a
n
a
lg
or
it
hm
c
a
n
di
s
ti
ngui
s
h
b
e
t
w
e
e
n
nor
m
a
l
a
nd
a
nom
a
lo
u
s
e
v
e
nt
s
.
A
hi
ghe
r
A
U
C
va
lu
e
in
di
c
a
te
s
a
hi
ghe
r
li
ke
li
hood
th
a
t
th
e
m
ode
l
c
or
r
e
c
tl
y
id
e
nt
if
ie
s
a
nom
a
li
e
s
a
nd
nor
m
a
l
a
c
ti
vi
ti
e
s
.
D
ue
to
th
e
of
te
n
c
om
pl
e
x
a
nd
dyn
a
m
ic
na
tu
r
e
of
vi
de
o
d
a
ta
,
a
no
m
a
li
e
s
c
a
n
va
r
y
w
id
e
ly
in
a
ppe
a
r
a
nc
e
a
nd
be
ha
vi
or
,
m
a
ki
ng
th
e
A
U
C
a
va
lu
a
bl
e
m
e
a
s
ur
e
f
or
a
s
s
e
s
s
in
g
th
e
ge
ne
r
a
li
z
a
bi
li
ty
of
a
n
a
lg
or
it
hm
a
c
r
os
s
di
f
f
e
r
e
nt
ty
pe
s
of
a
nom
a
li
e
s
.
3.3. Re
s
u
lt
s
an
d
D
is
c
u
s
s
io
n
T
he
gr
a
ph i
n F
ig
ur
e
2 pr
e
s
e
nt
s
t
he
A
U
C
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
f
o
r
va
r
io
us
a
nom
a
ly
de
te
c
ti
on me
th
ods
a
ppl
ie
d
to
th
e
U
C
S
D
P
e
d1
da
ta
s
e
t.
T
he
P
S
m
e
th
od
le
a
ds
w
it
h
th
e
hi
ghe
s
t
A
U
C
of
96.34
%
,
in
di
c
a
ti
ng
it
s
s
upe
r
io
r
a
bi
li
ty
to
di
s
ti
ngui
s
h
b
e
twe
e
n
nor
m
a
l
a
nd
a
nom
a
lo
u
s
b
e
ha
vi
or
w
it
hi
n
th
e
vi
de
o
f
oot
a
ge
.
P
lu
g
a
nd
pl
a
y
C
N
N
a
ls
o
p
e
r
f
or
m
s
e
xc
e
pt
io
na
ll
y
w
e
ll
w
it
h
a
95.7%
A
U
C
.
I
n
c
o
nt
r
a
s
t,
m
e
th
ods
li
ke
m
ot
io
n
in
f
lu
e
nc
e
m
a
p
a
nd
M
a
r
c
he
nko
-
P
a
s
tu
r
pr
in
c
ip
a
l
c
om
pone
nt
a
na
ly
s
i
s
(
M
P
P
C
A
)
a
r
e
a
t
th
e
lo
w
e
r
e
nd
of
th
e
s
pe
c
tr
um
,
w
it
h A
U
C
s
of
61.9%
a
nd
66.8%
r
e
s
pe
c
ti
ve
ly
,
s
ugge
s
ti
ng
a
le
s
s
e
r
d
e
gr
e
e
of
a
c
c
ur
a
c
y
in
a
nom
a
ly
d
e
te
c
ti
on.
T
he
m
a
jo
r
it
y
of
m
e
th
ods
c
lu
s
te
r
be
twe
e
n
th
e
70%
to
95%
r
a
nge
,
r
e
f
le
c
ti
ng
a
w
id
e
va
r
ia
nc
e
in
e
f
f
e
c
ti
ve
n
e
s
s
,
w
it
h
s
e
v
e
r
a
l
m
e
th
ods
li
ke
s
pa
r
s
e
r
e
c
ons
tr
uc
ti
on
,
a
pp
e
a
r
a
nc
e
a
nd
m
ot
io
n
D
e
e
pN
e
t
(
A
M
D
N
)
,
a
nd
G
r
ow
in
gG
a
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
. 13, No. 4, D
e
c
e
m
be
r
2024:
4121
-
4128
4126
de
m
ons
tr
a
ti
ng
hi
gh
e
f
f
ic
a
c
y
,
a
ll
s
c
or
in
g
a
bove
90%
.
T
he
gr
a
ph
vi
s
ua
ll
y
e
nc
a
ps
ul
a
te
s
th
e
pe
r
f
or
m
a
nc
e
r
a
nge
a
c
r
os
s
th
e
m
e
th
ods
,
hi
ghl
ig
ht
in
g
th
e
s
ig
ni
f
ic
a
nt
di
f
f
e
r
e
nc
e
s
in
th
e
ir
a
bi
li
ty
to
m
ode
l
a
nd
de
te
c
t
a
nom
a
li
e
s
w
it
hi
n
th
e
da
ta
s
e
t.
T
a
bl
e
1 s
how
s
t
he
A
U
C
c
om
pa
r
is
on f
or
t
he
U
C
S
D
pe
d 1 da
ta
s
e
t.
T
a
bl
e
1.
A
U
C
c
om
pa
r
is
on f
or
U
C
S
D
pe
d1 da
ta
s
e
t
M
e
t
hod
AUC
(%)
M
P
P
C
A
[
16]
66.80
SF
[
17]
67.50
M
P
P
C
A
+S
F
[
17]
74.20
S
pa
r
s
e
r
e
c
ons
t
r
uc
t
i
on
[
18]
86.00
C
onvA
E
[
19]
81.00
C
onvL
S
T
M
-
A
E
[
20
]
81.50
M
ot
i
on
i
nf
l
ue
nc
e
m
a
p
[
21]
61.90
U
nm
a
s
ki
ng
[
22]
68.40
C
hong
a
nd
T
a
y
[
23]
89.90
A
M
D
N
[
24]
92.10
G
r
ow
i
ngG
a
s
[
25]
93.80
S
t
a
c
ke
d R
N
N
[
26]
83.10
F
r
a
m
e
-
pr
e
d
[
27]
81.10
P
l
ug a
nd
pl
a
y
C
N
N
[
28]
95.70
D
e
e
p
or
di
na
l
r
e
gr
e
s
s
i
on
[
29]
71.70
R
a
m
a
c
ha
ndr
a
e
t
al
.
[
30]
77.30
R
a
m
a
c
ha
ndr
a
a
nd
J
one
s
[
31]
86.00
G
e
or
ge
s
c
u
e
t
al
.
[
32]
76.30
E
S
[
33]
94.20
PS
96.34
F
ig
ur
e
1
.
A
U
C
c
ur
ve
f
or
U
C
S
D
pr
e
d 1 da
ta
s
e
t
4.
C
O
N
C
L
U
S
I
O
N
I
n
c
onc
lu
s
io
n,
th
e
s
tu
dy
s
uc
c
e
s
s
f
ul
ly
de
m
ons
tr
a
te
s
th
e
e
f
f
ic
a
c
y
of
th
e
d
e
e
p
H
T
S
N
e
t
f
or
a
nom
a
ly
de
te
c
ti
on
in
vi
de
o
s
ur
ve
il
la
nc
e
,
m
a
r
ki
ng
a
s
ig
ni
f
ic
a
nt
s
tr
id
e
in
a
ut
om
a
te
d
a
nom
a
ly
de
te
c
ti
on
te
c
hnol
ogy
.
B
y
in
nova
ti
ve
ly
s
e
gm
e
nt
in
g
vi
de
o
in
to
c
ons
ta
nt
-
le
ngt
h
s
e
gm
e
nt
s
a
nd
e
m
pl
oyi
ng
a
gr
a
ph
-
ba
s
e
d
f
e
a
tu
r
e
a
na
ly
s
is
,
d
e
e
p
H
T
S
N
e
t
not
onl
y
e
nha
n
c
e
s
th
e
a
c
c
ur
a
c
y
of
de
te
c
ti
ng
a
no
m
a
li
e
s
in
c
om
pl
e
x
a
nd
dyna
m
ic
e
nvi
r
onm
e
nt
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
E
nhanc
in
g v
id
e
o anomaly
de
te
c
ti
on f
or
hum
an
s
us
pi
c
io
us
b
e
ha
v
io
r
t
hr
ough de
e
p h
y
br
id
…
(
K
u
s
um
a Sr
i
r
am
)
4127
but
a
ls
o
s
ig
ni
f
ic
a
nt
ly
boos
t
s
c
om
put
a
ti
ona
l
e
f
f
ic
ie
n
c
y
.
T
he
m
e
th
od'
s
r
obus
tn
e
s
s
a
ga
in
s
t
noi
s
e
a
nd
f
a
ls
e
po
s
it
iv
e
s
is
a
not
a
bl
e
a
dva
nc
e
m
e
nt
, a
ddr
e
s
s
in
g ke
y c
ha
ll
e
nge
s
i
n t
he
f
ie
ld
.
T
he
c
om
pr
e
he
ns
iv
e
e
va
lu
a
ti
on of
t
he
U
C
S
D
P
e
d1
da
ta
s
e
t,
w
he
r
e
th
e
m
e
th
od
s
how
c
a
s
e
d
s
up
e
r
io
r
pe
r
f
or
m
a
nc
e
ove
r
e
xi
s
ti
ng
s
ta
te
-
of
-
th
e
-
a
r
t
te
c
hni
que
s
,
pa
r
ti
c
ul
a
r
ly
i
n t
e
r
m
s
of
A
U
C
, r
e
a
f
f
ir
m
s
i
ts
pot
e
nt
ia
l
f
or
pr
a
c
ti
c
a
l
im
pl
e
m
e
nt
a
ti
on.
T
hi
s
r
e
s
e
a
r
c
h pa
ve
s
t
he
w
a
y
f
or
m
or
e
r
e
s
pons
iv
e
a
nd
i
nt
e
ll
ig
e
nt
s
ur
ve
il
la
nc
e
s
y
s
te
m
s
,
c
ont
r
ib
ut
in
g t
o
he
ig
ht
e
ne
d publi
c
s
a
f
e
ty
a
nd s
e
c
ur
it
y
in
our
i
nc
r
e
a
s
in
gl
y di
gi
ta
l
a
nd ur
ba
ni
z
e
d w
or
ld
.
R
E
F
E
R
E
N
C
E
S
[
1]
C
. K
.
M
e
he
r
, R
.
N
a
ya
k,
a
nd U
.
C
. P
a
t
i
, “
D
u
a
l
s
t
r
e
a
m
v
a
r
i
a
t
i
ona
l
a
ut
oe
nc
ode
r
f
or
vi
de
o a
nom
a
l
y de
t
e
c
t
i
on i
n
s
i
ngl
e
s
c
e
ne
vi
de
o
s
,
”
2nd
O
di
s
ha
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
E
l
e
c
t
r
i
c
al
P
ow
e
r
E
ngi
ne
e
r
i
ng,
C
om
m
uni
c
at
i
on
and
C
om
put
i
ng
T
e
c
hnol
ogy
,
O
D
I
C
O
N
2022
, 2022, doi
:
10.1109/
O
D
I
C
O
N
54453.2022.10010086.
[
2]
S
.
Y
u,
C
.
W
a
ng,
Q
. M
a
o,
Y
. L
i
,
a
nd
J
.
W
u,
“
C
r
os
s
-
e
poc
h
l
e
a
r
ni
ng
f
or
w
e
a
kl
y
s
u
pe
r
vi
s
e
d
a
nom
a
l
y
de
t
e
c
t
i
on
i
n
s
ur
ve
i
l
l
a
nc
e
vi
de
os
,
”
I
E
E
E
Si
gnal
P
r
oc
e
s
s
i
ng L
e
t
t
e
r
s
, vol
. 28, pp. 2137
–
2141, 2021, doi
:
10.1109/
L
S
P
.2021.3117737.
[
3]
S
.
B
i
s
w
a
s
a
nd
R
.
V
.
B
a
bu,
“
R
e
a
l
t
i
m
e
a
nom
a
l
y
de
t
e
c
t
i
on
i
n
H
.264
c
om
pr
e
s
s
e
d
vi
de
os
,”
2013
4t
h
N
at
i
onal
C
onf
e
r
e
nc
e
on
C
o
m
put
e
r
V
i
s
i
on, P
at
t
e
r
n R
e
c
ogni
t
i
on, I
m
age
P
r
oc
e
s
s
i
ng and G
r
aphi
c
s
,
N
C
V
P
R
I
P
G
201
3
, 2013, doi
:
10.1109/
N
C
V
P
R
I
P
G
.2013.6776164.
[
4]
C
.
K
.
M
e
he
r
,
R
.
N
a
ya
k,
a
nd
U
.
C
.
P
a
t
i
,
“
V
i
de
o
a
nom
a
l
y
de
t
e
c
t
i
on
us
i
ng
va
r
i
a
t
i
ona
l
a
ut
oe
nc
ode
r
,”
2022
I
E
E
E
2nd
I
nt
e
r
nat
i
onal
Sy
m
pos
i
um
on
Sus
t
ai
nabl
e
E
ne
r
gy
,
Si
gnal
P
r
oc
e
s
s
i
ng
and
C
y
be
r
S
e
c
ur
i
t
y
,
i
SSSC
2022
,
2022,
doi
:
10.1109/
i
S
S
S
C
56467.2022.10051511.
[
5]
G
.
V
.
P
i
l
l
a
i
,
A
.
V
e
r
m
a
,
a
nd
D
.
S
e
n,
“
T
r
a
ns
f
or
m
e
r
ba
s
e
d
s
e
l
f
-
c
ont
e
xt
a
w
a
r
e
pr
e
di
c
t
i
on
f
or
f
e
w
-
s
hot
a
nom
a
l
y
de
t
e
c
t
i
on
i
n
vi
de
os
,”
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on I
m
age
P
r
o
c
e
s
s
i
ng, I
C
I
P
, pp. 3485
–
3489, 2022, doi
:
10.1109/
I
C
I
P
46576.2022.9897615.
[
6]
A
.
M
e
hm
ood,
“
E
f
f
i
c
i
e
nt
a
nom
a
l
y
de
t
e
c
t
i
on
i
n
c
r
ow
d
vi
de
os
u
s
i
ng
pr
e
-
t
r
a
i
ne
d
2D
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
,”
I
E
E
E
A
c
c
e
s
s
,
vol
. 9, pp. 138283
–
138295, 2021, doi
:
10.1109/
A
C
C
E
S
S
.2021.3118009.
[
7]
X
.
H
u
e
t
al
.
,
“
A
w
e
a
kl
y
s
upe
r
vi
s
e
d
f
r
a
m
e
w
or
k
f
or
a
bnor
m
a
l
be
ha
vi
or
de
t
e
c
t
i
on
a
nd
l
oc
a
l
i
z
a
t
i
on
i
n
c
r
ow
de
d
s
c
e
ne
s
,
”
N
e
ur
oc
om
put
i
ng
, vol
. 383, pp. 270
–
281, 2020, doi
:
10.1016/
j
.ne
uc
om
.2019.11.
087.
[
8]
X
.
Z
ha
ng,
Q
.
Y
u,
a
nd
H
.
Y
u,
“
P
hys
i
c
s
i
ns
pi
r
e
d
m
e
t
hods
f
or
c
r
ow
d
vi
de
o
s
ur
ve
i
l
l
a
nc
e
a
nd
a
na
l
ys
i
s
:
A
s
ur
ve
y,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
6, pp. 66816
–
66830, 2018, doi
:
10.1109/
A
C
C
E
S
S
.2018.2878733.
[
9]
S
.
C
ha
ndr
a
ka
l
a
,
K
.
D
e
e
pa
k,
a
nd
V
.
L
.K
.P
.,
“
B
a
g
-
of
-
e
ve
nt
-
m
ode
l
s
ba
s
e
d
e
m
be
ddi
ngs
f
or
de
t
e
c
t
i
ng
a
nom
a
l
i
e
s
i
n
s
ur
ve
i
l
l
a
nc
e
vi
de
os
,”
E
x
pe
r
t
Sy
s
t
e
m
s
w
i
t
h A
ppl
i
c
at
i
ons
, vol
. 190, 2022, doi
:
10.1016/
j
.e
s
w
a
.
2021.116168.
[
10]
W
.
L
i
n,
J
.
G
a
o,
Q
.
W
a
ng,
a
nd
X
.
L
i
,
“
L
e
a
r
ni
ng
t
o
de
t
e
c
t
a
nom
a
l
y
e
ve
nt
s
i
n
c
r
ow
d
s
c
e
ne
s
f
r
om
s
ynt
he
t
i
c
da
t
a
,”
N
e
u
r
oc
om
put
i
ng
,
vol
. 436, pp. 248
–
259, 2021, doi
:
10.1016/
j
.ne
uc
om
.2021.01.031.
[
11]
N
. L
i
, Y
. B
. H
ou, a
nd Z
. Q
. H
ua
ng
, “
I
m
pl
e
m
e
nt
a
t
i
on of
a
r
e
a
l
-
t
i
m
e
f
a
l
l
de
t
e
c
t
i
on a
l
gor
i
t
hm
ba
s
e
d on body’
s
a
c
c
e
l
e
r
a
t
i
on,”
J
our
nal
of
C
hi
ne
s
e
C
o
m
put
e
r
Sy
s
t
e
m
s
,
vol
.
33,
no.
11,
pp.
2410
–
2413,
2012
.
[
O
nl
i
ne
]
.
A
va
i
l
a
bl
e
:
ht
t
ps
:
/
/
kns
.c
nki
.ne
t
/
kc
m
s
/
de
t
a
i
l
/
de
t
a
i
l
.a
s
px?
F
i
l
e
N
a
m
e
=X
X
W
X
201211019&
D
b
N
a
m
e
=C
J
F
Q
2012
[
12]
D
.
P
a
n,
H
.
L
i
u,
D
.
Q
u,
a
nd
Z
.
Z
ha
ng,
“
H
um
a
n
f
a
l
l
i
ng
de
t
e
c
t
i
on
a
l
gor
i
t
hm
ba
s
e
d
on
m
ul
t
i
s
e
ns
or
da
t
a
f
us
i
on
w
i
t
h
S
V
M
,”
M
obi
l
e
I
nf
or
m
at
i
on Sy
s
t
e
m
s
, vol
. 2020, 2020, doi
:
10.1155/
2020/
8826088.
[
13]
Y
.
Z
hong,
X
.
C
he
n,
Y
.
H
u,
P
.
T
a
ng,
a
nd
F
.
R
e
n,
“
B
i
di
r
e
c
t
i
ona
l
s
pa
t
i
o
-
t
e
m
por
a
l
f
e
a
t
ur
e
l
e
a
r
ni
ng
w
i
t
h
m
ul
t
i
s
c
a
l
e
e
va
l
ua
t
i
on
f
or
vi
de
o
a
nom
a
l
y de
t
e
c
t
i
on,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
C
i
r
c
ui
t
s
and
Sy
s
t
e
m
s
f
o
r
V
i
de
o T
e
c
hnol
ogy
, vol
. 32, no. 12,
pp. 8285
–
8296, 2022,
doi
:
10.1109/
T
C
S
V
T
.2022.3190539.
[
14]
C
.
D
i
r
e
kogl
u,
“
A
bnor
m
a
l
c
r
ow
d
be
ha
vi
or
de
t
e
c
t
i
on
us
i
ng
m
ot
i
on
i
nf
or
m
a
t
i
on
i
m
a
ge
s
a
nd
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 80408
–
80416, 2020, doi
:
10.1109/
A
C
C
E
S
S
.2020.2990355.
[
15]
J
. L
i
, Q
. H
ua
ng, Y
.
D
u, X
. Z
he
n, S
. C
h
e
n, a
nd
L
. S
ha
o,
“
V
a
r
i
a
t
i
ona
l
a
bnor
m
a
l
be
ha
vi
or
de
t
e
c
t
i
on w
i
t
h m
ot
i
on c
ons
i
s
t
e
nc
y,”
I
E
E
E
T
r
ans
ac
t
i
ons
on I
m
age
P
r
oc
e
s
s
i
ng
, vol
. 31, pp. 275
–
286, 2022, doi
:
10.1109/
T
I
P
.2021.3130545.
[
16]
S
.
G
uo,
Q
.
B
a
i
,
S
.
G
a
o,
Y
.
Z
ha
ng,
a
nd
A
.
L
i
,
“
A
n
a
na
l
y
s
i
s
m
e
t
hod
of
c
r
ow
d
a
bnor
m
a
l
be
ha
vi
or
f
or
vi
de
o
s
e
r
vi
c
e
r
obot
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 7, pp. 169577
–
169585, 2019, doi
:
10.1109/
A
C
C
E
S
S
.2019.295454
4.
[
17]
Z
.
M
a
,
Y
.
L
uo,
C
.
B
.
Y
un,
H
.
P
.
W
a
n,
a
nd
Y
.
S
he
n,
“
A
n
M
P
P
C
A
-
ba
s
e
d
a
ppr
oa
c
h
f
or
a
nom
a
l
y
de
t
e
c
t
i
on
of
s
t
r
uc
t
ur
e
s
unde
r
m
ul
t
i
pl
e
ope
r
a
t
i
ona
l
c
ondi
t
i
ons
a
nd
m
i
s
s
i
ng
da
t
a
,”
St
r
uc
t
u
r
al
H
e
al
t
h
M
oni
t
or
i
ng
,
vol
.
22,
no.
2,
pp.
1069
–
1089,
2023,
doi
:
10.1177/
14759217221100708.
[
18]
T
.
A
.
M
a
ngol
i
,
C
.
S
uj
a
t
ha
,
a
nd
U
.
M
ude
n
a
gudi
,
“
A
nom
a
l
y
de
t
e
c
t
i
on
i
n
s
ur
ve
i
l
l
a
nc
e
vi
de
o
us
i
ng
m
ot
i
on
-
di
r
e
c
t
i
on
m
ode
l
,
”
P
r
oc
e
e
di
ngs
of
2018
2nd
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
A
dv
anc
e
s
i
n
E
l
e
c
t
r
oni
c
s
,
C
om
put
e
r
s
and
C
om
m
uni
c
at
i
on
s
,
I
C
A
E
C
C
2018
,
2018, doi
:
10.1109/
I
C
A
E
C
C
.2018.8479508.
[
19]
M
.
H
a
s
a
n,
J
.
C
hoi
,
J
.
N
e
um
a
nn,
A
.
K
.
R
.
-
C
how
dhur
y,
a
nd
L
.
S
.
D
a
vi
s
,
“
L
e
a
r
ni
ng
t
e
m
por
a
l
r
e
gul
a
r
i
t
y
i
n
vi
de
o
s
e
que
nc
e
s
,
”
P
r
oc
e
e
di
ngs
of
t
h
e
I
E
E
E
C
om
put
e
r
Soc
i
e
t
y
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
pp.
733
–
742,
2016,
doi
:
10.1109/
C
V
P
R
.2016.86.
[
20]
W
.
L
uo,
W
.
L
i
u,
a
nd
S
.
G
a
o,
“
R
e
m
e
m
be
r
i
ng
hi
s
t
or
y
w
i
t
h
c
onvol
ut
i
ona
l
L
S
T
M
f
or
a
nom
a
l
y
de
t
e
c
t
i
on,”
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on M
ul
t
i
m
e
di
a and E
x
po
, pp. 439
–
444, 2017, doi
:
10.1109/
I
C
M
E
.2
017.8019325.
[
21]
D
.
G
.
L
e
e
,
H
.
I
.
S
uk,
S
.
K
.
P
a
r
k,
a
nd
S
.
W
.
L
e
e
,
“
M
ot
i
on
i
nf
l
ue
nc
e
m
a
p
f
or
u
nus
ua
l
hum
a
n
a
c
t
i
vi
t
y
de
t
e
c
t
i
on
a
nd
l
oc
a
l
i
z
a
t
i
on
i
n
c
r
ow
de
d
s
c
e
ne
s
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
C
i
r
c
ui
t
s
and
Sy
s
t
e
m
s
f
o
r
V
i
de
o
T
e
c
h
nol
ogy
,
vol
.
25,
no.
10,
pp.
1612
–
1623,
2015,
doi
:
10.1109/
T
C
S
V
T
.2015.2395752.
[
22]
R
.
T
.
I
one
s
c
u,
S
.
S
m
e
ur
e
a
nu,
B
.
A
l
e
xe
,
a
nd
M
.
P
ope
s
c
u,
“
U
nm
a
s
ki
ng
t
he
a
bnor
m
a
l
e
ve
nt
s
i
n
vi
de
o,”
P
r
oc
e
e
di
ngs
of
t
he
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
, pp. 2914
–
2922, 2017, doi
:
10.1
109/
I
C
C
V
.2017.315.
[
23]
Y
.
S
.
C
hong
a
nd
Y
.
H
.
T
a
y,
“
A
bnor
m
a
l
e
ve
nt
d
e
t
e
c
t
i
on
i
n
vi
de
o
s
us
i
ng s
pa
t
i
ot
e
m
por
a
l
a
ut
oe
nc
ode
r
,”
A
dv
anc
e
s
i
n
N
e
u
r
al
N
e
t
w
o
r
k
s
-
I
SN
N
2017
, pp. 189
–
196, 2017, doi
:
10.1007/
978
-
3
-
319
-
59081
-
3_23.
[
24]
D
.
X
u,
Y
.
Y
a
n,
E
.
R
i
c
c
i
,
a
nd
N
.
S
e
be
,
“
D
e
t
e
c
t
i
ng
a
nom
a
l
ous
e
ve
nt
s
i
n
vi
de
os
by
l
e
a
r
ni
ng
de
e
p
r
e
pr
e
s
e
nt
a
t
i
ons
of
a
ppe
a
r
a
nc
e
a
n
d
m
ot
i
on,”
C
om
put
e
r
V
i
s
i
on and I
m
age
U
nde
r
s
t
andi
ng
, vol
. 156, pp. 117
–
127, 2
017, doi
:
10.1016/
j
.c
vi
u.2016.10.010.
[
25]
T
.
H
a
r
a
da
a
nd
H
.
L
i
u,
“
O
nl
i
ne
gr
ow
i
ng
ne
ur
a
l
ga
s
f
or
a
nom
a
l
y
de
t
e
c
t
i
on
i
n
c
ha
ngi
ng
s
ur
ve
i
l
l
a
nc
e
s
c
e
ne
s
,”
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
vol
. 64, pp. 187
–
201, 2017, doi
:
10.1016/
j
.pa
t
c
og.2016.09.016.
[
26]
W
. L
uo, W
. L
i
u, a
nd S
. G
a
o, “
A
r
e
vi
s
i
t
of
s
pa
r
s
e
c
odi
ng
-
ba
s
e
d a
nom
a
l
y de
t
e
c
t
i
on i
n s
t
a
c
ke
d R
N
N
f
r
a
m
e
w
or
k,”
P
r
oc
e
e
di
ngs
of
t
he
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on C
om
put
e
r
V
i
s
i
on
, pp. 341
–
349, 2017, doi
:
10.1109/
I
C
C
V
.2017.45.
[
27]
W
.
L
i
u,
W
.
L
uo,
D
.
L
i
a
n,
a
nd
S
.
G
a
o,
“
F
ut
ur
e
f
r
a
m
e
p
r
e
di
c
t
i
on
f
o
r
a
nom
a
l
y
de
t
e
c
t
i
on
-
a
ne
w
ba
s
e
l
i
ne
,”
P
r
oc
e
e
di
ngs
of
t
he
I
E
E
E
C
om
put
e
r
Soc
i
e
t
y
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
pp.
6536
–
6545,
2018,
doi
:
10.1109/
C
V
P
R
.2018.00684.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
. 13, No. 4, D
e
c
e
m
be
r
2024:
4121
-
4128
4128
[
28]
Z
.
Y
a
ng,
Y
. L
i
,
J
.
Y
a
ng,
a
nd
J
.
L
uo, “
A
c
t
i
on
r
e
c
ogni
t
i
on
w
i
t
h s
pa
t
i
o
–
t
e
m
por
a
l
v
i
s
ua
l
a
t
t
e
nt
i
on
on s
ke
l
e
t
on
i
m
a
g
e
s
e
que
nc
e
s
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
C
i
r
c
ui
t
s
and
Sy
s
t
e
m
s
f
o
r
V
i
de
o
T
e
c
hnol
ogy
,
vol
.
29,
no.
8,
pp.
2405
–
2415,
2019,
doi
:
10.1109/
T
C
S
V
T
.2018.2864148.
[
29]
G
.
P
a
ng,
C
.
Y
a
n,
C
.
S
he
n,
A
.
V.
D.
H
e
nge
l
,
a
nd
X
.
B
a
i
,
“
S
e
l
f
-
t
r
a
i
ne
d
de
e
p
or
di
na
l
r
e
gr
e
s
s
i
on
f
or
e
nd
-
to
-
e
nd
vi
de
o
a
nom
a
l
y
de
t
e
c
t
i
on,”
P
r
oc
e
e
di
ngs
of
t
he
I
E
E
E
C
om
put
e
r
Soc
i
e
t
y
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
pp.
12170
–
12179,
2020, doi
:
10.1109/
C
V
P
R
42600.2020.01219.
[
30]
B
.
R
a
m
a
c
ha
ndr
a
,
M
.
J
.
J
one
s
,
a
nd
R
.
R
.
V
a
t
s
a
va
i
,
“
L
e
a
r
ni
ng
a
di
s
t
a
nc
e
f
unc
t
i
on
w
i
t
h
a
S
i
a
m
e
s
e
ne
t
w
or
k
t
o
l
oc
a
l
i
z
e
a
nom
a
l
i
e
s
i
n
vi
de
os
,”
P
r
oc
e
e
di
ng
s
-
2020
I
E
E
E
W
i
nt
e
r
C
onf
e
r
e
nc
e
on
A
ppl
i
c
at
i
ons
of
C
om
p
ut
e
r
V
i
s
i
on,
W
A
C
V
2020
,
pp.
2587
–
2596,
2020,
doi
:
10.1109/
W
A
C
V
45572.2020.9093417.
[
31]
B
.
R
a
m
a
c
ha
ndr
a
a
nd
M
.
J
.
J
one
s
,
“
S
t
r
e
e
t
s
c
e
ne
:
a
ne
w
da
t
a
s
e
t
a
nd
e
va
l
ua
t
i
on
pr
ot
oc
ol
f
or
vi
de
o
a
nom
a
l
y
de
t
e
c
t
i
on,”
P
r
oc
e
e
di
ngs
-
2020
I
E
E
E
W
i
nt
e
r
C
onf
e
r
e
nc
e
on
A
ppl
i
c
at
i
ons
of
C
om
put
e
r
V
i
s
i
on,
W
A
C
V
2020
,
pp.
2558
–
2567,
2020,
doi
:
10.1109/
W
A
C
V
45572.2020.9093457.
[
32]
M
.
I
.
G
e
or
ge
s
c
u,
A
.
B
a
r
ba
l
a
u,
R
.
T
.
I
one
s
c
u,
F
.
S
.
K
ha
n,
M
.
P
ope
s
c
u,
a
nd
M
.
S
h
a
h,
“
A
nom
a
l
y
de
t
e
c
t
i
on
i
n
vi
de
o
vi
a
s
e
l
f
-
s
upe
r
vi
s
e
d
a
nd
m
ul
t
i
-
t
a
s
k
l
e
a
r
ni
ng,”
P
r
o
c
e
e
di
ngs
of
t
he
I
E
E
E
C
om
put
e
r
Soc
i
e
t
y
C
onf
e
r
e
nc
e
on C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
pp
.
12737
–
12747, 2021, doi
:
10.1109/
C
V
P
R
46437.2021.01255.
[
33]
S
. Z
ha
ng
e
t
al
.
, “
I
nf
l
ue
nc
e
-
a
w
a
r
e
a
t
t
e
nt
i
on ne
t
w
or
ks
f
or
a
nom
a
l
y de
t
e
c
t
i
on i
n s
ur
ve
i
l
l
a
nc
e
vi
de
os
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on C
i
r
c
ui
t
s
and Sy
s
t
e
m
s
f
o
r
V
i
de
o T
e
c
hnol
ogy
, vol
. 32, no. 8, pp. 5427
–
5437, 2022, doi
:
10.1109/
T
C
S
V
T
.2022.3148392.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
Mrs. Kus
uma
Sriram
r
eceived B.
E
.
degree under
VTU in 2004,
an
d M.Tech
.
from
RVCE in the
year
2010. She
is an
active
acade
mician
having 1
8
+
yea
rs of te
aching e
xperie
nce
.
Currentl
y
she
is
working
at
Ramaiah
Institute
of
Technol
ogy,
Be
ngaluru,
as
an
Assistant
Profes
sor
in
the
Department
of
Information
Science
and
Engineering.
She
is
p
ursuing
Ph.D
.
in
the
area
of
video
processing/machine
learning
.
She
is
a
trained
Fac
ulty
of
Infosys
InfyTQ
t
o
motivate
students
for
i
nfosys
job
opportunities.
She
has
worked
as
tr
ainer
for
infosys
campus
connect pro
gram. Sh
e can be con
tacted at
email:
kusumas
_12@
rediffmail
.com.
Dr.
Kiran
Purusotham
currently
working
as
HOD,
Departme
nt
of
Compute
r
Scienc
e
and
Engine
ering
.
with
total
experience
of
20
years.
He
has
done
research
on
privacy
preserving
data
mining
with
focus
on
detection
of
sensitivity
patterns
and
was
awarded
Ph.D
.
from
VTU
in
2014.
His
research
interests
include
cryptograp
hy,
rand
omizati
on,
anonymi
zation
methods
in
generaliz
ation,
indexing
techniques
and
design
patterns
.
He
has
guided
several
M.Tech
.
and B.E
.
projects &
internships
and is cu
rrently g
uiding
four
research scho
lars at V
TU.
He can be contacted at email
:
kiranpmys@
gmail.com.
Evaluation Warning : The document was created with Spire.PDF for Python.