I
n
d
o
n
e
s
i
a
n
J
o
u
r
n
a
l
o
f
E
l
e
c
tr
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
a
n
d
C
o
m
p
u
te
r
S
c
i
e
n
c
e
V
ol
.
7,
N
o.
3,
S
ept
em
ber
201
7,
pp
.
737
~
747
D
O
I
:
10.
115
91/
i
j
eec
s
.
v
7
.
i
3
.
pp
73
7
-
74
7
7
37
R
ec
ei
v
ed
M
ay
19
,
201
7
;
R
ev
i
s
ed
J
u
l
y
28,
201
7
;
A
c
c
ept
ed
A
u
gus
t
1
3,
2
017
Cr
o
w
d
A
n
o
m
al
y
Det
ec
t
io
n
Usi
n
g
M
o
t
i
o
n
Base
d
S
pa
ti
o
-
T
em
p
o
r
al F
eat
u
r
e
A
n
al
y
si
s
B
asav
ar
aj
G
M
*
1
,
A
s
h
o
k
K
u
s
a
g
u
r
2
1
D
epar
t
m
ent
of
E
l
ec
t
r
oni
c
s
an
d C
om
m
un
i
c
at
i
on
E
ngi
neer
i
ng
N
agar
j
u
na
C
ol
l
e
ge
o
f
E
ngi
nee
r
i
ng
a
nd
T
ec
h
nol
ogy
,
B
eng
al
ur
u,
K
ar
na
t
ak
a,
I
n
di
a
2
D
epar
t
m
ent
of
E
l
ec
t
r
i
c
al
a
nd
E
l
ec
t
r
oni
c
s
E
ngi
n
eer
i
ng
U
n
i
v
e
r
s
i
ty
B
.D
.T
C
ol
l
e
ge
of
E
n
gi
neer
i
ng
(
U
B
D
T
C
E
)
,
D
av
anager
e,
K
ar
nat
a
k
a,
I
nd
i
a
*
C
or
r
es
po
ndi
ng a
ut
hor
,
e
-
m
a
i
l:
bas
w
ar
aj
8
32@
g
m
ai
l
.
c
o
m
,
as
hok
.
k
u
s
agur
@
gm
a
i
l
.
c
om
A
b
st
r
act
R
ec
ent
l
y
,
t
h
e dem
and f
or
s
ur
v
ei
l
l
anc
e s
y
s
t
em
i
s
i
n
c
r
ea
s
i
n
g i
n r
eal
t
i
m
e appl
i
c
a
t
i
o
n t
o enha
nc
e t
he
s
ec
ur
i
t
y
s
y
s
t
em
.
T
he
s
e s
ur
v
e
i
l
l
anc
e s
y
s
t
em
s
ar
e m
a
i
nl
y
u
s
e
d i
n
c
r
ow
de
d pl
a
c
e
s
s
uc
h
as
s
hop
pi
n
g m
al
l
s
,
s
por
t
s
s
t
adi
um
et
c
.
I
n or
der
t
o s
up
por
t
e
nhan
c
e t
he s
ec
u
r
i
t
y
s
y
s
t
em
,
c
r
ow
d b
eha
v
i
or
an
al
y
s
i
s
ha
s
bee
n
pr
ov
e
n a
s
i
gn
i
f
i
c
ant
t
e
c
hn
i
qu
e w
hi
c
h
i
s
u
s
ed
f
or
c
r
ow
d
m
oni
t
or
i
ng,
v
i
s
ua
l
s
ur
v
ei
l
l
an
c
e et
c
.
F
or
c
r
ow
d
beha
v
i
or
ana
l
y
s
i
s
,
m
ot
i
o
n an
al
y
s
i
s
i
s
a
c
r
u
c
i
a
l
t
a
s
k
w
hi
c
h c
a
n be
ac
h
i
ev
ed w
i
t
h t
h
e he
l
p
of
t
r
aj
e
c
t
or
i
es
and
t
r
ac
k
i
ng of
ob
j
e
c
t
s
.
V
ar
i
ou
s
appr
oa
c
h
e
s
ha
v
e bee
n pr
op
os
ed f
or
c
r
ow
d beh
av
i
or
ana
l
y
s
i
s
w
hi
c
h ha
s
l
i
m
i
t
at
i
on
f
or
den
s
el
y
c
r
ow
d
ed
s
c
enar
i
os
,
a
new
ob
j
ec
t
e
nt
er
i
ng
t
he
s
c
ene
et
c
.
I
n
t
hi
s
w
or
k
,
w
e
pr
opo
s
e
a
new
appr
o
ac
h f
or
abn
or
m
al
c
r
ow
d be
hav
i
or
de
t
ec
t
i
on
.
P
r
opos
ed ap
pr
oa
c
h i
s
a m
ot
i
o
n bas
e
d s
p
a
ti
o
-
t
em
por
al
f
e
at
ur
e
a
nal
y
s
i
s
t
e
c
h
ni
que
w
h
i
c
h
i
s
c
apab
l
e
of
o
bt
a
i
ni
n
g
t
r
aj
e
c
t
or
i
es
of
ea
c
h
det
ec
t
ed
obj
ec
t
.
W
e
al
s
o
pr
es
ent
a
t
ec
h
ni
q
ue
t
o
c
ar
r
y
o
ut
t
he
ev
a
l
ua
t
i
on
of
i
ndi
v
i
d
ual
o
bj
e
c
t
an
d
gr
ou
p
of
obj
e
c
t
s
by
c
on
s
i
der
i
ng
r
el
at
i
o
nal
des
c
r
i
pt
or
s
b
as
e
d
on
t
h
e
i
r
en
v
i
r
onm
e
nt
al
c
ont
ex
t
.
F
i
nal
l
y
,
a
c
l
as
s
i
f
i
c
at
i
on
i
s
c
ar
r
i
ed
out
f
or
det
e
c
t
i
on of
ab
nor
m
al
or
nor
m
al
c
r
ow
d be
ha
v
i
or
b
y
f
ol
l
ow
i
ng pat
c
h ba
s
ed pr
oc
e
s
s
.
I
n t
he r
es
u
l
t
s
,
w
e hav
e r
epor
t
ed t
hat
pr
opo
s
ed m
odel
i
s
abl
e
t
o a
c
hi
ev
e
b
et
t
er
p
er
f
or
m
an
c
e w
he
n c
om
p
ar
e
d
t
o ex
i
s
t
i
ng
t
ec
h
ni
qu
es
i
n t
er
m
s
of
c
l
as
s
i
f
i
c
at
i
on
ac
c
ur
a
c
y
,
t
r
ue
po
s
i
t
i
v
e r
a
t
e,
an
d f
a
l
s
e
po
s
i
t
i
v
e r
at
e.
Ke
y
w
o
rd
s
:
S
p
at
i
o
-
t
em
por
al
,
v
i
s
ua
l
s
ur
v
e
i
l
l
an
c
e,
c
r
ow
d
beha
v
i
or
anal
y
s
i
s
,
S
t
r
eam
l
i
ne,
S
t
r
e
ak
l
i
ne
C
o
p
y
r
i
g
h
t
©
2
01
7
I
n
s
t
i
t
u
t
e
o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
r
i
n
g
a
n
d
S
c
i
e
n
c
e
.
A
l
l
r
i
g
h
t
s r
es
er
ved
.
1.
I
n
tr
o
d
u
c
ti
o
n
R
ec
ent
l
y
,
v
i
deo s
ur
v
ei
l
l
anc
e t
ec
hno
l
o
g
y
h
as
gr
o
w
n d
ue t
o i
t
s
i
m
por
t
anc
e f
or
s
ec
ur
i
t
y
r
equi
r
em
ent
a
nd
det
ec
t
i
o
n
of
ev
e
nt
s
i
n
p
ub
l
i
c
p
l
ac
es
s
uc
h
as
s
t
r
eet
s
,
s
h
opp
i
n
g
m
al
l
s
,
s
ub
w
a
y
s
t
at
i
o
n
et
c
.
V
i
de
o
s
ur
v
e
i
l
l
a
nc
e
s
y
s
t
em
i
s
w
i
de
l
y
us
ed
t
o
m
oni
t
or
t
he
c
r
o
w
d
ac
t
i
v
i
t
i
es
dur
i
ng
a
n
y
pub
l
i
c
ev
ent
s
.
A
ut
om
at
ed an
al
y
s
i
s
a
nd
det
ec
t
i
o
n
of
anom
al
i
es
i
n
c
r
o
w
d
a
c
t
i
v
i
t
i
es
i
s
a
c
hal
l
eng
i
n
g
i
s
s
ue
i
n
v
i
de
o s
ur
v
ei
l
l
anc
e s
y
s
t
em
.
A
c
c
or
di
ng
t
o t
he s
t
ud
y
pr
es
ent
e
d
i
n
[
1]
,
abnor
m
al
ac
t
i
v
i
t
i
es
ar
e
det
er
m
i
ned
bas
ed on
d
ev
i
at
i
o
n f
r
o
m
t
he
nor
m
al
or
abnor
m
al
s
t
andar
d
.
B
y
c
o
ns
i
d
er
i
n
g t
h
i
s
i
nt
er
pr
et
at
i
on,
abn
or
m
al
ac
t
i
v
i
t
i
es
ar
e
def
i
n
ed b
as
ed o
n
t
h
e
dev
i
at
i
o
n
f
r
o
m
nor
m
al
ac
t
i
v
i
t
i
es
.
I
n
r
e
al
-
w
or
l
d
s
c
enar
i
os
s
uc
h
a
s
pedes
t
r
i
an
w
al
k
i
ng,
s
ubj
ec
t
s
f
o
llo
w
nei
ghb
or
i
n
g s
ubj
ec
t
s
ai
m
i
ng f
or
t
he s
a
m
e des
t
i
nat
i
o
n
.
A
bn
or
m
al
c
r
ow
d ac
t
i
v
i
t
i
e
s
af
f
ec
t
publ
i
c
s
af
et
y
s
uc
h as
a
n ex
p
l
os
i
o
n,
f
i
r
e,
d
i
s
as
t
er
s
et
c
.
T
hi
s
t
ec
hni
qu
e
of
c
r
ow
d
be
hav
i
or
a
nal
y
s
i
s
i
s
w
i
de
l
y
us
ed
i
n
r
e
al
-
t
i
m
e
appl
i
c
at
i
ons
s
uc
h
as
:
(
a)
V
is
u
a
l s
u
r
v
ei
l
l
anc
e
C
r
ow
d b
eha
v
i
or
s
ana
l
y
s
i
s
i
s
appl
i
c
ab
l
e f
or
v
ar
i
o
us
v
i
s
ual
s
ur
v
e
i
l
l
anc
e
s
c
enar
i
os
s
uc
h as
s
hopp
i
ng
m
al
l
s
,
r
ai
l
w
a
y
s
t
a
t
i
ons
et
c
.
C
on
v
ent
i
o
na
l
m
et
hods
ar
e
n
ot
c
a
pab
l
e
of
pr
ov
i
di
ng
bet
t
er
ef
f
i
c
i
enc
y
du
e t
o
a h
uge
de
ns
i
t
y
of
t
he c
r
o
w
d
.
(
b)
C
r
ow
d m
anagem
en
t
T
hi
s
appr
oac
h
i
s
ut
i
l
i
z
e
d
f
o
r
a
m
as
s
gat
her
i
n
g
of
a
c
r
o
w
d
s
uc
h
as
s
por
t
s
ev
ent
,
m
u
s
i
c
f
es
t
i
v
al
et
c
.
U
s
i
ng
t
h
i
s
a
pp
r
oac
h,
t
h
e
gat
h
er
i
n
g ar
e
as
c
an b
e a
na
l
y
z
e
d
a
nd
t
he
c
r
ow
d
c
an
be
as
s
i
s
t
ed f
or
m
ov
em
ent
.
(c
)
P
ub
l
i
c
s
p
ac
e des
i
gn
C
r
ow
d
beh
av
i
or
an
al
y
s
i
s
c
an
be
us
ed
des
i
gn
i
ng
p
ubl
i
c
s
pac
es
dur
i
ng
m
ov
em
ent
of
t
he
c
r
ow
d
i
n
a r
a
i
l
w
a
y
s
t
at
i
on,
s
hopp
i
n
g m
al
l
s
t
o
i
nc
r
eas
e t
he s
af
et
y
and
ef
f
i
c
i
enc
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
7
,
N
o.
3,
S
e
pt
em
ber
2017
:
737
–
7
47
738
D
ur
i
n
g l
as
t
dec
ad
e,
v
ar
i
o
u
s
m
et
hods
hav
e b
een d
e
v
e
l
op
ed t
o ad
dr
es
s
t
he i
s
s
ue of
abnor
m
al
c
r
o
w
d b
eha
v
i
or
.
I
n or
der
t
o a
n
a
l
y
z
e
abn
or
m
al
c
r
ow
d b
eha
v
i
or
,
c
r
o
w
d
m
odel
i
ng
i
s
a
c
hal
l
eng
i
n
g t
as
k
f
or
r
es
ear
c
her
s
.
W
i
t
h t
he
hel
p of
c
r
ow
d m
odel
i
ng,
be
ha
v
i
or
pr
edi
c
t
i
on
an
d
gr
oup f
or
m
at
i
on of
peopl
e
c
an be an
al
y
z
e
d e
as
i
l
y
w
hi
c
h ut
i
l
i
z
es
i
nf
or
m
at
i
on
r
el
at
e
d t
o t
h
e
geogr
aph
i
c
al
or
l
ogi
c
a
l
s
t
at
e of
af
f
ec
t
ed r
egi
on.
I
n [
2]
,
an ef
f
i
c
i
ent
m
odel
i
s
pr
es
ent
f
or
c
r
ow
d
beha
v
i
or
m
odel
i
ng
b
y
pr
op
os
i
ng
bo
t
t
om
-
up m
et
hodol
og
y
.
S
i
m
i
l
ar
l
y
,
dec
i
s
i
on
m
a
k
i
ng
f
or
c
r
ow
d
abnor
m
al
i
t
y
i
s
a
c
r
uc
i
al
t
as
k
.
I
n [
3]
,
B
r
aun
e
t
al
.
ad
dr
es
s
ed
t
hi
s
i
s
s
ue
w
i
t
h
t
he h
el
p
of
H
e
lb
in
g
m
et
hod.
T
hi
da
et
al
[
4]
ad
dr
es
s
ed t
h
e
i
s
s
ue of
c
r
o
w
d l
oc
al
i
z
at
i
on
and
d
et
ec
t
i
o
n of
ab
nor
m
al
c
r
ow
d
ac
t
i
v
i
t
i
es
.
R
ec
e
nt
l
y
,
R
ao
et
al
[
5]
de
v
e
l
o
ped
a
ne
w
m
odel
f
or
abn
or
m
al
c
r
ow
d b
eha
v
i
or
det
ec
t
i
on
bas
e
d
o
n
t
he
m
ot
i
on
pat
t
er
n
ex
t
r
ac
t
i
o
n.
T
hi
s
w
or
k
m
ai
n
l
y
a
im
s
a
t
o
p
t
ic
a
l
f
lo
w
f
r
a
m
ew
or
k
f
or
m
ot
i
on anal
y
s
i
s
.
F
ur
t
her
m
or
e,
t
hi
s
appr
oac
h i
s
ex
t
ende
d t
o obt
ai
n
a c
l
as
s
i
f
i
c
at
i
on
of
m
er
gi
ng an
d s
pl
i
t
t
i
ng
o
f
t
he c
r
ow
d.
A
n
ot
h
er
s
i
m
i
l
ar
appr
o
ac
h f
or
c
l
as
s
i
f
i
c
at
i
on of
m
ov
i
ng
obj
ec
t
i
s
c
ar
r
i
e
d
ou
t
w
i
t
h
t
h
e
he
l
p
of
opt
i
c
a
l
f
l
o
w
m
eas
ur
em
ent
.
A
c
c
or
di
n
g
t
o
t
hi
s
m
et
hodol
o
g
y
,
obj
ec
t
s
egm
ent
at
i
on
and
t
r
ac
k
i
ng ar
e p
er
f
or
m
ed b
y
a
ppl
y
i
ng
bac
k
gr
ound s
u
bt
r
a
c
t
i
on,
K
a
l
m
an
f
i
l
t
er
,
s
i
l
ho
uet
t
e c
om
put
at
i
on
an
d c
or
r
el
at
i
on
c
om
put
at
i
o
n b
et
w
e
en
ap
pear
anc
e m
odel
s
.
S
egm
ent
ed
o
bj
ec
t
s
ar
e
c
l
as
s
i
f
i
ed
b
y
a
pp
l
y
i
ng
a
c
o
m
bi
nat
i
on
of
s
pat
i
ot
em
por
al
an
d
s
t
at
i
c
f
eat
ur
es
b
y
ut
i
l
i
z
i
n
g
app
ear
anc
e
an
d
m
ot
i
on
pat
t
er
n
m
ode
l
w
her
e
s
pa
t
i
o
-
t
em
por
al
f
eat
ur
es
ar
e
c
o
m
put
ed
us
i
ng
ada
pt
i
v
e
b
l
oc
k
gr
adi
ent
i
n
t
ens
i
t
i
es
an
d
H
oG
f
eat
ur
es
.
B
ut
t
hes
e
s
t
udi
es
s
uf
f
er
f
r
om
t
h
e
i
s
s
ue
of
obj
ec
t
boundar
i
es
and
a
per
t
ur
e
pr
o
bl
em
w
hi
c
h
d
egr
ad
es
t
he
per
f
or
m
anc
e
o
f
ana
l
y
s
i
s
.
Z
h
ou
e
t
al
[
7]
pr
es
ent
e
d
a
n
e
w
m
odel
f
or
c
r
ow
d
beh
av
i
or
a
na
l
y
s
i
s
b
y
u
t
i
l
i
z
i
ng
pedes
t
r
i
an
t
r
aj
ec
t
or
i
es
f
r
ag
m
ent
.
T
hi
s
m
odel
s
uf
f
er
s
f
r
om
t
he
i
s
s
ue
of
e
f
f
i
c
i
ent
t
r
a
c
k
i
ng
c
aus
ed
b
y
i
nc
om
pl
et
e
i
nf
or
m
at
i
on
abou
t
f
r
agm
ent
t
r
aj
ec
t
or
i
es
,
m
i
s
s
i
ng
obj
ec
t
t
r
aj
ec
t
or
i
es
,
ent
er
i
ng
of
a
ne
w
o
bj
ec
t
i
n t
h
e s
c
ene
e
t
c
.
S
t
i
l
l
,
t
h
er
e ar
e
v
ar
i
o
us
c
hal
l
eng
es
pr
es
en
t
i
n t
h
i
s
f
i
el
d of
c
r
o
w
d
beha
v
i
or
de
t
ec
t
i
on,
t
r
ac
k
i
ng,
an
d ac
t
i
v
i
t
y
c
l
as
s
i
f
i
c
at
i
o
n
.
C
r
o
w
de
d s
c
enes
ar
e ex
t
r
em
el
y
c
l
ut
t
er
ed
and i
nd
uc
ed
b
y
v
ar
i
ous
oc
c
l
us
i
ons
w
her
e
c
o
nv
ent
i
on
al
m
et
hods
c
an
not
pr
ov
i
d
e
t
he
s
i
gn
i
f
i
c
ant
per
f
or
m
anc
e of
det
ec
t
i
on a
nd c
l
as
s
i
f
i
c
at
i
on.
A
c
c
or
d
i
n
g t
o
l
i
t
er
at
ur
e
s
t
ud
y
pr
es
e
nt
ed b
y
A
l
i
i
n [
8]
,
it
is
c
o
n
c
lu
d
e
d t
h
at
n
at
ur
e
of
hum
an c
r
ow
d i
s
c
om
pl
ex
w
h
i
c
h c
ons
i
s
t
s
of
ps
y
c
hol
ogi
c
a
l
a
nd
d
y
n
am
i
c
c
har
ac
t
er
i
s
t
i
c
s
.
T
hi
s
na
t
ur
e
of
hum
an c
r
ow
d
m
a
k
es
i
s
m
or
e c
o
m
pl
ex
t
o es
t
i
m
at
e t
h
e
gr
anu
l
ar
i
t
y
l
e
v
e
l
f
or
t
he
d
y
n
am
i
c
c
r
ow
d.
2.
R
el
at
ed
W
o
r
k
I
n t
hi
s
w
or
k
,
our
m
ai
n ai
m
i
s
t
o a
ddr
es
s
t
he i
s
s
ue of
an
al
y
z
i
ng c
r
o
w
d b
eha
v
i
or
i
n v
ar
i
ou
s
s
ur
v
ei
l
l
anc
e s
c
enar
i
os
.
I
n o
r
der
t
o per
f
or
m
t
hi
s
,
w
e
ha
v
e de
v
el
ope
d a r
ob
us
t
appr
o
ac
h bas
ed o
n
s
e
m
at
i
c
c
onc
ept
w
hi
c
h
i
s
a
bl
e
t
o
pr
o
v
i
de
a
r
e
l
at
i
ons
h
i
p
bet
w
ee
n
pe
op
l
e
pr
es
ent
i
n
t
he
c
r
o
w
ded
s
c
ene dep
en
d
i
n
g up
on t
h
ei
r
en
v
i
r
o
nm
ent
al
c
ont
ex
t
.
I
n t
he pr
o
pos
e
d m
odel
,
t
he des
c
r
i
p
t
or
bas
ed m
odel
i
s
a
l
s
o de
v
e
l
o
ped
w
h
i
c
h r
es
ul
t
s
i
n i
nf
or
m
at
i
o
n ex
t
r
ac
t
i
on be
t
w
ee
n i
n
di
v
i
d
ua
l
obj
ec
t
and s
c
ene
.
T
he nov
el
t
y
of
pr
opos
ed
appr
oac
h i
s
t
ha
t
i
t
i
s
c
ap
abl
e of
ex
t
r
ac
t
i
n
g
t
he
r
el
at
i
ona
l
f
eat
ur
e
i
n
a
n
aut
om
at
i
c
pr
o
c
edur
e
w
i
t
hou
t
per
f
or
m
i
ng
an
y
m
anual
ann
ot
at
i
ons
.
A
c
l
as
s
i
f
i
c
at
i
on
m
odel
i
s
dev
e
l
o
ped b
as
ed
on t
he pa
t
c
hes
of
es
t
i
m
at
ed t
r
aj
ec
t
or
y
a
nd
m
ot
i
on
p
ar
am
et
er
s
w
i
t
h
t
he h
el
p of
s
pat
i
ot
em
por
al
f
eat
ur
e
ex
t
r
ac
t
i
o
n
w
hi
c
h
pr
o
v
i
de
s
be
t
t
er
p
er
f
or
m
anc
e f
or
c
l
as
s
i
f
i
c
at
i
on
of
c
r
ow
d b
eh
av
i
or
a
na
l
y
s
i
s
.
R
es
t
of
t
he ar
t
i
c
l
es
i
s
or
g
ani
z
e
d as
f
ol
l
o
w
s
:
a
br
i
ef
l
i
t
er
at
ur
e r
e
v
i
e
w
i
s
pr
es
en
t
ed
i
n
s
ec
t
i
on
I
I
,
pr
opos
e
d
appr
oa
c
h
i
s
depi
c
t
ed
i
n
s
ec
t
i
on
I
I
I
and
s
ec
t
i
on
I
V
pr
ov
i
d
es
an
ex
per
i
m
ent
al
st
ud
y
of
pr
opos
e
d m
odel
a
nd f
i
na
l
l
y
c
onc
l
ud
i
n
g r
em
ar
k
s
ar
e pr
es
ent
e
d i
n s
ec
t
i
on
V
.
T
hi
s
s
ec
t
i
on pr
ov
i
des
a br
i
e
f
s
t
ud
y
of
r
ec
ent
t
ec
hn
i
qu
e
s
w
hi
c
h ar
e de
v
el
ope
d r
ec
e
nt
l
y
i
n
t
he f
i
el
d of
c
r
ow
d beh
av
i
or
ana
l
y
s
i
s
.
S
.
Y
i
e
t
al
[
9
]
de
v
el
o
ped a n
e
w
m
odel
f
or
c
r
o
w
d
m
odel
i
ng i
n
v
i
deo s
ur
v
ei
l
l
anc
e s
y
s
t
em
s
f
or
pedes
t
r
i
an
w
a
l
k
i
ng s
c
enar
i
o.
F
or
s
t
at
i
o
nar
y
c
r
o
w
d s
c
enar
i
os
,
w
al
k
i
ng
obj
ec
t
s
ar
e
i
gnor
e
d
w
h
i
c
h
i
s
a
n af
f
ec
t
i
ng
par
am
et
er
f
or
c
r
ow
d
beh
av
i
or
ana
l
y
s
i
s
.
T
hi
s
w
or
k
pr
opos
es
a
nov
e
l
m
o
del
f
or
pedes
t
r
i
an
b
eha
v
i
or
anal
y
s
i
s
.
I
n
[
10]
a
n
e
w
m
et
ho
d
ex
pl
or
ed
f
or
c
r
ow
d beh
av
i
or
an
al
y
s
i
s
b
y
i
nc
or
p
or
at
i
ng
v
i
r
t
u
al
e
nv
i
r
onm
ent
s
w
i
t
hi
n i
nf
or
m
at
i
on s
p
ac
e of
c
o
m
put
at
i
on.
T
hi
s
w
or
k
r
epr
es
ent
s
be
ha
v
i
or
m
i
ni
ng
a
ppr
oac
h
f
or
c
r
ow
d
be
ha
v
i
o
r
anal
y
s
i
s
an
d
dea
l
s
w
i
t
h c
r
o
w
d m
er
gi
ng
a
nd s
pl
i
t
t
i
ng
s
c
enar
i
o.
I
n [
11]
C
hen
et
al
.
pr
op
os
ed
a no
v
e
l
a
l
gor
i
t
hm
f
or
c
r
o
w
d
be
ha
v
i
or
det
ec
t
i
on
ba
s
ed on
ac
c
el
er
at
i
on
f
eat
ur
es
of
t
he
c
r
ow
d.
U
nl
i
k
e
s
t
at
e
-
of
-
ar
t
t
ec
hni
ques
,
t
hi
s
w
or
k
us
es
l
oc
al
f
eat
ur
es
and r
e
l
at
i
o
ns
hi
p
i
s
ex
t
r
ac
t
ed b
et
w
e
en t
he c
ur
r
en
t
a
nd pr
e
v
i
ous
s
t
at
e of
be
ha
v
i
or
f
or
ac
t
i
v
i
t
y
ana
l
y
s
i
s
.
S
i
nc
e
t
h
i
s
m
et
hod
us
es
opt
i
c
a
l
f
l
o
w
bas
ed
c
om
put
at
i
on
ap
pr
oac
h
w
hi
c
h
c
aus
es
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
C
r
ow
d
A
no
ma
l
y
D
et
e
c
t
i
on
U
s
i
ng Mot
i
on
B
as
e
d S
pat
i
o
…
(
B
as
av
ar
a
j
G
M
)
739
i
ns
t
ab
i
l
i
t
y
i
n f
eat
ur
e a
nal
y
s
i
s
.
F
or
det
ec
t
i
o
n,
f
or
egr
ou
n
d ex
t
r
ac
t
i
on
bas
ed m
odel
i
s
dev
el
ope
d b
y
us
i
ng
ac
c
el
er
at
i
on c
om
put
at
i
o
n m
et
ho
d.
F
or
an
om
al
y
d
et
ec
t
i
on,
obj
ec
t
s
e
gm
ent
at
i
on
i
s
a
l
s
o
app
l
i
e
d
i
n
t
hi
s
w
or
k
w
hi
c
h
m
a
k
e
s
i
t
m
or
e
r
obus
t
f
or
r
eal
t
i
m
e
s
c
enar
i
os
an
d
f
i
na
l
l
y
c
l
as
s
i
f
i
c
at
i
o
n
of
ac
t
i
v
i
t
i
es
i
s
c
ar
r
i
ed
out
w
i
t
h t
h
e he
l
p
of
t
hr
es
ho
l
d a
na
l
y
s
i
s
m
et
hod.
C
.
Y
.
C
h
en
et
al
[
1
2]
d
i
s
c
us
s
ed
t
he
ab
nor
m
al
c
r
ow
d
beha
v
i
or
an
al
y
s
i
s
m
et
hod.
I
n
t
hi
s
w
or
k
behav
i
or
,
det
ec
t
i
o
n a
nd l
oc
al
i
z
at
i
on
c
ar
r
i
ed
out
b
y
us
i
n
g d
i
v
er
gen
t
c
ent
er
s
of
an
y
v
i
d
eo
s
ur
v
ei
l
l
anc
e s
y
s
t
em
.
F
or
m
ot
i
o
n an
al
y
s
i
s
,
a
w
el
l
-
k
now
n t
ec
hn
i
qu
e i
s
us
e
d k
now
n
as
opt
i
c
a
l
f
l
o
w
es
t
i
m
at
i
on.
T
he m
ot
i
on of
gi
v
en s
e
que
nc
e i
s
m
odel
e
d b
y
obt
ai
ni
n
g m
agni
t
u
de,
di
r
ec
t
i
on a
n
d
pos
i
t
i
on of
det
ec
t
ed obj
e
c
t
s
and v
e
l
oc
i
t
y
ar
e c
om
put
e
d b
y
i
nc
or
por
at
i
ng m
ot
i
o
n v
e
l
oc
i
t
y
c
o
m
put
at
i
on.
H
.
S
.
W
ong
et
al
[
13]
d
ev
e
l
op
ed
a
n
e
w
f
r
a
m
ew
or
k
f
or
anom
al
y
det
ec
t
i
o
n
bas
e
d
on
t
he
B
a
y
es
i
a
n
m
odel
.
T
hi
s
w
or
k
m
ai
nl
y
a
i
m
s
of
es
c
ape
det
ec
t
i
on
i
n
t
he
c
r
ow
d
b
y
p
er
f
or
m
i
ng
c
r
ow
d
m
odel
i
ng.
T
hi
s
m
et
hod
us
e
s
a
s
i
m
i
l
ar
c
onc
ept
of
di
v
er
gent
c
e
nt
er
as
di
s
c
us
s
ed
b
ef
or
e
w
hi
c
h
i
s
us
ed f
or
m
ot
i
on c
h
ar
ac
t
er
i
z
at
i
on
and
a
pr
ob
abi
l
i
t
y
dens
i
t
y
f
unc
t
i
on
i
s
f
or
m
ul
at
ed
bas
e
d o
n
opt
i
c
a
l
f
l
o
w
c
om
put
at
i
on
.
A
s
w
e
ha
v
e
di
s
c
us
s
ed
bef
o
r
e t
hat
s
pat
i
ot
em
por
al
f
eat
ur
e has
be
en
pr
ov
en
a s
i
g
ni
f
i
c
ant
t
ec
hni
que
f
or
c
r
ow
d
beh
av
i
or
an
al
y
s
i
s
b
y
es
t
i
m
at
i
on m
ot
i
o
n f
l
o
w
.
B
y
t
ak
i
ng t
hi
s
i
nt
o
c
ons
i
der
a
t
i
o
n,
Z
hen
g e
t
a
l
[
14]
de
v
e
l
op
ed
a
ne
w
m
et
hod
t
o m
odel
t
h
e c
r
o
w
d
m
ot
i
o
n p
at
t
er
ns
w
i
t
h
t
h
e he
l
p of
s
pat
i
o
-
t
em
por
al
v
i
s
c
ous
f
l
ui
d f
i
el
d
f
ea
t
ur
e
c
om
put
at
i
on.
T
hi
s
he
l
ps
t
o
c
o
m
put
e
and
es
t
i
m
at
e t
he
m
ot
i
on f
or
dens
e
l
y
c
r
o
w
d
ed s
c
en
a
r
i
os
.
A
c
c
or
d
i
ng
t
o
t
hi
s
m
odel
,
t
h
e s
pat
i
o
-
t
em
por
al
m
a
t
r
i
x
i
s
c
o
m
put
ed f
i
r
s
t
w
hi
c
h pr
o
v
i
des
t
he
m
eas
ur
e
m
ent
of
l
oc
al
f
l
uc
t
uat
i
ons
i
n t
h
e
v
i
deo d
at
a
b
y
c
o
ns
i
der
i
ng
s
pat
i
a
l
an
d t
em
por
al
d
om
ai
n.
A
f
t
er
c
om
put
i
ng t
he s
pat
i
o
-
t
em
por
al
m
at
r
i
x
,
f
l
ui
d f
i
e
l
d s
c
hem
e i
s
appl
i
e
d t
o
ex
t
r
ac
t
t
he
i
n
f
or
m
at
i
on bas
e
d on
ei
gen
v
a
l
ue ana
l
y
s
i
s
m
et
hod and f
i
nal
l
y
,
a c
od
eb
ook
i
s
c
ons
t
r
uc
t
ed b
y
ut
i
l
i
z
i
ng c
l
us
t
er
i
ng ap
pr
oac
h i
n s
pat
i
o
-
t
em
por
al
f
eat
ur
e s
i
m
i
l
ar
i
t
y
.
T
he c
l
as
s
i
f
i
c
at
i
o
n m
odel
i
s
de
v
el
ope
d b
y
ap
pl
y
i
ng D
i
r
i
c
h
l
et
m
od
el
.
S
t
i
l
l
,
t
h
e v
ar
i
ous
c
hal
l
e
ng
i
n
g
t
as
k
i
s
pr
es
ent
i
n
t
hi
s
r
es
e
ar
c
h
f
i
el
d.
I
n t
hi
s
l
i
t
er
at
u
r
e,
w
e
hav
e pr
es
en
t
ed m
os
t
r
ec
ent
w
or
k
s
.
F
r
o
m
t
hi
s
s
t
ud
y
i
s
c
onc
l
ud
ed t
h
at
m
os
t
of
t
hes
e appr
oac
h
es
us
e
opt
i
c
a
l
f
l
o
w
bas
ed
m
ot
i
o
n
es
t
i
m
at
i
on
m
et
hod
w
hi
c
h
i
s
not
c
ap
abl
e
of
ex
t
r
ac
t
i
ng
t
h
e
boun
dar
y
of
gi
v
e
n
obj
ec
t
.
S
i
nc
e c
r
o
w
ds
ar
e a
l
w
a
y
s
u
ns
t
r
uc
t
ur
ed
dur
i
ng m
as
s
g
at
her
i
ng
w
h
i
c
h
c
aus
es
v
ar
i
o
us
oc
c
l
us
i
o
ns
and am
bi
gui
t
i
es
w
h
i
c
h c
a
nnot
b
e addr
es
s
ed b
y
us
i
ng s
t
at
e
-
of
-
ar
t
t
ec
hni
ques
.
S
t
at
e
-
of
-
ar
t
t
ec
hni
q
ue
s
uf
f
er
s
f
r
o
m
t
he
i
s
s
ue
of
per
f
or
m
anc
e
ac
c
ur
ac
y
.
T
hes
e
i
s
s
ues
m
ot
i
v
at
e
us
t
o
de
v
e
l
op a
n ef
f
i
c
i
ent
m
odel
f
or
c
r
o
w
d b
eha
v
i
or
a
nal
y
s
i
s
.
T
abl
e 1:
N
ot
at
i
o
n us
e
d i
n p
aper
−
ℎ
−
(
)
−
ℋ
−
ℎ
ℎ
−
ℎ
ℎ
(
)
,
(
)
−
−
ℎ
−
ℎ
ℎ
ℎ
⊺
−
,
ℎ
−
−
{
Γ
}
−
−
ℳ
−
−
(
)
,
ℎ
(
)
−
−
ℎ
(
,
)
−
(
)
,
ℎ
(
)
−
−
(
,
)
−
−
(
)
−
(
)
−
−
ℎ
−
(
ℒ
)
−
−
−
,
,
−
(
)
−
(
)
−
ℎ
−
3.
P
r
o
p
o
s
e
d
M
o
d
e
l
T
hi
s
s
ec
t
i
on des
c
r
i
bes
pr
op
os
ed an ap
pr
oac
h f
or
abno
r
m
al
c
r
ow
d be
ha
v
i
or
det
ec
t
i
on.
I
n
or
der
t
o m
odel
t
he m
ot
i
on
f
l
ow
d
y
n
am
i
c
s
i
s
us
ed.
F
l
o
w
des
c
r
i
pt
or
s
ar
e us
ed
t
o des
c
r
i
be t
he
m
ot
i
ons
w
hi
c
h
ar
e f
or
m
ed b
y
c
o
ns
i
d
er
i
n
g
v
ar
i
ous
r
ef
er
enc
e f
r
am
es
and t
em
por
al
s
t
r
uc
t
ur
e
of
d
y
n
am
i
c
s
w
hi
c
h r
epr
es
en
t
s
t
he t
r
aj
ec
t
or
y
.
L
agr
a
ng
i
an m
odel
i
n
g i
s
us
ed t
o
es
t
i
m
at
e t
he
m
ov
e
m
ent
and t
r
ac
k
i
ng of
par
t
i
c
l
es
w
h
i
c
h
ar
e
pr
es
ent
i
n
m
ot
i
on
f
l
o
w
w
hi
c
h
h
el
p
s
us
t
o
obt
ai
n
f
l
ow
def
or
m
at
i
on and c
om
pl
et
e m
ov
em
ent
of
m
ot
i
on.
T
he E
ul
er
i
an
t
ec
hn
i
q
ue
pr
ov
i
des
f
l
o
w
c
ov
er
ag
e.
P
ar
t
i
c
l
es
of
f
l
o
w
ar
e
c
om
put
ed
f
or
f
i
x
ed
pos
i
t
i
ons
w
h
i
c
h
r
es
u
l
t
i
n
o
v
er
a
l
l
f
l
o
w
ex
t
r
ac
t
i
on.
S
i
nc
e,
t
h
e m
ot
i
on
v
ec
t
or
i
s
t
i
m
e de
p
en
de
nt
f
i
el
ds
w
h
i
c
h c
on
t
ai
n d
i
s
t
i
nc
t
i
v
e
c
ur
v
es
s
uc
h as
s
t
r
eam
l
i
ne
c
ur
v
e
,
s
t
r
eak
l
i
ne,
pat
hl
i
n
e a
nd
t
i
m
el
i
ne c
ur
v
es
.
P
a
t
hl
i
n
e a
nd s
t
r
eam
l
i
n
e
c
ur
v
es
ar
e us
e
d t
o
def
i
n
e t
he t
ang
ent
c
ur
v
e of
a
v
ec
t
or
f
i
e
l
d.
A
f
t
er
c
o
m
put
i
ng f
l
o
w
m
ap,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
7
,
N
o.
3,
S
e
pt
em
ber
2017
:
737
–
7
47
740
S
t
r
eak
l
i
n
e
i
s
c
om
put
ed
b
y
c
o
ns
i
der
i
ng
s
pat
i
a
l
an
d
t
e
m
por
al
gr
adi
en
t
s
of
c
o
m
put
ed
f
l
o
w
m
ap.
I
n
t
he
c
as
e
of
a
bnor
m
al
ac
t
i
v
i
t
i
es
,
pos
i
t
i
o
n
a
nd
t
i
m
e
ar
e
c
r
uc
i
al
par
am
et
er
s
w
hi
c
h
a
f
f
ec
t
f
l
ow
m
ap
c
o
m
put
at
i
ons
.
T
o ov
er
c
om
e t
hi
s
i
s
s
ue,
w
e de
v
e
l
o
p an
adap
t
i
v
e r
e
pr
es
ent
at
i
on
of
s
t
r
eak
l
i
ne f
l
o
w
w
i
t
h
t
he
he
l
p of
s
t
r
eam
l
i
ne r
epr
es
ent
at
i
on
of
t
he m
ot
i
on
.
3.
1.
R
e
p
r
e
s
e
n
ta
ti
o
n
o
f M
o
ti
o
n
V
e
c
to
r
F
i
e
l
d
F
or
an
y
gi
v
e
n m
ot
i
on s
equenc
e par
t
i
c
l
es
ar
e c
om
put
ed
w
hi
c
h ar
e pr
es
ent
i
n t
he f
l
ow
f
i
el
d.
Mo
t
i
o
n i
s
ana
l
y
z
e
d us
i
ng t
h
e m
ov
e
m
ent
of
p
ar
t
i
c
l
es
w
h
i
c
h ar
e c
om
put
e
d us
i
ng d
ens
e
opt
i
c
a
l
f
l
o
w
.
A
c
c
or
di
ng t
o
t
hi
s
m
odel
,
a
v
i
d
eo s
eq
ue
nc
e i
s
c
ons
i
der
e
d as
a
n i
nput
w
hi
c
h
i
s
deno
t
ed
b
y
c
ons
i
der
i
ng
3
-
di
m
ens
i
ona
l
ar
r
a
y
den
ot
e
d
as
×
ℋ
×
w
her
e
w
i
d
t
h
of
f
r
a
m
e
i
s
deno
t
ed b
y
,
hei
g
ht
of
f
r
am
e
i
s
denot
ed
as
ℋ
and
de
not
es
t
o
t
a
l
num
ber
of
f
r
a
m
es
w
hi
c
h
has
opt
i
c
al
f
l
o
w
s
uc
h as
(
)
,
ℎ
(
)
w
her
e
,
ℎ
and
ar
e ex
pr
es
s
ed as
∈
[
1
,
]
,
ℎ
∈
[
1
,
ℋ
]
and
∈
[
1
.
−
1
]
.
F
i
r
s
t o
f
a
l
l
par
t
i
c
l
e
p
o
s
i
t
i
o
n ar
e
c
om
put
ed
at
c
om
put
at
i
o
n gr
i
d
po
i
nt
f
or
a t
i
m
e
w
hi
c
h c
a
n b
e de
not
ed
as
:
(
+
1
)
=
(
(
)
,
ℎ
(
)
,
)
+
(
)
ℎ
(
+
1
)
=
(
(
)
,
ℎ
(
)
,
)
+
ℎ
(
)
(
1)
P
ar
t
i
c
l
e p
os
i
t
i
ons
ar
e de
not
ed b
y
(
)
,
ℎ
(
)
w
her
e
,
ℎ
de
not
es
gr
i
d
poi
nt
of
f
l
o
w
.
B
y
c
om
put
i
ng t
h
i
s
pr
oc
e
s
s
i
n an i
t
er
at
i
v
e a
ppr
oa
c
h,
c
ur
v
es
ar
e obt
ai
ned
w
hi
c
h
r
epr
es
ent
par
t
i
c
l
e
s
et
w
hi
c
h
c
ont
ai
ns
t
r
aj
ec
t
or
y
i
nf
or
m
at
i
on.
A
s
di
s
c
us
s
ed
bef
or
e
t
hat
f
or
uns
t
ab
l
e
f
l
o
w
w
e us
e
s
t
r
ea
k
l
i
ne,
pa
t
hl
i
n
e an
d s
t
r
eam
l
i
ne f
or
f
l
ow
i
nf
or
m
at
i
on ex
t
r
ac
t
i
on
.
3.
2.
C
o
m
p
u
ta
ti
o
n
o
f
S
t
r
ea
kl
i
n
e
I
n t
h
e
pr
ev
i
o
us
s
ec
t
i
o
n,
w
e
ha
v
e
d
i
s
c
us
s
ed t
h
at
s
t
r
eak
l
i
n
e f
l
o
w
c
om
put
at
i
on
i
s
t
i
m
e a
nd
m
ot
i
on
dep
end
ent
.
D
ur
i
ng
c
o
m
put
at
i
on
of
s
t
r
eak
l
i
ne
f
l
o
w
,
s
ha
pe
i
nc
ons
i
s
t
enc
y
a
nd
m
ot
i
on
ar
e
t
he m
ai
n f
ac
t
or
s
w
h
i
c
h af
f
ec
t
t
he c
om
put
at
i
on
of
s
t
r
eak
l
i
ne f
l
o
w
.
I
n
pr
ev
i
o
us
w
or
k
s
,
opt
i
c
al
f
l
ow
[
15
]
i
s
us
ed
f
or
m
ot
i
on
i
nf
or
m
at
i
on
ex
t
r
ac
t
i
o
n
bu
t
f
ai
l
s
t
o
pr
ov
i
de
b
et
t
er
r
es
ul
t
s
i
n
t
er
m
s
of
m
ot
i
on i
nf
or
m
at
i
on
w
her
ea
s
s
t
r
eak
l
i
ne f
l
o
w
pr
es
ent
s
bet
t
er
r
es
ul
t
s
w
hi
c
h
c
an
b
e ob
t
ai
ned
b
y
i
nt
e
gr
at
i
ng t
i
m
e of
f
i
el
d of
v
el
oc
i
t
y
w
h
i
c
h r
es
ul
t
s
i
n
bet
t
er
a
na
l
y
s
i
s
of
f
as
t
er
dy
n
am
i
c
f
l
ow
of
m
ot
i
on and
he
l
ps
t
o
r
epr
es
ent
p
ar
t
i
c
l
es
ef
f
i
c
i
ent
l
y
o
v
er
t
he c
om
put
at
i
on
gr
i
d.
A
not
her
p
ar
adi
gm
c
ons
i
der
ed
her
e
i
s
k
now
n as
s
t
r
e
am
l
i
ne
w
hi
c
h i
s
us
ed f
or
m
ot
i
on
es
t
i
m
at
i
on.
A
br
i
ef
des
c
r
i
pt
i
o
n i
s
pr
es
ent
ed
i
n n
ex
t
s
ec
t
i
on.
3.
3.
C
o
m
p
u
ta
ti
o
n
o
f
S
tr
e
a
m
l
i
n
e
S
t
r
eam
l
i
ne
of
an
y
m
ot
i
on
v
ec
t
or
c
a
n b
e c
om
put
ed
b
y
per
f
or
m
i
ng t
h
e b
i
-
di
r
e
c
t
i
ona
l
i
nt
e
gr
at
i
on.
A
s
d
i
s
c
us
s
ed
bef
or
e
,
s
t
r
eam
l
i
ne c
a
n b
e
def
i
ne
d as
c
ur
v
es
w
h
i
c
h
ar
e t
an
ge
nt
i
n
nat
ur
e t
o v
ec
t
or
f
i
e
l
d at
a gi
v
en
po
i
nt
of
m
ot
i
on
f
l
o
w
.
T
hi
s
i
nt
egr
at
i
o
n
c
om
put
at
i
on
i
n
i
t
i
al
i
z
ed
f
r
om
a
pr
e
-
def
i
ne
d
s
eed
po
i
nt
and
e
nds
w
hen
i
t
ac
hi
e
v
es
b
oun
dar
y
of
t
he c
l
os
ed pat
h.
T
hi
s
pr
oc
es
s
i
s
m
ai
nl
y
c
at
e
gor
i
z
e
d i
nt
o t
hr
e
e s
ubs
ec
t
i
o
ns
as
m
ent
i
oned
be
l
o
w
[
16]
:
a)
I
ni
t
i
a
l
p
l
ac
em
ent
of
s
eeds
b)
D
at
a di
f
f
us
i
on
c)
F
i
na
l
s
t
op
pi
ng c
r
i
t
er
i
a
B
y
c
ons
i
der
i
ng
t
hes
e
t
w
o
t
ec
hni
ques
,
i
n
nex
t
s
ec
t
i
on,
w
e
de
v
e
l
op
a
n
ef
f
i
c
i
ent
m
o
del
f
or
m
ot
i
on ana
l
y
s
i
s
w
h
i
c
h i
s
us
ed f
o
r
c
r
ow
d ac
t
i
v
i
t
y
det
ec
t
i
on.
4.
S
y
st
em
M
o
d
el
I
n t
h
i
s
s
ec
t
i
on
w
e d
es
c
r
i
be
pr
opos
e
d s
y
s
t
em
m
odel
f
or
m
ot
i
on ana
l
y
s
i
s
.
A
c
c
or
d
i
ng t
o
l
i
t
er
at
ur
e
s
t
ud
y
pr
es
ent
ed
i
n
s
ec
t
i
on
I
I
,
i
t
c
a
n
be
c
o
nc
l
uded
t
h
at
ex
i
s
t
i
ng
appr
o
ac
hes
f
or
m
ot
i
on
ana
l
y
s
i
s
f
eat
ur
e s
uc
h as
m
ov
i
ng
obj
ec
t
det
ec
t
i
on
,
o
bj
ec
t
t
r
ac
k
i
ng and t
r
ac
k
anal
y
s
i
s
et
c
.
but
w
hen
t
h
er
e
i
s
uns
t
a
bl
e
m
ot
i
on
pr
es
e
nt
,
i
n
t
hat
c
as
e
e
x
i
s
t
i
ng
m
odel
s
f
ai
l
t
o
pr
o
v
i
de
t
he
ef
f
i
c
i
ent
r
es
ul
t
s
.
I
n or
der
t
o a
ddr
es
s
t
hi
s
i
s
s
ue,
her
e
w
e pr
es
ent
a ne
w
m
et
hod
w
h
i
c
h i
s
c
apabl
e of
ex
t
r
a
c
t
i
ng
m
ot
i
o
n
i
nf
or
m
at
i
on
f
r
om
uns
t
abl
e
m
ov
em
ent
a
nd
e
nc
od
es
s
pat
i
al
and
t
em
por
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
C
r
ow
d
A
no
ma
l
y
D
et
e
c
t
i
on
U
s
i
ng Mot
i
on
B
as
e
d S
pat
i
o
…
(
B
as
av
ar
a
j
G
M
)
741
v
ar
i
at
i
on of
m
ot
i
on.
I
n or
d
er
t
o ex
t
r
ac
t
p
ar
t
i
c
l
e i
nf
or
m
at
i
on,
t
he t
em
por
al
dom
ai
n
of
m
ot
i
on
i
s
ev
a
l
u
at
ed
un
der
E
ul
er
v
i
e
w
.
I
n b
el
o
w
f
i
gur
e
1 c
om
pl
et
e f
l
o
w
of
pr
op
os
ed m
odel
i
s
des
c
r
i
be
d
.
A
c
c
or
di
n
g t
o t
he
F
ig
u
r
e
1,
t
he v
i
deo s
eq
uenc
e i
s
c
ons
i
der
ed as
i
n
put
w
h
i
c
h i
s
c
onv
er
t
e
d
i
nt
o
f
r
a
m
es
bef
or
e
pr
oc
es
s
i
ng.
I
n
nex
t
s
t
age,
w
e
app
l
y
s
am
pl
i
ng
al
ong
w
i
t
h
m
ot
i
on
es
t
i
m
at
i
on.
I
n or
der
t
o es
t
i
m
at
e t
he m
ot
i
on of
gi
v
en s
equ
enc
e
s
pat
i
o
-
te
m
por
al
i
nt
egr
at
i
o
n and s
pat
i
o
-
t
em
por
al
i
n
t
egr
a
t
i
o
n ar
e
c
om
put
ed.
T
hi
s
s
t
age
pr
o
v
i
des
a c
om
pl
et
e
m
ot
i
on s
t
r
uc
t
ur
e of
i
nput
s
equenc
e
.
A
f
t
er
m
ot
i
on es
t
i
m
at
i
on,
f
i
l
t
er
i
ng
and
c
el
l
di
s
t
r
i
b
ut
i
on
of
dat
a
i
nt
o
v
a
r
i
ous
c
e
l
l
s
ar
e
app
l
i
e
d.
W
i
t
h t
he hel
p of
m
ot
i
o
n di
s
t
r
i
but
i
on
,
c
om
pl
et
e i
nf
or
m
at
i
on of
m
ot
i
on i
s
c
om
put
ed.
U
s
i
ng
t
hi
s
d
y
n
am
i
c
s
s
t
at
i
s
t
i
c
s
,
s
t
r
eak
l
i
ne
an
d s
t
r
eam
l
i
ne f
l
o
w
i
s
c
om
put
ed
w
hi
c
h r
es
u
l
t
s
i
n
s
egm
ent
at
i
on
a
nd f
i
n
al
l
y
,
t
he m
ot
i
on
i
s
c
l
as
s
i
f
i
ed
i
n
t
o n
or
m
al
and
ab
nor
m
al
w
hi
c
h
ai
m
s
at
a
ct
i
vi
t
y
cl
a
ssi
f
i
ca
t
i
on.
F
ig
ur
e
1.
P
r
opos
e
d M
et
h
od
C
r
o
w
d A
nom
al
y
D
et
ec
t
i
o
n
F
or
t
hi
s
s
t
ud
y
i
npu
t
v
i
d
eo
s
equenc
e
i
s
de
not
ed as
w
hi
c
h c
ons
i
s
t
s
of
num
b
er
of
f
r
a
m
es
,
3 di
m
ens
i
onal
ar
r
ay
den
ot
e
d as
×
ℋ
×
w
her
e
w
i
dt
h of
f
r
a
m
e i
s
denot
ed
b
y
,
hei
ght
of
f
r
a
m
e
i
s
de
not
e
d
as
ℋ
.
V
o
l
um
e
of
eac
h
f
r
a
m
e
i
s
di
v
i
d
ed
i
nt
o
di
f
f
er
ent
c
el
l
s
w
hi
c
h
ar
e
and
ha
v
e
s
i
z
e
of
×
ℎ
×
.
den
ot
es
w
i
dt
h
of
c
el
l
,
h
ei
ght
i
s
de
not
ed
b
y
ℎ
a
nd
deno
t
es
t
ot
al
n
um
ber
of
f
r
a
m
e
s
i
n
c
el
l
.
V
i
de
o
i
s
c
om
pos
ed
b
y
ut
i
l
i
z
i
n
g
t
hes
e
s
u
b
pat
c
hes
as
ℳ
=
{
}
.
T
r
a
j
ec
or
t
y
of
v
i
deo
s
e
que
nc
e
i
s
gi
v
en
as
Γ
=
(
)
,
(
)
,
t
r
aj
ec
t
or
y
des
c
r
i
pt
or
s
ar
e
gi
v
en as
(
)
an
d s
pat
i
a
l
c
oor
di
n
at
es
of
t
r
aj
ec
t
or
y
i
s
gi
v
en as
(
)
=
(
)
,
(
)
. D
e
te
c
te
d
t
r
aj
ec
t
or
i
es
ar
e
r
epr
es
e
nt
ed
i
n
a
s
et
w
h
i
c
h
i
s
r
e
pr
es
ent
e
d as
⊺
=
{
Γ
}
.
T
hi
s
pr
oc
es
s
i
s
s
u
m
m
ar
i
z
ed as
m
ent
i
o
ned
:
S
t
ep
1:
I
ni
t
i
at
e c
om
put
at
i
on
S
t
ep
2:
F
r
om
1:
T
ot
al
f
r
am
e
s
S
t
ep
3:
P
er
f
or
m
s
a
m
pl
i
ng
a
nd pr
o
v
i
des
k
e
y
p
oi
n
t
s
S
t
ep
4:
Mot
i
on
E
s
t
i
m
at
i
o
n
S
t
ep
4:
C
om
put
e m
ot
i
on f
l
o
w
v
ec
t
or
St
e
p
5:
A
v
er
ag
e F
l
o
w
m
ap c
o
m
put
at
i
on
S
t
ep
6:
S
p
at
i
al
di
s
t
r
i
but
i
on
S
t
ep
7:
c
om
put
at
i
on
of
f
eat
ur
e pat
c
h
es
S
t
ep
8:
Q
u
ant
i
z
at
i
on
and
c
l
us
t
er
i
ng
S
t
ep
9:
Mot
i
on
ad
v
ec
t
i
on c
om
put
at
i
on
S
t
ep
10:
A
v
er
a
ge s
t
r
eak
m
a
p c
om
put
at
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
7
,
N
o.
3,
S
e
pt
em
ber
2017
:
737
–
7
47
742
S
t
ep
11:
F
l
o
w
i
nt
er
p
ol
a
t
i
o
n
and
i
nt
egr
at
i
o
n
c
om
put
at
i
o
n
S
t
ep
12:
T
r
aj
ec
t
or
y
es
t
i
m
at
i
on
F
i
r
s
t
v
i
de
o s
am
pl
e i
s
c
onv
er
t
e
d i
nt
o f
r
a
m
es
.
F
or
e
ac
h f
r
a
m
e
s
a
m
pl
i
ng an
d m
ot
i
on
es
t
i
m
at
i
on pr
oc
es
s
has
i
m
pl
em
ent
ed.
S
am
pl
i
n
g pr
oc
es
s
pr
ov
i
des
a num
ber
of
k
e
y
po
i
nt
s
an
d
m
ot
i
on
es
t
i
m
at
i
on
def
i
n
es
t
he
i
ns
t
ant
f
l
o
w
m
ap.
F
ur
t
h
er
f
i
l
t
er
i
n
g
i
s
ap
pl
i
ed
w
i
t
h
t
he
he
l
p
of
k
e
y
poi
nt
s
an
d i
ns
t
ant
f
l
o
w
m
ap.
A
f
t
er
f
i
l
t
er
i
ng
pr
oc
es
s
,
di
s
t
r
i
but
i
on
of
f
l
ow
v
ec
t
or
s
i
s
pr
oc
es
s
ed
t
hr
oug
h
t
h
e
s
pat
i
al
l
y
enc
l
o
s
ed
c
el
l
.
F
or
eac
h
c
e
l
l
qua
nt
i
z
at
i
on
and
c
l
us
t
er
i
ng
h
a
s
per
f
or
m
ed
t
o
get
r
ep
r
es
ent
a
t
i
o
n of
f
l
o
w
v
ec
t
or
.
Mot
i
on
ad
v
ec
t
i
on
i
s
c
om
put
ed t
hr
ou
gh
a
v
er
ag
e f
l
o
w
m
ap
of
pr
ev
i
ous
f
r
am
e
and
c
ur
r
ent
f
r
a
m
e.
N
o
w
av
er
ag
e
s
t
r
ea
k
m
ap
of
al
l
m
i
ni
b
at
c
hes
h
as
pr
oc
es
s
ed.
W
i
t
h t
he
hel
p of
m
ot
i
on f
l
o
w
v
ec
t
or
r
epr
es
e
nt
at
i
o
n,
w
e ex
t
r
ac
t
i
nt
er
pol
at
e
d f
l
o
w
m
ap.
A
f
t
er
w
ar
d
s
w
e
c
om
put
e
t
he
s
t
r
eak
f
l
ow
l
i
ne
w
h
i
c
h
i
s
c
om
bi
nat
i
o
n
bet
w
ee
n
a
v
er
age
s
t
r
eak
f
l
o
w
a
nd
f
i
ne
t
o
-
c
oar
s
e s
t
r
eak
f
l
ow
.
A
ppl
y
i
ng d
i
f
f
us
i
on t
ec
h
ni
que,
w
e
ev
al
uat
e
s
t
r
eam
l
i
n
es
w
h
i
c
h f
or
m
s
v
i
deo
t
r
aj
ec
t
or
y
.
4.
1.
D
at
a
S
a
m
p
l
i
n
g
a
n
d
E
s
ti
m
a
ti
o
n
o
f
M
o
ti
o
n
H
er
e i
n t
h
i
s
s
ec
t
i
on,
w
e p
er
f
or
m
dat
a s
am
pl
i
ng
w
her
e
k
e
y
p
oi
n
t
s
ar
e
ex
t
r
ac
t
ed f
r
o
m
t
he
i
np
ut
s
eque
nc
e b
y
ap
pl
y
i
ng
s
par
s
e or
dens
e di
s
t
r
i
but
i
o
ns
.
D
ur
i
ng pr
oc
es
s
i
n
g of
v
i
deo,
s
equ
enc
e
noi
s
e
i
s
i
nd
uc
ed
w
h
i
c
h i
s
e
s
t
i
m
at
ed b
y
ap
pl
y
i
ng da
t
a s
am
pl
i
ng an
d c
om
pl
e
x
i
t
y
i
s
r
educ
ed
.
Lat
er
m
ot
i
on f
l
o
w
i
s
es
t
i
m
at
ed b
y
c
om
put
i
ng f
r
a
m
e t
o f
r
a
m
e s
pat
i
o
t
em
por
al
d
i
s
pl
ac
em
e
nt
an
al
y
s
i
s
.
4.
2.
F
i
l
te
r
i
n
g
O
nc
e dat
a i
s
s
am
pl
ed
t
h
en
v
ec
t
or
f
l
o
w
i
s
ex
pr
es
s
ed i
s
=
(
,
,
,
)
,
s
am
pl
i
ng
poi
nt
s
ar
e
den
ot
ed
b
y
(
,
)
an
d m
ot
i
on v
ec
t
or
s
ar
e
den
o
t
ed as
(
,
)
in
an
d
di
r
ec
t
i
on
r
es
pec
i
v
el
y
.
I
n or
d
er
t
o b
ui
l
d f
i
l
t
er
i
n
g m
odel
,
e
ac
h v
e
c
t
or
f
l
ow
c
o
ns
i
d
er
s
k
e
y
p
oi
nt
l
oc
a
t
i
on b
y
c
oni
s
der
i
ng i
n
i
t
i
al
pos
i
t
i
o
n
v
ec
t
or
w
hi
c
h
i
s
ex
pr
es
s
ed a
s
(
)
=
(
)
,
(
)
.
H
er
e i
t
i
s
as
s
um
ed
t
hat
k
e
y
p
oi
nt
s
and
f
l
o
w
v
e
c
t
or
s
ar
e
equal
a
nd
f
i
nal
l
y
m
edi
an
f
i
l
t
er
i
n
g
i
s
app
l
i
ed
her
e
t
o
per
f
or
m
f
i
l
t
er
i
n
g of
m
ot
i
on v
ec
t
or
.
4.
3.
C
e
l
l
D
i
s
tr
i
b
u
ti
o
n
T
hi
s
s
ubs
ec
t
i
on
pr
o
v
i
des
c
el
l
d
i
s
t
r
i
b
ut
i
on
m
et
hod
ol
og
y
di
s
c
r
et
i
on.
A
s
m
ent
i
one
d
bef
or
e
t
hat
dat
a
i
s
di
s
t
r
i
b
ut
e
d
s
pat
i
o
t
em
por
al
l
y
w
her
e
e
ac
h
f
r
am
e
i
s
di
v
i
de
d
i
nt
o
a
gr
i
d.
T
he
r
es
ol
ut
i
o
n
of
t
he
i
m
age
d
epe
nds
o
n
f
r
am
e
s
i
z
e
an
d
v
i
d
eo
dur
at
i
o
n
i
s
d
i
v
i
d
ed
t
em
por
al
l
y
.
I
n
t
hi
s
w
or
k
,
eac
h
s
pat
i
o
t
em
por
al
v
ec
t
or
i
s
s
t
or
ed i
nt
o c
el
l
an
d c
ons
i
s
t
m
ot
i
on f
l
o
w
v
ec
t
or
s
.
E
nc
o
di
n
g of
f
l
o
w
v
ec
t
or
i
s
gi
v
e
n as
=
(
,
,
,
,
)
w
her
e
s
am
pl
i
n
g
p
oi
nt
s
ar
e de
not
e
d as
(
,
)
,
f
lo
w
m
agni
t
ude
l
engt
h
i
s
d
enot
e
d
b
y
,
de
not
es
f
l
o
w
ang
l
e
w
h
i
c
h
i
s
r
e
pr
es
ent
ed
w
i
t
h
r
es
pec
t
t
o
−
f
or
f
r
a
m
e.
T
hes
e s
t
eps
ar
e ap
pl
i
ed f
or
e
ac
h f
r
a
m
e of
i
np
ut
s
eq
uenc
e.
4.
4.
C
o
m
p
u
ta
ti
o
n
o
f
M
o
ti
o
n
F
l
o
w
a
n
d
S
p
a
ti
o
T
e
m
p
o
r
a
l
I
n
te
r
p
o
l
a
ti
o
n
F
or
m
ot
i
on
es
t
i
m
at
i
o
n
a
n
d
c
om
put
at
i
o
n,
t
he
dens
e
gr
i
d
i
s
c
ons
i
der
e
d
w
hi
c
h
has
par
t
i
c
l
es
i
n t
he gr
i
d
w
her
e eac
h par
t
i
c
l
e c
ont
ai
ns
t
h
e i
nf
or
m
at
i
on abo
ut
f
l
ui
d a
nd t
hei
r
pos
i
t
i
o
n i
n
t
he
gr
i
d
.
A
n
a
v
er
a
ge
f
l
o
w
m
ap
i
s
c
o
m
put
ed
b
y
i
nt
e
gr
at
i
n
g
al
l
f
l
o
w
v
ec
t
or
s
bas
ed
on
t
he
t
i
m
e
of
s
equenc
e
.
F
or
an,
eac
h t
i
m
e
-
s
t
ep par
t
i
c
l
es
ar
e
p
os
i
t
i
on
ed at
and
o
l
d
par
t
i
c
l
es
ar
e
i
m
i
t
at
ed
f
r
o
m
s
a
m
e pos
i
t
i
on as
f
l
ow
f
i
el
d.
T
hi
s
pr
oc
es
s
c
an be ex
pr
es
s
ed as
equat
i
o
n 1.
H
er
e i
t
i
s
as
s
u
m
ed t
ha
t
s
t
r
eak
l
i
ne
i
s
a c
ol
l
ec
t
i
on
of
v
ar
i
o
us
p
ar
t
i
c
l
es
w
hi
c
h
ar
e
ac
h
i
e
v
e
d f
r
o
m
m
ot
i
on
es
t
i
m
at
i
on r
es
u
l
t
i
ng
i
n r
e
pr
es
ent
at
i
o
n o
f
dat
a
i
n t
h
e d
i
r
ec
t
i
on
.
H
er
e
w
e c
om
put
e s
t
r
eam
l
i
ne of
v
ec
t
or
f
l
o
w
i
n
t
er
m
s
of
pr
oba
bi
l
i
t
y
i
s
def
i
ne
d as
:
(
ℒ
)
=
(
)
,
,
,
∈
(
)
(2
)
unar
y
pot
ent
i
a
l
s
ar
e d
ef
i
ne
d as
,
c
andi
dat
e
s
t
r
eam
l
i
ne
i
s
den
ot
e
d b
y
;
,
,
.
A
n ap
pear
a
nc
e m
odel
i
s
f
or
m
ul
at
ed h
er
e b
y
c
ons
i
der
i
ng s
i
m
i
l
ar
i
t
y
b
et
w
een t
r
ac
k
and s
t
eak
f
l
ow
w
hi
c
h
i
s
de
not
e
d b
y
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
C
r
ow
d
A
no
ma
l
y
D
et
e
c
t
i
on
U
s
i
ng Mot
i
on
B
as
e
d S
pat
i
o
…
(
B
as
av
ar
a
j
G
M
)
743
=
1
(
(
−
−
1
=
0
)
−
−
+
0
+
)
(
)
(3
)
(
)
deno
t
es
t
r
ac
k
i
nf
or
m
at
i
on,
c
os
i
ne
of
s
t
r
eak
f
l
ow
i
s
c
om
put
e
d at
t
i
m
e
,
(
)
de
not
es
w
e
i
ght
m
eas
ur
e
m
ent
of
t
r
ac
k
i
nf
or
m
at
i
on.
N
or
m
al
i
z
at
i
on f
ac
t
o
r
i
s
gi
v
en
as
:
=
−
−
+
(
)
−
1
=
0
(4
)
A
nd s
i
m
i
l
ar
i
t
y
i
s
d
ef
i
ned
as
:
=
e
xp
(
−
1
2
(5
)
D
ur
i
n
g m
ot
i
on es
t
i
m
at
i
on
v
el
oc
i
t
y
v
ar
i
at
i
ons
ar
e c
ons
i
der
ed
w
h
er
e po
i
n
t
-
to
-
p
o
in
t
v
e
lo
c
it
y
di
f
f
er
enc
e i
s
c
om
put
ed,
t
h
i
s
c
an be
ex
pr
es
s
ed
as
:
=
−
−
+
(
)
−
1
=
0
(6
)
U
s
i
ng t
hi
s
v
e
l
oc
i
t
y
i
nf
or
m
at
i
o
n of
eac
h
f
r
a
m
e and
eac
h p
ar
t
i
c
l
e m
ot
i
on
i
s cl
a
ssi
f
i
e
d
ac
c
or
di
ng
t
o t
he s
i
m
i
l
ar
i
t
y
m
eas
ur
e
m
ent
of
v
ec
t
or
f
l
ow
.
5.
E
x
p
e
r
i
m
e
n
ta
l
r
e
s
u
l
ts
a
n
d
d
i
s
c
u
s
s
i
o
n
I
n t
h
i
s
s
ec
t
i
o
n,
w
e
pr
es
en
t
c
o
m
par
at
i
v
e ex
p
er
i
m
ent
al
r
es
ul
t
s
a
nd a
na
l
y
s
i
s
of
abno
r
m
al
c
r
ow
d
be
ha
v
i
or
det
ec
t
i
o
n u
s
i
ng pr
o
pos
e
d m
odel
.
5.
1.
D
at
aset
I
n o
r
der
t
o ev
al
uat
e t
he p
er
f
or
m
anc
e,
pr
opos
ed m
o
del
i
s
t
es
t
e
d on pu
bl
i
c
l
y
a
v
a
i
l
a
bl
e
abnor
m
al
c
r
o
w
d
dat
as
e
t
.
T
hes
e d
at
as
et
s
ac
qui
r
ed f
r
om
U
ni
v
er
s
i
t
y
of
Mi
nnes
ot
a
(
U
MN
)
.
T
hi
s
dat
as
et
c
ons
i
s
t
s
of
11
v
i
deos
f
r
om
di
f
f
er
ent
s
c
enes
c
ons
i
d
er
i
n
g
t
h
e
i
ndo
or
and
out
doo
r
env
i
r
onm
ent
.
M
ot
i
on f
i
e
l
d
v
ec
t
or
i
s
ex
t
r
ac
t
ed
b
y
par
t
i
t
i
on
i
ng
t
he
i
m
age i
nt
o gr
i
d s
i
z
e of
8
×
8
bl
oc
k
s
.
I
n t
he
U
MN
da
t
as
et
,
w
e
h
av
e c
on
duc
t
e
d t
w
o
e
x
per
i
m
ent
s
f
or
abnor
m
al
c
r
o
w
d
beh
a
v
i
o
r
ana
l
y
s
i
s
.
C
l
as
s
i
f
i
c
at
i
on
of
ac
t
i
v
i
t
y
i
s
o
bt
a
i
n
ed b
y
t
he
m
et
hod w
hi
c
h
i
s d
i
scu
sse
d
i
n
se
ct
i
o
n
I
I
I
.
F
i
gur
e
2
a
nd 3
s
ho
w
s
am
pl
e f
r
a
m
e of
i
nput
v
i
deo
s
equ
enc
e.
F
i
gur
e
2.
N
or
m
al
C
r
ow
d ac
t
i
v
i
t
y
s
am
pl
e f
r
o
m
D
S1
F
i
gur
e
3.
N
or
m
al
C
r
ow
d A
c
t
i
v
i
t
y
f
r
om
D
S
2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
7
,
N
o.
3,
S
e
pt
em
ber
2017
:
737
–
7
47
744
5.
2.
E
xp
e
r
i
m
en
t
al
A
n
al
ysi
s
I
n t
h
i
s
s
ec
t
i
on,
w
e
s
ho
w
ex
per
i
m
ent
al
s
t
ud
y
f
o
r
c
r
o
w
d
beha
v
i
or
a
na
l
y
s
i
s
us
i
ng
pr
o
pos
ed
appr
o
ac
h.
I
n
i
t
i
al
l
y
,
w
e c
ons
i
der
s
equ
enc
e
1 (
c
or
r
i
dor
)
a
s
t
he f
i
r
s
t
i
n
put
f
or
pr
oc
es
s
i
ng.
A
c
c
or
di
n
g
t
o pr
opos
e
d
m
odel
,
t
h
e v
i
deo s
equ
enc
e i
s
c
onv
e
r
t
ed i
nt
o t
he f
r
a
m
e.
F
i
r
s
t
f
r
a
m
e
and
c
or
r
es
pondi
ng
s
t
eps
ac
c
or
di
n
g t
o
pr
opos
e
d m
odel
ar
e
an
al
y
z
e
d
her
e
.
I
n
F
i
gur
e
3,
i
n
i
t
i
al
f
r
am
e i
s
s
ho
w
n.
T
hi
s
f
r
a
m
e i
s
pr
oc
es
s
ed f
ur
t
her
and m
ot
i
o
n f
l
o
w
m
ap i
s
c
o
m
put
ed
as
dep
i
c
t
ed
i
n
F
i
gur
e 4.
F
i
gur
e
4.
N
or
m
al
C
r
ow
d ac
t
i
v
i
t
y
s
am
pl
e f
r
o
m
D
S1
F
i
gur
e
5.
M
ot
i
on F
l
o
w
Ma
p
F
i
gur
e
6.
x
-
di
r
ec
t
i
o
n m
ot
i
o
n v
ec
t
or
F
i
gur
e
7.
y
-
di
r
ec
t
i
on
m
ot
i
on
v
ec
t
or
F
i
gur
e
8.
G
r
ad
i
e
nt
i
n x
-
d
ir
e
c
t
io
n
F
i
gur
e
9.
G
r
ad
i
e
nt
i
n
y
-
d
ir
e
c
t
io
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
C
r
ow
d
A
no
ma
l
y
D
et
e
c
t
i
on
U
s
i
ng Mot
i
on
B
as
e
d S
pat
i
o
…
(
B
as
av
ar
a
j
G
M
)
745
F
i
gur
e
10.
Mot
i
o
n D
y
nam
i
c
s
i
nf
or
m
at
i
on
F
i
gur
e
11.
Mot
i
o
n i
nf
or
m
at
i
on
F
i
gur
e
12.
Mot
i
on
S
e
gm
ent
at
i
o
n
F
i
gur
e
13.
D
et
ec
t
ed
F
or
t
hi
s
s
equ
enc
e
i
n
put
v
i
deo f
r
am
e i
s
gi
v
e
n
i
n
F
i
g
ur
e 3,
m
ot
i
on m
ap c
om
put
at
i
on
c
or
r
es
pondi
ng
t
o
t
hi
s
f
r
am
e m
ot
i
on f
l
o
w
m
ap i
s
c
o
m
put
ed b
y
c
ons
i
der
i
ng x
and
y
-
di
r
e
c
t
i
on
m
ot
i
on v
ec
t
or
w
hi
c
h
i
s
pr
es
ent
e
d
i
n F
i
g
ur
e 6
a
nd
F
i
g
ur
e 7.
B
as
ed
on
t
h
e m
ot
i
on
m
ap,
gr
adi
e
nt
s
ar
e
c
om
put
ed
i
n
x
an
d
y
-
di
r
ec
t
i
on
gr
adi
ent
s
ar
e
c
om
put
ed
as
s
ho
w
n
i
n
F
i
gur
e
8
and
9
.
Mot
i
o
n
d
y
n
am
i
c
s
and
m
ot
i
on
i
nf
or
m
at
i
on
i
s
ex
t
r
ac
t
ed
an
d
de
t
ec
t
ed
i
n
F
i
gur
e
10
and
f
i
g
ur
e
11.
F
i
n
al
l
y
,
t
he s
e
gm
ent
ed m
ot
i
on
i
s
p
r
es
ent
ed
i
n F
i
gur
e
12
an
d
F
i
g
ur
e 1
3 s
h
o
w
s
de
t
ec
t
i
o
n of
ab
nor
m
al
ac
t
i
v
i
t
y
us
i
ng t
hi
s
pr
oc
es
s
.
T
he per
f
or
m
anc
e o
f
pr
opos
ed
m
odel
i
s
c
om
put
ed i
n
t
er
m
s
o
f
f
al
s
e pos
i
t
i
v
e r
at
e,
t
r
ue
pos
i
t
i
v
e r
at
e a
nd c
l
as
s
i
f
i
c
at
i
on ac
c
ur
ac
y
f
or
v
ar
i
ed s
c
en
ar
i
os
.
P
er
f
or
m
anc
e
of
pr
opos
ed m
odel
i
s
c
o
m
par
ed
w
i
t
h s
t
at
e of
ar
t
t
ec
hni
ques
w
hi
c
h
ar
e pr
es
e
nt
i
n [
1
7]
a
nd [
18]
F
i
gur
e
14.
R
O
C
P
er
f
or
m
anc
e c
om
par
i
s
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
7
,
N
o.
3,
S
e
pt
em
ber
2017
:
737
–
7
47
746
I
n
F
i
gur
e
14,
a
c
om
par
at
i
v
e
s
t
ud
y
i
s
pr
es
ent
e
d
f
or
c
ons
i
der
e
d
s
c
enar
i
o
1
as
m
e
nt
i
o
ned
bef
or
e.
T
hi
s
anal
y
s
i
s
i
s
c
ar
r
i
ed
out
b
y
c
om
put
i
ng
f
al
s
e
pos
i
t
i
v
e r
at
e and t
r
ue po
s
i
t
i
v
e
r
at
e
an
d
c
o
m
par
ed
w
i
t
h
v
i
s
c
ous
f
l
ui
d
f
i
el
d
m
et
hod
and
d
ens
e
t
r
aj
ec
t
or
y
-
bas
ed
m
et
hod.
P
r
opos
ed
m
odel
i
s
ab
l
e t
o ac
h
i
e
v
e b
et
t
er
pe
r
f
or
m
anc
e w
he
n c
om
par
ed w
i
t
h
t
hes
e
t
ec
hn
i
qu
es
.
F
i
gur
e
15.
R
O
C
P
er
f
o
r
m
anc
e c
om
par
i
s
on f
or
c
as
e 2
I
n
F
i
g
ur
e 1
5 a
nd
F
i
gur
e 16 w
e
s
ho
w
t
he c
om
par
i
s
on
a
nal
y
s
i
s
i
n t
er
m
s
of
t
r
ue pos
i
t
i
v
e
r
at
e an
d f
al
s
e
pos
i
t
i
v
e r
at
e
b
y
c
o
ns
i
der
i
ng
t
w
o
t
es
t
c
as
es
as
m
ent
i
one
d i
n
F
i
g
ur
e
2 an
d
F
i
g
ur
e 3
.
F
i
gur
e
16.
C
l
as
s
i
f
i
c
at
i
on
P
e
r
f
or
m
anc
e c
o
m
par
i
s
on
6.
C
o
n
c
l
u
s
i
o
n
I
n t
hi
s
w
or
k
,
w
e ha
v
e ad
d
r
es
s
ed t
he i
s
s
ue of
abnor
m
al
c
r
ow
d be
ha
v
i
or
ana
l
y
s
i
s
f
o
r
s
ur
v
ei
l
l
anc
e
s
c
en
ar
i
os
.
T
hi
s
i
s
ac
hi
ev
ed
b
y
us
i
n
g
s
e
m
ant
i
c
appr
oac
h
b
y
g
l
ob
al
dens
e
f
l
o
w
an
d
l
oc
al
m
ot
i
on
i
nf
or
m
at
i
on o
f
v
i
de
o
dat
a
.
s
pa
t
i
o
-
t
em
por
a
l
f
eat
ur
e
a
nal
y
s
i
s
t
ec
hn
i
que has
us
ed
l
oc
al
m
ot
i
o
n i
nf
or
m
at
i
on.
I
n t
hi
s
w
or
k
,
t
he r
el
at
i
on
al
des
c
r
i
pt
or
i
s
pr
es
ent
e
d f
or
c
l
as
s
i
f
i
c
at
i
o
n
pur
pos
e
w
hi
c
h
ut
i
l
i
z
es
t
h
e r
el
a
t
i
o
ns
hi
p be
t
w
e
en e
ac
h s
c
ene a
nd
i
nd
i
v
i
d
ua
l
obj
ec
t
s
.
A
n
aut
om
at
ed
pr
oc
es
s
of
f
eat
ur
e
ex
t
r
a
c
t
i
on
i
s
dev
el
o
pe
d
her
e
f
or
behav
i
or
ana
l
y
s
i
s
of
c
r
ow
d.
F
or
ev
e
nt
c
l
as
s
i
f
i
c
at
i
o
n,
t
he
pa
t
c
h bas
e
d pr
oc
es
s
i
s
per
f
o
r
m
ed w
hi
c
h
s
ho
w
s
pr
om
i
s
i
ng r
es
u
l
t
s
f
or
d
y
n
am
i
c
m
ot
i
on f
eat
ur
e a
n
d c
an
be
us
ed f
or
r
e
al
-
t
i
m
e
app
l
i
c
at
i
on
s
c
enar
i
os
.
E
x
pe
r
i
m
ent
al
s
t
ud
y
sh
o
w
s t
h
a
t
pr
o
pos
e
d m
odel
i
s
a
bl
e t
o
pr
o
v
i
de
bet
t
er
p
e
r
f
or
m
anc
e w
he
n c
om
par
ed
t
o s
t
at
e of
ar
t
t
ec
hni
ques
.
Evaluation Warning : The document was created with Spire.PDF for Python.