T
E
L
KO
M
NIK
A
, V
ol
.
17
,
No.
4,
A
ug
us
t
20
1
9,
p
p.2
04
8
~
2
05
7
IS
S
N: 1
69
3
-
6
93
0
,
accr
ed
ited
F
irst
Gr
ad
e b
y K
em
en
r
istekdikti,
Decr
ee
No: 2
1/E/
K
P
T
/20
18
DOI:
10.12928/TE
LK
OM
N
IK
A
.v
1
7
i
4
.
12759
◼
20
48
Rec
ei
v
ed
A
ug
us
t
29
,
20
1
8
;
Rev
i
s
ed
F
eb
r
ua
r
y
1
9
, 2
01
9
; A
c
c
ep
t
ed
Ma
r
c
h 1
3
, 2
01
9
Road
marker
s
c
lassifica
tio
n u
si
ng
bi
nar
y
sc
an
n
ing
and
sl
op
e
con
to
urs
Z
amani M
d
S
ani*
1
,
H
adh
r
ami
A
b
d
G
h
ani
2
,
Ro
sli
Be
sa
r
3
,
A
z
iz
u
l
A
z
iz
an
4
1
Fa
c
u
l
ty
o
f
E
l
e
c
tr
i
c
a
l
En
g
i
n
e
e
ri
n
g
,
Uni
v
e
r
s
i
ti
T
e
k
n
i
k
a
l
M
a
l
a
y
s
i
a
M
e
l
a
k
a
,
M
a
l
a
y
s
i
a
2
,3
Fa
c
u
l
ty
o
f
En
g
i
n
e
e
r
i
n
g
a
n
d
T
e
ch
n
o
l
o
g
y
,
M
u
l
ti
m
e
d
i
a
Un
i
v
e
rs
i
ty
,
M
a
l
a
y
s
i
a
4
Ad
v
a
n
c
e
d
I
n
f
o
rm
a
ti
c
s
S
c
h
o
o
l
,
Un
i
v
e
rs
i
t
i
T
e
k
n
o
l
o
g
i
M
a
l
a
y
s
i
a
,
M
a
l
a
y
s
i
a
*C
o
rre
s
p
o
n
d
i
n
g
a
u
th
o
r,
e
-
m
a
i
l
:
z
a
m
a
n
i
s
a
n
i
@ut
e
m
.e
d
u
.
m
y
1
,
h
a
d
h
ra
m
i
.a
b
d
g
h
a
n
i
@m
m
u
.e
d
u
.
m
y
2
,
ro
s
l
i
@
m
m
u
.e
d
u
.m
y
3
,
a
z
i
z
u
l
a
z
i
z
a
n
@ut
m
.
m
y
4
Ab
strac
t
Roa
d
m
a
rk
e
rs
g
u
i
d
e
t
h
e
d
r
i
v
e
r
whi
l
e
d
r
i
v
i
n
g
o
n
th
e
ro
a
d
t
o
c
o
n
tro
l
th
e
tra
ff
i
c
fo
r
t
h
e
s
a
fe
ty
o
f
th
e
r
o
a
d
u
s
e
rs
.
Wi
th
th
e
b
o
o
m
i
n
g
a
u
t
o
n
o
m
o
u
s
c
a
r
te
c
h
n
o
l
o
g
y
,
t
h
e
ro
a
d
m
a
r
k
e
rs
c
l
a
s
s
i
fi
c
a
ti
o
n
i
s
i
m
p
o
rt
a
n
t
i
n
i
t
s
v
i
s
i
o
n
s
e
g
m
e
n
t
to
n
a
v
i
g
a
te
th
e
a
u
to
n
o
m
o
u
s
c
a
r.
A
n
e
w
m
e
th
o
d
i
s
p
r
o
p
o
s
e
d
i
n
t
h
i
s
p
a
p
e
r
to
c
l
a
s
s
i
f
y
fi
v
e
ty
p
e
s
o
f
ro
a
d
m
a
rk
e
r
s
n
a
m
e
l
y
d
a
s
h
e
d
,
s
i
n
g
l
e
,
d
o
u
b
l
e
,
s
o
l
i
d
-
d
a
s
h
e
d
a
n
d
d
a
s
h
e
d
-
s
o
l
i
d
whi
c
h
a
r
e
c
o
m
m
o
n
l
y
fo
u
n
d
o
n
t
h
e
two
l
a
n
e
s
i
n
g
l
e
c
a
r
ri
a
g
e
way
.
Th
e
c
l
a
s
s
i
fi
c
a
ti
o
n
i
s
u
s
i
n
g
u
n
i
q
u
e
f
e
a
tu
re
a
c
q
u
i
r
e
d
fro
m
th
e
b
i
n
a
r
y
i
m
a
g
e
b
y
s
c
a
n
n
i
n
g
o
n
e
a
c
h
o
f
th
e
i
m
a
g
e
s
t
o
c
a
l
c
u
l
a
te
th
e
fr
e
q
u
e
n
c
y
o
f
b
i
n
a
ry
tr
a
n
s
i
ti
o
n
.
An
o
th
e
r
fe
a
tu
re
wh
i
c
h
i
s
th
e
s
l
o
p
e
s
b
e
twe
e
n
th
e
two
c
e
n
tro
i
d
s
whi
c
h
a
l
l
o
w
t
h
e
p
ro
p
o
s
e
d
m
e
th
o
d
,
t
o
p
e
rfo
rm
t
h
e
c
l
a
s
s
i
fi
c
a
t
i
o
n
wi
th
i
n
t
h
e
s
a
m
e
v
i
d
e
o
fra
m
e
p
e
r
i
o
d
.
Th
i
s
p
ro
p
o
s
e
d
m
e
th
o
d
h
a
s
b
e
e
n
o
b
s
e
r
v
e
d
t
o
a
c
h
i
e
v
e
a
n
a
c
c
u
r
a
c
y
v
a
l
u
e
o
f
a
t
l
e
a
s
t
9
3
%
,
whi
c
h
i
s
h
i
g
h
e
r
t
h
a
n
th
e
a
c
c
u
ra
c
y
v
a
l
u
e
a
c
h
i
e
v
e
d
b
y
th
e
e
x
i
s
ti
n
g
m
e
th
o
d
s
.
Key
w
ords
:
a
u
to
n
o
m
o
u
s
c
a
r,
b
i
n
a
ry
s
c
a
n
,
r
o
a
d
m
a
rk
e
r,
s
l
o
p
e
c
o
n
t
o
u
r
Copy
righ
t
©
2
0
1
9
Uni
v
e
rsi
t
a
s
Ahm
a
d
D
a
hl
a
n.
All
rig
ht
s
r
e
s
e
rve
d
.
1.
Int
r
o
d
u
ctio
n
T
he
r
oa
d
m
ar
k
ers
are
es
s
en
ti
al
to
c
o
ntrol
th
e
r
oa
d
tr
af
f
i
c
an
d
a
v
oi
d
tr
af
f
i
c
ha
z
a
r
ds
b
y
gu
i
di
n
g
t
he
r
o
ad
us
ers
to
s
af
el
y
de
c
i
de
ei
t
he
r
t
o
m
ai
nt
ai
n
the
c
ou
r
s
e
i
n
t
he
m
i
dd
l
e
of
the
l
a
ne
or
to
o
v
ertak
e
the
f
r
on
t
v
e
hi
c
l
es
.
Roa
d
m
ar
k
ers
are
c
ate
g
ori
z
ed
i
n
di
f
f
erent
t
y
p
es
s
ub
j
ec
t
to
the
t
y
pe
s
of
the
r
oa
ds
s
uc
h
as
urban,
no
n
-
ur
ba
n
a
nd
hi
gh
wa
y
,
on
whi
c
h
t
he
s
i
z
es
of
the
r
oa
d
m
ar
k
e
r
s
m
a
y
s
l
i
g
htl
y
v
ar
y
.
E
x
i
s
ti
ng
r
es
ea
r
c
h
wor
k
ha
v
e
be
e
n
ex
ten
s
i
v
e
l
y
c
arr
i
e
d
ou
t
to
s
ol
v
e
the
l
an
e
d
ete
c
t
i
on
an
d
t
r
ac
k
i
ng
prob
l
em
s
f
or
A
uto
-
A
s
s
i
s
t
Dr
i
v
i
n
g
S
y
s
t
em
(
A
DS
)
[1
–
6
].
T
he
s
c
he
m
es
propos
ed
ar
e
m
ai
nl
y
a
bo
ut
de
tec
ti
ng
a
nd
tr
ac
k
i
ng
the
l
a
ne
s
,
w
i
th
ou
t
f
oc
us
i
ng
on
c
l
as
s
i
f
y
i
ng
t
he
r
oa
d
m
ar
k
ers
.
O
the
r
s
,
ha
d
w
ork
ed
on
the
r
oa
d
s
y
m
bo
l
s
,
al
p
ha
b
ets
an
d
c
us
tom
i
z
e
m
ar
k
ers
[7
-
10
]
whi
c
h
a
l
s
o
us
e
d
as
the
i
nf
orm
ati
on
to
a
l
ert
th
e
dri
v
er
s
.
Nev
ert
he
l
es
s
,
r
oa
d
m
ar
k
er
c
l
as
s
i
f
i
c
ati
on
s
ti
l
l
ha
v
e
s
om
e
r
oo
m
f
or
i
m
prov
e
m
en
t
as
the
r
e
are
di
f
f
erent
t
y
p
es
of
ma
r
k
e
r
s
c
an
be
f
ou
nd
gl
o
b
al
l
y
[
11
,
12
].
A
f
r
eq
u
en
c
y
a
na
l
y
s
i
s
m
eth
od
f
or
th
e
urb
an
r
o
ad
m
ar
k
e
r
c
l
as
s
i
f
i
c
ati
on
ha
d
be
en
propos
e
d
i
n
[
13
]
whi
c
h
us
e
s
the
i
nv
ers
e
p
ers
pe
c
ti
v
e
m
ap
pi
ng
(
I
P
M)
to
prod
uc
e
a
b
i
r
d
-
e
y
e
v
i
e
w
of
the
r
oa
d
a
nd
prod
uc
e
a
m
od
i
f
i
ed
Hou
gh
T
r
an
s
f
or
m
(
H
T
)
f
or
be
tte
r
l
an
e
de
t
ec
ti
o
n.
T
he
n
the
p
o
w
er
s
p
ec
tr
um
an
d
F
ou
r
i
er
a
na
l
y
s
i
s
are
a
pp
l
i
ed
to
d
ete
c
t
thre
e
t
y
p
e
s
of
m
ar
k
er
s
,
na
m
el
y
;
s
i
n
gl
e
s
ol
i
d,
d
as
he
d,
an
d
m
erged
r
oa
d
m
ar
k
er
s
w
h
i
c
h
are
c
om
m
on
l
y
f
ou
n
d
at
th
e
urba
n
r
oa
d.
S
c
h
ub
ert
et
a
l
.
r
ep
ort
ed
i
n
[
14
]
ab
o
ut
a
l
an
e
c
h
a
ng
i
ng
ap
proac
h
ba
s
ed
o
n
t
h
e
r
oa
d
tr
af
f
i
c
s
urr
ou
nd
i
ng
s
i
t
ua
t
i
o
n
i
nc
l
u
di
n
g
the
t
y
p
e
of
the
r
o
a
d
m
ar
k
er.
How
ev
er
the
m
ar
k
e
r
t
y
p
e’
s
c
l
as
s
i
f
i
c
ati
on
i
s
l
i
m
i
ted
to
on
l
y
da
s
he
d
an
d
s
ol
i
d
r
oa
d
m
ar
k
ers
.
In
a
w
ork
pres
en
te
d
b
y
Li
n
dn
er
et
al
.
[
15
],
an
ed
ge
de
tec
t
or
tec
h
ni
q
ue
w
a
s
de
s
i
g
ne
d
to
s
ub
s
e
qu
e
ntl
y
s
e
arc
h
f
or
a
di
f
f
erent
ob
j
ec
ts
on
the
r
oa
d
t
o
d
ete
c
t
t
he
r
o
ad
m
ark
ers
.
E
v
en
t
ho
u
gh
f
ou
r
t
y
p
es
of
r
oa
d
m
ar
k
e
r
s
c
an
be
de
t
ec
ted
,
i
t
on
l
y
c
on
c
e
ntrat
es
on
t
he
d
as
he
d
l
i
ne
s
w
i
th
d
i
f
f
erent s
i
z
es
.
In
an
oth
er
ap
proac
h
b
y
S
u
c
hi
tr
a
e
t
a
l
.
[
16
],
a
m
od
ul
ar
ap
pro
ac
h
i
s
us
ed
t
o
c
l
as
s
i
f
y
thr
ee
r
oa
d
m
ar
k
ers
na
m
el
y
;
da
s
he
d,
s
o
l
i
d
an
d
z
i
g
z
ag
.
T
o
c
l
as
s
i
f
y
ei
t
he
r
d
as
he
d
or
s
ol
i
d,
B
as
i
c
La
ne
Ma
r
k
i
ng
(
B
LM
)
,
w
h
i
c
h
i
s
ba
s
ed
on
c
o
nti
nu
i
t
y
pro
pe
r
ti
es
i
s
ap
p
l
i
e
d.
T
hi
s
ap
pr
oa
c
h
ho
wev
er,
ap
p
l
i
es
t
he
t
em
po
r
al
i
nf
or
m
ati
on
i
n
t
he
c
l
as
s
i
f
i
c
ati
on
op
erat
i
o
n,
w
h
i
c
h
r
en
de
r
s
s
l
o
w
er
d
ete
c
ti
on
when
th
e
r
oa
d
m
ar
k
er
t
y
p
e
c
ha
n
ge
s
whi
l
s
t
dr
i
v
i
n
g
o
n
th
e
r
o
ad
.
Ned
ev
s
c
h
i
et
al
.
[
17
]
a
bl
e
to
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
930
◼
Road
ma
r
k
ers
c
l
as
s
i
f
i
c
ati
on
us
i
ng
bi
na
r
y
s
c
an
n
i
ng
…
(
Z
am
an
i
Md
S
an
i
)
2049
c
l
as
s
i
f
y
r
o
ad
m
ar
k
ers
i
nto
f
ou
r
t
y
pe
s
na
m
el
y
do
ub
l
e
s
ol
i
d,
s
i
n
gl
e
s
ol
i
d,
m
erged,
a
nd
da
s
h
ed
us
i
ng
a p
e
r
i
o
di
c
h
i
s
tog
r
am
an
d
eg
o
-
l
oc
al
i
z
at
i
on
.
P
au
l
a
et
a
l
[
18
-
20
]
h
ad
w
o
r
k
ed
on
th
e
f
i
v
e
t
y
p
es
of
r
oa
d
m
ar
k
er.
A
de
tec
ti
on
m
eth
o
d
ba
s
ed
o
n
B
a
y
es
i
an
ap
pr
oa
c
h
f
or
c
l
as
s
i
f
y
i
n
g
the
r
o
ad
m
ar
k
e
r
s
ha
s
be
en
prop
os
e
d.
A
l
tho
ug
h
i
t
i
s
ab
l
e
to
c
l
as
s
i
f
y
f
i
v
e
t
y
pe
s
of
r
oa
d
m
ar
k
ers
,
the
m
eth
od
s
uf
f
ers
ou
tl
i
er
prob
l
em
s
es
pe
c
i
a
l
l
y
at
tr
an
s
i
ti
on
s
be
t
ween
t
wo
t
y
pe
s
of
m
ar
k
ers
.
T
he
r
es
ul
t
s
of
the
c
l
as
s
i
f
i
c
ati
on
w
ere
f
ou
nd
to
ha
v
e
c
ha
ng
e
d
ab
r
up
t
l
y
o
n
e
ac
h
f
r
a
m
e,
w
hi
c
h
c
au
s
ed
i
nc
on
s
i
s
ten
t
r
es
ul
ts
whi
l
e
dri
v
i
n
g.
Ma
th
i
be
l
a
e
t
al
.
[
21
]
c
on
c
en
tr
ate
on
th
e
r
oa
d
m
ar
k
ers
f
ou
nd
at
the
urba
n
ar
ea
an
d
ab
l
e
to
c
l
as
s
i
f
y
s
e
v
en
t
y
p
es
,
na
m
el
y
;
d
ou
b
l
e
bo
u
nd
ar
y
,
s
i
n
gl
e
bo
un
d
ar
y
,
z
i
g
-
z
ag
,
s
e
pa
r
ato
r
,
bo
x
e
d
j
un
c
ti
on
,
i
nte
r
s
ec
ti
o
na
nd
s
pe
c
i
a
l
l
an
es
b
y
us
i
ng
a
un
i
qu
e
s
et
of
ge
om
etri
c
fea
tures
whi
c
h
f
un
c
ti
on
wi
th
i
n
a
pro
ba
b
i
l
i
s
ti
c
RU
S
B
oo
s
t
a
nd
C
on
di
t
i
on
al
R
an
do
m
F
i
el
d
(
C
RF
)
ne
t
w
ork
.
Ho
w
e
v
er,
t
hi
s
r
es
e
arc
h
i
s
m
ai
nl
y
c
arr
i
ed
o
ut
o
n
a
s
tat
i
c
i
m
ag
e.
T
he
l
ate
s
t
p
ub
l
i
c
ati
on
f
r
o
m
Z
am
an
i
et
a
l
.
[
22
,
2
3]
us
i
n
g
th
e
f
ea
t
ures
ex
tr
ac
ted
f
r
om
the
r
oa
d
l
an
e
,
an
d
f
ed
i
n
to
t
he
A
r
t
i
f
i
c
i
al
Neura
l
Net
wor
k
(
A
NN)
.
T
he
i
n
i
t
i
al
r
es
e
arc
h
s
tarts
wi
th
t
wo
t
y
p
es
of
l
an
e
m
ark
ers
an
d
l
at
er
ex
pa
nd
ed
t
o
f
i
v
e
t
y
p
es
of
m
ar
k
er
s
c
l
as
s
i
f
i
c
ati
on
[
2
4]
whi
c
h
t
he
f
ea
tures
ex
tr
ac
ted
f
r
o
m
the
c
us
t
om
i
z
ed
Reg
i
on
of
I
nte
r
es
t
(
RO
I)
.
Ne
v
erth
el
es
s
,
the
a
l
go
r
i
thm
r
eq
ui
r
es
te
m
po
r
al
v
a
l
ue
s
f
r
o
m
to
l
ot
of
i
m
ag
e
s
eq
u
en
c
e
be
f
ore
i
t
c
an
m
a
k
e
a
d
e
c
i
s
i
on
on
t
y
pe
’
s
tr
a
ns
i
t
i
on
d
ue
t
o
i
ts
s
m
al
l
RO
I.
T
hi
s
m
a
y
l
a
gs
th
e
i
nf
orm
ati
on
to
al
ert th
e d
r
i
v
ers
.
T
he
af
orem
en
ti
on
ed
m
eth
od
s
ha
v
e
be
en
propos
e
d t
o
ad
dres
s
di
f
f
erent t
y
p
es
of
r
oa
d m
ar
k
ers
s
u
bj
ec
t to
t
he
ph
y
s
i
c
al
f
ea
ture
s
of
th
e roa
ds
i
nc
l
u
di
ng
th
e
d
i
m
en
s
i
on
,
c
ol
o
ur
an
d
s
i
z
e
of
the
m
ark
er
s
.
In
ou
r
r
es
ea
r
c
h,
t
he
t
y
p
es
of
r
oa
d
m
ar
k
e
r
s
w
hi
c
h
are
n
orm
al
l
y
f
ou
nd
at
t
he
n
on
-
urb
an
c
arr
i
ag
e
wa
y
are
c
l
as
s
i
f
i
e
d
u
s
i
ng
a
m
ac
hi
ne
v
i
s
i
on
s
y
s
tem
att
ac
he
d
t
o
a
c
ar.
T
he
r
oa
d
m
ark
ers
i
nc
l
ud
e
s
o
l
i
d,
da
s
he
d
,
d
as
he
d
-
s
o
l
i
d
,
do
ub
l
e
-
s
o
l
i
d
an
d s
ol
i
d
-
d
as
h
ed
, a
s
s
h
o
w
n
i
n F
i
gu
r
e
1.
(
a)
(
b)
(
c
)
(
d)
(
e)
F
i
gu
r
e
1
.
R
oa
d
Ma
r
k
ers
f
ou
nd
at
t
w
o
l
a
ne
s
i
ng
l
e c
arr
i
a
ge
w
a
y
(
a)
d
as
he
d
(
D)
(
b)
da
s
he
d
-
s
o
l
i
d
(
DS
)
(
c
)
d
ou
b
l
e s
o
l
i
d
(
DD)
(
d)
s
ol
i
d
-
d
as
he
d
(
S
D)
(
e)
s
i
n
gl
e
s
ol
i
d
(
S
S
)
2.
Re
se
a
r
ch M
eth
o
d
T
he
propos
ed
c
l
as
s
i
f
i
c
ati
o
n
proc
es
s
i
s
us
i
n
g
a
t
wo
-
l
a
y
er
c
l
as
s
i
f
i
er
as
s
ho
wn
as
i
n
F
i
gu
r
e
2.
F
i
r
s
t
l
a
y
er
wi
l
l
c
l
as
s
i
f
y
t
hree
c
l
as
s
es
of
m
ar
k
ers
w
he
r
e
as
the
ne
x
t
l
a
y
er
wi
l
l
c
l
a
s
s
i
f
y
an
ot
he
r
t
w
o
m
ore
c
l
as
s
es
.
T
he
f
i
r
s
t
l
a
y
er
i
s
us
i
n
g
the
s
c
an
n
i
ng
the
bi
n
ar
y
i
m
ag
e
to
ge
t
the
pe
r
c
en
tag
e
of
the
bi
n
a
r
y
tr
a
ns
i
t
i
on
a
nd
th
e
s
ec
on
d
l
a
y
er
c
l
as
s
i
f
i
er
i
s
us
i
ng
t
he
s
l
op
e
v
a
l
u
e
f
r
o
m
the
tw
o
c
on
tou
r
s
de
t
ec
ted
i
n
the
s
am
e
i
m
ag
e.
B
ef
ore
al
l
th
es
e
proc
es
s
c
an
be
d
on
e,
the
Re
gi
on
of
Int
eres
t
(
RO
I)
wi
l
l
b
e
de
t
erm
i
ne
d
f
r
o
m
the
v
i
d
eo
i
m
ag
e,
an
d
th
i
s
woul
d
r
eq
ui
r
e
the
c
am
era l
oc
at
i
on
to
b
e a
dj
us
ted
at
th
e
f
r
on
t o
f
th
e c
am
era.
F
i
gu
r
e
2
. T
w
o
-
l
a
y
er c
l
as
s
i
f
i
er
F
i
r
s
t
L
a
y
e
r
C
l
a
s
s
i
f
i
e
r
S
I
N
G
L
E
S
O
L
I
D
D
A
S
H
E
D
D
O
U
B
L
E
S
O
L
I
D
S
e
c
o
n
d
l
a
y
e
r
C
l
a
s
s
i
f
i
e
r
S
O
L
I
D
-
D
A
S
H
E
D
D
A
S
H
E
D
-
S
O
L
I
D
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
4
,
A
ug
us
t
20
19
:
20
4
8
-
20
5
7
2050
2.1
. C
ame
r
a
S
etu
p
f
o
r
R
e
g
ion
of
Int
er
es
t
(
RO
I)
s
ele
ctio
n
A
n
e
w
m
eth
od
i
n
s
e
l
ec
t
i
ng
the
RO
I
b
as
ed
on
i
ts
c
al
i
br
ate
d
c
am
era’
s
he
i
gh
t
a
nd
F
i
el
d
of
V
i
e
w
(
F
O
V
)
are
us
e
d.
F
or
the
i
n
i
ti
al
s
etu
p
,
a
c
am
era
wi
th
a
r
es
o
l
ut
i
o
n
of
1
28
0x
72
0
l
oc
at
ed
at
the
c
en
t
er
of
the
l
a
ne
,
i
s
c
al
i
bra
ted
on
i
t
s
p
os
i
t
i
on
wi
th
i
ts
F
O
V
ad
j
us
te
d
to
w
ar
ds
the
pl
an
ar
r
oa
d
s
urf
ac
e
c
ap
tured
b
y
th
e
c
a
m
era
as
s
ho
w
n
i
n
F
i
gu
r
e
s
3
(
a)
an
d
3
(
b).
T
o
c
al
i
brate
the
p
os
i
t
i
on
of
the
c
am
era,
tw
o
h
ori
z
on
t
al
l
i
n
es
,
h1
a
nd
h2
,
are
s
et
on
the
F
O
V
to
e
qu
a
l
l
y
d
i
v
i
de
i
nto
3
s
ec
ti
o
ns
as
i
n
F
i
gu
r
e
3
(
c
)
,
w
he
r
e
t
he
h1
ho
r
i
z
o
nta
l
l
i
ne
i
n
t
he
i
m
ag
e
m
us
t
c
r
os
s
the
i
nt
e
r
s
ec
ti
on
po
i
nt
be
t
w
e
en
th
e
r
i
gh
tm
os
t
l
an
e,
the
of
f
-
r
oa
d
r
eg
i
o
n
an
d
the
bo
r
d
er
of
the
i
m
ag
e
to
m
ax
i
m
i
z
e
the
RO
I.
T
he
RO
I
s
el
ec
t
i
o
n
i
s
the
n
ap
p
l
i
e
d
t
o
t
he
v
i
de
o
a
nd
th
e
ou
tpu
t
f
or
the
RO
I
i
s
l
ab
e
l
l
e
d
as
r
(
x
,y
)
as
i
n
F
i
gu
r
e
4
w
h
i
c
h
c
on
tai
ns
the
i
nf
orm
ati
on
of
the
m
ar
k
er,
b
y
d
i
v
i
d
i
n
g
the
i
m
ag
e
i
nto
t
wo
ha
l
v
es
be
f
ore s
e
l
ec
t
i
ng
t
w
o
th
i
r
d
of
f
r
o
m
th
e ri
g
ht
,
as
i
n
di
c
ate
d
b
y
th
e re
d
bo
x
,
whi
c
h i
s
bo
r
de
r
ed
b
y
the
t
w
o
h
ori
z
on
t
al
l
i
ne
s
.
W
i
t
h
thi
s
m
eth
od
,
th
e
RO
I
c
on
tai
ns
th
e
l
arges
t
v
i
s
i
b
l
e
r
oa
d
m
ar
k
er
o
f
the
i
m
ag
e.
(
a)
(
b)
(
c
)
F
i
gu
r
e
3
. T
he
ca
m
era
s
etu
p
(
a)
c
am
era po
s
i
ti
o
n
(
b) 3
D
c
am
era c
oo
r
di
na
t
e s
y
s
te
m
(
c
)
c
a
m
era he
i
gh
t
ad
j
us
tm
en
t
w
i
th
h
1 a
nd
h
2 h
ori
z
on
t
al
l
i
ne
s
F
i
g
ure
4
. RO
I
s
e
l
ec
t
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
930
◼
Road
ma
r
k
ers
c
l
as
s
i
f
i
c
ati
on
us
i
ng
bi
na
r
y
s
c
an
n
i
ng
…
(
Z
am
an
i
Md
S
an
i
)
2051
T
he
RO
I
,
r
(
x
,y
)
w
hi
c
h
i
s
i
n
t
wo
di
m
en
s
i
on
(
2D)
w
i
l
l
r
e
pres
en
ts
the
v
i
de
o
i
m
ag
es
i
n
pi
x
el
s
an
d
wi
l
l
go
t
hroug
h
the
pre
-
proc
es
s
i
ng
pri
or
to
the
f
ea
ture
ex
tr
ac
ti
o
n
an
d
c
l
as
s
i
f
i
c
ati
o
n
proc
es
s
.
T
he
r
aw
i
m
ag
e
i
n
R
ed
G
r
ee
n
B
l
ue
(
RG
B
)
v
i
de
o
i
m
ag
es
w
i
l
l
be
f
i
l
t
ered
t
hrou
gh
the
G
a
us
s
i
an
f
i
l
ter
f
or
2D
as
i
n
(1
).
T
he
p
aram
ete
r
x
i
s
th
e
ho
r
i
z
o
nta
l
l
e
ng
t
h f
r
om
th
e o
r
i
gi
n
a
nd
y
i
s
th
e
v
ert
i
c
al
l
en
gth
f
r
om
th
e o
r
i
g
i
n a
nd
σ
i
s
th
e
s
tan
d
ard d
ev
i
at
i
o
n o
f
th
e Ga
us
s
i
an
di
s
tr
i
bu
t
i
on
[
25
]
.
(
,
)
=
1
2
2
−
2
+
2
2
2
(
1)
T
hi
s
G
au
s
s
i
an
op
erat
i
o
n
w
i
l
l
r
ed
uc
e
the
n
oi
s
e
g
en
er
at
ed
i
n
t
he
s
eg
m
en
tat
i
on
pro
c
es
s
to
pro
du
c
e
a
g
oo
d
b
i
n
ar
y
i
m
ag
e
f
or
thres
ho
l
di
n
g
op
era
ti
on
on
Hue
,
S
at
urati
on
a
nd
Lu
m
i
na
nc
e
f
or
m
at
(
HS
L
)
.
In
th
i
s
pro
c
es
s
,
the
no
r
m
al
RG
B
c
o
l
ou
r
i
m
ag
e
i
s
c
on
v
erte
d
t
o
HS
L
.
S
i
nc
e
the
t
y
p
i
c
al
l
y
w
h
i
te
l
a
ne
m
ark
er
i
s
on
a
gr
e
y
ba
c
k
ground
,
th
e
l
um
i
na
nc
e
v
a
l
ue
i
s
ad
j
u
s
te
d
as
the
t
hres
ho
l
d,
us
e t
o c
o
nv
e
r
t th
e
i
m
ag
e t
o b
i
na
r
y
f
orm
a
t.
2.
2
.
E
xtr
ac
t
ing
T
h
e Road
M
ar
king
Cu
es
A
f
ter
i
n
v
es
t
i
ga
t
i
n
g
th
e
b
es
t
RO
I
f
or
proc
es
s
i
ng
the
i
m
ag
e,
the
n
ex
t
s
tep
i
s
t
o
ex
tr
ac
t
the
r
oa
d
m
ar
k
i
ng
c
ue
s
.
T
h
e
RO
I
w
i
l
l
b
e
s
c
an
ne
d
f
r
o
m
the
to
p
t
o
th
e
bo
t
tom
an
d
fr
om
the
l
ef
t
to
the
r
i
gh
t
as
s
ho
wn
b
y
t
he
arr
ow
s
i
n
F
i
gu
r
e
5
(
b)
.
A
bi
na
r
y
c
o
l
um
n
v
ec
tor
,
w
h
i
c
h
i
s
ex
tr
ac
ted
f
r
o
m
the
x
-
th
c
ol
um
n
of
the
bi
n
ar
y
i
m
ag
e
h
av
i
ng
b
i
n
ar
y
p
i
x
el
s
of
(
,
)
∈
{
0
,
1
}
for
=
0
,
1
,
⋯
,
and
=
0
,
1
,
⋯
,
,
i
s
d
ef
i
ne
d
as
f
ol
l
o
w
s
[
]
=
{
0
,
(
,
)
=
0
1
,
(
,
)
=
1
(
2)
T
hi
s
bi
na
r
y
c
o
l
um
n
v
ec
tor
t
el
l
s
h
o
w
m
an
y
whi
t
e
pi
x
el
s
f
ou
nd
on
th
e
x
-
th
c
ol
um
n.
W
h
e
n
the
r
e
ar
e
r
oa
d
m
ar
k
ers
c
r
os
s
i
ng
th
e
c
ol
um
n,
the
nu
m
be
r
of
w
h
i
te
pi
x
el
s
wi
l
l
i
nc
r
ea
s
e.
In
ord
er
to
r
ed
uc
e
the
c
om
pu
tat
i
o
na
l
c
om
pl
ex
i
t
y
,
no
t
al
l
c
o
l
um
ns
wi
l
l
be
r
ea
d
,
or
h
ori
z
on
t
al
l
y
s
c
an
ne
d
ov
er
the
i
m
ag
e.
It
i
s
s
uf
f
i
c
i
en
t
t
o
ho
r
i
z
on
ta
l
l
y
s
c
an
t
he
i
m
ag
e
f
or
ob
ta
i
n
i
ng
the
v
a
l
u
es
o
v
er
a
c
o
ns
tan
t
ho
r
i
z
on
t
al
i
n
terv
al
of
δx
as
i
n
F
i
gu
r
e
5
(
b)
,
w
h
i
c
h
i
s
t
he
n
um
be
r
of
pi
x
el
c
o
l
um
ns
of
the
b
i
na
r
y
i
m
ag
e
w
hi
c
h
are
s
k
i
pp
ed
a
nd
n
ot
s
c
an
ne
d,
as
l
o
ng
as
the
r
oa
d
m
ar
k
er
c
an
be
c
o
r
r
ec
tl
y
c
l
as
s
i
f
y
.
E
x
am
pl
e
of
s
c
an
n
i
n
g
pr
oc
es
s
on
the
i
m
ag
e
s
ho
w
n
i
n
F
i
gu
r
e
5
,
wi
th
v
ario
us
t
y
p
e
of
r
oa
d
m
ar
k
e
r
s
an
d d
i
f
f
erent v
al
u
e
of
δx
.
B
y
us
i
n
g
an
ex
am
pl
e
of
a
4
20
x
26
0
pi
x
e
l
s
i
z
e
f
or
an
R
O
I,
i
t
wi
l
l
r
eq
ui
r
es
4
1
ti
m
es
of
l
i
ne
s
s
c
an
ni
n
g
us
i
ng
wi
th
i
nte
r
v
a
l
s
i
z
e
of
δx
=
10
pi
x
e
l
s
,
pro
d
uc
i
ng
41
bi
na
r
y
c
ol
um
n
v
ec
tors
,
w
he
r
e
=
260
(
the
He
i
g
ht
s
i
z
e).
E
v
er
y
bi
n
ar
y
c
o
l
um
n
v
ec
tor,
wi
l
l
be
proc
es
s
ed
to
de
tec
t
a
nd
c
ou
nt
the
n
um
be
r
of
tr
an
s
i
ti
on
s
,
f
r
o
m
0
to
1
an
d
f
r
o
m
1
to
0
i
n
i
t
b
y
us
i
ng
a
b
i
t
w
i
s
e
ex
c
l
us
i
v
e
O
R
op
erat
i
on
be
t
ween
i
ts
e
l
f
,
and
,
w
h
i
c
h
i
s
a
v
ec
tor
produc
ed
f
r
om
s
hi
f
ti
ng
by
on
e
bi
n
ar
y
b
i
t
or el
em
en
t a
s
s
ho
w
n
i
n a
n e
x
am
pl
e
.
-
bi
t
wi
s
e
s
hi
f
t
to
the
r
i
gh
t
b
y
ad
di
n
g
0
to
th
e
f
i
r
s
t
bi
t
an
d
s
hi
f
t
the
r
es
t
t
o
the
r
i
g
ht
=
[0
00
11
1
10
0
0],
=
[
00
0
01
11
10
0]
-
pe
r
f
or
m
th
e e
x
c
l
us
i
v
e OR
wi
th
orig
i
n
al
v
a
l
ue
r
es
ul
t
i
n
g
on
l
y
tra
ns
i
t
i
on
bi
ts
are l
ef
t
-
=
⨁
=
[
00
01
11
10
00
]
⊕
[
00
00
1
11
10
0]
=
[
00
01
00
01
00
]
-
c
ou
nt
t
he
n
um
be
r
of
trans
i
ti
on
s
on
ev
er
y
s
c
an
l
i
ne
or b
i
na
r
y
c
o
l
um
n v
ec
tor
=
∑
[
]
=
1
=
2
T
he
nu
m
be
r
s
of
tr
an
s
i
ti
o
ns
i
n
ea
c
h
b
i
n
ar
y
c
o
l
um
n
v
e
c
tor
ob
ta
i
n
ed
f
r
om
the
s
c
an
ni
ng
l
i
n
e
are
s
av
ed
i
nto
a
v
ec
tor
,
.
A
n
ex
am
pl
e
of
thi
s
v
ec
to
r
produc
ed
f
r
o
m
s
c
an
ni
ng
a
n
RO
I
wi
th
an
i
nte
r
v
a
l
s
i
z
e
of
10
p
i
x
el
s
i
s
gi
v
e
n
i
n
F
i
gu
r
e
6.
T
he
nu
m
be
r
of
tr
an
s
i
ti
on
s
are
i
n
e
ac
h
b
i
n
ar
y
c
ol
um
n
v
e
c
tor
i
s
o
bs
erv
ed
to
b
e
e
i
th
er
0,
2
or
4.
F
or
t
he
r
o
ad
m
ark
er
S
S
,
th
ere
are
at
m
os
t
2
tr
an
s
i
ti
on
s
i
n
e
ac
h
v
ec
t
or
whereas
f
or
D,
the
n
um
be
r
of
tr
an
s
i
t
i
o
ns
i
s
e
i
th
er
0
or
2.
F
or
DD,
the
n
um
be
r
of
tr
an
s
i
ti
on
s
i
s
4,
w
he
r
e
as
f
or
S
D
an
d
DS
,
the
n
um
be
r
of
tr
an
s
i
t
i
on
s
i
s
ei
the
r
2
or
4.
T
o
di
f
f
erenti
ate
b
et
w
e
en
S
D
a
nd
D
S
,
ad
d
i
ti
on
al
f
e
atu
r
es
w
i
l
l
be
ex
tr
ac
ted
t
o
di
f
f
erenti
ate
be
t
w
e
en
t
he
s
e t
wo c
l
as
s
es
.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
4
,
A
ug
us
t
20
19
:
20
4
8
-
20
5
7
2052
(
a)
(
b)
(
c
)
(
d)
F
i
gu
r
e
5
.
E
x
am
pl
e o
f
s
c
an
ni
ng
i
m
ag
e
w
i
t
h d
i
f
f
erent i
nt
e
r
v
al
s
i
z
e
wi
th
di
f
f
erent road
m
ar
k
e
r
s
V
ec
tor
w
i
l
l
b
e
proc
es
s
ed
w
h
ere
th
e
nu
m
be
r
of
tr
an
s
i
t
i
on
s
w
i
l
l
be
gro
up
e
d
an
d
the
pe
r
c
en
t
ag
e
of
,
or
the
n
um
be
r
s
of
tr
an
s
i
ti
o
ns
wi
l
l
b
e
s
tore
d
i
n
v
ec
tor
as
f
orm
ul
at
ed
b
y
p
in
(
3),
whi
c
h
i
s
ap
p
l
i
ed
to
c
al
c
ul
ate
t
he
pe
r
c
en
tag
e
of
f
or
f
i
v
e
d
i
f
f
erent
c
l
as
s
es
as
tab
u
l
at
ed
i
n T
ab
l
e 1
.
A
c
c
ordi
n
g t
o
thi
s
t
ab
l
e,
1
=
0
,
2
=
2
and
3
=
4
.
p
=
Fr
e
q
u
e
n
c
y
o
f
Total
f
r
e
q
u
e
n
c
y
×
100%
(
3)
T
ab
l
e
1
.
E
x
am
pl
e o
f
T
r
an
s
i
ti
on
P
e
r
c
en
tag
es
f
r
o
m
(
f
r
om
DS
I
m
ag
e Fi
gu
r
e 6
)
Fr
e
q
u
e
n
c
y
o
f
p
(
%
)
0
1
2
.
4
4
2
26
6
3
.
4
1
4
14
3
4
.
1
5
=
[2
, 2
,
2,
2
, 2
,
2,
2
, 2
,
2,
2
,
2,
2,
2,
2
, 2
,
2,
2,
2,2
,
2,
4
, 4
,
4,
4
, 4
,
4,
4,
4,
4,
4,
4,
4,
4,
4,
2,
2,
2,
2,
2,
2,
0]
F
i
gu
r
e
6
.
E
x
am
pl
e o
f
r
es
ul
t
f
or
af
ter s
c
an
ni
ng
proc
es
s
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
930
◼
Road
ma
r
k
ers
c
l
as
s
i
f
i
c
ati
on
us
i
ng
bi
na
r
y
s
c
an
n
i
ng
…
(
Z
am
an
i
Md
S
an
i
)
2053
2.
3.
T
h
e
F
ir
st
and
S
ec
o
n
d
La
y
er
Cla
ss
if
ic
atio
n
2.
3
.1
.
De
sign
ing
t
h
e
F
ilte
r
ing
T
able
f
o
r
F
ir
st L
a
ye
r
Clas
s
if
ica
t
ion
T
he
pe
r
c
en
ta
ge
s
of
ob
ta
i
ne
d
whe
n
proc
e
s
s
i
ng
10
0
f
r
a
m
es
f
r
o
m
ea
c
h
of
the
f
i
v
e
c
l
as
s
es
are
pl
ott
e
d
i
n
F
i
gu
r
e
7
.
T
he
X
-
ax
i
s
r
ep
r
es
en
ts
where
as
Y
-
ax
i
s
r
e
pres
en
t
s
p
,
w
h
i
c
h
is
the
pe
r
c
en
t
ag
e
of
.
In
o
ur
m
eth
od
,
w
e
us
e
10
p
i
x
e
l
s
as
the
i
nte
r
v
a
l
s
i
z
e
f
or
t
he
s
c
an
ni
ng
proc
es
s
.
F
r
o
m
ob
s
erv
at
i
on
,
a
thres
ho
l
d
c
a
n
be
s
et
f
r
o
m
the
da
ta
to
e
na
b
l
e
t
he
c
l
as
s
i
f
i
c
ati
on
f
or
the
m
ark
er
s
.
A
t
t
hres
ho
l
d
v
a
l
ue
of
20
%
as
s
ho
wn
a
s
r
ed
da
s
h
l
i
ne
,
a
f
i
l
ter
i
ng
tab
l
e
c
an
be
de
s
i
g
ne
d
as
i
n
T
ab
l
e
2.
A
t
l
e
as
t
2
t
y
p
es
of
m
ar
k
ers
c
an
b
e
c
l
as
s
i
f
i
ed
,
where
as
S
S
an
d
D
S
/S
D
c
l
as
s
es
c
ou
l
d
pro
du
c
ed
th
e
s
am
e
r
es
ul
t.
T
o
i
m
prov
e
th
i
s
pro
bl
em
,
an
ad
d
i
t
i
o
na
l
pe
r
c
en
tag
e
v
a
l
ue
i
s
i
ntrod
uc
ed
,
w
h
i
c
h
i
s
4
p
=
6
,
w
h
i
c
h
i
s
a
m
ul
ti
p
l
i
c
at
i
o
n
of
2
p
=
2
an
d
3
p
=
4
an
d
l
ate
r
gai
n
b
y
2
as
f
or
m
ul
ate
d
i
n
(
4)
are
pl
o
tte
d
i
n
F
i
gu
r
e
s
7
(
a)
-
(
e)
as
i
ns
i
de
the
b
l
ue
d
as
h
bo
x
.
F
r
om
thi
s
ad
d
i
t
i
on
al
pe
r
c
e
nta
g
e
v
a
l
ue
,
t
he
f
i
l
teri
ng
ta
bl
e
c
a
n
b
e
i
m
prov
ed
as
i
n
T
ab
l
e
2
to
c
l
as
s
i
f
y
three
r
oa
d
m
ar
k
er
t
y
pe
s
na
m
el
y
D,
DD
an
d
S
S
.
A
s
f
or
S
D
an
d
DS
r
o
ad
m
ar
k
er
t
y
pe
,
the
a
l
g
orit
hm
wi
l
l
proc
e
ed
t
o t
he
s
ec
on
d c
l
as
s
i
f
i
er as
d
es
c
r
i
be
d
i
n t
he
ne
x
t s
ec
ti
on
.
4
p
=
(
2
p
×
3
p
)
∗
2
(
4)
(
a)
(
b)
(
c
)
(
d)
(
e)
F
i
gu
r
e
1
.
P
l
o
tti
ng
P
erc
en
t
a
ge
of
(
%)
f
or 5 di
f
f
erent c
l
as
s
es
w
i
t
h 2
0% thres
h
ol
d l
i
n
e
w
i
t
h
,
4
p
=
6
i
nc
l
ud
e
d a
s
i
ns
i
d
e t
h
e b
l
ue
da
s
h b
ox
(
a) DD
(
b)D
(
c
)
D
S
(
d) S
D (
e)S
S
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
4
,
A
ug
us
t
20
19
:
20
4
8
-
20
5
7
2054
T
ab
l
e
2
.
F
i
l
t
erin
g T
ab
l
e
C
o
ns
tr
uc
ti
on
f
r
om
th
e Gr
ap
h
O
bs
erv
at
i
on
Inc
l
us
i
v
e o
f
4
p
=
6
C
o
n
d
it
ion
f
o
r
c
la
s
s
if
y
ing
p
>
2
0
%
B
ina
r
y
t
r
a
n
s
i
t
ion
0
2
4
6
D
a
s
h
e
d
(
D
)
C1
Y
e
s
Y
e
s
/
N
o
No
No
D
o
u
b
le
S
o
li
d
(
D
D
)
C2
No
No
Y
e
s
Y
e
s
/
N
o
S
ing
le
(
S
S
)
C3
No
Y
e
s
No
No
S
o
li
d
-
D
a
s
h
e
d
(
S
D
)
C4
No
Y
e
s
Y
e
s
/
N
o
Y
e
s
D
a
s
h
e
d
-
S
o
li
d
(
D
S
)
C5
No
Y
e
s
Y
e
s
/
N
o
Y
e
s
2.
3
.2
.
Co
n
t
o
u
r
s
S
lop
e (
S
e
con
d
La
y
e
r
Cla
ss
if
i
er
)
In
the
s
ec
o
nd
l
a
y
er
c
l
as
s
i
f
i
c
ati
on
,
the
ai
m
i
s
to
s
ep
arate
th
e
DS
an
d
S
D
c
l
as
s
e
s
w
h
i
c
h
r
eq
ui
r
es
th
e
proc
es
s
to
de
t
ec
t
th
e
c
o
nto
urs
(
l
on
g
an
d
s
ho
r
t
m
ar
k
ers
)
.
A
DS
c
l
as
s
i
s
r
ec
og
n
i
z
ed
b
y
t
he
l
oc
ati
on
of
the
s
ho
r
t
er
c
on
tou
r
(
of
the
da
s
he
d
m
ar
k
e
r
)
be
l
o
w
th
e
l
on
g
er
c
on
to
ur
(
the
s
ol
i
d
m
ar
k
e
r
)
an
d a
n
S
D
c
l
as
s
i
s
r
ec
og
n
i
z
ed
v
i
c
e
v
ers
a.
In
F
i
gu
r
e
8,
the
l
o
ng
er c
o
nto
ur i
s
i
nd
i
c
ate
d b
y
two
po
i
nts
,
A
an
d
B
,
an
d
i
ts
bo
th
en
ds
.
T
he
s
ho
r
ter
c
on
tou
r
i
s
i
nd
i
c
ate
d
b
y
p
oi
nt
C,
w
h
i
c
h
i
s
the
c
en
tr
o
i
d
of
the
s
ho
r
ter
c
on
tou
r
.
He
nc
e
b
y
k
no
w
i
n
g
the
l
oc
a
ti
o
n
of
C,
w
h
eth
er
i
t’
s
be
l
o
w
or
ab
o
v
e t
he
l
on
ge
r
c
on
tou
r
, t
he
c
l
as
s
S
D or D
S
c
an
be
i
de
nt
i
f
i
ed
.
(
a)
(
b)
F
i
gu
r
e
2
. T
hree p
oi
n
ts
to
c
a
l
c
ul
a
te
t
he
s
l
op
e (
a) DS c
l
a
s
s
(
b) SD c
l
as
s
T
he
s
l
op
e
of
=
−
−
,
be
t
w
ee
n
A
a
nd
C
wi
l
l
be
s
m
al
l
er
tha
n
th
e
s
l
o
pe
of
=
−
−
be
t
w
e
en
A
an
d
B
,
i
f
the
s
ho
r
ter
c
on
to
ur
i
s
be
l
o
w
t
he
l
on
g
er
c
on
to
ur
w
hi
c
h
i
s
f
ou
nd
i
n
D
S
c
l
as
s
.
O
n
t
he
oth
er
ha
n
d,
t
he
s
l
op
e
of
,
wi
l
l
b
e
l
arg
er
tha
n
t
he
s
l
op
e,
,
i
f
the
s
ho
r
ter
c
on
t
ou
r
i
s
be
l
o
w
th
e
l
o
ng
er
c
on
tou
r
f
or
S
D
c
l
as
s
.
If
bo
th
and
are
m
ul
ti
pl
i
ed
wi
th
(
)
(
)
x
x
x
x
B
A
C
A
−
−
, th
e
n t
h
e res
u
l
t o
f
th
e
m
ul
ti
pl
i
c
a
ti
o
n,
(
)
(
)
(
)
(
)
x
x
y
y
x
x
x
x
AC
B
A
C
A
B
A
C
A
m
−
−
=
−
−
,
(
5)
wi
l
l
s
ti
l
l
be
s
m
al
l
er th
an
the
r
es
ul
t o
f
an
oth
er m
ul
ti
pl
i
c
ati
on
f
or DS c
l
as
s
.
(
)
(
)
(
)
(
)
x
x
y
y
x
x
x
x
AB
C
A
B
A
B
A
C
A
m
−
−
=
−
−
,
(
6)
If
the
di
f
f
erenc
e
be
twe
en
the
s
e
t
wo
m
ul
ti
pl
i
c
a
ti
on
s
i
s
de
no
t
ed
as
,
the
f
ol
l
o
wi
ng
eq
ua
ti
o
n
y
i
e
l
ds
(
)
(
)
(
)
(
)
x
x
y
y
x
x
y
y
C
A
B
A
B
A
C
A
−
−
−
−
−
=
,
(
7)
the
n
the
v
al
ue
of
c
an
be
u
s
ed
as
t
he
f
ea
t
ure
t
o
d
ete
r
m
i
ne
or
s
eg
r
eg
ate
be
t
wee
n
S
D
an
d
D
S
c
l
as
s
es
.
If
the
<
0,
th
en
th
e
r
oa
d
m
ar
k
er
i
s
c
l
as
s
i
f
i
ed
as
DS
c
l
as
s
.
If
,
0
the
n
the
r
oa
d
m
ar
k
e
r
i
s
c
l
as
s
i
f
i
ed
as
S
D
c
l
as
s
.
S
i
nc
e
th
e
c
l
as
s
i
f
i
c
at
i
on
i
s
on
l
y
c
arr
i
ed
ou
t
t
o
i
d
en
ti
f
y
be
t
ween
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
930
◼
Road
ma
r
k
ers
c
l
as
s
i
f
i
c
ati
on
us
i
ng
bi
na
r
y
s
c
an
n
i
ng
…
(
Z
am
an
i
Md
S
an
i
)
2055
S
D
a
nd
D
S
f
or
th
i
s
s
ec
on
d
l
a
y
er
c
l
as
s
i
f
i
c
ati
o
n,
th
e
v
al
ue
of
wi
l
l
no
t
e
qu
i
v
a
l
e
nt
a
s
z
ero
as
C
wi
l
l
n
ev
er be
l
oc
at
ed
on
t
he
s
tr
ai
gh
t
l
i
ne
be
t
wee
n A
an
d B
.
2.
3
.3
.
O
u
t
lie
r
s a
n
d
Imp
r
o
v
ing
t
h
e
Clas
sificatio
n
R
es
u
lt
B
y
o
bs
erv
i
ng
th
e
grap
hs
i
n
F
i
gu
r
e
7,
t
he
r
e
are
p
ot
en
ti
al
o
utl
i
ers
ge
n
erate
d
d
ue
to
the
v
al
u
es
c
al
c
ul
ate
d
ne
ar
to
the
thres
h
ol
d
v
a
l
ue
s
a
t
20
%
as
i
nd
i
c
at
ed
b
y
the
r
ed
c
i
r
c
l
e
i
n
F
i
gu
r
e
s
7
(
a),
(
c
)
and
(
e).
O
the
r
tha
n
i
ntr
od
uc
i
ng
4
p
=
6
,
thi
s
po
s
s
i
bl
e
f
al
s
e
err
or
c
an
al
s
o
be
r
es
ol
v
ed
b
y
us
i
ng
tem
po
r
al
v
al
ue
s
be
t
wee
n
the
v
i
de
o
i
m
ag
es
.
T
he
m
ar
k
e
r
s
ten
d
to
b
e
the
s
i
m
i
l
arl
y
l
ab
el
e
d
f
or
a
s
et
of
ad
j
ac
en
t
f
r
a
m
es
.
B
y
u
s
i
n
g
th
i
s
po
s
s
i
b
i
l
i
t
y
,
the
pr
ev
i
ou
s
c
l
as
s
i
f
i
c
ati
on
r
es
u
l
t
w
o
ul
d
be
s
tored
an
d
i
f
the
r
e
i
s
a
tr
an
s
i
ti
on
of
a
c
l
as
s
or
m
ar
k
er
t
y
pe
,
i
t
wi
l
l
c
on
f
i
r
m
on
the
ne
x
t
te
n
(
10
)
f
r
a
m
es
o
f
s
eq
ue
nc
e.
T
hi
s
m
eth
od
al
s
o
i
m
prov
ed
t
he
i
s
s
ue
r
el
ate
d
to
the
v
e
hi
c
l
e
b
l
oc
k
i
ng
t
h
e
r
oa
d
m
ar
k
er
as
the
r
e
i
s
no
f
i
r
m
c
l
as
s
i
f
i
c
ati
on
r
es
ul
t
wh
en
th
i
s
oc
c
urs
an
d
the
r
es
ul
t
wi
l
l
ho
l
d
t
he
l
as
t
c
l
as
s
i
f
i
c
ati
o
n
r
es
u
l
t
as
r
ef
erenc
e.
He
nc
e
t
he
v
a
l
i
da
t
i
o
n
f
or
tr
an
s
i
ti
on
t
y
p
es
of
m
ar
k
er r
eq
u
i
r
es
10
*
0.0
33
s
=
0.3
3s
at
th
e m
os
t.
3.
Re
sult
s
a
n
d
A
n
al
y
s
is
T
he
propos
ed
t
w
o
-
l
a
y
er
c
l
as
s
i
f
i
c
ati
on
m
eth
od
s
are
i
m
pl
e
m
en
ted
to
c
l
as
s
i
f
y
th
e
r
oa
d
m
ar
k
e
r
s
fr
o
m
a
nu
m
be
r
of
v
i
de
o
c
l
i
ps
r
ec
orde
d
whe
n
dr
i
v
i
ng
at
d
i
f
f
erent
r
oa
d
s
c
en
ario
s
wi
th
di
f
f
erent
r
oa
d
m
ar
k
ers
.
T
h
e
s
pe
ed
of
th
e
c
ar
w
h
i
l
e
t
he
s
e
v
i
de
o
c
l
i
ps
w
ere
tak
en
was
be
t
w
e
en
70
-
80
km
/hr.
In
ord
er
to
r
u
n
the
prop
os
ed
r
oa
d
m
ark
er
c
l
as
s
i
f
i
c
ati
o
n
m
od
el
,
the
f
ea
tures
f
r
o
m
the
f
i
r
s
t
three
v
i
de
o
c
l
i
ps
gi
v
en
i
n
T
ab
l
e
3
are
ex
tr
ac
ted
an
d
a
na
l
y
z
ed
to
s
e
e
the
s
i
gn
i
f
i
c
an
t
pa
tte
r
ns
f
or
the
c
l
as
s
i
f
i
c
at
i
on
proc
es
s
an
d
i
t
c
on
ta
i
n
al
l
t
y
pe
s
of
th
e
r
o
ad
m
ar
k
ers
w
hi
c
h
are
ai
m
ed
to
be
c
l
as
s
i
f
i
e
d.
Ma
nu
a
l
c
om
pa
r
i
s
on
i
s
pe
r
f
orm
ed
on
th
e
pred
i
c
ted
r
oa
d
m
ar
k
er
t
y
pe
of
the
a
l
go
r
i
thm
w
i
th
th
e
gro
un
d
tr
u
th
o
n
e
v
er
y
f
r
am
e
an
d
t
he
ac
c
urac
y
i
s
c
al
c
u
l
ate
d
as
i
n
(
8)
where
t
he
t
ota
l
f
r
am
es
c
l
as
s
i
f
i
ed
c
orr
ec
tl
y
f
r
om
the
g
r
ou
nd
tr
uth
∑
i
s
d
i
v
i
de
d
wi
th
the
t
ota
l
f
r
a
m
es
,
∑
.
It
c
an
be
ob
s
er
v
ed
th
at
t
he
ac
c
urac
y
ac
h
i
ev
e
d
i
n
t
he
tr
a
i
n
i
n
g
v
i
d
eo
s
are
w
e
l
l
ab
o
v
e
98
%
ex
c
ep
t
c
l
i
p
n
um
be
r
3
at
93
%
whi
c
h
d
ue
t
o
th
e
f
ai
nt
l
an
e
m
ar
k
ers
de
tec
t
ed
i
n
the
v
i
de
o.
T
hi
s
i
s
es
s
en
t
i
a
l
to
e
n
s
ure
t
ha
t
the
c
l
as
s
i
f
i
c
a
ti
on
pe
r
f
orm
an
c
e
ac
hi
e
v
e
d
when
r
un
n
i
n
g
thi
s
m
od
el
o
n t
h
e t
es
t
i
ng
v
i
de
os
i
s
as
go
o
d a
s
p
os
s
i
b
l
e.
=
∑
∑
100%
(
8)
T
ab
l
e
3.
V
ari
ou
s
C
l
i
ps
an
d
A
c
c
urac
y
C
li
p
N
a
m
e
R
e
s
o
lut
ion
FP
S
Ty
p
e
Fr
a
m
e
s
Fea
t
u
r
e
s
A
n
a
ly
s
i
s
Tes
t
A
c
c
u
r
a
c
y
(
%
)
1
V
id2
4
1
2
8
0
x
7
2
0
30
D
-
DD
-
SD
1840
x
x
9
8
.
0
4
%
2
V
id2
5
1
2
8
0
x
7
2
0
30
DD
-
DS
1099
x
x
100%
3
V
id2
3
1
2
8
0
x
7
2
0
30
DD
-
S
357
x
x
9
3
.
8
4
%
4
V
id_
1
_
Tes
t
1
2
8
0
x
7
2
0
30
DD
-
DS
-
D
2019
x
9
7
.
1
7
%
5
V
id_
2
_
Tes
t
1
2
8
0
x
7
2
0
30
DD
-
D
796
x
9
4
.
4
7
%
6
V
id_
3
_
Tes
t
1
2
8
0
x
7
2
0
30
D
-
SD
-
DD
1287
x
100%
7
V
id_
4
_
Tes
t
1
2
8
0
x
7
2
0
30
DD
-
D
-
SD
1547
x
9
7
.
5
4
%
8
V
id_
5
_
Tes
t
1
9
2
0
x
1
0
8
0
30
SS
763
x
100%
T
he
propos
ed
c
l
as
s
i
f
i
c
ati
o
n
m
od
el
w
as
l
ate
r
t
es
ted
wi
th
f
i
v
e
v
i
de
o
c
l
i
ps
,
w
h
i
c
h
are
nu
m
be
r
ed
f
r
o
m
4
to
8
i
n
T
ab
l
e
3.
I
n
v
i
d
eo
c
l
i
p
V
i
d_
2
_T
es
t,
w
h
i
c
h
c
on
t
ai
ns
79
6
f
r
a
m
es
ha
v
i
ng
r
oa
d
m
ar
k
ers
of
ty
pe
s
DD
an
d
D,
th
e
r
ec
orded
ac
c
urac
y
i
s
94
.
47
%,
w
h
i
c
h
s
ti
l
l
h
i
gh
er
tha
n
9
0%
bu
t
s
tan
ds
as
the
l
o
w
es
t
am
on
gs
t
al
l
ot
he
r
ac
c
urac
y
v
al
u
es
ac
hi
ev
ed
.
T
hi
s
i
s
m
ai
nl
y
du
e
to
the
qu
a
l
i
t
y
of
th
i
s
v
i
de
o
c
l
i
p
whi
c
h
i
s
m
ore
bl
urr
y
c
o
m
pa
r
ed
on
t
he
oth
er
c
l
i
ps
.
T
he
m
ar
k
e
r
tr
an
s
i
ti
on
h
ad
s
ho
wn
m
i
ni
m
u
m
e
f
f
ec
t
on
the
c
l
as
s
i
f
i
c
ati
on
ac
c
urac
y
as
the
al
go
r
i
th
m
s
uc
c
es
s
f
ul
l
y
r
e
v
a
l
i
d
ate
d
be
t
w
e
en
th
e
f
al
s
e
an
d
tr
ue
tr
an
s
i
ti
on
a
nd
c
on
tr
i
b
ute
d
on
the
ac
c
uc
arc
y
c
a
l
c
ul
ati
on
.
A
s
f
or
the
c
o
nf
us
i
on
m
atri
x
as
i
n
T
ab
l
e
4,
we
c
a
n
f
i
nd
o
ut
th
at
m
os
t
err
or
i
s
c
om
i
ng
f
r
o
m
the
DD c
l
as
s
du
e t
o
the
th
r
es
ho
l
di
n
g
proc
es
s
r
es
ul
t
i
n
g
th
e
o
utp
u
t
of
c
om
bi
ne
d c
o
nto
ur
at
th
e
to
p
of
the
l
a
ne
s
w
h
i
c
h
i
s
v
er
y
ha
r
d
to
di
s
t
i
n
g
ui
s
h
i
ts
thres
ho
l
d
v
al
u
e.
E
x
am
pl
es
of
the
thres
ho
l
d
ed
i
m
ag
es
are
as
i
n
F
i
g
ure
9
as
i
n
th
e
bl
u
e
c
i
r
c
l
e.
I
n
t
hi
s
c
i
r
c
l
e
are
a,
t
he
s
c
an
ni
ng
pr
oc
es
s
wi
l
l
r
es
ul
ted
th
i
s
as
=
2
i
ns
tea
d o
f
=
2
an
d t
he
c
l
as
s
i
f
i
c
ati
o
n
w
o
ul
d
l
ea
d t
o
S
D
i
ns
te
ad
of
DD.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
4
,
A
ug
us
t
20
19
:
20
4
8
-
20
5
7
2056
F
i
gu
r
e
9
. T
hres
ho
l
di
n
g o
utp
ut
f
or DD
T
ab
l
e
4
.
C
on
f
us
i
o
n M
a
tr
i
x
T
ab
l
e
P
r
e
d
ic
t
e
d
C
las
s
D
a
s
h
e
d
D
o
u
b
le
S
o
li
d
S
ing
le
S
o
li
d
-
D
a
s
h
e
d
D
a
s
h
e
d
-
S
o
li
d
A
c
t
u
a
l
C
la
s
s
C1
C2
C3
C4
C5
C1
1254
32
0
0
0
9
7
.
4
4
%
C2
0
2476
0
128
0
9
4
.
8
3
%
C3
0
0
763
0
0
100%
C4
0
0
0
1080
0
100%
C5
0
0
0
0
679
100%
4.
Co
n
clus
ion
T
he
pa
pe
r
pres
en
ted
a
ne
w
ap
proac
h
f
or
l
an
e
m
ar
k
er
c
l
as
s
i
f
i
c
ati
o
n.
T
he
RO
I
s
e
l
ec
ti
on
i
s
ba
s
ed
on
th
e
c
am
era
p
os
i
t
i
on
to
c
ap
t
ure
the
r
o
ad
i
m
ag
e.
T
he
RO
I
i
s
pr
oc
es
s
ed
a
nd
the
f
ea
tures
of
the
l
a
ne
m
ar
k
i
ng
are
ex
tr
ac
te
d
f
or
the
tr
a
i
n
i
ng
pu
r
p
os
es
an
d
l
ate
r
f
or
the
c
l
as
s
i
f
i
c
ati
on
proc
es
s
.
T
hi
s
ne
w
al
go
r
i
t
hm
w
as
tes
ted
w
i
t
h
the
l
oc
al
n
e
w
v
i
d
eo
s
an
d
the
gro
un
d
tr
uth
.
E
x
pe
r
i
m
en
ta
l
r
es
ul
ts
ha
v
e
s
ho
wn
tha
t
t
he
ac
c
urac
y
v
a
l
u
es
ac
hi
e
v
e
d
b
y
the
pro
po
s
ed
ne
w
m
eth
od
ac
hi
e
v
e
d
m
ore
tha
n
93
%.
T
he
err
ors
f
r
o
m
the
c
l
as
s
i
f
i
c
ati
o
n
op
erat
i
on
are
m
ai
nl
y
du
e
to
the
f
ai
n
t
r
oa
d
m
ar
k
er
w
hi
c
h
c
au
s
ed
w
r
on
g
pr
e
-
proc
es
s
i
ng
r
es
ul
t,
s
ud
de
n
i
l
l
u
m
i
na
ti
on
c
ha
ng
es
an
d
th
e d
i
f
f
erenc
e i
n t
h
e d
i
m
en
s
i
on
of
th
e
l
an
e m
ar
k
ers
.
A
c
kno
w
ledg
ement
s
T
he
au
th
ors
w
ou
l
d
l
i
k
e
to
ac
k
no
wl
ed
ge
t
he
f
un
di
ng
s
up
po
r
t
r
ec
ei
v
e
d
f
r
om
Uni
v
ers
i
ti
T
e
k
ni
k
al
Ma
l
a
y
s
i
a
Me
l
ak
a
(
UT
eM
)
fr
o
m
P
J
P
/20
14
/FK
E
(
15
D)
/
S
01
35
7
–
Ro
ad
L
an
e
Rec
og
ni
t
i
o
n
Us
i
ng
V
i
s
i
on
an
d
A
r
t
i
f
i
c
i
al
I
nte
l
l
i
ge
nc
e
f
or Au
to
A
s
s
i
s
t
Dr
i
v
er
A
l
ert S
y
s
tem
.
Ref
er
en
ce
s
[1
]
Doy
l
e
DD
,
J
e
n
n
i
n
g
s
AL
,
Bl
a
c
k
J
T
.
O
p
ti
c
a
l
F
l
o
w
Ba
c
k
g
r
o
u
n
d
Es
t
i
m
a
ti
o
n
f
o
r
Rea
l
-
T
i
m
e
Pa
n
/T
i
l
t
Cam
e
ra
Ob
j
e
c
t
T
ra
c
k
i
n
g
.
M
e
a
s
u
re
m
e
n
t
2
0
1
4
;
48
:
1
9
5
–
2
0
7
.
[2
]
L
i
u
W
,
Zh
a
n
g
Z
,
L
i
S,
T
a
o
D.
Roa
d
Det
e
c
ti
o
n
By
U
s
i
n
g
A
G
e
n
e
ra
l
i
z
e
d
Ho
u
g
h
T
ra
n
s
fo
r
m
.
Rem
o
t
e
Se
n
s
.
2
0
1
7
;
9
(6
)
:
5
9
0
.
[3
]
So
n
J
,
Y
o
o
H,
Ki
m
S
,
So
h
n
K.
Rea
l
-
t
i
m
e
Il
l
u
m
i
n
a
t
i
o
n
In
v
a
ri
a
n
t
L
a
n
e
Det
e
c
ti
o
n
f
o
r
L
a
n
e
Dep
a
rtu
r
e
W
a
r
n
i
n
g
Sy
s
te
m
.
E
x
p
e
rt
Sy
s
t
Ap
p
l
.
2
0
1
5
;
4
2
(
4
)
:
1
8
1
6
–
1
8
2
4
.
[4
]
D
o
rj
B,
L
e
e
DJ
.
A
Pre
c
i
s
e
L
a
n
e
Det
e
c
ti
o
n
A
l
g
o
r
i
th
m
B
a
s
e
d
o
n
T
o
p
Vi
e
w
Im
a
g
e
T
ra
n
s
fo
r
m
a
ti
o
n
a
n
d
L
e
a
s
t
-
Sq
u
a
r
e
Ap
p
r
o
a
c
h
e
s
.
J
Se
n
s
o
r
s
.
2
0
1
6
;
2
0
1
6
.
[5
]
Sa
tz
o
d
a
,
R
K
,
T
ri
v
e
d
i
,
M
M
.
Vi
s
i
o
n
-
B
a
s
e
d
L
a
n
e
An
a
l
y
s
i
s
:
E
x
p
l
o
r
a
ti
o
n
o
f
Is
s
u
e
s
a
n
d
Ap
p
ro
a
c
h
e
s
f
o
r
Em
b
e
d
d
e
d
Rea
l
i
z
a
ti
o
n
.
IEE
E
Com
p
u
te
r
So
c
i
e
ty
Con
fe
r
e
n
c
e
o
n
Com
p
u
te
r
Vi
s
i
o
n
a
n
d
Pa
tt
e
r
n
Rec
o
g
n
i
ti
o
n
W
o
rk
s
h
o
p
s
.
2013
:
604
–
609
.
[6
]
An
X
,
Sh
a
n
g
E,
So
n
g
J
,
L
i
J
,
He
H.
Rea
l
-
T
i
m
e
L
a
n
e
De
p
a
rt
u
re
W
a
rn
i
n
g
Sy
s
t
e
m
Ba
s
e
d
O
n
A
S
i
n
g
l
e
FPGA
.
EUR
ASIP
J
o
u
rn
a
l
o
n
I
m
a
g
e
a
n
d
Vi
d
e
o
Pro
c
e
s
s
i
n
g
.
2
0
1
3
:
3
8
.
[7
]
Chi
ra
IM
,
Chi
b
u
l
c
u
te
a
n
A
,
Da
n
e
s
c
u
RG
.
Rea
l
-
Ti
m
e
De
te
c
ti
o
n
o
f
Ro
a
d
M
a
rk
i
n
g
s
f
o
r
Dri
v
i
n
g
As
s
i
s
ta
n
c
e
Ap
p
l
i
c
a
ti
o
n
s
.
Com
p
u
te
r
E
n
g
i
n
e
e
ri
n
g
a
n
d
Sy
s
te
m
s
ICCE
S
2
0
1
0
I
n
te
rn
a
ti
o
n
a
l
Co
n
fe
r
e
n
c
e
.
2
0
1
0
:
158
–
1
6
3
.
[8
]
W
u
T
,
Ran
g
a
n
a
th
a
n
A
.
A
Pra
c
ti
c
a
l
S
y
s
t
e
m
Fo
r
Roa
d
M
a
rk
i
n
g
Det
e
c
t
i
o
n
An
d
Rec
o
g
n
i
ti
o
n
.
IEE
E
In
te
l
l
i
g
e
n
t
Ve
h
i
c
l
e
s
Sy
m
p
o
s
i
u
m
,
Pro
c
e
e
d
i
n
g
s
,
2
0
1
2
:
25
–
3
0
.
[9
]
Kh
e
y
ro
l
l
a
h
i
A,
Bre
c
k
o
n
T
P.
Au
to
m
a
ti
c
re
a
l
-
t
i
m
e
ro
a
d
m
a
rk
i
n
g
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
a
fe
a
tu
re
d
ri
v
e
n
a
p
p
ro
a
c
h
.
M
a
c
h
i
n
e
V
i
s
i
o
n
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
.
2
0
1
2
;
2
3
(1
)
:
1
2
3
–
1
3
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
930
◼
Road
ma
r
k
ers
c
l
as
s
i
f
i
c
ati
on
us
i
ng
bi
na
r
y
s
c
an
n
i
ng
…
(
Z
am
an
i
Md
S
an
i
)
2057
[1
0
]
M
c
Cal
l
J
C,
T
ri
v
e
d
i
M
M
.
Vi
d
e
o
-
Ba
s
e
d
L
a
n
e
E
s
ti
m
a
t
i
o
n
a
n
d
T
ra
c
k
i
n
g
f
o
r
Dri
v
e
r
A
s
s
i
s
ta
n
c
e
:
Su
rv
e
y
,
Sy
s
te
m
,
a
n
d
Ev
a
l
u
a
ti
o
n
.
IEEE
Tra
n
s
a
c
t
i
o
n
s
o
n
I
n
te
l
l
i
g
e
n
t
T
ra
n
s
p
o
r
ta
ti
o
n
S
y
s
te
m
s
.
2
0
0
6
;
7
(
1
)
:
20
–
3
7
.
[1
1
]
Y
e
n
i
k
a
y
a
S,
Y
e
n
i
k
a
y
a
G
,
Dü
v
e
n
E.
K
e
e
p
i
n
g
th
e
Ve
h
i
c
l
e
o
n
th
e
R
o
a
d
:
A
Su
rv
e
y
o
n
O
n
-
ro
a
d
L
a
n
e
Det
e
c
t
i
o
n
Sy
s
t
e
m
s
.
ACM
Co
m
p
u
t.
Su
r
v
.
2
0
1
3
;
4
6
(1
):
4
3
.
[1
2
]
Ba
r
Hil
l
e
l
A
,
L
e
r
n
e
r
R,
L
e
v
i
D,
Raz
G
.
Rec
e
n
t
Pro
g
re
s
s
In
Roa
d
An
d
L
a
n
e
Det
e
c
t
i
o
n
:
A
Su
rv
e
y
.
M
a
c
h
i
n
e
Vi
s
i
o
n
a
n
d
Ap
p
l
i
c
a
ti
o
n
s
.
2
0
1
4
;
2
5
(3
)
:
7
2
7
–
7
4
5
.
[1
3
]
Col
l
a
d
o
J
M
,
Hil
a
ri
o
C,
De
L
a
Es
c
a
l
e
ra
A,
Ar
m
i
n
g
o
l
J
M
.
Det
e
c
ti
o
n
An
d
Cl
a
s
s
i
f
i
c
a
ti
o
n
o
f
R
o
a
d
L
a
n
e
s
Wi
th
A
Fr
e
q
u
e
n
c
y
An
a
l
y
s
i
s
.
I
n
:
I
EEE I
n
te
l
l
i
g
e
n
t
V
e
h
i
c
l
e
s
Sy
m
p
o
s
i
u
m
,
Pro
c
e
e
d
i
n
g
s
.
2
0
0
5
:
78
–
83.
[1
4
]
Sc
h
u
b
e
rt
R,
S
c
h
u
l
z
e
K,
W
a
n
i
e
l
i
k
G
.
Si
t
u
a
ti
o
n
A
s
s
e
s
s
m
e
n
t
Fo
r
Au
to
m
a
t
i
c
L
a
n
e
-
Cha
n
g
e
M
a
n
e
u
v
e
rs
.
IEEE
Tra
n
s
I
n
t
e
l
l
Tra
n
s
p
S
y
s
t
.
2
0
1
0
;
11
:
6
0
7
–
6
1
6
[1
5
]
L
i
n
d
n
e
r
P,
Ri
c
h
te
r
E,
W
a
n
i
e
l
i
k
G
,
T
a
k
a
g
i
K,
Is
o
g
a
i
A.
M
u
l
ti
-
Cha
n
n
e
l
L
i
d
a
r
Pr
o
c
e
s
s
i
n
g
f
o
r
L
a
n
e
Det
e
c
t
i
o
n
A
n
d
E
s
ti
m
a
ti
o
n
.
I
n
:
IEEE
Con
fe
r
e
n
c
e
o
n
In
t
e
l
l
i
g
e
n
t
T
ra
n
s
p
o
rta
ti
o
n
Sy
s
t
e
m
s
,
Pr
o
c
e
e
d
i
n
g
s
,
IT
SC
.
2
0
0
9
:
02
–
2
0
7
.
[1
6
]
Su
c
h
i
tra
S,
Sa
tz
o
d
a
RK,
Sri
k
a
n
t
h
a
n
T
.
Id
e
n
t
i
fy
i
n
g
L
a
n
e
Ty
p
e
s
:
A
M
o
d
u
l
a
r
Ap
p
r
o
a
c
h
.
In
:
IEEE
Con
fe
re
n
c
e
o
n
I
n
te
l
l
i
g
e
n
t
T
ra
n
s
p
o
rt
a
ti
o
n
Sy
s
te
m
s
,
Pro
c
e
e
d
i
n
g
s
,
IT
SC
.
2
0
1
3
:
1
9
2
9
–
1
9
3
4
.
[1
7
]
Dan
e
s
c
u
R,
So
b
o
l
S,
Ned
e
v
s
c
h
i
S,
G
ra
f
T
.
Ste
re
o
v
i
s
i
o
n
-
B
a
s
e
d
Si
d
e
L
a
n
e
An
d
G
u
a
rd
ra
i
l
Det
e
c
t
i
o
n
.
2
0
0
6
I
EEE
I
n
te
l
l
T
ra
n
s
p
Sy
s
t
Con
f
.
2
0
0
6
;
1156
–
1
1
6
1
[1
8
]
M
B
d
e
Pa
u
l
a
,
CR
J
u
n
g
.
Rea
l
-
Ti
m
e
Det
e
c
t
i
o
n
a
n
d
Cla
s
s
i
fi
c
a
ti
o
n
o
f
Roa
d
L
a
n
e
M
a
rk
i
n
g
s
.
XX
VI
Con
f.
G
ra
p
h
.
Pa
tt
e
rn
s
I
m
a
g
e
s
.
2
0
1
3
:
83
–
9
0
.
[1
9
]
M
B
De
Pa
u
l
a
,
CR
J
u
n
g
,
L
G
Da
Si
l
v
e
i
ra
.
Au
to
m
a
t
i
c
On
-
T
h
e
-
Fl
y
E
x
tri
n
s
i
c
Ca
m
e
ra
C
a
l
i
b
r
a
ti
o
n
o
f
O
n
b
o
a
rd
V
e
h
i
c
u
l
a
r Ca
m
e
ra
s
.
Ex
p
e
rt
S
y
s
t.
Ap
p
l
.
2
0
1
4
;
41
(
4
PART
2
):
1997
–
2
0
0
7
.
[2
0
]
M
B
De
Pa
u
l
a
,
CR
J
u
n
g
.
Au
to
m
a
t
i
c
Det
e
c
t
i
o
n
a
n
d
Cla
s
s
i
fi
c
a
t
i
o
n
o
f
Roa
d
L
a
n
e
M
a
rk
i
n
g
s
Us
i
n
g
O
n
b
o
a
rd
V
e
h
i
c
u
l
a
r Ca
m
e
ra
s
.
I
EEE T
ra
n
s
I
n
te
l
l
T
r
a
n
s
p
Sy
s
t
.
2
0
1
5
;
16
:
3
1
6
0
–
3
1
6
9
.
[2
1
]
M
a
th
i
b
e
l
a
B,
New
m
a
n
P,
Po
s
n
e
r
I.
Rea
d
i
n
g
th
e
Roa
d
:
Ro
a
d
M
a
rk
i
n
g
Cl
a
s
s
i
f
i
c
a
ti
o
n
a
n
d
I
n
te
rp
re
ta
ti
o
n
.
IEEE
Tra
n
s
I
n
t
e
l
l
Tra
n
s
p
S
y
s
t
.
2
0
1
5
;
16
:
2
0
7
2
–
2
0
8
1
.
[2
2
]
Z
M
Sa
n
i
,
HA.
G
h
a
n
i
,
L
W
S
e
n
R
.
Day
ti
m
e
Roa
d
M
a
rk
e
r
Re
c
o
g
n
i
ti
o
n
U
s
i
n
g
G
ra
y
s
c
a
l
e
Hi
s
to
g
ra
m
a
n
d
Pi
x
e
l
Va
l
u
e
s
.
In
te
r
n
e
two
r
k
i
n
g
I
n
d
o
n
e
s
i
a
J
o
u
rn
a
l
.
2
0
1
6
;
8
(1
)
:
1
1
-
16
.
[2
3
]
ZM
Sa
n
i
,
HA
G
h
a
n
i
,
R
B
e
s
a
r
,
L
W
Se
n
.
Rea
l
-
T
i
m
e
Day
ti
m
e
Roa
d
M
a
rk
e
r
Rec
o
g
n
i
t
i
o
n
Us
i
n
g
F
e
a
tu
r
e
s
Ve
c
to
r
s
An
d
N
e
u
ra
l
Net
w
o
rk
.
IEEE
Con
fe
r
e
n
c
e
o
n
S
u
s
t
a
i
n
a
b
l
e
Ut
i
l
i
z
a
ti
o
n
An
d
D
e
v
e
l
o
p
m
e
n
t
I
n
En
g
i
n
e
e
ri
n
g
a
n
d
T
e
c
h
n
o
l
o
g
y
(
CSU
DET)
.
2015
:
1
-
6
.
[2
4
]
ZM
Sa
n
i
,
HA G
h
a
n
i
,
R Be
s
a
r,
A Az
i
z
a
n
,
H Ab
a
s
.
Rea
l
-
T
i
m
e
Vi
d
e
o
Pro
c
e
s
s
i
n
g
u
s
i
n
g
Co
n
to
u
r Nu
m
b
e
r
s
a
n
d
An
g
l
e
s
f
o
r
Non
-
u
r
b
a
n
Roa
d
M
a
rk
e
r
Cla
s
s
i
fi
c
a
t
i
o
n
.
In
te
rn
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
E
l
e
c
tri
c
a
l
a
n
d
Com
p
u
te
r En
g
i
n
e
e
ri
n
g
(I
J
ECE
)
.
2018
;
4
(8
)
:
2
5
4
0
-
2
5
4
8
.
[2
5
]
JC
Ru
s
s
,
J
C
Rus
s
.
In
tr
o
d
u
c
ti
o
n
to
Im
a
g
e
Pro
c
e
s
s
i
n
g
a
n
d
An
a
l
y
s
i
s
.
CR
C
Pre
s
s
,
B
o
c
a
Rat
o
n
.
2008
:
72
-
79
.
Evaluation Warning : The document was created with Spire.PDF for Python.