I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
3
7
,
No
.
2
,
Feb
r
u
ar
y
20
2
5
,
p
p
.
1
18
3
~
1
2
0
0
I
SS
N:
2
502
-
4
7
52
,
DOI
: 1
0
.
1
1
5
9
1
/ijee
cs
.v
3
7
.
i
2
.
pp
1
18
3
-
1
2
0
0
1183
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs
.
ia
esco
r
e.
co
m
M
o
dified
-
v
ehicle
detec
tion a
nd loc
a
liza
tion mo
del f
o
r
a
utono
mo
us v
ehicle t
ra
f
fic
sy
ste
m
Am
it
J
uy
a
l
1
,
Sa
chin Sh
a
rm
a
2
,
Sh
uchi
B
ha
du
la
2
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
G
r
a
p
h
i
c
Er
a
D
e
e
me
d
t
o
b
e
U
n
i
v
e
r
s
i
t
y
,
G
r
a
p
h
i
c
Er
a
H
i
l
l
U
n
i
v
e
r
si
t
y
,
D
e
h
r
a
d
u
n
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
G
r
a
p
h
i
c
Er
a
D
e
e
me
d
t
o
b
e
U
n
i
v
e
r
s
i
t
y
,
D
e
h
r
a
d
u
n
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Mar
25
,
2
0
2
4
R
ev
is
ed
Sep
20
,
2
0
2
4
Acc
ep
ted
Sep
30
,
2
0
2
4
Th
e
m
o
d
ifi
c
a
ti
o
n
o
f
v
e
h
icle
s
fo
r
fin
a
n
c
ial
g
a
in
is
a
n
e
v
o
lv
in
g
ten
d
e
n
c
y
o
b
se
rv
e
d
in
I
n
d
ia.
Re
c
o
g
n
izin
g
a
n
d
d
e
tec
ti
n
g
o
f
th
e
se
m
o
d
ifi
e
d
il
l
icit
c
a
rs
is
a
n
imp
o
rtan
t
b
u
t
c
rit
ica
l
tas
k
i
n
a
u
t
o
n
o
m
o
u
s
v
e
h
icle
s
(AV
)
.
I
t
is
a
lwa
y
s
p
o
ss
ib
le f
o
r
a
c
y
c
li
st o
r
p
e
d
e
stria
n
to
trav
e
rse
o
b
sta
c
les
o
r
o
t
h
e
r
fix
e
d
o
b
jec
ts
th
a
t
a
p
p
e
a
r
in
fr
o
n
t
o
f
a
n
y
m
o
v
i
n
g
v
e
h
icle
.
Ve
h
icle
s
th
a
t
a
re
a
u
to
n
o
m
o
u
s
o
r
se
lf
-
d
riv
i
n
g
re
q
u
ire
a
d
iffere
n
t
sy
ste
m
to
q
u
ick
l
y
i
d
e
n
ti
f
y
b
o
th
sta
ti
o
n
a
ry
a
n
d
m
o
v
in
g
o
b
jec
ts.
A
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l
n
a
m
e
d
y
o
u
o
n
l
y
l
o
o
k
o
n
c
e
v
e
rsio
n
5
(
YO
LOv
5
)
-
c
o
n
v
o
lu
ti
o
n
a
l
b
l
o
c
k
a
tt
e
n
ti
o
n
m
o
d
u
le
(
CBAM
)
is
p
r
o
p
o
se
d
h
e
re
fo
r
t
h
e
I
n
d
ia
n
traffic
s
y
ste
m
wh
i
c
h
is
b
a
se
d
o
n
YO
LOv
5
m
.
T
h
e
p
ro
p
o
se
d
a
lg
o
rit
h
m
,
YO
LOv
5
-
CBAM,
h
a
s
th
re
e
m
a
jo
r
c
o
m
p
o
n
e
n
ts.
T
h
e
firs
t
m
o
d
u
le,
th
e
b
a
c
k
b
o
n
e
m
o
d
u
le
is
e
m
p
l
o
y
e
d
fo
r
fe
a
tu
re
e
x
trac
ti
o
n
.
T
h
e
se
c
o
n
d
m
o
d
u
le
is
to
d
e
tec
t
sta
ti
c
a
s
we
ll
a
s
d
y
n
a
m
ic
o
b
jec
ts
a
t
th
e
sa
m
e
ti
m
e
a
n
d
th
e
t
h
ir
d
CBAM
m
o
d
u
le
is
a
d
o
p
ted
in
th
e
b
a
c
k
b
o
n
e
a
n
d
n
e
c
k
p
a
rt,
w
h
ic
h
m
a
in
l
y
fo
c
u
se
s
o
n
t
h
e
m
o
re
p
ro
m
in
e
n
t
fe
a
tu
re
s.
Two
c
ro
ss
sta
g
e
p
a
rt
ial
(
CS
P
)
m
o
d
u
les
we
re
u
se
d
a
fter
e
v
e
ry
c
o
n
v
o
lu
ti
o
n
a
l
la
y
e
r
re
su
lt
in
g
in
a
n
a
d
d
it
i
o
n
a
l
h
e
a
d
to
t
h
e
p
r
o
p
o
se
d
m
o
d
e
l.
F
o
u
r
h
e
a
d
m
o
d
u
les
e
q
u
i
p
p
e
d
with
a
n
c
h
o
r
b
o
x
e
s
p
e
rfo
rm
e
d
t
h
e
fin
a
l
d
e
tec
ti
o
n
.
F
o
r
th
e
p
re
se
n
t
d
a
tas
e
t,
t
h
e
p
r
o
p
o
se
d
m
o
d
e
l
sh
o
we
d
9
8
.
2
%
m
e
a
n
a
v
e
ra
g
e
p
re
c
isio
n
(
m
AP
)
,
9
8
.
4
%
p
re
c
isi
o
n
,
a
n
d
9
4
.
8
%
re
c
a
ll
a
s c
o
m
p
a
re
d
to
t
h
e
o
ri
g
in
a
l
YO
LOv
5
m
.
K
ey
w
o
r
d
s
:
Au
to
n
o
m
o
u
s
v
eh
icle
C
B
AM
C
SP
YOL
O
YOL
Ov
5
-
C
B
AM
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Am
it Ju
y
al
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
Gr
a
p
h
ic
E
r
a
Dee
m
ed
to
b
e
Un
iv
er
s
ity
Gr
ap
h
ic
E
r
a
Hill Un
iv
er
s
ity
Deh
r
ad
u
n
,
I
n
d
ia
E
m
ail:
am
itju
y
al2
6
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Self
-
d
r
iv
in
g
is
an
in
tellig
en
t
au
to
n
o
m
o
u
s
d
r
iv
in
g
f
ea
tu
r
e
th
at
im
p
r
o
v
es
s
af
ety
an
d
c
o
n
v
en
ie
n
ce
b
esid
e
th
e
co
n
tr
o
ls
.
So
m
e
in
ter
n
atio
n
al
au
to
m
a
k
er
s
h
av
e
alr
ea
d
y
lau
n
ch
e
d
an
d
ar
e
o
p
er
atin
g
s
elf
-
d
r
iv
in
g
au
to
m
o
b
iles
.
T
esla,
W
ay
m
o
,
Z
o
o
x
,
an
d
C
r
u
is
e
ar
e
a
f
ew
ex
am
p
les.
C
o
m
p
ar
ed
to
d
e
v
elo
p
ed
n
atio
n
s
,
tr
af
f
ic
co
n
d
itio
n
s
ar
e
m
o
r
e
d
if
f
ic
u
lt
in
em
er
g
in
g
n
atio
n
s
.
I
n
d
ia
is
th
e
m
o
s
t
p
o
p
u
lo
u
s
an
d
d
ev
elo
p
in
g
n
atio
n
.
Au
to
n
o
m
o
u
s
v
eh
icles
(
AV)
ar
e
th
e
m
o
s
t
ch
allen
g
in
g
t
ask
s
in
th
e
p
r
esen
t
I
n
d
ian
tr
af
f
ic
s
ce
n
ar
io
.
I
f
au
to
m
ak
er
s
ar
e
s
u
cc
ess
f
u
l
in
m
an
u
f
ac
tu
r
in
g
AV
wh
ile
co
n
ce
n
tr
atin
g
o
n
th
e
tr
af
f
ic
c
o
n
d
i
tio
n
s
in
I
n
d
ia,
th
e
n
o
th
er
n
atio
n
s
ca
n
s
im
p
ly
em
b
r
ac
e
v
eh
icles
as
well.
T
h
e
cu
r
r
en
t
s
tu
d
y
aim
s
to
ass
es
s
th
e
s
u
cc
ess
o
f
AV
in
th
e
I
n
d
ian
tr
a
f
f
ic
s
ce
n
ar
io
.
T
o
a
ch
iev
e
s
af
e
an
d
tr
u
s
tab
le
au
t
o
n
o
m
o
u
s
d
r
iv
in
g
m
an
y
in
tellig
en
t
s
o
f
twar
e
an
d
d
if
f
er
en
t
h
ar
d
war
e
h
av
e
b
ee
n
d
ev
elo
p
e
d
to
d
ate
an
d
s
o
m
eti
m
es
f
o
r
b
etter
d
r
iv
i
n
g
ac
cu
r
ac
y
,
th
e
in
teg
r
atio
n
o
f
in
tellig
en
t
s
o
f
twar
e
an
d
h
ar
d
war
e
is
n
ee
d
ed
.
T
h
e
o
v
e
r
all
in
tellig
en
t
d
r
iv
in
g
s
y
s
tem
in
cl
u
d
es
C
am
er
as,
r
ad
io
d
etec
tio
n
,
a
n
d
r
an
g
i
n
g
(
R
ADAR
)
,
an
d
lig
h
t
d
etec
tio
n
a
n
d
r
an
g
in
g
(
L
iDAR
)
as
th
e
p
r
i
m
ar
y
s
en
s
o
r
s
t
h
at
ar
e
u
s
ed
to
co
llect
d
ata
ab
o
u
t
n
ea
r
b
y
t
r
af
f
ic
as
d
ep
icted
in
Fig
u
r
e
1
.
T
h
e
v
is
u
al
in
f
o
r
m
atio
n
is
p
r
o
v
id
ed
b
y
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
7
,
No
.
2
,
Feb
r
u
a
r
y
20
2
5
:
1
18
3
-
1
2
0
0
1184
ca
m
er
a,
th
e
d
is
tan
ce
a
n
d
v
elo
cities
o
f
o
b
jects
ar
e
d
eter
m
in
ed
b
y
R
ADAR,
an
d
th
e
th
r
ee
-
d
im
en
s
io
n
al
(
3D
)
p
o
in
t
clo
u
d
s
o
f
n
ea
r
b
y
tr
af
f
ic
ar
e
p
r
o
d
u
ce
d
b
y
L
iDAR
.
So
f
twar
e
b
ased
o
n
d
ee
p
lear
n
in
g
,
m
ac
h
in
e
lear
n
in
g
,
p
o
s
e
-
esti
m
atio
n
-
b
ased
m
o
d
els,
etc
f
u
r
th
er
p
r
o
ce
s
s
es
th
is
in
f
o
r
m
atio
n
to
esti
m
ate
th
e
o
b
je
ct
an
d
in
ten
tio
n
o
f
th
e
o
b
s
er
v
e
d
o
b
ject.
B
ased
o
n
th
ese
o
u
tp
u
ts
,
AV
co
n
tr
o
ls
its
elf
.
Alth
o
u
g
h
ca
r
m
a
n
u
f
ac
t
u
r
er
s
in
I
n
d
ia
h
av
e
alr
ea
d
y
d
e
p
lo
y
ed
ad
v
an
ce
d
d
r
i
v
er
ass
is
tan
ce
s
y
s
tem
(
ADAS
)
,
f
u
lly
AV
s
till
r
eq
u
ir
e
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
in
th
e
co
m
p
licated
s
y
s
tem
o
f
AV
.
On
I
n
d
ian
r
o
ad
s
,
a
wi
d
e
v
ar
iety
o
f
v
eh
icles
a
r
e
in
u
s
e.
Peo
p
le
m
o
d
if
y
th
e
ir
v
eh
icles
to
m
ee
t
th
eir
s
p
e
cif
ic
r
eq
u
ir
em
e
n
ts
.
Fig
u
r
e
2
d
em
o
n
s
tr
ates
th
e
m
o
d
if
ied
v
e
h
icles
th
at
ca
n
n
o
t
b
e
ea
s
ily
r
ec
o
g
n
is
ed
as
an
i
llu
s
tr
atio
n
o
f
s
u
ch
v
eh
icles
is
s
ee
n
in
Fig
u
r
e
s
2
(
a)
a
n
d
2
(
b
)
.
T
h
er
ef
o
r
e,
an
i
n
tellig
en
t
v
e
h
icle
s
y
s
tem
th
at
ca
n
r
ec
o
g
n
ize
an
d
d
etec
t a
ll c
ateg
o
r
ies o
f
v
eh
icle
s
,
is
n
ec
ess
ar
y
f
o
r
s
af
e
d
r
iv
i
n
g
in
AV
.
Fig
u
r
e
1
.
I
n
tellig
en
t sy
s
tem
in
AV
f
o
r
s
af
e
d
r
i
v
in
g
(
a)
(
b
)
Fig
u
r
e
2
.
Dis
p
lay
s
th
e
m
o
d
if
ie
d
au
to
m
o
b
iles
th
at
ar
e
d
if
f
icu
l
t to
r
ec
o
g
n
ize,
as seen
in
th
e
i
m
ag
es
;
(
a)
an
d
(
b
)
illu
s
tr
ate
ex
am
p
les o
f
m
o
d
if
ie
d
v
eh
icles
b
y
attac
h
i
n
g
th
e
ca
r
r
ier
to
th
e
m
o
to
r
cy
cle
,
tr
an
s
f
o
r
m
ed
in
to
a
ca
r
g
o
o
r
tr
an
s
it v
eh
icle
T
h
e
s
af
e
o
p
er
atio
n
o
f
an
a
u
to
n
o
m
o
u
s
v
eh
icle
d
e
p
en
d
s
o
n
th
e
ac
cu
r
ate
d
etec
tio
n
o
f
f
ix
ed
a
n
d
m
o
v
in
g
o
b
jects
as
ill
u
s
tr
ated
in
Fig
u
r
e
3
.
T
h
e
illu
s
tr
atio
n
s
h
o
ws
th
at
AV
is
f
o
llo
win
g
its
p
r
ed
eter
m
in
e
d
p
ath
.
T
h
e
AV
’
s
p
lan
n
ed
r
o
u
te
m
ay
in
clu
d
e
c
r
o
s
s
r
o
ad
s
.
Fix
e
d
o
b
je
cts
lik
e
b
ar
r
icad
es
an
d
tr
a
f
f
ic
co
n
es
m
ay
b
e
u
s
ed
to
r
estrict
o
n
e
o
r
b
o
th
s
id
es
o
f
th
e
in
ter
s
ec
tio
n
.
B
ar
r
ica
d
e
s
an
d
tr
a
f
f
ic
c
o
n
es
ca
n
r
estri
ct
th
e
f
l
o
w
o
f
ca
r
s
,
tr
u
ck
s
,
b
u
s
es,
an
d
o
t
h
er
v
eh
icles,
b
u
t
c
y
clis
ts
an
d
p
e
d
estrian
s
ca
n
b
y
p
ass
th
ese
o
b
s
t
ac
les.
C
y
clis
ts
an
d
p
ed
estrian
s
ca
n
cr
o
s
s
in
f
r
o
n
t
o
f
an
AV
b
y
g
o
in
g
o
v
er
p
er
m
an
en
t
b
ar
r
ier
s
lik
e
b
ar
r
icad
e
s
an
d
tr
af
f
ic
co
n
es
.
Acc
id
en
ts
m
ay
o
cc
u
r
d
u
e
to
th
is
cir
cu
m
s
tan
ce
,
co
n
s
eq
u
e
n
t
ly
,
in
AV
f
ix
ed
an
d
m
o
v
in
g
o
b
ject
d
etec
tio
n
is
ess
en
tial f
o
r
s
af
e
d
r
iv
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Mo
d
ified
-
ve
h
icle
d
etec
tio
n
a
n
d
lo
ca
liz
a
tio
n
mo
d
el
fo
r
a
u
to
n
o
mo
u
s
ve
h
icle
tr
a
ffic sys
tem
(
A
mit J
u
ya
l
)
1185
Fig
u
r
e
3
.
C
r
o
s
s
r
o
ad
s
with
two
s
id
es b
lo
ck
ed
b
y
f
ix
e
d
o
b
ject
s
s
u
d
d
en
ly
ap
p
ea
r
ed
i
n
th
e
r
o
u
te
o
f
th
e
AV
T
h
e
o
b
ject
d
etec
tio
n
ch
allen
g
e
in
v
o
lv
es
id
en
tify
in
g
an
d
lo
ca
tin
g
m
an
y
o
b
jects
in
a
d
ig
ital
im
ag
e.
T
o
r
eso
lv
e
th
e
is
s
u
e
o
f
o
b
ject
d
etec
tio
n
till
n
o
w
v
a
r
io
u
s
m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
tech
n
iq
u
es
h
av
e
b
ee
n
ap
p
lied
.
An
im
p
o
r
tan
t
c
o
m
p
u
ter
-
v
is
io
n
is
s
u
e
is
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
,
esp
ec
ially
wh
en
it
co
m
es
to
AV
s
o
r
au
to
n
o
m
o
u
s
d
r
i
v
in
g
.
C
o
m
p
ar
ed
to
co
n
v
en
tio
n
al
tech
n
iq
u
es,
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN
)
-
b
ased
tech
n
iq
u
es
with
g
o
o
d
m
ea
n
a
v
er
ag
e
p
r
ec
is
io
n
(
m
AP
)
a
r
e
b
ei
n
g
f
o
cu
s
ed
o
n
r
ea
l
-
tim
e
tr
af
f
ic
d
etec
tio
n
in
au
t
o
n
o
m
o
u
s
d
r
i
v
in
g
.
T
h
e
o
b
ject
d
etec
tio
n
m
eth
o
d
is
o
f
two
t
y
p
es:
two
-
s
tag
e
d
etec
tio
n
an
d
o
n
e
-
s
tag
e
d
etec
tio
n
.
T
h
e
m
eth
o
d
s
l
ik
e
R
-
C
NN
b
ased
o
n
th
e
C
NN
p
er
f
o
r
m
d
etec
tio
n
in
two
s
tep
s
,
with
f
ir
s
t
r
eg
io
n
p
r
o
p
o
s
als
ar
e
g
e
n
er
ated
u
s
in
g
s
eg
m
en
tatio
n
an
d
s
elec
tiv
e
s
ea
r
ch
a
n
d
s
ec
o
n
d
,
it
tr
ies
to
f
in
d
th
e
o
b
jects
in
ea
ch
r
eg
io
n
.
So
m
e
t
wo
-
s
tag
e
d
etec
to
r
m
o
d
els
th
at
h
av
e
b
ee
n
u
s
ed
in
a
u
to
n
o
m
o
u
s
d
r
iv
in
g
co
u
ld
b
e
d
escr
ib
ed
as:
−
R
eg
io
n
-
b
ased
C
NN
(
R
C
NN)
[
1
]
:
i
t
is
a
two
-
s
tep
d
etec
t
o
r
th
at
u
s
es
th
e
s
elec
tiv
e
s
e
ar
ch
f
o
r
r
eg
io
n
p
r
o
p
o
s
al,
C
NN
ex
tr
ac
ts
th
e
f
ea
tu
r
e
m
ap
s
an
d
f
in
ally
li
n
ea
r
SVM
ca
lcu
lates
th
e
weig
h
ts
f
o
r
ea
ch
la
y
er
in
o
b
ject
class
if
icatio
n
.
−
Fas
t
R
-
C
NN
[
2
]
:
t
o
cr
ea
te
a
f
ea
tu
r
e
m
ap
,
th
e
co
m
p
lete
im
a
g
e
with
th
e
r
eg
i
o
n
o
f
in
ter
est
(
R
OI
)
was
f
ed
in
to
n
u
m
er
o
u
s
co
n
v
o
lu
tio
n
al
l
ay
er
s
.
T
h
e
n
,
u
s
in
g
a
n
R
OI
p
o
o
lin
g
lay
e
r
,
a
f
ix
e
d
-
len
g
t
h
v
e
cto
r
o
f
f
ea
t
u
r
es
was p
r
o
d
u
ce
d
f
r
o
m
th
e
f
ea
t
u
r
e
m
ap
f
o
r
ea
ch
r
eg
io
n
p
r
o
p
o
s
al.
−
Fas
ter
R
C
NN
[
3
]
:
i
ts
wo
r
k
in
g
is
lik
e
f
ast
R
C
NN
ex
ce
p
t
th
at
it
g
en
er
ates
r
e
g
io
n
p
r
o
p
o
s
als
m
o
r
e
q
u
ick
l
y
u
s
in
g
a
s
ep
ar
ate
r
e
g
io
n
p
r
o
p
o
s
al
n
etwo
r
k
.
B
ec
au
s
e
o
f
th
e
two
s
tag
es
in
o
b
ject
d
etec
tio
n
,
R
C
NN
f
am
ily
m
eth
o
d
s
ar
e
co
n
s
id
er
ed
s
lo
w
in
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
.
T
o
s
p
ee
d
u
p
th
e
d
etec
tio
n
,
o
ne
-
s
tag
e
d
etec
tio
n
m
eth
o
d
s
s
u
ch
as YO
L
O
[
4
]
an
d
SS
D
[
5
]
ar
e
p
r
ef
er
r
ed
.
YOL
O
an
d
SS
D
ar
e
co
n
s
id
er
ed
f
aster
th
an
two
-
s
tag
e
r
eg
i
o
n
-
b
ased
C
NN
m
eth
o
d
s
as
th
e
y
tr
an
s
f
o
r
m
th
e
d
etec
tio
n
p
r
o
b
le
m
in
to
a
u
n
if
ied
r
eg
r
ess
io
n
p
r
o
b
lem
.
So
m
e
o
n
e
-
s
tag
e
d
etec
t
o
r
m
o
d
els
th
at
ar
e
b
ein
g
u
s
ed
i
n
Au
to
n
o
m
o
u
s
d
r
i
v
in
g
co
u
ld
b
e
d
escr
ib
ed
as
:
−
SS
D:
in
SS
D,
th
e
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
(
VGG)
-
1
6
[
6
]
n
etw
o
r
k
p
r
o
ce
s
s
es
all
th
e
im
ag
es
to
cr
ea
te
f
ea
tu
r
e
m
ap
s
.
C
o
n
v
o
lu
tio
n
al
lay
er
s
u
tili
ze
th
ese
f
ea
tu
r
e
m
ap
s
af
t
er
war
d
,
p
e
r
f
o
r
m
in
g
th
e
ac
t
u
a
l
d
etec
tio
n
an
d
p
r
o
d
u
cin
g
m
u
ltip
le
b
o
u
n
d
in
g
b
o
x
es f
o
r
ea
ch
o
b
ject.
−
YOL
O:
a
n
o
b
ject
d
etec
tio
n
m
eth
o
d
ca
lled
“
y
o
u
o
n
ly
lo
o
k
o
n
ce
”
was
f
ir
s
t
p
u
t
o
u
t
b
y
R
ed
m
o
n
et
a
l
.
[4
].
T
h
e
en
tire
im
ag
e
is
s
p
lit
in
to
a
g
r
id
b
y
YOL
O
an
d
th
e
n
g
e
n
er
ates
th
e
an
c
h
o
r
b
o
x
es
f
o
r
ea
ch
g
r
i
d
ce
ll.
I
t
is
th
e
f
astes
t
m
eth
o
d
in
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
[
7
]
.
YOL
O
is
a
well
-
k
n
o
wn
o
b
ject
d
etec
tio
n
m
eth
o
d
b
ec
au
s
e
o
f
th
e
way
it q
u
ick
ly
an
d
p
r
ec
is
ely
d
etec
ts
o
b
jects.
Usi
n
g
th
e
Py
to
r
ch
f
r
am
ewo
r
k
,
J
o
ch
er
et
a
l.
[
8
]
u
n
v
eiled
YOL
Ov
5
.
T
h
e
ex
is
tin
g
YOL
Ov
5
ar
ch
itectu
r
e
is
s
h
o
wn
in
Fig
u
r
e
4
.
YOL
Ov
5
au
to
m
atica
lly
ex
tr
ac
ts
f
ea
tu
r
es
f
o
r
o
b
ject
d
etec
tio
n
f
r
o
m
in
p
u
t
im
ag
es.
T
h
ese
f
ea
tu
r
es
ar
e
th
en
p
ass
e
d
to
a
p
r
ed
ictio
n
s
y
s
tem
to
f
o
r
m
b
o
x
es
ar
o
u
n
d
o
b
jects
to
r
ec
o
g
n
ize
th
eir
clas
s
es.
T
h
e
YOL
O
is
th
e
f
ir
s
t
en
d
-
to
-
en
d
o
b
ject
d
etec
tio
n
n
etwo
r
k
th
at
ca
n
p
r
e
d
ict
b
o
u
n
d
in
g
b
o
x
es in
ad
d
itio
n
to
class
lab
els.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
7
,
No
.
2
,
Feb
r
u
a
r
y
20
2
5
:
1
18
3
-
1
2
0
0
1186
Fig
u
r
e
4
.
Or
ig
i
n
al
YOL
Ov
5
ar
ch
itectu
r
e
T
h
e
YOL
Ov
5
n
etwo
r
k
is
co
m
p
r
is
ed
o
f
th
r
ee
cr
itical
co
m
p
o
n
e
n
ts
.
C
NN
s
an
d
o
p
tim
ized
s
p
atial
p
y
r
am
id
p
o
o
lin
g
(
SP
PF
)
g
en
er
ate
v
is
u
al
f
ea
tu
r
es a
t v
ar
io
u
s
g
r
an
u
lar
ities
is
th
e
b
ac
k
b
o
n
e.
N
ec
k
,
s
ev
er
al
lay
er
s
th
at
in
teg
r
ate
a
n
d
m
i
x
v
is
u
al
f
ea
tu
r
es
b
e
f
o
r
e
p
ass
in
g
th
e
m
f
o
r
p
r
e
d
ictio
n
.
T
h
e
h
ea
d
p
ar
t
o
f
th
e
n
etwo
r
k
p
r
o
ce
s
s
es
th
e
f
ea
tu
r
es
f
r
o
m
th
e
n
ec
k
to
p
er
f
o
r
m
b
o
u
n
d
in
g
b
o
x
p
r
ed
ictio
n
with
class
lab
els.
Ho
wev
er
,
in
r
ea
l
-
tim
e
d
etec
tio
n
s
y
s
tem
s
lik
e
in
AV
,
m
o
r
e
ac
cu
r
ate
an
d
ag
ile
d
etec
tio
n
is
r
e
q
u
ir
e
d
f
o
r
s
af
e
d
r
iv
in
g
.
T
h
e
p
r
esen
t
s
tu
d
y
f
o
cu
s
es
o
n
d
ev
elo
p
i
n
g
a
d
ee
p
-
lear
n
i
n
g
m
o
d
el
th
at
ca
n
d
etec
t
b
o
th
s
tatio
n
ar
y
an
d
m
o
v
in
g
o
b
jects
f
o
r
s
af
er
d
r
iv
i
n
g
.
Seco
n
d
ly
,
th
e
s
tu
d
y
aim
s
t
o
d
etec
t
th
e
i
n
ten
t
io
n
o
f
cy
clis
ts
o
r
p
ed
estrian
s
to
cr
o
s
s
th
e
r
o
ad
.
B
ased
o
n
th
e
liter
atu
r
e
r
ev
ie
w
o
f
p
r
ev
io
u
s
wo
r
k
,
th
e
p
r
e
s
en
t
s
tu
d
y
s
u
g
g
ested
th
e
u
s
e
o
f
an
im
p
r
o
v
is
ed
YOL
Ov
5
m
o
d
el.
I
t
h
as
b
ee
n
r
ep
o
r
ted
t
h
at
th
e
o
r
i
g
in
al
Y
OL
Ov
5
h
as
3
h
ea
d
s
f
o
r
d
ete
ctio
n
,
wh
ile
i
n
th
e
im
p
r
o
v
is
ed
YOL
Ov
5
-
c
o
n
v
o
lu
tio
n
al
b
lo
ck
atten
tio
n
m
o
d
u
le
(
C
B
AM
)
m
o
d
el
1
an
o
th
er
h
ea
d
h
as
also
b
ee
n
ad
d
ed
in
th
e
p
r
esen
t
wo
r
k
f
o
r
b
etter
d
etec
tio
n
o
f
id
e
n
tical
an
d
s
m
all
o
b
jects.
O
th
er
im
p
r
o
v
is
atio
n
s
co
n
tr
ib
u
ted
:
i)
B
ased
o
n
th
e
ar
ch
itectu
r
e
o
f
Y
OL
Ov
5
,
an
im
p
r
o
v
ed
m
o
d
el
h
as
b
ee
n
p
r
esen
ted
to
d
etec
t
1
8
d
if
f
er
en
t
ty
p
es
o
f
o
b
jects.
ii)
A
co
m
p
lex
s
elf
-
d
r
iv
in
g
I
n
d
ian
v
eh
icle
d
ataset
with
1
8
ty
p
es
o
f
d
y
n
am
ic
an
d
f
ix
ed
o
b
ject
im
ag
es
is
p
r
esen
ted
.
iii)
C
r
o
s
s
-
s
tag
e
lo
ca
l
n
etwo
r
k
(
C
SP
)
[
9
]
an
d
atten
tio
n
m
o
d
u
le
C
B
AM
[
1
0
]
ar
e
ad
d
e
d
to
en
h
an
ce
th
e
p
r
ed
ictio
n
s
o
f
an
ch
o
r
b
o
x
es.
T
h
e
p
r
esen
t
s
tu
d
y
ela
b
o
r
ate
d
o
n
th
e
d
ev
elo
p
m
en
t,
f
u
n
c
tio
n
in
g
,
an
d
o
u
tc
o
m
es
o
f
i
m
p
r
o
v
is
ed
/en
h
an
ce
d
YOL
Ov
5
-
C
B
AM
.
T
h
e
o
b
s
er
v
atio
n
s
d
escr
ib
ed
th
at
th
e
u
s
e
o
f
im
p
r
o
v
is
ed
m
o
d
els
th
at
h
av
e
b
ee
n
im
p
lem
en
ted
o
n
th
e
cu
s
to
m
d
ataset
p
r
o
d
u
ce
s
b
etter
r
esu
l
ts
f
o
r
f
ix
ed
an
d
d
y
n
am
ic
I
n
d
ian
tr
af
f
ic
o
b
ject
d
etec
tio
n
.
Ho
wev
er
,
it
is
m
o
r
e
ch
allen
g
in
g
to
r
ec
o
g
n
ize
th
e
o
b
jects,
esp
ec
ially
to
ev
alu
a
te
th
e
in
ten
tio
n
s
o
f
m
u
lticlas
s
o
b
jects
lik
e
v
eh
icles,
b
ar
r
ier
s
,
tr
af
f
ic
co
n
es,
cy
clis
ts
,
an
d
p
ed
estrian
s
,
as th
ey
all
b
elo
n
g
to
d
if
f
er
en
t
class
es.
So
th
e
s
tu
d
y
co
n
tin
u
e
d
with
t
h
e
u
s
e
o
f
th
e
PAF
m
o
d
el
alo
n
g
with
an
im
p
r
o
v
is
ed
d
ee
p
-
lear
n
i
n
g
m
o
d
el
f
o
r
b
etter
d
etec
tio
n
.
2.
RE
VI
E
W
O
F
L
I
T
E
RA
T
UR
E
R
ec
en
tly
,
th
e
f
ield
o
f
ar
tific
ial
in
tellig
en
ce
in
au
to
n
o
m
o
u
s
d
r
iv
in
g
h
as
s
ee
n
a
lo
t
o
f
s
tu
d
y
in
ter
est
in
th
e
p
r
o
b
lem
o
f
m
u
lti
-
o
b
ject
i
d
en
tific
atio
n
.
T
h
e
o
n
e
-
s
tag
e
YOL
O
tech
n
iq
u
e
was
in
itially
d
ev
elo
p
ed
to
tack
le
o
b
ject
d
etec
tio
n
as
a
r
eg
r
ess
io
n
p
r
o
b
lem
.
On
e
o
f
th
e
m
o
s
t
r
ec
en
t
d
ev
elo
p
m
en
ts
,
YOL
O
,
ca
n
d
etec
t
o
b
jects
with
a
g
o
o
d
co
n
f
id
en
ce
s
co
r
e,
b
u
t
th
e
m
o
d
el
’
s
s
p
a
tial
co
n
s
tr
ain
t
r
ed
u
ce
s
th
e
n
u
m
b
er
o
f
o
b
jects
th
at
m
ay
b
e
p
r
ed
icted
.
T
o
p
r
ed
ict
th
e
m
o
t
io
n
o
f
n
ea
r
b
y
tr
a
f
f
ic
o
b
jects,
[
1
1
]
p
r
esen
ts
a
GAM
M
m
o
d
e
l
f
o
r
AV
v
eh
icles.
T
h
e
m
o
d
el
is
ca
p
ab
le
o
f
f
o
r
e
ca
s
tin
g
tr
af
f
ic
ag
en
ts
’
m
o
v
e
m
en
ts
u
n
d
er
v
ar
io
u
s
r
estrictio
n
s
lik
e
k
in
em
atics,
g
eo
m
etr
ic,
a
n
d
b
eh
av
i
o
r
al.
Ha
n
et
a
l.
[
1
2
]
co
m
b
in
e
L
iDAR
an
d
co
l
o
r
im
a
g
es
f
o
r
o
b
ject
d
etec
tio
n
to
im
p
r
o
v
e
th
e
YOL
O
m
o
d
el.
YOL
O
is
im
p
r
o
v
e
d
to
d
etec
t
lo
w
-
in
te
n
s
ity
o
b
jects
lik
e
p
ed
estrian
s
an
d
f
ix
ed
o
b
jects.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Mo
d
ified
-
ve
h
icle
d
etec
tio
n
a
n
d
lo
ca
liz
a
tio
n
mo
d
el
fo
r
a
u
to
n
o
mo
u
s
ve
h
icle
tr
a
ffic sys
tem
(
A
mit J
u
ya
l
)
1187
Or
ig
in
al
YOL
O
was
tr
ain
ed
to
g
et
im
p
o
r
ta
n
t
p
ar
am
eter
s
.
C
o
l
o
r
im
ag
es
an
d
d
ep
th
i
m
ag
es
ar
e
u
s
ed
to
im
p
r
o
v
e
ac
cu
r
ac
y
.
Dr
eo
s
s
i
et
a
l.
[
1
3
]
p
r
esen
ted
C
NN
-
b
ased
f
r
am
e
wo
r
k
t
o
class
if
y
v
eh
icles.
T
h
e
f
r
am
ew
o
r
k
u
s
es
a
n
im
ag
e
g
en
e
r
ato
r
f
o
r
s
y
n
t
h
etic
im
ag
es
wh
ich
is
u
s
ed
to
v
alid
ate
th
e
C
NN
m
o
d
el.
Fas
ter
YOL
O
b
y
[
1
4
]
co
m
b
in
es
th
e
p
r
o
p
o
s
ed
r
an
d
o
m
k
er
n
el
co
n
v
o
lu
tio
n
al
n
etwo
r
k
with
au
to
-
e
n
co
d
er
h
id
d
en
lay
er
s
n
etwo
r
k
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
.
Z
h
an
g
et
a
l.
[
1
5
]
in
tr
o
d
u
ce
a
n
o
b
s
tacle
-
d
etec
tio
n
alg
o
r
ith
m
.
YOL
O
co
m
b
in
es
lig
h
t
f
ield
ca
m
er
a
im
ag
es
th
at
co
n
tain
d
ep
th
im
ag
es
an
d
h
ig
h
-
r
eso
l
u
tio
n
im
ag
es.
YOL
O
tak
es
r
ed
,
g
r
ee
n
,
an
d
b
lu
e
(
R
GB
)
im
ag
es
as
in
p
u
t
a
n
d
g
en
er
ates
o
b
ject
d
etec
tio
n
i
n
f
o
r
m
atio
n
u
s
in
g
a
d
ep
th
m
ap
.
An
o
b
ject
d
etec
tio
n
s
y
s
tem
f
o
r
AV
u
s
in
g
m
o
d
if
ied
YOL
O
with
s
ev
en
co
n
v
o
lu
tio
n
al
lay
er
s
was p
r
o
p
o
s
ed
[
1
6
]
.
An
o
th
er
r
esear
ch
e
r
p
r
o
p
o
s
ed
a
YOL
O
v
3
-
liv
e
m
o
d
el
f
o
r
r
ea
l
-
tim
e
o
b
ject
d
e
tectio
n
[
1
7
]
.
Fo
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
m
u
lti
-
s
ca
le
r
ec
ep
tiv
e
f
ield
s
wer
e
u
s
ed
.
T
h
e
o
v
er
all
m
o
d
el
was
co
n
s
tr
u
cted
b
y
m
o
d
if
y
in
g
th
e
s
tr
u
ctu
r
e
an
d
n
etwo
r
k
p
ar
am
eter
s
.
W
o
n
g
et
a
l.
[
1
8
]
i
n
tr
o
d
u
ce
YOL
O
Nan
o
f
o
r
o
b
j
ec
t
d
etec
tio
n
.
T
h
e
n
etwo
r
k
ar
c
h
itectu
r
e
o
f
YOL
O
N
an
o
is
b
ased
o
n
s
in
g
le
-
s
h
o
t
o
b
ject
d
etec
tio
n
n
etwo
r
k
ar
ch
itectu
r
e.
W
ith
a
h
ig
h
ly
co
m
p
a
ct
n
etwo
r
k
d
esig
n
,
YOL
O
n
an
o
ac
h
iev
ed
6
9
.
1
%
m
AP.
I
n
an
o
th
er
s
tu
d
y
,
m
u
l
tis
p
ec
tr
al
im
ag
es
f
o
r
o
b
ject
d
etec
tio
n
wer
e
u
s
ed
[
1
9
]
.
Mu
ltis
p
ec
tr
al
im
ag
es
c
o
n
tain
m
u
ltil
ater
al
in
f
o
r
m
ati
o
n
th
at
ca
n
h
elp
d
etec
t
o
b
je
cts
th
at
ca
n
n
o
t
b
e
v
is
u
alize
d
in
n
o
r
m
al
R
GB
im
ag
es.
T
h
e
p
r
o
p
o
s
ed
m
u
ltis
p
ec
tr
al
en
s
em
b
le
d
etec
tio
n
p
ip
elin
e
is
d
iv
id
ed
in
t
o
two
s
in
g
le
s
p
ec
tr
al
d
etec
tio
n
m
o
d
els
an
d
an
e
n
s
em
b
le
p
a
r
t.
Sq
u
ee
ze
Det
wh
ich
is
a
f
u
lly
C
NN
f
o
r
o
b
je
ct
d
etec
tio
n
was a
ls
o
p
r
o
p
o
s
ed
i
n
a
s
tu
d
y
[
2
0
]
.
Featu
r
e
m
a
p
,
cl
ass
p
r
o
b
ab
ilit
y
,
an
d
b
o
u
n
d
in
g
b
o
x
wer
e
g
e
n
er
ated
u
s
in
g
co
n
v
o
lu
tio
n
al
la
y
er
s
.
3.
M
E
T
H
O
D
R
ea
l
-
tim
e
d
etec
tio
n
o
f
f
ix
e
d
o
b
jects
as
well
as
m
o
v
in
g
o
b
jects
is
an
im
p
o
r
ta
n
t
task
f
o
r
s
a
f
e
d
r
iv
i
n
g
.
T
h
e
p
r
im
a
r
y
r
is
k
co
n
ce
r
n
f
o
r
s
elf
-
d
r
iv
in
g
i
n
r
ea
l
tr
a
f
f
ic
s
ce
n
ar
io
s
is
th
e
d
etec
tio
n
o
f
m
u
l
ticlass
o
b
jects
lik
e
d
y
n
am
ic
o
b
jects
(
cy
clis
ts
,
p
ed
estrian
s
,
an
d
v
ar
i
o
u
s
ty
p
es
o
f
v
eh
icles)
as
well
as
s
tati
c
o
r
f
ix
e
d
o
b
jects
(
b
ar
r
icad
es,
tr
a
f
f
ic
co
n
es).
Mu
lticlas
s
ob
ject
lo
ca
lizatio
n
(
MO
L
)
is
an
ess
en
tial
p
r
o
b
lem
in
c
o
m
p
u
te
r
v
is
io
n
co
n
tain
in
g
n
u
m
er
o
u
s
ap
p
licatio
n
s
li
k
e
s
elf
-
d
r
iv
in
g
ca
r
s
,
s
ec
u
r
ity
s
y
s
tem
s
,
X
-
r
ay
s
,
an
d
m
a
n
y
o
th
er
s
.
T
h
e
c
h
allen
g
e
d
er
iv
ed
f
r
o
m
th
e
ab
o
v
e
d
is
cu
s
s
io
n
s
is
k
n
o
wn
as
MO
L
an
d
it
p
lay
s
an
im
p
o
r
ta
n
t
r
o
le
f
o
r
th
e
au
to
n
o
m
o
u
s
v
eh
icle
b
ec
a
u
s
e
d
u
e
to
th
is
t
ec
h
n
iq
u
e
v
ar
i
o
u
s
o
b
jects
p
r
e
s
en
t
in
th
e
s
u
r
r
o
u
n
d
in
g
s
o
f
th
e
v
e
h
icle
ca
n
b
e
d
etec
ted
,
class
if
ied
,
an
d
lo
ca
te
d
in
ter
m
s
o
f
th
eir
ex
ac
t
p
o
s
itio
n
.
Fig
u
r
e
5
,
d
is
p
lay
s
th
e
en
tire
d
iag
r
am
o
f
th
e
s
u
g
g
ested
f
r
a
m
ewo
r
k
f
o
r
th
e
d
etec
tio
n
o
f
m
u
lticlas
s
d
y
n
am
ic
a
n
d
f
ix
ed
o
b
jects
in
r
ea
l
tr
af
f
ic
s
ce
n
a
r
io
s
.
I
n
th
e
f
ir
s
t
s
tep
m
o
v
in
g
an
d
f
ix
e
d
o
b
jects
ar
e
d
etec
ted
u
s
in
g
an
im
p
r
o
v
ed
YOL
Ov
5
-
C
B
AM
m
o
d
el.
T
o
e
n
ab
le
AV
to
o
p
er
ate
s
af
ely
,
th
e
n
ex
t
s
tag
e
is
to
d
etec
t
th
e
o
b
jects
th
at
h
av
e
b
ee
n
m
o
d
i
f
ied
.
Ad
d
itio
n
ally
,
r
ec
o
g
n
izin
g
s
tatio
n
ar
y
o
b
jects
lik
e
tr
af
f
ic
co
n
es,
b
ar
r
icad
es,
an
d
tr
af
f
ic
s
ig
n
s
is
cr
u
cial
f
o
r
s
elf
-
d
r
iv
in
g
a
u
to
m
o
b
iles
.
Fig
u
r
e
5
.
T
h
e
o
v
er
all
f
u
n
ctio
n
ality
o
f
th
e
s
u
g
g
ested
tech
n
iq
u
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
7
,
No
.
2
,
Feb
r
u
a
r
y
20
2
5
:
1
18
3
-
1
2
0
0
1188
Alth
o
u
g
h
b
ar
r
icad
es
an
d
tr
a
f
f
i
c
co
n
es
a
r
e
u
s
ed
as
f
ix
ed
item
s
to
co
n
tr
o
l
t
r
af
f
ic,
it
h
as
b
ee
n
o
b
s
er
v
e
d
th
at
p
ed
estrian
s
,
c
y
clis
ts
,
an
d
two
-
wh
ee
ler
r
id
er
s
o
f
ten
d
is
r
eg
ar
d
th
ese
ty
p
es
o
f
b
ar
r
ier
s
an
d
ca
n
em
e
r
g
e
ab
r
u
p
tly
in
f
r
o
n
t o
f
an
y
ca
r
.
As a
r
esu
lt,
in
th
e
I
n
d
ian
tr
af
f
ic
s
ce
n
e,
p
ed
estrian
s
,
cy
clis
ts
,
an
d
two
-
wh
ee
ler
s
ar
e
s
ee
n
as
p
o
ten
tial
ac
cid
en
t
ca
u
s
es.
T
h
er
ef
o
r
e,
it
is
cr
u
cial
f
o
r
s
af
e
d
r
iv
in
g
to
d
etec
t
p
ed
estrian
s
,
cy
clis
ts
,
an
d
two
-
wh
ee
ler
s
.
3.
1
.
I
m
pro
v
ed
m
o
del
T
h
e
o
r
i
g
in
al
YOL
Ov
5
o
b
ject
d
etec
tio
n
n
etwo
r
k
n
o
r
m
ally
c
o
m
p
r
is
es
th
e
n
ec
k
,
t
h
e
b
ac
k
b
o
n
e,
an
d
th
e
d
etec
tin
g
h
ea
d
.
T
h
e
f
ea
tu
r
es
r
etr
iev
ed
in
th
e
m
id
d
le
n
etwo
r
k
ar
e
n
o
ticea
b
ly
lo
w,
wh
ich
ca
u
s
es
th
e
d
etec
tio
n
r
esu
lts
o
f
YOL
Ov
5
f
o
r
th
e
I
n
d
ian
tr
af
f
ic
s
ce
n
es
to
b
e
less
t
h
an
o
p
tim
al.
E
v
e
n
th
o
u
g
h
th
e
o
r
ig
in
al
YOL
Ov
5
c
an
d
etec
t
a
n
d
id
en
tify
au
to
m
o
b
iles
in
v
e
h
icle
im
ag
es
with
g
o
o
d
ac
cu
r
ac
y
,
it
o
n
ly
r
eq
u
ir
e
s
a
s
m
all
n
u
m
b
er
o
f
p
ar
am
eter
s
to
s
u
it
th
e
in
f
o
r
m
atio
n
b
ec
au
s
e
th
er
e
a
r
e
s
im
ilar
v
eh
icle
im
ag
es
th
at
ca
n
b
e
d
etec
ted
with
f
ewe
r
p
ar
am
eter
s
.
T
h
e
tech
n
iq
u
e
’
s
k
ey
g
o
al
is
to
en
h
an
ce
r
ec
o
g
n
itio
n
o
f
b
o
th
s
tatic
an
d
m
o
v
in
g
I
n
d
ian
v
eh
ic
u
lar
tr
af
f
ic
item
s
.
YOL
Ov
5
h
as
b
ee
n
m
o
d
if
ied
to
in
cr
ea
s
e
ac
cu
r
ac
y
w
h
ile
co
n
s
id
er
in
g
th
e
n
ee
d
f
o
r
less
co
m
p
u
tatio
n
.
C
er
tain
ch
an
g
es h
av
e
b
ee
n
m
a
d
e
in
t
h
e
o
r
ig
in
al
YOL
Ov
5
to
p
r
o
p
o
s
e
o
u
r
im
p
r
o
v
is
ed
m
o
d
el
YOL
Ov
5
-
C
B
AM
f
o
r
th
e
p
r
esen
t
s
tu
d
y
.
T
h
e
p
r
o
p
o
s
ed
YOL
Ov
5
-
C
B
AM
m
o
d
el
ad
d
r
ess
es
th
e
in
ad
e
q
u
ac
i
es
o
f
th
e
YOL
Ov
5
m
o
d
el
in
th
e
v
eh
icle
d
etec
tio
n
jo
b
,
an
d
p
r
o
v
id
es th
r
ee
m
ajo
r
co
n
tr
ib
u
tio
n
s
as f
o
llo
ws:
3
.
1
.
1
.
I
nc
lus
io
n
o
f
CSP
net
w
o
rk
As
s
h
o
wn
in
Fig
u
r
e
6
,
a
cr
o
s
s
-
s
tag
e
lo
ca
l
n
etwo
r
k
ca
lled
C
SP
is
em
p
lo
y
ed
to
in
teg
r
ate
a
n
d
g
en
er
ate
im
ag
e
f
ea
tu
r
es
o
n
m
an
y
lig
h
t
weig
h
t
C
NN
s
f
o
r
b
etter
ad
ju
s
tin
g
th
e
d
ep
th
o
f
th
e
n
etwo
r
k
.
T
h
e
b
ase
lay
er
’
s
f
ea
tu
r
e
m
a
p
is
s
p
lit
in
to
two
p
o
r
tio
n
s
u
s
in
g
th
e
C
SP
ap
p
r
o
ac
h
,
wh
ich
th
en
in
teg
r
ates
th
e
m
u
tili
zin
g
a
cr
o
s
s
-
s
tag
e
h
ier
ar
ch
y
.
Un
lik
e
th
e
YOL
Ov
4
n
etwo
r
k
,
th
e
YOL
Ov
5
-
C
B
AM
n
etwo
r
k
s
tr
u
ctu
r
e
h
as
two
C
SP
s
tr
u
ctu
r
es,
C
SP
1
an
d
C
SP
2
,
with
th
e
f
ir
s
t
ty
p
e
m
o
s
tly
u
til
ized
th
r
o
u
g
h
o
u
t
th
e
b
ac
k
b
o
n
e
an
d
th
e
o
th
e
r
o
n
e
b
ein
g
p
r
im
ar
ily
u
s
ed
th
r
o
u
g
h
o
u
t
th
e
Nec
k
.
T
h
e
YOL
Ov
5
-
C
B
AM
n
etwo
r
k
p
r
o
d
u
ce
s
4
o
u
tp
u
t
f
ea
tu
r
e
m
ap
s
h
av
in
g
v
ar
io
u
s
s
izes in
co
m
p
a
r
is
o
n
to
th
e
YOL
Ov
5
n
etwo
r
k
.
C
SP
:
-
th
e
C
SP
n
etwo
r
k
is
a
tech
n
iq
u
e
c
o
m
m
o
n
ly
u
s
ed
i
n
C
NN
s
p
ec
if
ically
in
ef
f
ici
en
tn
et
to
im
p
r
o
v
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
n
etwo
r
k
t
h
r
o
u
g
h
s
eg
m
en
tat
io
n
an
d
f
u
s
io
n
o
f
f
ea
tu
r
e
m
a
p
s
.
T
h
e
C
SP
n
etwo
r
k
ad
d
r
ess
es
th
e
p
r
o
b
lem
o
f
v
an
i
s
h
in
g
g
r
ad
ien
t,
a
p
r
o
b
lem
witn
ess
ed
in
a
n
eu
r
al
n
etwo
r
k
w
h
er
e
th
e
g
r
ad
ien
ts
o
f
th
e
lo
s
s
f
u
n
ctio
n
co
n
ce
r
n
i
n
g
t
h
e
p
ar
a
m
eter
s
r
ed
u
ce
to
in
s
ig
n
if
ican
t
am
o
u
n
ts
ar
e
th
e
g
r
ad
i
en
ts
ar
e
tr
an
s
p
o
r
ted
b
ac
k
war
d
s
th
r
o
u
g
h
th
e
n
etwo
r
k
d
u
r
in
g
th
e
o
p
tim
izatio
n
s
ta
g
e
o
f
g
r
ad
ien
t
d
escen
t.
C
o
m
p
u
tin
g
e
f
f
icien
cy
is
also
in
cr
ea
s
ed
th
r
o
u
g
h
p
u
t
i
n
p
r
o
ce
s
s
in
g
o
n
ly
th
e
p
ar
titi
o
n
ed
p
ar
t.
T
h
e
C
SP
ca
n
b
e
m
ath
e
m
atica
lly
ex
p
r
ess
ed
as:
A
n
i
m
ag
e
r
ep
r
esen
ted
b
y
f
ea
tu
r
e
v
ec
to
r
I
is
an
in
p
u
t f
ea
tu
r
e
m
ap
.
∈
{
=
ℎ
,
=
ℎ
,
=
ℎ
ℎ
}
(
1
)
T
o
g
eth
er
t
h
e
ch
an
n
el
d
im
en
s
i
o
n
d
iv
i
d
es th
e
in
co
m
in
g
f
ea
tu
r
e
m
ap
in
to
two
p
ar
ts
:
,
=
2
w
h
er
e
∈
2
an
d
∈
2
.
E
x
ec
u
te
a
s
eq
u
e
n
ce
o
f
c
o
n
v
o
l
u
tio
n
al
o
p
e
r
atio
n
s
(
ƒ
1
)
o
n
I
a.
1
=
ƒ1
(
∶
1
)
(
2
)
I
d
en
tity
m
ap
p
in
g
:
r
etain
th
e
v
alu
e
o
f
X2
with
o
u
t a
n
y
ch
an
g
es.
2
=
(
3
)
C
o
n
ca
t
(
1
)
an
d
(
2
).
=
1
(
+
)
2
(
4
)
∈
T
r
an
s
itio
n
:
n
ex
t,
p
r
o
ce
e
d
with
th
e
ap
p
licatio
n
o
f
an
ad
d
itio
n
al
co
n
v
o
l
u
tio
n
al
lay
er
to
th
e
c
o
m
b
in
ed
o
u
tp
u
t.
=
ƒ2
(
∶
2
)
(
5
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Mo
d
ified
-
ve
h
icle
d
etec
tio
n
a
n
d
lo
ca
liz
a
tio
n
mo
d
el
fo
r
a
u
to
n
o
mo
u
s
ve
h
icle
tr
a
ffic sys
tem
(
A
mit J
u
ya
l
)
1189
W
h
er
e
t
h
e
tr
a
n
s
itio
n
co
n
v
o
l
u
tio
n
al
lay
e
r
is
r
e
p
r
esen
ted
b
y
ƒ
2
,
wh
ich
h
as
p
a
r
am
eter
s
d
en
o
te
d
b
y
.
C
SP
b
lo
ck
as a
n
en
tire
ly
is
d
en
o
te
d
as.
=
ƒ2
(
(
ƒ1
(
(
)
:
1
)
+
(
)
)
:
2
)
(
6
)
3
.
1
.
2.
I
nc
lus
io
n
o
f
CSP
net
w
o
rk
T
h
e
ef
f
icien
c
y
with
wh
ic
h
th
e
in
ter
m
ed
iar
y
lay
er
s
e
x
tr
ac
t
th
e
cr
u
cial
d
ata
d
eter
m
in
es
t
h
e
s
u
cc
ess
o
f
o
b
ject
d
etec
tio
n
.
T
h
e
YOL
Ov
5
b
ac
k
b
o
n
e
m
o
d
u
le
ex
tr
ac
t
s
th
e
im
ag
e
’
s
f
ea
tu
r
es
an
d
u
s
es
th
em
to
d
etec
t
o
b
jects
in
th
e
d
ig
ital
im
ag
e
.
Fro
m
th
e
ex
t
r
ac
ted
f
ea
tu
r
es,
s
o
m
e
f
ea
tu
r
es
h
a
v
e
a
g
r
ea
ter
in
f
lu
en
ce
o
n
t
h
e
o
u
tco
m
es th
an
o
t
h
er
s
.
As a
r
esu
lt,
we
r
eq
u
ir
e
a
m
o
d
u
le
th
at
c
an
s
o
lely
co
n
ce
n
tr
ate
o
n
k
ey
f
ea
tu
r
es.
T
h
er
ef
o
r
e,
we
h
av
e
ad
d
e
d
a
n
atten
tio
n
c
o
m
p
o
n
en
t
in
YOL
Ov
5
m
to
f
o
cu
s
o
n
m
o
r
e
im
p
o
r
tan
t
f
ea
tu
r
e
s
.
Fo
r
th
e
atten
tio
n
m
o
d
u
le,
th
e
C
NN
d
is
tr
ib
u
tes
h
ig
h
er
weig
h
ts
to
m
o
r
e
im
p
o
r
tan
t
f
ea
tu
r
es
an
d
lo
wer
weig
h
ts
to
le
s
s
e
s
s
en
tial
in
f
o
r
m
atio
n
.
T
h
e
weig
h
ted
d
at
a
is
th
en
co
m
b
in
ed
an
d
ev
alu
a
ted
.
T
h
r
ee
ca
teg
o
r
ies
ca
n
b
e
u
s
ed
to
class
if
y
th
e
co
m
m
o
n
atten
tio
n
m
o
d
u
les,
th
e
h
y
b
r
id
atten
tio
n
m
o
d
u
le
[
2
1
]
,
th
e
s
p
atial
atten
tio
n
m
o
d
u
le,
an
d
th
e
ch
a
n
n
el
atten
tio
n
m
o
d
u
le.
T
h
e
h
y
b
r
i
d
atten
tio
n
m
o
d
u
le
is
a
lig
h
tweig
h
t
atten
tio
n
m
o
d
u
le
th
at
g
en
er
ates
a
ch
an
n
el
atten
tio
n
m
ap
an
d
u
s
es
th
is
in
f
o
r
m
atio
n
,
to
g
en
er
ate
a
f
in
al
f
in
e
-
tu
n
e
d
f
ea
tu
r
e
in
th
e
s
p
atial
atten
tio
n
m
o
d
u
le.
I
n
th
e
c
h
an
n
el
atten
tio
n
m
o
d
u
le,
ea
ch
o
f
th
e
c
h
an
n
els
o
f
a
f
ea
tu
r
e
m
ap
is
v
iewe
d
lik
e
a
f
e
atu
r
e
d
etec
to
r
,
th
at
co
n
ce
n
tr
ates
o
n
“
wh
at
”
is
im
p
o
r
tan
t
g
iv
en
an
in
p
u
t
im
ag
e,
an
d
th
e
s
p
atial
atten
tio
n
m
o
d
u
le
is
r
esp
o
n
s
ib
le
f
o
r
allo
ca
tin
g
weig
h
ts
f
o
r
t
h
e
s
p
a
tial
d
ata.
T
h
e
C
B
AM
,
wh
ich
is
m
ad
e
u
p
o
f
th
e
ch
an
n
el
an
d
s
p
atial
atten
tio
n
m
o
d
u
les
C
AM
an
d
SAM
r
esp
ec
tiv
ely
,
is
th
e
atten
tio
n
m
o
d
u
le
th
at
is
ad
d
ed
to
t
h
e
Y
OL
Ov
5
m
n
etwo
r
k
ar
ch
itectu
r
e.
I
n
th
e
I
m
p
r
o
v
e
d
YOL
Ov
5
C
B
AM
,
th
e
C
B
AM
lay
er
is
ad
d
ed
i
n
t
h
e
b
ac
k
b
o
n
e
af
ter
ev
er
y
C
SP
1
lay
er
,
an
d
in
th
e
Nec
k
p
ar
t,
it
is
ad
d
ed
af
ter
ev
er
y
C
SP
2
lay
er
s
o
th
at
th
e
n
etwo
r
k
c
an
f
o
cu
s
o
n
m
o
r
e
im
p
o
r
tan
t
f
ea
t
u
r
es.
T
h
e
p
r
im
a
r
y
m
o
tiv
atio
n
f
o
r
u
tili
zin
g
C
B
AM
in
th
e
im
p
r
o
v
ed
YOL
Ov
5
C
B
AM
i
s
th
at
it
is
an
im
p
r
ess
iv
e
atten
tio
n
m
o
d
u
le
f
o
r
C
NN
th
at
ca
n
ef
f
icien
tly
ex
tr
ac
t
in
f
o
r
m
atio
n
ab
o
u
t
ch
an
n
el
an
d
s
p
atial
atten
tio
n
.
T
h
e
C
B
AM
m
o
d
u
le
is
an
ea
s
y
-
to
-
i
n
teg
r
ate,
less
co
m
p
u
tatio
n
al
co
s
t
co
m
p
o
n
e
n
t
o
f
C
NN
s
.
C
B
AM
tr
ain
in
g
h
as
m
in
im
al
o
v
er
h
ea
d
s
an
d
ca
n
b
e
tr
ai
n
ed
en
d
-
to
-
en
d
to
g
eth
e
r
with
th
e
n
etwo
r
k
b
ac
k
b
o
n
e.
E
v
er
y
C
SP
m
o
d
u
le
is
f
o
llo
wed
b
y
t
h
e
ad
d
itio
n
o
f
C
B
AM
,
wh
ich
ass
ig
n
s
h
ig
h
er
weig
h
ts
to
f
ea
tu
r
es
th
at
ar
e
m
o
r
e
cr
u
cial
to
co
n
ce
n
tr
ate
o
n
f
ea
t
u
r
es
th
at
ar
e
cr
u
cial
f
o
r
im
p
r
o
v
in
g
th
e
ac
cu
r
ac
y
o
f
c
o
m
p
a
r
ab
le
tr
af
f
ic
o
b
ject
d
etec
tio
n
.
C
B
AM
:
a
r
elativ
ely
s
m
all
lig
h
tweig
h
t
m
o
d
u
le
ca
lled
th
e
C
B
AM
.
C
NN,
in
p
ar
ticu
lar
,
ca
n
ea
s
ily
in
co
r
p
o
r
ate
C
B
AM
as
a
ty
p
e
o
f
n
eu
r
al
n
etwo
r
k
.
I
t
is
a
s
tr
o
n
g
e
n
h
an
ce
m
en
t
s
tr
ateg
y
t
h
a
t
ca
n
h
elp
ac
h
iev
e
m
o
r
e
im
p
o
r
tan
t
an
d
r
elev
a
n
t
f
ea
tu
r
es
o
n
th
e
ch
an
n
el
an
d
i
n
th
e
s
p
atia
l
d
o
m
ain
.
C
B
AM
an
d
DSAM
m
ain
ly
h
av
e
two
s
tr
u
ctu
r
es n
am
ely
C
AM
an
d
SAM.
I
n
(
C
AM
)
th
e
g
lo
b
al
av
er
ag
e
p
o
o
lin
g
(
GAP)
an
d
g
lo
b
al
m
ax
p
o
o
lin
g
(
GM
P)
p
r
o
p
o
s
ed
ar
e
u
s
ed
o
n
th
e
in
p
u
t
f
ea
tu
r
e
m
a
p
an
d
o
n
th
e
ch
an
n
el
o
n
s
p
atial
in
f
o
r
m
atio
n
wh
ich
co
n
tain
s
m
o
r
e
ef
f
ec
tiv
e
f
ea
tu
r
es.
E
x
tr
ac
ted
f
ea
tu
r
es
f
r
o
m
GAP
an
d
GM
P
ar
e
th
en
d
is
p
atch
ed
to
th
e
s
h
ar
ed
m
u
lti
-
lay
er
p
er
ce
p
tr
o
n
(
ML
P)
.
T
h
ese
f
ea
tu
r
es
ar
e
p
ass
ed
th
r
o
u
g
h
th
e
v
ar
i
o
u
s
h
i
d
d
en
lay
e
r
s
o
f
ML
P
an
d
p
r
o
d
u
ce
two
f
ea
tu
r
e
m
ap
s
as
its
o
u
tp
u
t.
T
o
o
b
tain
th
e
f
i
n
al
c
h
an
n
el
atten
t
io
n
m
a
p
,
th
e
o
b
t
ain
ed
m
ap
s
a
r
e
s
u
m
m
e
d
o
v
er
ch
an
n
els
a
n
d
t
h
en
p
ass
ed
th
r
o
u
g
h
th
e
s
ig
m
o
id
ac
tiv
atio
n
.
Fin
ally
,
th
e
m
o
s
t
s
ig
n
if
ican
t
ch
an
n
els
(
f
ea
tu
r
e
m
ap
s
)
ar
e
o
b
tain
ed
f
r
o
m
th
e
m
u
ltip
licatio
n
o
f
th
e
ch
an
n
el
atten
tio
n
m
ap
an
d
th
e
o
r
ig
in
al
in
p
u
t
f
ea
tu
r
e
m
ap
.
Ma
th
em
atica
lly
C
AM
is
r
ep
r
esen
ted
in
(
7
)
.
(
)
=
(
(
(
)
)
+
(
(
)
)
(
7
)
W
h
er
e
ch
an
n
el
atten
tio
n
m
ap
,
is
d
en
o
ted
b
y
C
m
ap
(
F),
Sm
l
p
is
a
s
h
ar
ed
m
u
lti
-
lay
er
p
er
c
ep
tr
o
n
a
n
d
s
ig
m
o
id
f
u
n
ctio
n
is
d
e
n
o
ted
b
y
σ
.
T
h
e
m
o
tiv
atio
n
b
eh
in
d
th
e
s
p
atial
atten
tio
n
m
o
d
u
le
is
s
im
p
ly
to
d
r
aw
a
f
o
c
u
s
o
n
im
p
o
r
tan
t
ar
ea
s
o
f
th
e
f
ea
tu
r
e
m
ap
.
T
h
e
n
,
a
v
er
a
g
e
p
o
o
lin
g
an
d
m
ax
p
o
o
lin
g
o
p
er
atio
n
s
a
r
e
ap
p
lied
ac
r
o
s
s
ch
an
n
el
d
im
en
s
io
n
s
s
o
,
we
h
av
e
two
2
D
m
ap
s
.
T
h
ese
m
ap
s
ar
e
co
n
ca
ten
ated
an
d
p
ass
ed
th
r
o
u
g
h
a
co
n
v
o
lu
tio
n
al
lay
er
f
o
llo
wed
b
y
a
s
ig
m
o
id
ac
tiv
atio
n
to
o
b
tain
a
s
p
atial
atten
ti
o
n
m
ap
.
T
h
e
s
p
atial
atten
tio
n
m
ap
is
th
en
elem
en
t
-
wis
e
m
u
ltip
lied
b
y
th
e
f
ea
tu
r
e
m
ap
wh
ich
h
as
alr
ea
d
y
p
ass
ed
th
r
o
u
g
h
c
h
an
n
el
atten
tio
n
i
n
o
r
d
er
to
f
u
r
th
er
en
h
a
n
ce
th
e
r
elev
an
t a
r
ea
s
.
Ma
th
e
m
atica
lly
SAM
is
r
ep
r
esen
ted
in
(
8
)
.
(
)
=
(
ƒ7
7
(
[
+
]
)
)
(
8
)
W
h
er
e
s
p
atial
atten
tio
n
m
o
d
u
l
e
is
Sm
ap
(
F),
σ
is
s
ig
m
o
id
f
u
n
ctio
n
,
ƒ
7
X7
7
X7
f
ilter
co
n
v
o
lu
tio
n
o
p
e
r
atio
n
,
+
is
co
n
ca
ten
ati
on.
I
n
th
e
p
r
o
p
o
s
ed
im
p
r
o
v
e
d
Y
OL
Ov
5
-
C
B
AM
,
th
e
b
ac
k
b
o
n
e
an
d
n
ec
k
p
ar
t
o
f
th
e
m
o
d
el
ar
ch
itectu
r
e
in
clu
d
ed
C
SP
an
d
C
B
AM
m
o
d
u
les
to
en
h
a
n
ce
t
h
e
m
o
d
el
’
s
ac
cu
r
ac
y
.
T
h
e
ad
d
itio
n
o
f
C
SP
an
d
C
B
AM
in
b
ac
k
b
o
n
e
is
m
ath
em
atica
lly
r
e
p
r
esen
ted
in
th
e
(
9
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
7
,
No
.
2
,
Feb
r
u
a
r
y
20
2
5
:
1
18
3
-
1
2
0
0
1190
=
(
(
−
1
)
)
(
9
)
I
n
(
3
)
,
f
o
r
th
e
it
h
C
SP
b
lo
ck
,
,
is
th
e
r
ep
r
esen
tatio
n
o
f
t
h
e
i
th
C
SP
b
lo
ck
with
C
B
AM
.
X
i
-
1
is
th
e
o
r
ig
in
al
f
ee
d
-
f
o
r
war
d
in
p
u
t to
th
is
b
lo
ck
is
th
e
in
p
u
t f
ea
tu
r
e
m
ap
.
T
h
e
ad
d
itio
n
o
f
C
SP
an
d
C
B
AM
in
Nec
k
is
m
ath
em
atica
lly
r
ep
r
esen
ted
in
th
e
(
1
0
)
.
=
(
(
)
)
(
10
)
I
n
(
4
)
,
f
o
r
th
e
jth
n
ec
k
lay
er
,
,
is
th
e
r
ep
r
esen
tatio
n
o
f
th
e
jth
Nec
k
lay
e
r
with
C
B
AM
,
,
is
th
e
f
ea
tu
r
e
m
ap
f
r
o
m
th
e
b
ac
k
b
o
n
e,
wh
ich
is
th
e
co
r
r
esp
o
n
d
i
n
g
f
ea
tu
r
e
m
ap
h
er
e
f
o
r
th
e
in
p
u
t
im
ag
e.
Fig
u
r
e
6
.
Pro
p
o
s
ed
YOL
Ov
6
-
C
B
AM
m
o
d
el
f
o
r
r
ea
l
-
tim
e
I
n
d
ian
tr
af
f
ic
o
b
ject
d
etec
tio
n
3
.
1
.
3
.
I
nclus
io
n o
f
hea
d
I
n
th
e
p
r
esen
t
s
tu
d
y
we
h
av
e
ad
d
ed
a
n
ex
tr
a
h
ea
d
in
th
e
i
m
p
r
o
v
e
d
YOL
Ov
5
C
B
AM
to
g
en
er
ate
m
o
r
e
f
ea
tu
r
es.
YOL
Ov
5
-
C
B
AM
with
f
o
u
r
h
ea
d
s
,
i
n
Y
OL
Ov
5
,
wh
ich
was
o
r
ig
in
al
ly
d
ev
el
o
p
ed
,
th
r
ee
d
etec
tio
n
h
ea
d
s
m
ak
e
it
p
o
s
s
ib
le
to
d
etec
t
o
b
jects
o
f
d
if
f
er
e
n
t
s
izes.
I
n
t
h
e
im
p
r
o
v
e
d
YO
L
Ov
5
-
C
B
AM
th
er
e
ar
e
f
o
u
r
h
ea
d
d
etec
tio
n
lay
e
r
s
,
r
eg
ar
d
in
g
o
b
ject
d
etec
tio
n
,
th
e
n
u
m
b
er
o
f
d
etec
tio
n
h
e
ad
s
is
eq
u
al
to
th
e
n
u
m
b
er
o
f
f
ea
tu
r
e
m
ap
s
/s
ca
les
o
n
wh
ich
th
e
p
r
ed
ictio
n
s
a
r
e
m
ad
e.
T
h
e
s
izes
o
f
o
b
jects
in
im
ag
es
ca
n
b
e
r
eg
u
lated
i
n
a
l
a
r
g
e
r
an
g
e.
T
h
u
s
,
in
tr
o
d
u
cin
g
o
n
e
m
o
r
e
h
ea
d
with
th
e
o
th
er
th
r
ee
d
etec
tio
n
h
ea
d
s
will
im
p
r
o
v
e
th
e
d
etec
tio
n
ca
p
ab
ilit
y
o
f
th
e
m
o
d
el,
an
d
th
e
m
o
d
el
ca
n
ef
f
ec
tiv
ely
p
r
ed
ict
s
m
aller
o
b
je
cts
as
well
as
lar
g
e
o
b
jects.
T
h
e
d
etec
tio
n
h
ea
d
is
u
s
ed
to
p
r
ed
ict
t
h
e
b
o
u
n
d
in
g
b
o
x
es,
wh
ich
m
ea
n
s
,
t
h
e
r
e
g
io
n
s
o
f
in
ter
est
o
f
th
e
o
b
ject
an
d
t
h
e
class
p
r
o
b
ab
ilit
i
es a
t sev
er
al
s
ca
les.
,
=
(
)
(
11
)
I
n
(
11
)
,
,
is
th
e
o
u
tp
u
t
o
f
p
r
ed
ictio
n
s
(
b
o
u
n
d
in
g
b
o
x
es,
c
lass
s
co
r
es)
f
r
o
m
th
e
ith
d
etec
tio
n
h
ea
d
,
is
th
e
f
ea
tu
r
e
m
ap
f
r
o
m
th
e
p
r
ev
io
u
s
m
o
d
u
le
Nec
k
.
I
n
th
e
m
o
d
if
ied
YOL
Ov
5
C
B
AM
,
th
e
f
ea
tu
r
e
g
r
ap
h
ex
tr
ac
t
ed
af
ter
th
e
f
ir
s
t
C
SP
1
_
1
is
co
n
ca
t
alo
n
g
with
an
u
p
s
am
p
le
f
ea
tu
r
e
f
r
o
m
C
SP
2
_
1
in
th
e
Nec
k
p
ar
t
wh
ich
co
r
r
esp
o
n
d
s
to
an
ad
d
itio
n
al
1
6
0
×
1
6
0
×
2
5
6
f
ea
tu
r
e
d
etec
tio
n
h
ea
d
1
.
T
h
e
r
ed
esig
n
ed
n
etwo
r
k
in
clu
d
es
f
o
u
r
p
r
e
d
ictio
n
h
ea
d
s
w
ith
d
im
en
s
io
n
s
o
f
20
×
20
×
2
5
5
,
4
0
×
40
×
2
5
5
,
8
0
×
80
×
2
5
5
,
an
d
1
6
0
×
1
6
0
×
2
5
5
f
o
r
d
etec
tio
n
.
Ad
d
itio
n
ally
,
we
u
s
e
co
n
ca
ten
atio
n
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Mo
d
ified
-
ve
h
icle
d
etec
tio
n
a
n
d
lo
ca
liz
a
tio
n
mo
d
el
fo
r
a
u
to
n
o
mo
u
s
ve
h
icle
tr
a
ffic sys
tem
(
A
mit J
u
ya
l
)
1191
co
m
b
in
e
a
p
r
ev
io
u
s
n
etwo
r
k
f
ea
tu
r
e
m
ap
with
o
u
r
u
p
s
a
m
p
led
f
ea
tu
r
es.
T
h
is
allo
ws
u
s
to
o
b
tain
m
o
r
e
s
ig
n
if
ican
t
s
em
an
tic
in
f
o
r
m
atio
n
b
ased
o
n
th
e
u
p
s
am
p
le
d
f
e
atu
r
es
as
well
as
f
in
er
-
g
r
ain
e
d
in
f
o
r
m
atio
n
f
r
o
m
th
e
p
r
e
v
io
u
s
f
ea
tu
r
e
m
ap
.
W
h
ich
h
elp
s
th
e
n
etwo
r
k
to
d
if
f
e
r
en
tiate
b
etwe
en
s
im
ilar
v
eh
icl
es
(
m
o
d
if
ie
d
lo
a
d
er
v
eh
icles
u
s
in
g
m
o
to
r
c
y
cles)
.
T
h
en
,
u
s
in
g
th
is
m
er
g
ed
f
e
atu
r
e
m
ap
,
we
p
r
o
ce
s
s
it
wit
h
a
f
ew
ad
d
itio
n
al
co
n
v
o
l
u
tio
n
al
lay
er
s
an
d
f
in
all
y
p
r
ed
ic
t a
c
o
m
p
a
r
ab
le
ten
s
o
r
.
3
.
2
.
E
s
t
im
a
t
io
n o
f
t
he
po
s
e
inte
ntio
n o
f
a
cy
clis
t
a
nd
a
pe
des
t
ria
n
I
n
th
is
s
ec
tio
n
,
b
ased
o
n
th
e
d
etec
tio
n
o
f
b
o
t
h
cy
clis
ts
an
d
p
ed
estrian
s
,
th
e
tec
h
n
iq
u
e
o
f
u
s
in
g
p
o
s
e
esti
m
ate
as
th
e
p
r
im
ar
y
in
f
o
r
m
atio
n
to
ascer
tain
wh
eth
er
t
h
e
cy
clis
t
an
d
p
ed
estrian
s
will
cr
o
s
s
th
e
b
lo
c
k
ed
r
o
ad
u
s
in
g
f
ix
e
d
o
b
jects
is
in
v
esti
g
ated
.
I
n
th
e
p
r
o
p
o
s
ed
m
o
d
el,
t
h
e
in
ten
tio
n
r
ec
o
g
n
itio
n
alg
o
r
ith
m
is
in
teg
r
ated
with
th
e
o
b
ject
d
et
ec
tio
n
alg
o
r
ith
m
to
a
n
ticip
ate
th
e
in
ten
tio
n
o
f
d
etec
ted
cy
cli
s
ts
an
d
p
ed
es
tr
ian
s
.
Fig
u
r
e
7
s
h
o
ws th
e
f
l
o
w
o
f
o
b
j
ec
t d
etec
tio
n
alo
n
g
with
p
o
s
e
esti
m
atio
n
u
s
in
g
th
e
PAF m
eth
o
d
.
T
o
co
n
s
tr
u
ct
a
2
D
p
o
s
e
f
o
r
b
o
th
th
e
p
e
d
estrian
a
n
d
t
h
e
c
y
clis
t,
th
e
f
ir
s
t
s
tep
is
to
u
s
e
YOL
Ov
5
-
C
B
AM
to
d
etec
t
a
p
ed
estrian
an
d
th
en
i
n
p
u
t
th
at
lo
ca
tio
n
in
to
th
e
p
ar
t
a
f
f
in
ity
f
ield
s
(
P
AFs
)
m
eth
o
d
[
2
2
]
.
T
h
e
o
r
ien
tatio
n
an
d
p
o
s
itio
n
o
f
p
ed
estrian
lim
b
s
ar
e
r
ep
r
esen
ted
b
y
2
D
v
ec
to
r
f
ield
s
o
f
PAF.
T
h
e
PAF
ar
ch
itectu
r
e
is
u
s
ed
in
th
e
p
o
s
e
esti
m
atio
n
p
r
o
ce
s
s
to
ac
q
u
ir
e
th
e
2
D
f
ea
t
u
r
es
o
f
th
e
in
d
iv
id
u
al
p
r
esen
t
in
Fig
u
r
e
8
.
Fig
u
r
e
7
.
Po
s
e
esti
m
atio
n
an
d
d
etec
tio
n
m
eth
o
d
u
s
in
g
p
r
o
p
o
s
ed
YOL
Ov
5
-
C
B
AM
with
PA
F m
eth
o
d
Fig
u
r
e
8
.
Po
s
e
esti
m
atio
n
u
s
in
g
th
e
PAF m
eth
o
d
T
h
e
PAF
n
etwo
r
k
is
d
iv
id
ed
i
n
to
two
b
r
an
c
h
es,
with
th
e
t
o
p
b
r
a
n
ch
p
r
ed
ictin
g
th
e
c
o
n
f
id
en
ce
m
ap
s
an
d
th
e
b
o
tto
m
b
r
an
c
h
p
r
ed
ic
tin
g
th
e
af
f
in
ity
f
ield
s
.
T
h
e
f
i
r
s
t
s
tag
e
o
f
ea
ch
b
r
an
c
h
r
ec
ei
v
es
a
s
et
o
f
f
ea
t
u
r
e
m
ap
s
th
at
wer
e
p
r
o
d
u
ce
d
as
a
r
esu
lt
o
f
th
e
c
o
n
v
o
lu
tio
n
a
l
n
etwo
r
k
’
s
a
n
aly
s
is
o
f
th
e
p
ictu
r
e
(
wh
ic
h
was
in
itiated
b
y
th
e
f
ir
s
t
1
0
lay
er
s
o
f
th
e
VGG
-
1
9
[
2
3
]
an
d
th
en
f
in
etu
n
ed
)
.
E
v
er
y
PAF
r
elate
s
to
a
l
im
b
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
7
,
No
.
2
,
Feb
r
u
a
r
y
20
2
5
:
1
18
3
-
1
2
0
0
1192
co
n
s
is
ts
o
f
two
s
ig
n
als(c
h
an
n
els),
o
n
e
b
ein
g
th
e
x
-
co
o
r
d
in
ate
an
d
t
h
e
o
t
h
er
b
ein
g
th
e
y
-
co
o
r
d
i
n
ate
o
f
th
e
in
v
o
lv
ed
v
ec
to
r
.
Fo
r
th
e
i
n
p
u
t
im
ag
e
I
,
f
ir
s
t
o
b
tain
th
e
c
o
n
f
id
en
ce
m
ap
f
o
r
ea
c
h
b
o
d
y
p
ar
t
p
.
Mp
(
x
,
y
)
illu
s
tr
ates
th
e
co
n
f
id
en
ce
m
ap
o
f
k
ey
p
o
in
t
p
i
n
th
e
b
o
d
y
o
r
jo
in
t.
Fo
r
ea
c
h
lim
b
p
o
i
n
t
ty
p
e
c,
T
h
e
PAF
is
Vc
(
x
,
y
)
=
(
Vc
(
x
,
y
)
,
Vc
(
x
,
y
)
)
.
Vc
(
x
,
y
)
,
Vc
(
x
,
y
)
ar
e
2
D
v
ec
to
r
s
at
p
ix
el
(
x
,
y
)
.
T
h
e
C
NN
h
as
two
b
r
an
ch
es:
o
n
e
f
o
r
th
e
co
n
f
id
e
n
ce
m
ap
s
an
d
o
n
e
f
o
r
th
e
PAFs
.
T
h
ese
n
et
wo
r
k
s
will
h
av
e
a
s
im
ilar
s
tr
u
ctu
r
e,
an
d
th
ey
will
co
n
s
is
t
o
f
th
r
ee
lay
e
r
s
.
Af
ter
co
n
f
id
en
t
m
ap
s
an
d
PAFs
ar
e
cr
ea
ted
,
th
e
m
atch
g
r
ap
h
s
b
e
twee
n
th
e
d
etec
ted
b
o
d
y
p
ar
ts
an
d
h
u
m
an
s
k
el
eto
n
s
ar
e
o
b
tain
ed
b
y
u
s
in
g
a
b
ip
a
r
tite
m
atch
in
g
s
ch
e
m
e.
No
n
-
m
ax
im
u
m
s
u
p
p
r
ess
io
n
,
o
n
th
e
co
n
f
id
en
ce
m
ap
s
(
Mp
)
,
is
ap
p
lied
to
d
etec
t
k
ey
p
o
in
ts
f
o
r
ea
c
h
o
f
th
e
b
o
d
y
p
ar
ts
p
.
C
an
d
id
ate
co
n
n
ec
tio
n
s
ar
e
cr
ea
ted
b
etwe
en
th
e
k
ey
p
o
in
ts
f
o
r
ea
c
h
b
o
d
y
p
ar
t
(
m
,
n
)
b
a
s
ed
o
n
PAFs
Mp
,
wh
ich
is
co
n
n
ec
ted
b
y
lim
b
ty
p
e
c.
T
o
ca
lcu
late
th
e
s
co
r
e
o
f
ca
n
d
id
ate
co
n
n
ec
tio
n
is
s
co
r
ed
b
y
in
teg
r
atio
n
o
f
PAF v
alu
es c
o
r
r
esp
o
n
d
in
g
to
t
h
e
lin
e
s
eg
m
en
t c
o
n
n
ec
tin
g
p
m
an
d
p
n
.
(
,
)
=
∫
Mp
(
pm
+
t
(
pn
−
pm
)
)
1
0
.
pn
−
pm
‖
pn
−
pm
‖
Fin
ally
b
ip
ar
tite
m
atch
in
g
alg
o
r
ith
m
is
u
s
ed
to
g
en
er
ate
th
e
g
r
ap
h
wh
ich
ass
o
ciate
s
th
e
d
i
tecte
d
k
ey
p
o
i
n
ts
to
th
e
lim
b
s
.
I
t is cle
ar
ly
s
h
o
wn
i
n
3
.
3
.
Cus
t
o
m
ized
da
t
a
s
et
f
o
r
t
esting
a
nd
t
ra
ini
ng
o
f
im
pr
o
v
ed
m
o
del
YOL
Ov
5
alr
ea
d
y
h
as
g
o
o
d
p
er
f
o
r
m
an
ce
in
v
eh
icle
d
etec
tio
n
.
T
h
e
u
n
ce
r
tain
ty
in
I
n
d
i
a
’
s
tr
af
f
ic
s
ce
n
ar
io
m
ak
es
it
m
o
r
e
ch
al
len
g
in
g
f
o
r
YOL
Ov
5
in
r
ea
l
-
tim
e
tr
af
f
ic
o
b
jects.
Fig
u
r
e
9
,
d
em
o
s
tr
ates
th
e
m
o
d
if
ied
v
e
h
icles
th
at
ar
e
d
et
ec
t
wr
o
n
g
ly
as
Fig
u
r
e
s
9
(
a
)
to
9
(
c)
,
c
lear
l
y
illu
s
tr
ates
th
at
YOL
Ov
5
p
r
o
d
u
ce
s
f
alse
r
esu
lts
d
u
e
to
a
lack
o
f
I
n
d
ian
v
eh
icle
in
f
o
r
m
atio
n
.
I
n
Fig
u
r
e
9
(
a
)
,
YOL
Ov
5
p
r
ed
ic
ted
th
e
o
b
ject
as
a
“
Mo
to
r
cy
cle
”
b
u
t
it
is
a
m
o
d
if
ied
v
eh
icle
wh
ich
is
b
u
ilt
u
s
i
n
g
an
o
ld
m
o
t
o
r
cy
cle.
Similar
ly
in
Fig
u
r
e
s
9
(
b
)
an
d
9
(
c
)
,
YOL
Ov
5
wr
o
n
g
ly
p
r
ed
icted
o
b
jects
as
“
T
r
u
ck
”
al
th
o
u
g
h
in
r
ea
l
b
o
t
h
ar
e
I
n
d
ian
v
eh
icles
“
Vik
r
am
”
an
d
“
E
-
R
ick
Sh
a
”
r
esp
ec
tiv
el
y
.
Fo
r
th
e
p
r
esen
t
s
tu
d
y
,
th
e
cu
s
to
m
d
ataset
is
r
eq
u
ir
ed
s
o
th
at
o
u
r
p
r
o
p
o
s
ed
YOL
Ov
5
b
ased
m
o
d
el
ca
n
d
etec
t
n
o
r
m
al
I
n
d
ian
v
e
h
icle
s
b
u
t
also
m
o
d
if
ied
a
u
to
m
o
b
iles
.
T
o
ass
es
s
th
e
ef
f
ec
tiv
en
ess
o
f
o
u
r
s
u
g
g
ested
m
eth
o
d
o
lo
g
y
,
we
cr
ea
te
d
an
e
n
tire
ly
n
o
v
el
s
et
o
f
I
n
d
ian
tr
af
f
ic
o
b
jects.
I
m
a
g
es
o
f
v
eh
icles
an
d
s
tatio
n
ar
y
o
b
je
cts
f
r
o
m
I
n
d
ian
tr
af
f
ic
s
itu
atio
n
s
h
av
e
b
ee
n
co
llected
in
Deh
r
ad
u
n
,
I
n
d
ia.
2
,
7
4
0
co
lo
r
im
ag
es h
a
v
e
b
ee
n
g
ath
er
ed
.
T
h
e
d
ata
au
g
m
en
tatio
n
p
r
o
ce
s
s
is
d
o
n
e
to
in
c
r
ea
s
e
th
e
n
u
m
b
er
o
f
im
ag
es
in
th
e
d
ataset.
Af
ter
d
ata
au
g
m
en
tatio
n
im
ag
e
co
u
n
t
in
cr
ea
s
es
f
r
o
m
2
,
7
4
0
to
5
,
0
9
8
im
ag
es.
B
y
u
tili
zin
g
d
ata
au
g
m
en
tatio
n
m
eth
o
d
s
,
th
e
m
o
d
el
is
im
p
r
o
v
ed
a
n
d
m
o
r
e
ac
cu
r
ate
b
ec
au
s
e
th
e
d
ata
is
p
len
tifu
l
an
d
en
o
u
g
h
.
Data
au
g
m
en
tatio
n
m
eth
o
d
s
r
o
tatio
n
,
g
r
a
y
s
ca
le,
s
atu
r
atio
n
,
b
lu
r
,
n
o
is
e,
etc
wer
e
u
s
ed
.
I
n
r
o
tatio
n
9
0
°
clo
ck
wi
s
e
an
d
9
0
°
c
o
u
n
ter
-
clo
ck
wis
e
im
ag
es
wer
e
r
o
tated
.
I
m
a
g
es
wer
e
c
o
n
v
e
r
ted
f
r
o
m
R
GB
to
g
r
ay
s
ca
le
in
2
5
%
o
f
im
ag
es,
b
etwe
en
-
2
5
%
an
d
+
2
5
%
s
atu
r
atio
n
w
as
d
o
n
e,
an
d
i
m
ag
es
wer
e
b
l
u
r
r
ed
u
p
to
2
.
5
p
x
,
f
r
o
m
th
e
to
t
al
p
ix
els
5
%
p
ix
els
n
o
is
e
was
ad
d
ed
in
th
e
im
ag
e.
T
h
e
d
ataset
i
s
d
iv
id
ed
in
to
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
in
g
.
T
h
er
e
ar
e
a
to
tal
o
f
1
8
class
es
in
th
e
d
ataset.
T
h
e
v
eh
icles
an
d
o
th
e
r
s
tatic
o
b
jects
ar
e
m
an
u
ally
an
n
o
tated
as
s
h
o
wn
in
Fig
u
r
e
1
0
.
C
las
s
n
am
es:
[
‘
am
b
u
lan
ce
’
,
‘
au
to
’
,
‘
b
ar
r
ier
’
,
‘
b
u
s
’
,
‘
car
’
,
‘
cy
clis
t
’
,
‘
e
-
r
ik
s
h
aw
’
,
‘
illeg
al
v
eh
icle
’
,
‘
l
o
ad
er
’
,
‘
m
o
to
r
b
ik
e
’
,
‘
p
ed
estrian
’
,
‘
s
co
o
ter
’
,
‘
tr
ac
to
r
’
,
‘
tr
a
f
f
ic
c
o
n
e
’
,
‘
tr
af
f
ic
p
o
lice
s
tan
d
’
,
‘
tr
u
ck
’
,
‘
v
ik
r
am
’
,
‘
p
ed
estrian
’
]
.
(
a)
(
b
)
(
c)
Fig
u
r
e
9
.
T
h
e
o
r
i
g
in
al
YOL
Ov
5
,
wh
en
ap
p
lied
to
d
etec
t th
e
m
o
d
if
ied
v
eh
icles th
en
it is
ab
le
to
d
etec
t th
e
v
eh
icle
b
u
t in
ac
c
u
r
ately
class
if
ies th
em
,
f
o
r
illu
s
tr
atio
n
,
(
a)
m
o
d
if
ied
v
eh
icle
is
r
ec
o
g
n
ize
d
as m
o
to
r
c
y
cle,
I
n
d
ian
p
u
b
lic
tr
a
n
s
p
o
r
t v
e
h
icle
“
Vik
r
am
”
is
m
is
id
en
tifie
d
as (
b
)
tr
u
c
k
an
d
E
-
R
ik
s
h
aw
is
r
ec
o
g
n
ized
as (
c)
tr
u
ck
Evaluation Warning : The document was created with Spire.PDF for Python.