TELKOM
NIKA
, Vol.13, No
.2, June 20
15
, pp. 563 ~ 5
7
0
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i2.980
563
Re
cei
v
ed
No
vem
ber 1
1
, 2014; Re
vi
sed
March 11, 20
15; Accepted
April 3, 2015
A New Method of Trajectory Restoration at Intersection
Zheng Ke
*
1,2
, Song Xiangbo
3
, Zhu Du
n
y
ao
1,
3
1
Intellige
n
t T
r
a
n
sport S
y
stem
s Researc
h
Ce
nter ,W
uhan U
n
iversit
y
of T
e
chno
log
y
,
W
uhan 4
3
0
063
, China
2
School of Co
mputer an
d Informatio
n
Engi
n
eeri
ng, Nan
y
a
n
g
Institute of
T
e
chn
o
lo
g
y
,
Nan
y
a
ng 473
0
04,
Chi
n
a
3
W
uhan KOT
E
I Informatics Co., L
T
D.,
W
uha
n 430
07
4, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: zanghs
hu
nn
y@16
3.com
A
b
st
r
a
ct
F
l
oatin
g car
d
a
t
a (F
CD) is
ve
chiel
’
s
positi
o
n
trace th
at is c
o
mes fro
m
Gl
o
bal
Positi
oni
ng
Syste
m
receiv
ers.coll
e
c
ted at the d
i
s
c
rete time. F
CD contai
ns
mu
ch infor
m
ati
o
n
of traffic and
road-
netw
o
rk. But
there ar
e differ
ent lev
e
l of traj
ectory shape
dam
a
ges because of t
he
affection of sample frequency and
runn
ing sp
ee
d. So before the
F
CD mi
nin
g
, the first thi
ng sh
oul
d be d
one i
s
to restore track to make u
p
fo
r
the loss. T
h
is
pap
er an
aly
z
e
s
the spat
ia
l-te
mp
oral c
har
acteristics of float
i
ng car traj
ecto
ry at intersecti
on,
and
b
u
il
ds
an
error r
e
cog
n
iti
o
n
mo
del
a
n
d
a
n
a
d
just
m
ent
a
l
gorit
hm.
Exp
e
riments s
how
ed
that th
is
meth
o
d
can i
m
prove
a
ccuracy of traj
ectory restorati
on a
nd
it
s perf
o
rmanc
e is b
e
t
t
er than ex
istin
g
metho
d
. T
h
u
s
,
the prop
ose
d
meth
od is pr
ac
tical for further data
min
i
ng.
Ke
y
w
ords
: floating car data, trajectory
resto
r
ation, inters
ec
tion, spat
ia
l-te
mp
oral c
haract
e
ristics
1.
Introduc
tion
Floating
Ca
r
refers to
the
moving
ca
r o
n
the
ro
ad
wi
th a va
riety o
f
sen
s
o
r
s, wh
ich
ca
n
periodically collect information su
ch
a
s
it
s po
sit
i
on,
v
e
locit
y
an
d di
rectio
n. The
car spatial a
nd
driving info
rm
ation is called
Floating Ca
r
Data or F
C
D for sho
r
t. Flo
a
ting ca
r like
mobile sen
s
o
r
s
distrib
u
ted in
road net
work [1] to capture the tr
affic informatio
n (such a
s
co
ng
estion, tran
sp
ort
road
spe
ed)
in current time. FCD ha
s become o
ne of the key technologi
e
s
in the field of
intelligent tra
n
sp
ortation,
whi
c
h i
s
wid
e
l
y use
d
in
th
e current
-time traffic surv
eillan
c
e
and
traffic
manag
eme
n
t. Comp
ared
with othe
r m
e
thod
s of tr
a
ffic informati
on colle
ction,
FCD
ha
s the
advantag
es o
f
low co
st, hi
gh cove
ra
ge,
and st
ron
g
real-time
cap
a
bility. On the other h
and, t
h
e
track of car
outline
s
the road t
opolo
g
y of an area [2], so FCD i
s
also u
s
e
d
as to extract
and
update the u
r
ban ro
ad info
rmation [3].
FCD i
s
coll
ected at the discrete time poi
nt
s. Sampling
interval and the vehicle
sp
eed wil
l
lead to the ra
ndom lo
ss of
track [4]. So
before t
he F
CD mini
ng, the first thing should be d
o
n
e
is
to re
store tra
c
k to m
a
ke u
p
for the l
o
ss.
The
r
e
are two ways i
n
track
re
storatio
n, one
i
s
matchi
ng
the tra
ck to t
he ro
ad o
n
the map,
and
the othe
r
is
fusing m
u
ltipl
e
traje
c
tori
es from the
sa
me
path. But this loss
will
cause the
position deviati
on
of intersection when
multi
p
le tracks fusing.
The fusi
on e
r
ror m
u
st b
e
reco
gni
zed a
n
d
rep
a
ire
d
in
orde
r to imp
r
ove the accu
racy of traje
c
t
o
r
y
resto
r
atio
n, and cla
s
sify tra
ck info
rmatio
n effectively.
Curre
n
tly, th
e mining ba
sed FCD is mainly
use
d
to get traffic information
and ro
ad
netwo
rk info
rmation. The system ge
ne
rally consi
s
ts
of four levels: Track Colle
ction and Sto
r
e,
Tra
c
k Regi
stration, Inform
ation Mini
ng
and Ap
plicat
ions. Tra
c
k Registration m
ean
s
to re
sto
r
e or
repai
r the l
o
ss shap
e of track cau
s
ed
b
y
colle
ct
ion freque
ncy a
nd
vehicle
sp
ee
d. There a
r
e t
w
o
kind
s of track regi
stration
method, on
e is
Map Ma
tching, an
d anothe
r is M
u
lti-tra
ck Fu
sion
[5],[6].
The Map M
a
tchin
g
method
matche
s the
track to a ro
ad at first, an
d then ma
ke
s up for
the tra
c
k lo
ss by u
s
ing
the
co
rrespon
din
g
road
shape
. Due
to the
slo
w
er up
dat
e spee
d of
m
ap,
not
all of
the trajecto
rie
s
h
a
s
th
eir own matche
d
roa
d
s or pla
c
e
s
. The
traje
c
tory
that
match with
the map
can
be u
s
ed to mi
ne the traffic i
n
formatio
n
of
the co
rre
sp
o
nding
roa
d
o
n
the ba
sis
of its
property information [4],[7]-[9], s
u
c
h
as traffic
flow, traffic
cos
t
s
and traffic
peak
periods
,
et
c
.
Whe
n
the tra
j
ectory that
cannot mat
c
h
the roa
d
on t
he map o
r
th
ere i
s
availa
ble map, T
r
a
ck
Regi
stratio
n
need to integ
r
ate a plu
r
alit
y of tracks fro
m
the same
road to co
mpe
n
sate the
sha
p
e
loss of track.
The re
sea
r
chers u
s
e ma
inly clus
te
rin
g
algorith
m
to fuse multi
p
le floating car
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 563 – 57
0
564
tracks, the literatu
r
e [2],[10]-[12]
su
gge
st that clu
s
tering algo
rithm
may use the
spatial di
sta
n
ce
betwe
en the
trajecto
ry poi
nts to mea
s
u
r
e whethe
r the tra
ck i
s
from the sa
me
road, an
d the
literature [13]-[16] sug
g
e
s
t that t
he sha
pe of the trajectory may u
s
e a
s
the si
milarity metri
c
to
identify a roa
d
on the sam
e
track.
Wheth
e
r the
clu
s
terin
g
al
g
o
rithm i
s
poi
nt-ba
s
ed
o
r
curve
-
b
a
sed,
the traje
c
tory sha
pe
after fusi
ng i
s
depe
nd o
n
the freq
uen
cy
sam
p
ling
a
n
d
the vehi
cle
spe
ed. At Crossro
ad a
nd
T-
junctio
n
, as t
he impa
ct of traffic re
gulati
on
and
drivin
g habits, fe
wer tra
ck
point
s ca
n be
colle
cted
at the intersection, and it
will resu
lt in that traj
ectory
integration
which is gai
ned by the traj
ectory
fusion al
gorit
hm deviates f
r
om the a
c
tua
l
location.
To imp
r
ove
the a
c
curacy
of tra
c
k re
sto
r
ati
on, it
nee
d to id
entify and
adju
s
t
such
erro
r.
The literature
[3] borrows
the co
ncept
of a sn
ake from the do
ma
in of image
pro
c
e
ssi
ng, a
nd
comp
utes an
app
roximati
on for th
e i
n
tersectio
n
p
o
int usi
ng a
simpl
e
sta
r
-sha
ped
co
ntou
r
suffice
s. In this pa
pe
r, we
pre
s
ent a
re
cogniti
on m
o
d
e
l and give a
n
adju
s
tment
algorith
m
for t
h
is
fusion e
r
ror b
a
se
d on the
spatial
-
temp
o
r
al characte
ri
stics of traje
c
tory at intersection. We h
a
ve
analyzed the
spatial
-
temp
oral
cha
r
a
c
te
ristics of t
he
car traje
c
tory
at the intersection
with the
intersectio
n
g
eometry a
nd
the v
ehicl
e
status va
riation
,
and b
u
ilt an
error
re
co
gni
tion mod
e
l an
d
a resto
r
atio
n algorith
m
for cro
s
sing p
o
sit
i
on deviati
on
that is gene
ra
ted by the trajectory fu
sion.
2. Res
earc
h
Method
The pu
rpo
s
e
of track reg
i
stration i
s
to rest
o
r
e the vehicle a
c
tua
l
traveling pa
th, the
shape of these tracks
correspondi
ng to the
road shape. Theref
ore,
fusing a
plurality of tracks
that come from the same
road
ca
n be
obtai
ned th
e
road
network topolo
g
y. There
are
so
me
related te
rms
by the followi
ng definition:
Defini
tion 1.
SHAPE_POINT, the point that repr
es
ent
s
the road s
h
ape and in this
pape
r
can
be a
pproximate u
nde
rstood
a
s
t
he
sam
p
ling
point
s of
traje
c
tory.
Each
SHAPE_POINT incl
ude m
u
ch informati
on, such
as position, speed, and directi
on etc.
Defini
tion 2.
NO
DE, a
sp
ecial
no
de
o
b
ject
s that i
s
made
up
in
orde
r to
express
road
netwo
rk top
o
l
ogy and can
be und
erstoo
d as an a
c
tua
l
interse
c
tion.
Defini
tion 3.
LINK, a curve with dire
cti
on m
ean
s th
e passag
e
wa
y between NODES. A
LINK is
c
o
mpos
ed
of two
NO
D
ES (START
NO
DE a
n
d
EN
D
NO
DE)
and
s
o
me
SHAPE_POINTS, namely, it is
a collec
t
ion of
dots
.
It
c
a
n be unders
t
ood as
a road that
c
o
nnec
t
s
two
NO
DES. A LINK i
s
ref
e
rred
as EXIT LINK
wh
en
the
con
n
e
c
te
d NODE
is a
START
NO
DE,
and calle
d
E
N
TER
LINK whe
n
the co
nne
cted NO
DE is an E
N
D
NO
DE. Figure
1
sh
ows th
e
floating ca
r traje
c
tory aro
und no
de. If we co
ns
i
der NODE
_1 a
s
the target n
ode, LINK_A
is
ENTER LI
NK; meanwhil
e
, LINK_B
and
LINK_C a
r
e
EXIT LINK.
Figure 1. The
floating car trajecto
ry aro
u
nd NO
DE
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
A N
e
w
Me
thod
o
f
T
r
a
j
ec
to
ry
R
e
s
t
or
ation
at Interse
c
tio
n
(Zhe
ng Ke)
565
Defini
tion 4.
DIRE
CTIO
N
ANG
L
E
wh
en v
ehi
cle
e
n
ters
into
a
NO
DE. That
is
the
change of
track direct
ion at the nearest
SHAPE_POINT of
E
N
TE
R LINK to
a NODE.As shown
in Figure 1,
SHAPE_POINT A1 and A2 form a vector
，
at same time SHAPE
_POINT A2 and
NO
DE1 al
so
form a ve
ctor, the ve
ctor an
gl
e i
s
the DIRE
CTI
O
N A
N
GLE
of LINK_A in
to
NO
DE1.
Defini
tion 5.
The
vehi
cle’
s di
re
ction
a
ngle
wh
en it
leave
s
p
o
int
NO
DE,
whi
c
h is an
angle variety after the beginning NODE at t
he SHAPE POINT when the vehicle departs from
LINK. Ju
st a
s
Figu
re 1
h
a
s
sho
w
n, th
e point NODE and SHAP
E_POINT_B1
form an ve
ctor,
SHAPE_POINT_B1 and S
H
APE_POIN
T_B2 form another vec
t
or, the inters
ec
tion angle
β
is
the
sha
pe an
gle
whe
n
the veh
i
cle leave
s
po
int NODE.
Defini
tion 6.
The vehi
cle’
s dire
ction a
n
g
l
e at point
NODE is th
e a
ngle vari
ety whe
n
the
vehicle enters point
NODE and depart
s
from LI
NK. As Figure 1
shown, SHAPE_POINT_A1, the
nearest poi
nt from NODE
1 after the vehicl
e
enters LINK_A, and NO
DE1 form an vector.
And
SHAPE_POINT_B1, the
nearest
shape point f
r
om
NODE1
after the vehi
cl
e depart
s
from
LINK_B, and
NODE
1 form anothe
r vector, the two
vectors int
e
rsectio
n
an
gle is the ve
hicle’
s
dire
ction an
gl
e at point NO
DE.
Defini
tion 7.
One ve
hicle
enters LI
NK
and the
oth
e
r d
epa
rts from LINK
at the poi
nt
NO
DE, whi
c
h
form a pair o
f
LINKS. To each p
a
ir
of LI
NKS, link the
two sha
pe p
o
ints that clo
s
e
to the point
NO
DE mo
stly besid
es
poi
nt NO
DE it
se
lf to form a curve. And d
e
fine the an
gl
e to
measure the smooth d
e
g
r
ss of t
he cu
rve
.
Figure 2 sho
w
s the a
m
plifying trajecto
ry around n
o
d
e
.
Figure 2. The
amplifying trajecto
ry aro
u
nd NO
DE
Taking two
sha
pe p
o
ints
that enter LI
NK
and
nea
r
the end
point
to form a ve
ctor, and
then ta
king
two
sh
ape
po
ints that
dep
art fro
m
LI
NK and
nea
r t
he
start
poin
t
NO
DE to f
o
rm
anothe
r vecto
r
. The directi
on angl
e
of the two ve
ctors is a
ngle
θ
.
As Figu
re 2
shown, angle
θ
is
the interse
c
tion angle
of vectors which
are formed f
r
om SHAP
E_POINT_A2
to
SHAPE_POINT_A1 and
SHAPE_POIN
T_B1to SHAPE_POINT_B2. Ta
k
i
ng two s
h
ape points
that ente
r
LI
NK an
d n
e
a
r
the e
ndp
oint
to form
a ve
ctor,
and
the
n
taki
ng t
w
o
sha
pe
point
s
that
enter
LINK th
en d
epa
rt fro
m
LINK
and
near st
a
r
t p
o
i
n
t NO
DE to
form
an
anoth
e
r ve
ctor.
Th
e
dire
ction
angl
e of two ve
ct
ors i
s
a
ngle
δ
. As Fig
u
re 2
sho
w
n,
angl
e
δ
i
s
the
inte
rse
c
tion
angl
e
o
f
two vec
t
ors that are f
o
rmed from SHAPE_
POINT_A2 to
SHAPE_PO
INT_A1 and from
SHAPE_POINT_A1 to SH
APE_POINT_B1. The larger one of angle
θ
an
d angle
δ
can
rep
r
e
s
ent the
smooth de
gree of the LINKS as
ax(
,
)
M
(1)
Defini
tion 8.
After the traje
c
torie
s
mixing
together,
the
distan
ce of p
o
int NO
DE d
e
viates
the actual ro
de. The dista
n
ce of pe
rpe
ndicular
lin
e segm
ent that from point NODE to the lin
e
segm
ent is forme
d
by the
sha
pe p
o
int
s
ne
ar
th
e p
o
int NO
DE m
o
stly wh
en th
e vehicl
e ent
ers
LINK an
d d
e
parts from
LI
NK. As Fi
gure 2
sho
w
n, t
he di
stan
ce
of perpen
dicular li
ne
seg
m
ent
from NODE1
to line s
egment (SHAPE_POINT_A1
，
S
H
APE_POINT_B1) is
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 563 – 57
0
566
Whe
n
the
ve
hicle
turn
s at
the inte
rse
c
tion, be
ca
use of
the affecti
on
of sa
mple
interval,
vehicle
spe
e
d
and di
re
cti
on, the sam
p
ling site
ra
rely appea
rs
at the intersection. O
n
shape
recognitio
n
o
r
clu
s
terin
g
algorith
m
of float
ing car trajecto
ry and red
u
ctio
n
of roads a
nd
intersectio
n
s,
due
to the
r
e
are
le
ss traj
ectori
es
at th
e interse
c
tion
s, may le
ad
to deviation
of
intersectio
n
p
o
sition a
nd a
c
tual po
sition
after fusion.
The po
sition
deviation m
a
inly appe
ars at t
he inte
rse
c
tion
s
whi
c
h exist the
pairs of
LINKS that near the lin
er.
So in rep
a
ir o
f
the
NODE p
o
sition , we should d
epe
nd
on these pai
rs
of LINKS. If there
is deviat
i
on of
NO
DE
after t
he t
r
aje
c
tory fu
sion, t
hen th
e devia
tion po
sse
s
ses
the followin
g
feature
s
.
Spatial-temp
oral Char
acteristic 1
: After eliminatin
g the point NO
DE, the positio
n
deviation of the point
NO
DE is p
r
od
uced by the turning of the vehi
cl
e, acco
rding to which
,
we
sho
u
ld not take the traje
c
tory that appe
ars at t
he veh
i
cle turni
ng in
to acco
unt when we a
r
e d
o
ing
the position repair. That is
to say, we sh
ould
miss the
interse
c
tion
angle when it
becam
e clo
s
ely
to a right angl
e that enters and de
part
s
from the pai
r of LINKS.
Spatial-temp
oral Ch
arac
teristic 2
: A
c
cording to
th
e rule
of the
vehicle m
o
tio
n
state,
we can kn
ow that if the ve
hicle do
es n
o
t
turn
at the intersectio
n
, then
the angl
e of the vehicle
satisfie
s th
e
angle
chan
g
e
rule, that i
s
, if t
he ve
hi
cle
driving
al
ong
a
straig
ht line i
n
to t
he
intersectio
n
a
nd it
cha
nge
s its di
re
ction,
then it
will
ch
ange
its
dire
ction an
d b
a
ck to the
straig
ht
line again
wh
en it depart
s
from the intersection.
Spatial-temp
oral Ch
arac
teristic
3:
T
h
e traje
c
to
ry that be
produ
ced
by the v
ehicl
e
passe
s the in
terse
c
tion m
u
st have angl
e chan
ge
wh
en it depart
s
the point NO
DE if the angle
cha
nge
s whe
n
it enters the
point NODE
.
Hen
c
e, we can de
cide
wh
ether the
r
e i
s
a dev
iation
of the point
NO
DE on the
basi
s
of
the motion
trajecto
ry characteri
st
ic on
the straight li
ne at the
in
t
e
rsectio
n
of t
he vehi
cle. T
hat
mean
s
we
can de
cid
e
whether we n
e
ed to d
o
th
e
positio
n repai
r on
the
ba
sis of th
e spe
c
ific
con
d
ition of
LINK: (1
) If the poi
nt NO
DE only
co
n
nect
s
with
on
e LINK, then
the poi
nt NODE
can
not re
co
g
n
ize th
e po
sition erro
r; (2
) i
1
f the
point NODE conn
ect
s
with
seve
ral
LINKS, and
all
of them are b
e
long to the
same pattern that either
the
y
enter LINK
or de
part fro
m
LINK, or th
en
the groupi
ng
doe
s
not n
e
ed to
rep
a
ir
the poi
nt
NO
DE; (3
) If th
e poi
nt NODE con
n
e
c
ts
with
several LI
NK
S, and
som
e
of them e
n
ter LINK a
nd th
e othe
r d
epa
rt from LI
NK, then
we
sh
oul
d
do furthe
r jud
g
ment with e
a
ch LI
NK.
The p
o
int
NO
DE which h
a
s
b
een
re
pai
red
sho
u
ld
gu
arante
e
the
straight lin
e th
at nea
r it
is the smo
o
th
est one. So we should fin
d
out t
he sm
oothe
st pair o
f
LINKS before do the posit
ion
repai
r of the point NO
DE. And the sele
ction crite
r
ia of
the most sm
ooth pair of LI
NKS as follo
ws:
(1) If there is only one p
a
i
r
that enters
and
de
pa
rts f
r
om the LI
N
KS, then the pair of
LINKS is the smooth
e
st on
e.
(2) If there
are several
pai
rs that
enters and depa
rts
from
th
e
LINK
S
，
then the
smalle
st
angle of the L
I
NKS is the most sm
ooth o
ne. Figur
e 3 shows ho
w to resto
r
e the
NODE po
sition
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
A N
e
w
Me
thod
o
f
T
r
a
j
ec
to
ry
R
e
s
t
or
ation
at Interse
c
tio
n
(Zhe
ng Ke)
567
Figure 3. Re
store the NODE position
Cho
o
si
ng two
sh
ape
point
s that ne
ar th
e
point
NO
DE
mostly from
the mo
st
smo
o
th pai
r
of LINKS to
form
an
segm
ent, and
that
segm
ent
i
s
th
e po
sition
rep
a
ir
stan
dard
segm
ent
of th
e
point
NO
DE.
The
ne
w p
o
in
t NO
DE
sho
u
l
d
be
in
th
is st
anda
rd
se
gm
ent. As
sh
own in
Figu
re
3(a),
the s
h
ape point SHAPE_POINT_A1 and SHAPE_P
O
INT_B1 form an
s
e
gm
ent is
the s
t
andard
segm
ent.
At the ra
nge
of sta
nda
rd
se
gment,
choo
si
ng
ne
w point
NO
DE
from th
e alt
e
rnative
positio
ns
on t
he ba
si
s of
specifi
c
conditi
on that
the v
ehicl
e ente
r
s
and d
epa
rts f
r
om the
pai
r
of
LINKS.
(1)
Cal
c
ulatin
g the midpoi
nt as the first
alte
rnative p
o
sition of the
point NO
DE, as the
point C_
ND1
in Figure 3(a
)
.
(2) Exce
pt the pair of LINK
S,
as to the other LINKS th
at
conn
ect to unprocesse
d NO
DE,
we shoul
d find the two sh
ape poi
nts th
at near the
u
npro
c
e
s
sed
NO
DE mostly
from the LINKS.
Figure 4 sh
o
w
s the
can
d
id
ate NO
DE po
sition.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 563 – 57
0
568
Figure 4. The
candi
date po
sition
If the intersection in the
standa
rd segment, the
n
we will ta
ke the segm
ent from
cal
c
ulatin
g a
s
the
altern
ative positio
n, as the
point
C_
ND2 in Fi
gure
3 a
nd
C_
ND2 in Fi
gu
re
4(a
)
.
If the intersection i
s
not i
n
the standard
segm
ent, we will choose
the
endpoint of the
segm
ent that
near the inte
rse
c
tion a
s
th
e alternat
ive
positio
n of po
int NO
DE the
mid interse
c
t
i
on
INTE_POINT
is not in the stan
dard
segme
n
t, so we will
choo
se the n
eare
r
en
dpoi
nt
SHAPE_POINT_A1 as
the alternative pos
i
tion of the
point NODE.
(3) Ta
king al
l the alternati
v
e position p
o
ints into the
alternative collectio
n in turn. We
can
ea
sily fin
d
out
that, all
the
altern
ative NO
DE are
in
the stan
dard
segm
en
t,
either are
t
h
e
endp
oint of t
he
stand
ard
segm
ent
or t
he inte
rse
c
tion of t
he
se
gment, a
nd t
here
a
r
e
sa
me
coo
r
din
a
te value in the alte
rnative po
sition point
s.
3.
Experiments
and the Ana
l
y
s
is
The expe
rim
ent data is p
r
ovided by
Wuh
an
Kotei
Informatics CO., LTD. T
he floating
trajecto
ry ha
s been g
a
thered for two m
onths
a
nd th
e colle
ction
a
r
ea
wa
s abo
u
t
400,000
sq
uare
kilomete
r. Th
e traje
c
tory
collectio
n freq
uen
cy is 1/
30
HZ a
nd the
GPS accu
ra
cy is 10
-6 latit
ude
and lon
g
itude
. The data include
s time, position, ro
ad range a
nd the
Parcel ID.
The p
r
o
g
ra
m
is aim
ed to
u
s
e th
e o
r
igin
a
l
traje
c
tory
da
ta to extra
c
t the a
r
ea
ro
ad
netwo
rk
topology, b
u
t
the spe
c
ific
a
l
gorithm
is no
t been
el
a
b
o
r
ated in
this th
esi
s
whi
c
h i
s
been
repla
c
e
d
by simple i
n
trodu
ction. Alg
o
rithm p
r
o
c
e
s
s: first
to d
e
lete the ab
normal traje
c
tory
point an
d th
en
split the traje
c
tory on the
basi
s
of the parcel
that in
clud
es traject
o
ry, the road
range
and t
h
e
intervals bet
wee
n
the tra
j
ectory
point
s. Ta
ki
ng th
e pa
rcel a
s
an unit a
nd
according to
the
trajecto
ry
curve simila
rity to re
co
gni
ze t
he
simila
rity
of the traj
ect
o
rie
s
a
nd bl
e
nding th
em ,f
rom
whi
c
h we ca
n
get a temporary roa
d
network top
o
logy
and there are
1 ,898 ,000 p
o
ints of NO
DE.
The exp
e
rim
ent appli
e
s t
he ab
ove mo
del to
recogn
ize a
nd
rep
a
i
r
the
s
e
NO
DES. And
the NODES that after repair will be contrasted
with the contemporary
NODES position in the
p
r
ac
tic
a
l ma
p.
Figure 5 sh
o
w
s fou
r
typica
l cases
of effect pi
ctu
r
e
s
that before error ide
n
tificati
on re
pair.
The left pi
ctu
r
e of
ea
ch
ca
se
s i
s
the
po
sition e
r
ro
r p
o
ints th
at are
identified
an
d the
right o
n
e
is
the po
sition o
f
NODE afte
r repai
r. We ca
n note fr
om t
he figure that
the LINKS which at the
po
int
NO
DE positio
n after rep
a
ir
become mo
re
smooth.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
A N
e
w
Me
thod
o
f
T
r
a
j
ec
to
ry
R
e
s
t
or
ation
at Interse
c
tio
n
(Zhe
ng Ke)
569
Figure 5. The
experime
n
tal
result of som
e
typical ca
se
To highlig
ht the perfo
rma
n
ce of the p
r
opo
se
d met
hod, we
co
mpared the
prop
osed
method with [
3
]. Table 1 lists the com
parison results.
Table 1. The
experim
ental
results.
Method
Total
Point
Recognized
point
cor
r
e
ct
repaired
error
repaired
cor
r
e
ct
recognition rate
improvement
rate
The pro
posed
method
1898113
146432
106324
40108
72.6%
5.60%
Method in [3]
128526
74223
54303
57.7%
3.91%
It can be
se
e
n
in table
1 th
at the co
rrect
re
c
ognitio
n
rate of the p
r
o
posed meth
o
d
in this
pape
r i
s
b
e
tter tha
n
the
method i
n
[3]
by 14.9%
a
nd the i
m
pro
v
ement is 1.
59% hig
her.
The
comp
ari
s
o
n
i
ndicates th
at the propo
se
d met
hod
provides
better perfo
rman
ce than exi
s
ting
method o
n
wi
ng to the con
s
ide
r
ation
of spatial
-
tem
p
o
r
al characte
ri
stics of
floating ca
r traj
ect
o
ry
at the typical road.
4. Conclu
sion
Becau
s
e
of the restrain of
geomet
rical
sha
p
e
s
at th
e ro
ad a
nd th
e influen
ce
of cha
nge
rule of vehi
cl
e run
n
ing
stat
ue, the F
CD
has it
s o
w
n f
eature
s
of tim
e
and
sp
ace. The research
of
FCD’
s tim
e
-space featu
r
e
s
is
ben
ef
it of the rate im
pro
v
ement of
traj
ectory den
oicing,
re
gist
rati
on
and mi
ning. T
h
is p
ape
r di
re
cts
at the p
o
sition erro
r
cau
s
ed
by traje
c
t
o
ry reco
gnitio
n
fusio
n
at th
e
intersectio
n
a
nd b
u
ild
s the
interse
c
tion
error re
co
gnit
i
on m
odel
an
d repai
r m
a
n
ageme
n
t. Th
e
experim
ent result shows this mod
e
l an
d algorithm
can identify the intersectio
n
point which can
cau
s
e the
po
sition deviatio
n
and fro
m
which im
prov
e
the co
rre
ct ra
te of the interse
c
tion po
siti
on
after the trajecto
ry fusi
on. The re
sea
r
ch t
eam
plans to
explore th
e
spatial
-
tem
poral
characteri
stics of floating
car t
r
ajectory
at t
he typical road, whi
c
h will be used in traj
ect
o
ry
recognitio
n
, fusio
n
redu
cti
on a
nd th
e i
m
provem
ent
of
traje
c
tory redu
ction accura
cy
in a
b
e
tter
degree an
d service mi
ning
application b
a
se
d on
floati
ng ca
r traje
c
t
o
ry in a better way.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 563 – 57
0
570
Referen
ces
[1]
Messel
odi S, Mode
na CM, Zanin M, et al.
Intellige
n
t ext
end
ed flo
a
ting
car data coll
e
c
tion.
Exper
t
Systems w
i
th Appl
icatio
ns
. 2
009; 36(
3): 421
3-42
27.
[2]
Schroe
dl S,
W
agstaff K, R
o
g
e
rs S, et
al. M
i
nin
g
GPS tr
ac
es for m
ap r
e
fi
neme
n
t.
Data
mi
nin
g
an
d
know
led
ge Dis
covery
. 200
4; 9(1): 59-8
7
.
[3]
Jian
g G, Chan
g A, W
u
C.
T
r
a
ffic information
collecti
on met
hod
base
d
on
GPS equi
ppe
d
floating c
a
r
.
Journ
a
l of Jili
n Univers
i
ty (Eng
ine
e
rin
g
an
d T
e
chn
o
lo
gy Editi
on)
. 201
0; 4: 971-9
75.
[4]
Z
hang W
,
Xu
J, W
ang H. Urban traffic situ
at
ion ev
al
uatio
n methods
ba
sed on
prob v
ehicl
e dat
a.
Journ
a
l of T
r
an
sportatio
n
Systems E
ngi
ne
eri
ng an
d Infor
m
a
t
ion T
e
chn
o
l
o
g
y
. 2007; 1: 43-
49.
[5]
Guan B, Liu L,
Chen J. Usin
g relativ
e
dista
n
ce
an
d ha
us
dorff distanc
e to mine traject
o
r
y
cl
usters
.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2013; 1
1
(1): 1
15–
12
2.
[6]
Brakatsou
l
as S
,
Pfoser D, S
a
l
a
s R.
On
map-
match
i
n
g
ve
hic
l
e tracki
ng
dat
a
. Proce
edi
ngs
of the
3
1
s
t
intern
ation
a
l co
nferenc
e on Ve
r
y
lar
ge d
a
ta
b
a
ses, VLDB En
do
w
m
ent. 200
5: 853-8
64.
[7]
W
ang C, Qin J, Guo M, Xu Y.
Prediction of vehicl
e traje
c
tor
y
bas
ed o
n
F
u
zz
y
col
o
r
ed petri n
e
t.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2013; 1
1
(10):
582
4–
583
2.
[8]
Lorko
w
s
ki S,
Mieth P, Schäfer RP.
New
IT
S app
licati
ons
for metr
opo
lita
n
are
a
s b
a
sed
on F
l
o
a
tin
g
Car Data
. ECT
R
I Young R
e
se
archer Sem
i
na
r, Den Haa
g
. 2005.
[9]
Guo H.
Ne
ural
net
w
o
rk
a
i
ded
Kalm
an
filt
erin
g for
in
tegrated
GPS
/INS navi
gati
o
n s
y
stem.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2013; 1
1
(3): 1
221-
122
6.
[10]
Z
hang
C, Y
a
n
g
X, Ya
n
X.
Tr
affic data collection system based on
floating cars
and
its
application in
Shan
gh
ai
. Pro
c
eed
ing
of the
F
i
rst Internatio
nal C
onf
er
enc
e of T
r
ansportation E
ngi
ne
eri
ng, Ch
en
gdu
,
Chin
a. 20
07.
[11]
Alvaro
V, Sa
n
g
ram
R. Se
ns
or fusi
on
vi
a
br
ain
emoti
o
n
a
l
l
earn
i
ng
for
gro
und
ve
hicl
e.
TELKOMNIKA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2014; 1
2
(7): 5
324
–5
330.
[12]
De Fa
britiis
C,
Rag
ona
R, V
a
l
enti G.
T
r
affic
estimatio
n
a
n
d
pre
d
ictio
n
bas
ed
on r
e
a
l
ti
me
floati
ng c
a
r
data
. Proceedings of 11th Internat
ional IEEE Conferenc
e
on Intell
igent T
r
ansportation
S
y
stems,
IT
SC, IEEE. 2008: 197-
20
3.
[13]
Nakata T
,
T
a
keuch
i
J.
Mining traffic
dat
a from
pr
obe-car system
for travel time prediction
.
Procee
din
g
s of
the tenth AC
M SIGKDD internat
i
ona
l conf
erenc
e on K
n
o
w
l
e
d
ge
discov
e
r
y
and
dat
a
minin
g
, ACM. 200
4: 817-
822.
[14]
Z
hang
Q, Li
B, Z
h
a
ng
N
.
Rese
arch
o
n
a
u
tomatic
target track
i
n
g
bas
ed
on
P
T
Z
s
y
stem.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(7): 1
582
–1
587.
[1
5
]
N
a
nn
i
M, Pe
dre
sch
i
D
.
T
i
me
-fo
c
u
s
ed
de
nsi
t
y
-
base
d
c
l
usterin
g
of tra
j
ectories
of mo
ving
ob
jects.
Journal of Intelligent Inform
ation System
s
. 2006; 27(
3): 267
-289.
[16]
Guangr
on
g B, Hongs
he
n L,
Nin
g
h
u
i
H. Mu
ltisens
or data
proces
s
i
ng
an
d fusin
g
bas
e
d
on K
a
lma
n
filter.
T
E
LKOMNIKA Indon
esi
an Jour
nal
of Electrical E
ngi
ne
erin
g
. 201
3; 11
(3): 1266-
12
77
.
Evaluation Warning : The document was created with Spire.PDF for Python.