Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
3
,
June
2020, p
p.
3022~3
034
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
3
.
pp3022
-
30
34
3022
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Effici
ent
and
s
ecure r
eal
-
time
m
ob
ile robots c
oope
ration us
ing
visual se
rvoing
So
umi
a
B
ou
d
ra
1
, N
as
r
-
Ed
d
ine B
errache
d
2
, A
mi
ne
Dah
an
e
3
1,
2
Inte
lligen
t
S
y
s
te
m
s Re
sea
r
ch
L
abor
at
or
y
,
El
e
ct
r
onic
s Depa
r
tmen
t,
Univer
sit
y
of
Sc
ie
nc
es
and
Tech
nolog
y
of
Or
an
US
TO
-
MB
,
Alger
ia
3
Resea
rch
L
abor
at
or
y
in
Industr
i
al
Com
puti
ng
an
d
Networks (RII
R),
Univer
sit
y
of
Or
an
1
Ahm
ed
B
en
Bella, Alge
r
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
pr
13
, 201
9
Re
vised
N
ov
2
7
,
2019
Accepte
d
Dec
10, 201
9
Thi
s
pape
r
de
al
s
with
th
e
ch
al
l
en
ging
proble
m
of
navi
ga
ti
on
in
fo
rm
at
ion
of
m
obil
es
robots
fle
e
t.
For
tha
t
pu
rpose,
a
sec
ur
e
appr
oac
h
is
used
base
d
on
visual
servo
ing
to
cont
rol
velocit
i
es
(l
ine
ar
a
nd
angular)
of
the
m
ult
ip
l
e
robots.
To
construc
t
our
s
y
st
em,
we
dev
el
op
th
e
intera
ct
ion
m
at
rix
whic
h
combines
the
m
om
ent
s
in
the
i
m
age
with
robot
s
vel
ocitie
s
and
we
esti
m
ate
the
dept
h
be
twe
en
ea
ch
robot
a
nd
the
ta
rg
e
te
d
obje
c
t.
Thi
s
is
done
wi
thout
an
y
comm
unic
at
ion
bet
we
en
th
e
robots
which
el
iminate
the
proble
m
of
the
infl
uen
ce
of
ea
ch
robot
err
ors
on
the
who
le
.
For
a
succ
e
ss
ful
visual
servoing,
we
pr
opose
a
powerf
ul
m
ec
han
ism
to
execut
e
safe
l
y
the
robo
ts
navi
ga
ti
on,
exploiti
ng
a
robo
t
acci
den
t
rep
ort
ing
s
y
stem
using
raspbe
rr
y
Pi
3
.
Thi
s
rep
ort
ing
s
y
stem
te
stb
ed
is
u
sed
to
sen
d
an
a
cc
id
ent
noti
ficat
ion,
in
the
form
of
a
spec
ifica
l
m
essage
.
Expe
r
imental
result
s
are
pre
s
ent
ed
using
nonholonomic
m
obil
es
robots
with
on
-
b
oar
d
r
ea
l
t
ime
ca
m
er
as,
to
s
how
the
eff
e
ct
iv
ene
ss
of
th
e
proposed
m
et
hod.
Ke
yw
or
d
s
:
Kinem
at
ic
s
m
o
delin
g
Mult
i robo
ts
syst
e
m
Secu
re
nav
i
gation
Visu
al
se
r
vo
i
ng
Copyright
©
202
0
Instit
ute of
Ad
v
ance
d
Engi
ne
eri
ng
and
Sc
ie
n
ce
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Soum
ia
Bou
dr
a
,
In
te
ll
igent
Syst
e
m
s Resea
rch
Lab
or
at
ory
, El
ect
ronics D
e
pa
rtm
ent,
Un
i
ver
sit
y o
f S
ci
ences a
nd Te
chnolo
gy
of Or
an UST
O
-
MB,
El M
naou
r,
BP
1505, B
ir El
D
j
ir
3100
0 Or
a
n, Alge
ria
.
Em
a
il
:
so
um
ia
.
boudra
@univ
-
us
to
.d
z
1.
INTROD
U
CTION
Roboti
cs
is
a
com
plex
en
gine
erin
g
fiel
d
be
cause
as
so
ci
at
es
dee
p
knowl
edg
e
of
se
ve
ra
l
discipli
nes
su
c
h
as
el
ect
ronic,
m
echan
ic
and
s
of
t
war
e
e
ng
i
neer
i
ng.
Th
is
com
plexity
i
s
com
po
unde
d
with
the
transit
ion
to
m
ul
ti
-
ro
bot
sy
stem
s.
The
co
op
e
rati
ve
Mult
i
R
obot
Con
t
r
ol
S
yst
em
s
(
MR
CS)
has
gro
wn
co
ns
ide
r
ably
in
the
la
st
decad
e
,
due
to
the
ex
te
ns
ive
nee
d
to
su
ch
te
c
hnology
in
dif
fer
e
nt
fiel
ds
m
a
inly
:
bio
m
edical
sc
ie
nce
,
rescu
e
i
ng
,
dis
placem
ent
of
heav
y
it
em
s
,
su
r
veill
ance
f
or
e
xam
ple
cat
ch
ing
i
nv
a
de
r
s
unde
r
sur
ve
il
la
nce
areas
[
1],
se
nsor
netw
orks
a
nd
co
operati
ve
t
ran
s
port
.
T
he
i
dea
is
that
aut
onom
ou
s,
c
ollaborat
ive
r
ob
ot
s
can
achieve
bette
r
resu
lt
s
the
n
each
rob
ot
separa
te
ly
[2
]
.Coop
erati
on
m
eans
that
robo
ts
m
us
t
co
m
m
un
ic
at
e
to
exch
a
nge in
f
orm
at
ion
and c
oor
din
at
e t
heir
a
ct
ion
s i
n order
to accom
plish
a
com
m
on
ta
sk
[
3,
4]
.
Howe
ver,
buil
ding
a
c
ontrol
syst
e
m
fo
r
a
gro
up
of
aut
onom
ou
s
r
obots
is
a
ver
y
com
plex
wo
r
k.
Am
on
g
the
ad
van
ta
ges
of
th
e
MRCS
,
we
can
ci
te
:
1)
The
syst
e
m
can
react
qu
ic
kly
to
exter
nal
interf
eren
ce
du
e
t
o
the
r
obots
div
isi
on
of
la
bour
ba
sed
on
e
nv
i
ronm
ent
sh
arin
g,
sin
ce
the
lim
it
at
ion
of
the
fiel
d
of
visio
n
of
eac
h
ind
i
vi
du
al
m
ob
il
e
m
akes
alm
os
t
i
m
po
ssible
the
pe
rcep
ti
on
of
the
r
obot
entire
env
ir
onm
ent;
2)
The
t
rainin
g
of
MR
CS
m
ai
ntains
a
s
pec
ific
at
ta
ck
on
t
he
ou
tsi
de
that
c
an
e
nh
a
nce
def
e
ns
e
ca
pa
bili
ti
es;
3
)
It c
a
n
im
pr
ove the
ro
bu
st
ne
ss and e
ff
ic
ie
nc
y of t
he
syst
e
m
.
In
the
pr
ese
nt
stud
y,
we
wa
nt
to
i
m
ple
m
e
nt
a
colla
borat
ive
co
ntro
l
sy
stem
fo
r
a
gro
up
of
non
-
ho
l
onom
ic
m
o
bile
robo
ts
,
usi
ng
cam
eras
aimi
ng
at
tracking
obj
ect
s
a
nd
perform
ing
ta
sk
s
ba
sed
on
visu
a
l
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Eff
ic
ie
nt a
nd s
ecure re
al
-
ti
me
mob
il
e r
obots
cooper
atio
n us
ing
vis
ual serv
oing
(
So
um
i
a
Bo
ud
r
a
)
3023
inf
or
m
at
ion
in
real
-
ti
m
e
con
tr
ol.
T
he
ap
plica
ti
on
s
of
m
achine
vision
te
c
hn
i
qu
e
s
can
ge
ne
r
al
ly
be
cat
ego
r
iz
ed
into
tw
o
cl
ass
es
base
d
on
t
he
re
qu
i
rem
ent
of
real
-
ti
m
e
processin
g
;
Non
-
Tim
e
-
Crit
ic
a
l
Visio
n
A
pp
li
c
at
ion
s
(N
TC
VA)
a
nd
Tim
e
-
Crit
ic
a
l
Visio
n
A
pp
li
cat
ion
s
(TC
V
A)
.
NTCV
A
don
’
t
re
quire
visu
al
fee
db
ac
k
for
highb
a
nd
widt
h
real
-
tim
e
con
t
ro
l.
E
xam
ples
inclu
de
obj
ect
rec
ogni
ti
on
an
din
s
pe
ct
ion
of
packagin
g
qu
al
it
y
[5
-
7]
.
TCVA
re
quire
s
r
eal
-
ti
m
e
visu
al
fee
dback
.
Exam
ples
include
visi
on
-
g
ui
ded
pic
k
-
a
nd
-
place,
al
ign
m
ent
and
inse
rtio
n
[
8].
T
he
intr
oductio
n
t
o
3
-
Dco
m
pu
te
r
vi
sion
te
c
hn
i
ques
helpe
d
t
o
m
ake
the
te
chn
i
que m
or
e
so
phist
ic
at
ed
[
9]
.
A
sta
nd
a
r
d
intr
oduc
ti
on
to v
isual
s
ervoin
g
te
ch
ni
qu
e
s
we
re
al
so
of
g
reat
interest
[
10
]
.
Seve
ral
public
at
ion
s
hav
e
ap
pear
e
d
in
rece
nt
ye
ars
do
c
um
enting
diff
e
r
ent
aspects
t
o
cl
assify
the v
is
ual se
rvoin
g
syst
em
s
[11
-
13]
as
sho
w
n
in
T
a
ble
1.
Table
1.
Cl
assi
ficat
ion
of v
is
ua
l servoin
g sy
stem
s w
it
h
res
pe
ct
to
se
ver
al
a
sp
ect
s
1.
The p
o
sitio
n
of
the
ca
m
e
ra
as e
y
e
-
in
-
h
an
d
and
ey
e
-
to
-
h
a
n
d
2.
The f
eedb
ack repr
esen
tatio
n
m
o
d
e p
o
sitio
n
-
b
ased
,
i
m
a
g
e
-
b
ased
,
an
d
hybr
id
vis
u
al servo
in
g
3.
The co
m
b
in
atio
n
of
vis
io
n
sen
s
o
r
an
d
con
troller of
the j
o
in
t
-
d
y
n
a
m
ic loo
k
-
an
d
m
o
v
e sy
ste
m
an
d
direct vis
u
al servo
sy
ste
m
4.
The
u
se
o
f
th
e
v
isu
al
in
f
o
r
m
atio
n
(
co
n
trol
m
o
d
el
)
d
istin
g
u
ish
es
two
ty
p
es
o
f
v
isu
al
serv
o
in
g
syste
m
s:
k
in
e
m
atics
-
b
ase
d
v
isu
al sev
o
in
g
an
d
dyna
m
ic
vis
u
al se
rvo
in
g
Gr
eat
ef
forts
ha
ve
bee
n
devo
te
d
to
num
ero
us
ap
plic
at
ion
s
of
visu
al
ser
voin
g
in
r
obotic
s
in
the
la
st
decad
e
.
It
can
be
us
e
d
as
soo
n
as
a
visio
n
sens
or
is
avail
able
and
a
ta
sk
is
assigne
d
to
a
dynam
ic
s
yste
m
to
con
t
ro
l
it
s
m
otion
.
Nowa
days,
visua
l
servo
ing
is
wi
dely
us
e
d
in
dif
fer
e
nt
fiel
ds
m
ai
nl
y;
g
aze
con
tr
ol
for
ta
rg
et
tracki
ng,
n
avigati
on
of
a
m
ob
il
e ro
bot t
o
fo
ll
ow a w
al
l using
a
n
om
nid
irect
ion
al
v
is
ion
se
ns
or,
gr
a
sp
in
g
a
ball
with
a
hum
ano
id
r
obot
,
assem
bly
of
m
ic
ro
el
ect
ro
m
echan
ic
al
syst
e
m
(MEM
S)
an
d
film
of
a
dia
logue
within
t
he
c
onstrai
nts
of
a
s
cript
in
anim
atio
n.
Vi
si
on
-
ba
sed
rob
ot
co
nt
ro
l
m
et
ho
d
use
s
the
visu
al
da
ta
to
con
t
ro
l
t
he
m
otion
of
dynam
i
c
syst
e
m
s.
A
c
on
t
ro
l
la
w
has
to
be
co
ns
ide
r
ed
that
the
m
e
asur
em
ents
s(
t
)
reach
a
desire
d
val
ue
s
*
,
def
i
ning
e
xa
ct
reali
zat
ion
of
the
ta
s
k.
T
he
obj
e
ct
ive
is
t
o
m
ini
m
iz
e
the
dif
fer
e
nce
bet
ween
the
curre
nt
an
d
wa
nted
c
onf
igurat
ion
s
.O
ne
of
the
ce
ntral
qu
est
io
ns
that
m
us
t
be
proc
essed
in
m
ulti
-
r
obot
syst
e
m
,
reg
ar
dl
ess
of
the
a
pp
li
cat
ion
dom
ain
,
is
how
to
c
oope
rate
ef
fecti
vely
an
d
a
utom
at
ic
ally
m
an
y
robo
t
s
to
e
xecu
te
a
c
om
m
on
ta
sk
.
Be
cause
t
he
sta
bili
ty
of
t
he
visu
a
l
ser
vo
i
ng
syst
e
m
[14
,
1
5
]
,
w
e
m
us
t
bu
il
t
op
tim
a
l
visu
al
c
har
act
e
risti
cs.
O
ur
pro
po
sit
io
n
s
houl
d
sat
isfy
se
veral
crit
eri
a
m
a
inly
;
sta
bili
t
y
of
the
dy
nam
ic
sy
stem
,
rob
us
tness
to
cal
ibrati
on,
loc
al
m
ini
m
a
avo
idan
ce,
non
-
si
ngularit
y
and
m
axi
m
al
deco
up
li
ng
an
d
li
ne
ar
li
nk
betwee
n
the
v
isual
char
act
er
ist
ic
s
and
the
degrees
of
f
re
edo
m
.
W
e
de
velo
p
ed
a
c
ontrol
la
w
to
c
al
culat
e
the
velocit
y
co
m
po
nen
ts
of
th
e
r
obots
pro
vidi
ng
ex
pone
ntial
decay
of
the
error.
T
w
o
m
ain
as
pects
ha
ve
gr
ea
t
i
m
pact
on
the
beh
a
vio
r
of
an
y
visu
al
ser
vo
i
ng
sc
hem
e;
th
e
sel
ect
ion
of
the
v
is
ual
featu
res
us
e
d
as
in
put
of
the
con
t
ro
l
la
w
and
the
f
or
m
of
the
co
ntr
ol
schem
e.
Visu
al
info
rm
at
ion
obta
ined
from
t
he
i
m
age
pr
oc
essin
g
can
be
us
e
d
to
extracti
ng
2D
featur
e
s.
It
can
al
so
be
us
e
d
f
or
est
im
ating
pose
pa
ram
et
ers
by
e
m
plo
yi
ng
po
s
e
est
i
m
ation
al
gorithm
fr
om
c
om
pu
te
r
visio
n.
T
he
est
im
a
t
ed
pose
is
tra
ns
f
or
m
ed
into
the
3D
featu
r
es
an
d
the
2D
a
nd/o
r
3D
feat
ur
es
ar
e
then
use
d
in
the
c
on
t
ro
l
sc
hem
e
[16]
.
He
nce,
o
pti
m
iz
at
i
on
te
ch
niques,
rob
ot
dynam
ic
and
r
obot
kin
em
at
ics
are
us
e
d
in
t
he
m
od
el
ing
of
t
he
c
ontrol
sche
m
es.
The
ai
m
s
of
the
prese
nt
s
tud
y
are
fi
rstly
,
desi
gn
an
d
de
velo
p
a
ne
w
sec
ur
e
synopti
c
base
d
on
the
a
ppr
oa
ch
of
the
visua
l
servoin
g
t
o
con
t
rol
li
near
a
nd
a
ngula
r
velocit
ie
s
of
m
ulti
ple
rob
ots.
Sec
on
dly,
to
s
ho
w
cl
early
th
e
int
erest
of
t
he
a
cci
den
t
repor
ti
ng
syst
em
, esp
eci
al
ly
i
n
the
outd
oor
e
xp
e
rim
ents.
The
rem
ai
nin
g
sect
ion
s
a
re
outl
ined
as
f
ollow
s:
In
sect
io
n
2
we
presen
t
a
br
ie
f
s
ur
ve
y
of
relat
ed
works
on
vis
ua
l
servoin
g
sche
m
es.
S
ect
ion
3,
is
devoted
t
o
introd
uce
an
d
e
xp
la
in
t
he
r
ob
otic
visu
al
se
rvoin
g
syst
e
m
,
an
em
ph
a
sis
is
pu
t
on
th
e
im
ple
m
e
ntati
on
of
t
he
pro
po
se
d
syst
em
enco
m
passi
ng
se
ver
al
m
od
ule
s
.
Sect
ion
4,
we
pr
ese
nt
an
il
lustrati
on
of
th
e
resu
lt
s
obta
ined
in
order
t
o
achieve
our
ultim
at
e
go
al
wh
il
e
con
cl
us
io
ns
a
nd
per
s
pecti
ves
are
offer
e
d
in
s
ec
ti
on
5.
2.
RE
LATE
D
W
ORKS
The
inter
est
in
us
in
g
MR
CS
is
d
ue
to
their
cha
racteri
sti
cs
reali
zed
with
dif
fer
e
nt
ty
pes
of
auto
no
m
ou
s
ve
hicle
s
su
ch
as
gro
und
m
ob
il
e
ro
bots,
un
derwate
r
ve
hicle
s
[17],
unm
ann
ed
aerial
veh
ic
l
es,
and
ai
rcr
aft.
M
ulti
-
r
obot
co
ordi
nation
pur
pos
e
is
to
le
t
ro
bots
s
har
i
ng
any
inf
or
m
at
ion
bet
ween
them
.
Fo
r
e
xam
ple, a
ro
bot p
os
it
ion
can b
e sh
a
red
with o
the
rs
in o
r
der
to
com
pute
m
or
e p
recise other
robot p
osi
ti
on
and
av
oid
colli
sion
.
I
n
t
he
co
ntext
of
c
om
pu
te
r
visio
n
f
ra
m
ewo
r
k,
vis
ua
l
servoin
g
te
ch
niques
ca
n
be
us
e
d
t
o
adjust
the
tra
je
ct
or
ie
s
of
t
he
m
ob
il
e
ro
bot
s
fleet
.
T
he
present
w
ork
c
on
sist
s
of
de
fi
ning
a
m
et
hod
that
com
bin
es
v
is
ua
l
ser
vo
i
ng
an
d
c
ontrol,
to
tr
ack
a
gr
oup
of
m
ob
il
e
ro
bots.
The
be
hav
i
or
of
the
loop
syst
e
m
dep
e
nds
on
the
cho
ic
e
of
visua
l
inform
atio
n
and
t
he
ass
ociat
ed
co
ntr
ol
la
w
.
The
re
is
a
la
rg
e
num
ber
of
visu
a
l
pr
im
itive
inf
orm
at
ion
scat
egor
iz
ed
into
tw
o m
ai
n
gro
up
s
, t
hat
that ca
n be
us
e
d
in
v
is
ual s
ervoin
g
ta
s
k
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020 :
30
22
-
3034
3024
The
first
cat
eg
or
y
us
es
m
easur
em
ents
co
ntained
i
n
the
im
age
plane
an
d
will
be
re
fere
nced
un
de
r
the
acr
on
ym
IBV
S
(I
m
age
-
Ba
sed
Visu
al
Ser
vo
i
ng).
T
he
c
on
tr
ol
la
w
is
cal
culat
e
d
f
r
om
i
m
age
error
,
so
that
the
be
hav
i
or
in
3D
sp
ace
is
no
t
di
rectl
y
con
stra
ined.
T
his
ty
pe
of
co
ntr
ol
is
robu
st
to
s
um
m
ary
cal
ibrati
on
of
the
intrinsic
c
a
m
era
par
am
eter
s.
Geo
m
et
ric
pr
im
itives,
suc
h
as
points
,
segm
ents,
li
nes
an
d
el
li
ps
es,
wer
e
t
he
fi
rst
to
be
s
tud
ie
d.
[
18
]
Propose
d
a
co
ntr
ol
schem
e
based
on
sig
n
det
ect
ion
al
gorith
m
s
fo
r
m
ob
il
e
ro
bot
pl
at
fo
rm
us
ing
Harris
co
rn
e
r
po
i
nt
detect
or
with
ei
ge
nval
ue
s
dec
om
po
sit
ion
s
t
o
trac
k
th
e
path.
Accor
ding
to
[19],
Me
ba
rk
i
p
r
opos
e
d
m
ome
nt’s
i
m
age
al
lowing
acc
eptable
res
ults
in
the
2D
a
nd
3D
beh
a
viors
for
bi
nar
iz
ed
im
ages.
A
m
et
ho
d
f
or
co
ntr
olli
n
g
four
d
e
gr
ees o
f
f
reedom
has
be
en
pr
opos
e
d
by
[20],
requirin
g
im
age
processin
g
a
nd
ta
king
into
account
the
pi
xe
l
value
of
the
i
m
age.
H
om
ography
was
inte
gr
at
e
d
in
the
con
tr
ol
s
chem
e
to
con
tr
ol
the
six
degr
ees
of
f
reedom
of
rob
ot
[21]
.
In
s
pire
d
by
this
wor
k,
plan
ni
ng
was
pro
po
se
d
to
e
nsure
r
obus
tne
s
s
to
cal
ibrati
on
error
s
f
or
la
r
ge
ca
m
era
m
ov
e
m
ents
[22]
.
I
n
order
t
o
av
oid
these
i
m
age
processi
ng
ste
ps,
s
uch
as
segm
en
ta
ti
on
or
pr
im
itive
tracki
ng,
[
23
]
stud
ie
d
a
"di
rect"
vis
ual
se
rvoin
g
schem
e,
ta
kin
g
into
acc
ount
the
pix
el
val
ue
s
of
the
curre
nt
and
de
sire
d
i
m
ages
as
pr
im
itive
.
Kadhi
m
an
d
Abd
ulsahi
b
[
24]
desig
n
e
d
a
veh
ic
le
as
rob
ot
-
m
ou
nte
d
se
ns
ors
ca
pa
ble
of
ca
rr
yi
ng
the
sens
or
s
of
the
m
et
al
and
ob
sta
cl
e.
In
or
der
to
im
pr
ov
e
the
accuracy
of
ide
ntific
at
ion
of
obj
ect
in
dif
fe
ren
t
il
lum
inatio
n
a
nd
backg
rou
nd
c
onditi
ons,
howe
ver
a
uthor
s
do
no
t
stu
dy
the
ki
nem
a
ti
c
m
od
el
s
of
a
r
obot
w
hich
help
to
va
li
date
or
ve
rify
by
cal
culat
ion
the
m
echan
ic
al
pe
rfor
m
ances
of
a
syst
e
m
.
Lat
e
rone
,
m
utu
al
i
nfor
m
at
ion
betwee
n
the
cu
r
re
nt
im
age
a
nd
the
de
sired
im
age
was
pro
po
s
ed
in
or
der
t
o
i
nc
rease
rob
us
tne
ss
to
occ
ultat
ion
an
d
changes
of
il
lum
inati
on
[
25]
.
A
no
t
her
crit
erio
n
base
d
on
t
he
diff
e
re
nc
e
bet
ween
th
e
cu
rr
e
nt
im
a
ge
a
nd
the
ref
e
re
nce
im
age,
update
d
accor
ding
to
i
ll
u
m
inati
on
of
t
he
cu
rr
e
nt
im
age
,
was
pro
po
s
ed
t
o
m
anag
e
any
changes
in
li
gh
ti
ng
a
nd
m
ult
i
m
od
al
ity [26]
.
The
sec
ond
a
ppr
oac
h,
PB
VS
(positi
on
-
base
d
vis
ual
ser
vo
i
ng)
use
s
pose
est
i
m
atio
n
afte
r
local
iz
at
ion
al
gorithm
[2
7]
.
T
he
3D
s
pac
e
is
well
c
onst
raine
d
,
howe
ve
r
this
is
n
o
lo
ng
e
r
t
he
case
for
t
he
im
age
sp
ace,
wh
ic
h
m
ay
lead
to
o
utput
pri
m
it
ive
i
m
age.
Furthe
rm
or
e
,
to
properly
do
the
3D
w
or
k,
it
is
i
m
po
rtant
to
accuratel
y
cal
ibrate
the
c
am
era.
T
he
m
a
in
pro
blem
of
the
PB
VS
ap
proach
is
t
hat
th
e
geo
m
et
ric
m
od
el
of
the
obj
ect
s
hould
be
kn
own
for
pose
est
im
at
ion
,
w
hic
h
m
akes
it
a
“
m
od
el
-
based
”
m
et
ho
d
c
o
m
par
ed
t
o
the
im
age
-
base
d
m
et
ho
d.
T
he
cam
era
cal
ibrati
on
is
requir
ed
to
obta
in
th
e
unbiase
d
Ca
r
te
sia
n
posit
io
ni
ng
t
o
ov
e
rc
om
e
the
sensiti
vity
of
th
e
ca
m
era
cal
ibrati
on
e
rror
.
H
y
br
i
d
ap
proac
he
s
com
bin
ing
2D
an
d
3D
pri
m
it
ives
associat
ion
the
ir
adv
a
ntages
t
o
be
nef
it
from
a
bette
r
3D
be
hav
i
or
[
28]
or
to
keep
the
p
ri
m
itives
in
the
fiel
d
of
view
[
22]
.
Sel
ect
i
ng
go
od
vi
su
al
cha
racteri
sti
c
is
a
cru
ci
al
aspect
of
vis
ual
ser
vo
i
ng
a
s
it
is
nece
ssary
for
achievin
g
opti
m
al
velocit
ie
s
an
d
in
creasi
ng
accu
racya
nd
reli
abili
ty
of
im
age
m
easur
em
ents
,
aff
ect
s
perform
ance an
d
r
obus
t
ness of vis
ual ser
vo
ing
[
29
]
.
Im
aging
m
eas
ur
em
ents
are
ei
ther
use
d
di
rectl
y
in
the
con
t
ro
l
lo
op
or
us
e
d
for
r
el
at
ive
po
s
e
es
tim
ation
.
T
he
nu
m
ber
of
de
gr
ees
of
f
ree
do
m
(D
OF)
to
be
con
tr
olled
by
the
e
m
plo
ye
d
con
t
ro
l
sc
hem
e
determ
ines
the
m
ini
m
u
m
num
ber
of
i
nd
e
pe
nd
e
nt
feature
s
re
qu
ir
ed
.V
is
ual
feat
ur
es
ca
n
be
sel
ect
ed
in
2D
i
m
age
sp
ace
as
po
int
co
or
din
at
es,
par
am
et
ers
represe
nting
strai
ght
li
nes
or
el
li
ps
es,
reg
i
on
of
interest
and
co
ntour
s
[30,
31]
.
T
he
se
featu
res
a
r
e
def
i
ne
d
f
rom
i
m
age
m
ea
su
rem
ents
[
32]
.
In
ca
se
of
i
m
age
po
i
nts,
Ca
rtesi
an
c
oor
din
at
es
are
ge
ner
al
ly
us
e
d
howe
ve
r
,
it
is
possi
ble
to
us
e
t
heir
pola
r
a
nd
cy
li
ndrical
coor
din
at
es
[33]. In
ge
ner
al
,
al
l
par
am
et
ers
def
i
ning the
int
ern
al
cam
era c
al
ibrati
on are
re
qu
i
red.
Im
age
m
o
m
ents
can
al
so
be
us
ed
i
n
vis
ual
ser
vo
i
ng
[34]
giv
es
be
tt
er
resu
lt
s
co
m
par
ed
to
the
cl
assic
al
visu
al
servoin
g
schem
e
.
I
m
age
m
o
m
ents
al
lo
w
ge
ner
ic
re
presentat
ion
a
nd
are
able
to
ha
nd
l
e
si
m
ple
geo
m
etr
ic
al
pr
im
it
ives
and
al
so
co
m
plex
ob
j
ect
s
with
unknow
n
sh
a
pes.
It
is
sh
own
that
m
o
m
ent
inv
a
riants
can
be
us
e
d
to
des
ign
decou
pled
2D
visu
al
ser
voin
g
schem
e
and
t
o
m
ini
m
ize
the
nonlinea
r
it
y
of
the
interact
io
n
m
at
rix
relat
ed
to
the
sel
ec
te
d
visua
l
fea
tures
[
35]
.
Ge
ner
a
ll
y,
ob
j
ect
m
od
el
an
d
im
age
m
easur
em
ents
are
us
ed
to
cal
culat
e
or
the
r
el
at
ive
pose
be
tween
o
bj
ect
a
nd
cam
era
fr
a
m
es
in
the
Ca
r
te
sia
n
sp
ace
,
or
t
o
re
const
ru
ct
the
3D
co
ordinates
.
Ther
e
fore,
an
adv
a
nce
d
kn
owle
dge
ab
out
the
cam
era
cal
i
br
at
io
n
par
am
et
ers
are
require
d
[36]
.
In
t
his
sect
ion,
we
re
view
s
om
e
visu
al
s
asp
ect
s
and
pr
im
itives
inf
orm
ation
that
can
be
use
d
in
visu
al
servoin
g
ta
s
k
.
A
lso
we
st
udy
so
m
e
m
ulti
ro
bot
sce
na
rio
s
desig
n
and
analy
sis.
W
e
ca
n
e
num
erate
the contri
bu
ti
ons
of our p
ape
r
as foll
ows:
Pr
op
os
al
of
a
MR
CS
based
on
vis
ual
servoin
g,
in
cl
ud
in
g
the
desig
n
an
d
de
velo
pm
ent
of
an
e
xp
e
rim
ent
al
protoc
ol
al
lowi
ng
t
he
c
ollaborat
io
n
bet
w
een
m
ob
il
e
robo
ts
fo
ll
ow
i
ng
a
m
ast
er
ro
bo
t
that i
s b
ase
d at
f
irst
on the
d
et
erm
inati
on
of e
ach
rob
ot co
m
pu
te
d
posit
i
on
.
The
syst
e
m
is
able
to
al
lo
w
m
ul
ti
ro
bots
na
viguati
ng
in
form
ation
usi
ng
t
he
kin
em
at
ic
m
od
el
of
m
ob
il
e
rob
ot
s
an
d
a
c
a
m
era
to
c
onstruct
a
com
m
a
nd
l
aw
.
As
vis
ual
pri
m
i
ti
ves
in
the
ser
voin
g
lo
op,
we
use
the m
o
m
ents ex
tract
ed
fro
m
the c
urren
t i
m
age.
Our
wor
k
is
well
-
su
it
ed
f
or
real
a
pp
li
cat
ion
s
as
it
is
robu
st
,
a
nd
le
ad
s
to
a
fast
im
plem
entat
ion
of
the m
ulti
r
obots
vis
ual ser
vo c
on
t
ro
l i
ss
ue
.
It sho
ws
cl
earl
y t
he
interest
of a
rob
ot acci
de
nt d
et
ect
io
n r
eporti
ng syst
e
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Eff
ic
ie
nt a
nd s
ecure re
al
-
ti
me
mob
il
e r
obots
cooper
atio
n us
ing
vis
ual serv
oing
(
So
um
i
a
Bo
ud
r
a
)
3025
3.
R
OBOT
I
C VI
SUAL SE
RVOIN
G S
YS
TE
M
We
de
velo
pe
d
an
ex
per
im
ental
pr
oto
c
ol
to
con
tr
ol
a
gro
up
of
r
obots
base
d
on
the
inf
or
m
at
ion
pro
vid
e
d
by
c
a
m
eras
m
ou
nted
on
eac
h
robo
t.
A
desire
d
value
i
n
the
i
m
age
is
deter
m
ined
f
or
eac
h
r
obot
ind
e
pende
ntly
of
the
ot
her
r
obot.
The
vis
ual
servoin
g
a
i
m
is
to
con
trol
the
current
value
by
fo
ll
ow
i
ng
the
desire
d
val
ue
in
the
im
ag
e
by
est
i
m
a
ti
ng
the
de
pth
Z
betwee
n
the
obj
ect
an
d
th
e
c
a
m
era.
The
f
ollow
in
g
Figure
1
s
hows
the
diag
ram
o
f
our vis
ual ser
voin
g
syst
em
.
Figure
1.
Sc
he
m
at
ic
o
f
r
oboti
c v
is
ual ser
voing
syst
em
An
IB
VS
co
nt
ro
ll
er
is
m
ai
nl
y
i
m
po
rtant
to
con
ti
nuously
adjust
the
w
he
el
velocit
ie
s
a
nd
the
refore
adjust
the
r
obot
to
m
ov
e
th
e
i
m
age
coo
r
di
nates
of
the
tracke
d
ob
j
ect
to
the
desire
d
po
sit
io
n
in
the
i
m
age
plane.
T
he
des
ired
posit
ion
m
us
t
be
def
ine
d
as
(ud,vd).
We
then
get
an
error
e
qu
at
io
n
for
our
im
age
plane
coor
din
at
es:
d
d
uu
e
vv
(1)
3.1.
Visu
al
fe
at
u
res
The
inf
or
m
at
i
on
c
ollec
te
d
by
the
vision
sens
or
dec
rease
s
the
sta
bili
ty
pro
blem
s
if
th
e
m
ov
em
ent
carried
out
by
the
r
obot
is
com
plex.
T
he
refor
e
,
sel
ect
ing
t
he
a
ppr
opriat
e
in
form
a
ti
on
is
im
po
rt
ant
to
accuratel
y
ap
pl
yi
ng
the
re
quired
ta
s
k.
Wh
il
e
sever
al
c
ho
ic
es
of
s
e
xis
t,
we
ha
ve
c
hose
n
in
the
presen
t
wo
r
k
the
inv
a
riant
m
o
m
ents,
as
vi
su
al
inf
or
m
at
i
on.
Im
age
m
o
m
ents
can
be
com
pu
te
d
f
rom
a
set
of
po
ints
or
a
well
-
se
gm
ented
re
gion
in
t
he
im
age.
The
ge
om
et
ric
m
o
m
ents
of
a
2D
distrib
utio
n
f
un
ct
io
n
f(
x,y)
can
be
expresse
d
as:
(
,
)
pq
pq
R
m
x
y
f
x
y
d
x
d
y
(2)
w
ith
f(
x,
y
)
≥
0
be
real
bo
unde
d
f
un
ct
io
n
with
s
uppor
t
on
com
pact
reg
i
on
R
.
R
is
the
area
occ
upie
d
by
the
obj
ect
in
th
e
i
m
age
and
p
,
q
is
the
m
o
m
e
nt
order.
In
thi
s
pap
e
r,
bi
nar
y
i
m
age
fu
nctio
ns
(i.e
f(
x,
y)
ca
n
only
ta
ke
0
or
1
val
ue
)
or
im
age
reg
io
n
s,
def
i
ne
d
by
cl
os
e
d
c
o
ntours
a
re
co
nsi
der
e
d.
I
nvari
ance
to
tra
ns
la
ti
on
an
d
scal
ing
ca
n be
com
pu
te
d from
m
o
m
ents (
2) a
s foll
ow
s:
(
2
)
/
2
00
pq
pq
pq
m
(3)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020 :
30
22
-
3034
3026
Fo
r
ce
ntered m
om
ents µ
pq
is
def
ine
d by:
(
)
(
)
(
,
)
pq
p
q
g
g
R
x
x
y
y
f
x
y
d
x
d
y
(4)
w
he
re
:
1
0
0
1
0
0
0
0
gg
mm
x
a
n
d
y
mm
(5)
x
g
and
y
g
rep
re
sent the co
or
din
at
es o
f
the gr
avity
center,
an
d
m
00
=a=
n
is t
he
obj
ect
area.
T
he
m
a
in o
rie
ntati
on
is o
btained
from
the seco
nd
orde
r
ce
ntere
d m
o
m
ents
[37]
:
11
2
0
0
2
2
1
a
r
c
t
a
n
2
(6)
Fr
om
ell
ipse param
et
ers,
w
e
can e
xpress
the
centere
d
m
ome
nts
o
f o
rd
e
r 2,
w
e
ha
ve:
2
2
2
2
2
0
0
0
1
2
2
2
2
2
0
2
0
0
1
2
2
2
2
1
1
0
0
1
2
(
(
)
/
4
(
1
)
)
(
(
)
/
4
(
1
)
)
(
(
)
/
4
(
1
)
)
m
a
a
t
t
m
a
t
a
t
m
t
a
a
t
(7)
3.2.
N
onhol
onomic
m
ob
il
e
robot
The
un
ic
yc
le
m
ob
il
e
ro
bot
a
s
show
n
i
n
Fi
gure
2
is
ty
pical
ly
a
nonholon
om
ic
syst
e
m
[3
8]
an
d
ca
n
be
descr
i
bed
by
t
he
nonhol
onom
ic
con
strai
nt
s.
T
her
e
are
t
wo
ty
pes
of
c
on
st
raints:
the
ro
ll
in
g
a
nd
the
sli
ding
const
raint
s
.
Figure
2
.
U
nic
yc
le
m
ob
il
e robo
t
The
sta
te
ve
ct
or
q
=(
x,y,
)
de
no
te
s
t
he
po
s
ture
of
t
he
r
obot.
(
x,y
)
repr
esents
the
m
ass
center
of
the
r
obot.
is
t
he
or
ie
ntati
on
of
t
he
rob
ot
ac
cordin
g
t
o
the
horizo
ntal
axis
.
ν
(
t
)
an
d
ω
(
t
)
are
t
he
tra
ns
l
at
ion
vel
ocity
and
t
he
an
gu
la
r
vel
ocity
,
wh
ic
h
a
re
us
e
d
as
the
con
tr
ol
inputs
.
The
no
nholono
m
ic
con
strai
nt
for
the m
ob
il
e rob
ots is
giv
e
n by:
(
)
s
i
n
(
)
(
)
c
o
s
(
)
(
)
0
x
t
y
t
t
(8)
A
gen
e
ral
k
i
nem
at
ic
m
od
el
of
t
he
m
ob
il
e
rob
ot
obta
ine
d
from
the
nonho
l
onom
ic
const
raints
is
gi
ve
n
by
the foll
owin
g
e
qu
at
io
n:
(
)
c
o
s
(
)
0
()
(
)
s
i
n
(
)
0
()
(
)
0
1
x
t
t
t
y
t
t
t
t
(9)
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Eff
ic
ie
nt a
nd s
ecure re
al
-
ti
me
mob
il
e r
obots
cooper
atio
n us
ing
vis
ual serv
oing
(
So
um
i
a
Bo
ud
r
a
)
3027
3.3
.
C
ontr
ol law
The
ai
m
of
the
con
tr
ol
schem
e
is
to
eli
m
inate
the
err
ors
bet
ween
the
init
ia
l
and
the
desir
ed
posit
io
ns
of f
eat
ures
on t
he
im
age p
la
ne
.
T
he
e
xpressio
n of t
hese e
rro
r
s is d
e
fine
d
as:
(10)
s
is
the
vecto
r of vis
ual f
eat
ure. Mo
del
of
t
he
contr
oller is
ve
locit
y con
tr
oller a
nd it
can be
co
m
pu
te
d
a
s:
*
ˆ
()
s
L
s
s
(11)
w
he
re
re
pr
e
sents
the
veloc
it
y
vector
of
t
he
cam
era
including
tra
ns
la
ti
onal
an
d
ro
ta
ti
onal
com
ponen
t
s
an
d
is pse
udo
-
in
verse o
f
the
app
roxim
a
ti
on
of im
age Jac
obia
n m
at
rix
.
3.
4
.
In
tera
c
ti
on
m
atri
x
We
ass
um
e
that
the
obj
ect
i
s
a
co
ntin
uous
su
r
face,
t
her
e
fore,
t
he
de
pt
h
Z
of
each
3D
point
is
expresse
d
as
a
functi
on
of
t
he
x
a
nd
y
c
oor
din
at
es
of
it
s
pro
j
ect
ion i
n t
he
i
m
age.
Mo
re
s
pe
ci
fical
ly
, w
e ha
ve
:
0
,
0
1
pq
pq
pq
A
x
y
Z
(12)
In the case
wh
ere the
ob
j
ect
is p
la
nar
or
has a
surf
ace
of the
p
la
na
r
li
m
bs
, its eq
uation i
n
t
he
cam
era
fr
am
e is expres
sed by:
1
2
0
Z
X
Y
Z
(13)
us
in
g
t
he
e
qu
at
ion
s
of the
pe
rs
pecti
ve pr
oject
ion
,
XY
xy
ZZ
(14)
we
ca
n deduce
:
1
A
x
B
y
C
Z
(15)
with:
12
0
0
0
1
,,
A
B
C
Z
Z
Z
(16)
for
eac
h po
i
nt
with c
oor
din
at
es
x
=(
x
,
y
)
in
t
he
i
m
age whose
corres
pondin
g 3
d po
i
nt h
a
s
de
pth
Z
,
we ha
ve
:
x
x
L
v
(17)
wh
e
re,
the i
nteracti
on m
at
rix
L
x
is give
n by:
2
2
1
/
0
/
(
1
)
0
1
/
/
1
x
Z
x
Z
x
y
x
y
L
Z
y
Z
y
x
y
x
(18)
us
in
g (15
)
in
(18),
(17) ca
n b
e wri
tt
en
as:
2
2
(
)
.
(
)
.
.
(
1
)
.
.
(
)
.
(
)
.
(
1
)
.
.
.
xz
x
y
z
yz
x
y
z
x
A
x
B
y
C
v
x
A
x
B
y
C
v
x
y
w
x
w
y
w
y
A
x
B
y
C
v
y
A
x
B
y
C
v
y
w
x
y
w
x
w
(19)
*
(
)
(
)
e
t
s
t
s
s
L
ˆ
s
L
ˆ
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020 :
30
22
-
3034
3028
In
t
he
m
at
rix
Lx,
any
c
on
t
rol
schem
e
that
us
es
t
his
f
or
m
of
t
he
inte
racti
on
m
at
rix
m
us
t
est
i
m
at
e
or
appr
ox
im
at
e the v
al
ue
of Z
.
F
ro
m
this equat
ion,
we ob
ta
in:
(
2
)
2
(
2
)
2
x
z
x
y
y
y
z
x
y
x
A
v
A
x
B
y
C
v
y
w
x
w
x
y
B
v
A
x
B
C
v
y
w
x
w
y
(20)
we have
:
1
1
ij
ij
f
i
x
y
x
f
j
x
y
y
(21)
T
he
i
nteracti
on
m
a
trix ass
ociat
ed
with
the
m
om
ent
m
ij
can
then be
d
et
e
rm
i
ned
by
:
m
i
j
v
x
v
y
v
z
w
x
w
y
w
z
L
m
m
m
m
m
m
(22)
w
he
re:
1
,
1
1
,
1
,
1
,
1
1
,
,
1
,
1
,
1
1
,
1
,
1
,
1
1
,
1
()
()
(
3
)
(
)
(
3
)
(
3
)
v
x
i
j
i
j
i
j
i
j
v
y
i
j
i
j
i
j
i
j
v
z
i
j
i
j
i
j
i
j
w
x
i
j
i
j
w
y
i
j
i
j
w
z
i
j
i
j
m
i
A
m
B
m
C
m
A
m
m
j
A
m
B
m
C
m
B
m
m
i
j
A
m
B
m
C
m
C
m
m
i
j
m
j
m
m
i
j
m
i
m
m
i
m
j
m
(23)
Fr
om
the
ge
ne
ral
form
giv
en
in
(
22),
we
de
du
ce
t
he
inter
a
ct
ion
m
at
rix
of
the
surface
of
the
ob
j
ect
m
00
(
we
c
onsid
er
i=j
=
0),
a
nd
the
intera
ct
ion
m
at
rix
relat
ed
to
t
he
c
oor
dina
te
s
x
g
a
nd
y
g
of
the
gravit
y
center
of
a
n
ob
j
ect
in
the
i
m
age.
The
visu
al
se
rvoin
g
al
go
rit
hm
that
we
prese
nt
bel
ow
is
base
d
on
the
ideas
pr
opos
e
d
by [1
4
,
23
,
35]
with m
od
ific
at
ion
s
m
ade f
or
our
a
ppli
cat
ion
.
Algori
th
m
Be
gin
1:
Set t
he
p
a
ra
m
et
ers
us
e
d;//
λ=0.6
2:
Co
nn
ect
t
o
t
he
m
ast
er r
ob
ot
an
d C
li
ent 1 a
nd Cl
ie
nt 2
;
3:
Determ
ine c
a
m
era p
ar
am
eter
s
(cali
br
at
io
n);
4:
Ac
quirin
g
a
n
im
age;
5:
Im
age d
is
play
;
6:
Creat
e the
s
urface to
foll
ow (
t
he
ta
r
get);
7:
Ca
lc
ulate
th
e m
o
m
ents in
the im
age;
8:
Determ
ine the c
oor
din
a
te
s
of the
desire
d
t
arg
et
t
o
f
ollo
w (f
or each
robot
);
9:
Ca
lc
ulate
th
e interact
io
n
m
at
rix
L
x
=S
xd
;
10:
Determ
ine (
Z/Z*
);
//
Z*: t
he desire
d dep
t
h of t
he fo
rm
i
s learne
d
a
nd e
qu
al
t
o
the
init
ia
l dep
th;
11:
Ca
lc
ulate
//
Tran
s
form
at
io
n of t
he
cam
era f
ra
m
e to the
eff
ect
or
of the
m
ob
il
e ro
bot,
this tran
sf
or
m
at
ion
m
akes it p
os
sible t
o
cal
c
ulate
the
veloci
ti
es expresse
d
i
n
the
end
eff
ect
or
fr
am
e into th
e
cam
era f
ram
e;
12:
F
or (
i=
0
;i
<=n
; i
++)//
n
:
is t
he nu
m
ber
of im
ages in
th
e v
ide
o
13:
F
or (x=
1; x
<=y
;
x++)
//
y:
is the num
ber o
f
se
rv
e
r robo
ts
Be
gin
Be
gin
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Eff
ic
ie
nt a
nd s
ecure re
al
-
ti
me
mob
il
e r
obots
cooper
atio
n us
ing
vis
ual serv
oing
(
So
um
i
a
Bo
ud
r
a
)
3029
14:
A
cq
ui
rin
g
a
n
im
age;
15:
Ca
lc
ul
at
e the ar
ea t
o fo
ll
ow; (cal
c
ul
at
ing
the
m
o
m
ents in
the im
a
ge)
16:
U
pdat
e the c
urren
t
x
featur
e
s;
18:
U
pdat
e the inte
racti
on m
at
rix;
19:
U
pdat
e the
global er
r
or(s
-
s
*);
20
:
Ca
lc
ul
at
e fo
r
eac
h
se
rv
e
r rob
ot the c
on
t
ro
l l
a
w: V
=
-
λ*(L
*
c
V
e
*
e
J
e
)
+
*error(s
-
s
*)
;
2
1:
Se
nd t
he veloci
ti
es to
the
rob
ot;
E
nd F
or
E
nd F
or
End
.
4.
E
X
PERI
MEN
TAL
RES
ULTS
The
e
xper
i
m
ents
wer
e
p
e
rfo
r
m
ed
with
tw
o
rob
ots
Pio
nee
r
3
-
AT
an
d
Pio
neer
3
-
D
X
e
quip
ped
with
a
cam
era
.
The
i
m
age
reso
l
ution
in
th
e
ex
pe
r
i
m
ents
was
640x480.
All
co
m
pu
ta
ti
on
s,
e
xc
ept
f
or
the
lo
w
-
le
vel
con
t
ro
l,
wer
e
perform
ed
on
a
la
pto
p
with
2.5
G
Hz
In
te
l
Core
(T
M)
i
5
-
4200U
CPU
,
with
4
-
GB
RAM.
The
na
viga
ti
on
ex
per
im
ent
was
pe
rfo
rm
e
d
onli
ne.
O
ur
syst
e
m
is
i
m
p
l
e
m
ented
unde
r
Vis
ual
S
tud
i
o
with
the
hel
p
of
A
r
ia
and
O
pe
nC
V
li
brary.
H
oweve
r
,
the
r
obot
re
portin
g
sy
stem
is
i
m
ple
m
ented
wit
h
P
yt
ho
n
3.
The
ex
pe
rim
en
ts
wer
e
pe
rfo
r
m
ed
in
an
in
door
en
vir
onm
e
nt
inside
a
la
borat
or
y
an
d
a
corrid
or
with
λ
=
0.6.
The
pionee
r
3DX
an
d
3AT
m
ob
il
es
ro
bots
are
unic
yc
le
s
non
-
ho
l
onom
i
cs
with
dif
fe
r
entia
l
dr
i
ve
W
MR
.
P3
-
D
X
has
2
ind
e
pende
nt
dri
ving
w
heels,
t
hat
ca
n
n
o’
t
be
ste
ere
d
on
the
sam
e
axis
an
d
fr
ee
sw
ing
i
ng
off
-
ce
nter
w
he
el
.
The
w
ho
le
syst
e
m
su
m
m
a
rize
in
F
ig
ure
3.
W
e
dev
el
ope
a
serv
er
ap
pl
ic
at
ion
us
in
g
the
C+
+
pro
gr
am
m
ing
la
nguag
e
wh
ic
h
is
instal
le
d
in
the
robo
t
PC.
It
per
f
or
m
s
the
sever
a
l
fu
nctio
ns
,
m
ai
nly:
m
anag
in
g
the
com
m
un
ic
at
ion
to
and
from
the
m
edial
serv
e
r
by
sen
din
g
in
form
at
io
n
ab
ou
t
the
sta
te
of
the
r
obot
a
nd
receivin
g
in
str
uctions
f
ro
m
the
use
r
,
c
ontr
ol
li
ng
t
he
m
ob
il
e
ro
bot
m
ov
e
m
ents
and
m
anag
i
ng
the m
ob
il
e rob
ot se
ns
ors.
The
de
velo
pe
d
sys
tem
con
s
ist
s
of
acci
de
nt
repor
ti
ng
s
yst
e
m
.
Fo
rthe
reali
zat
ion
of
our
syst
e
m
,
we use
d
the
foll
ow
in
g
e
quipm
ent num
ber
ed
as sho
wn in Fi
gure
3
a
nd F
i
gure
4
:
(1)
P
ower
sup
pl
y 5V
.
(2)
P
ushbutt
on
.
(3)
Accele
r
ome
te
r
MPU
6050
.
(4)
GP
S
m
od
ul
e GY
-
NE
O6
M
.
(5)
Se
nsors
(
S
ho
c
k, flam
e).
(6)
F
TD
I
c
onve
rter.
(7)
A
Ra
s
pb
e
rry
Pi 3
ca
r
d.
(8)
S
IM8
08
G
SM m
od
ule.
(9)
P
ower
sup
pl
y 12
V.
Figure
3
.
Gl
obal
v
ie
w
of the
pro
po
se
d
s
che
m
e
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020 :
30
22
-
3034
3030
Ba
sed
on
the
RPi
3
wh
ic
h
i
s
us
ed
to
c
omm
un
ic
at
e
with
a
GS
M
and
a
glo
bal
posit
io
ning
syst
e
m
(
GPS
)
m
od
ule
us
in
g
the
U
A
RT
li
nk
,
di
ff
e
r
ent
sens
or
s
we
re
us
e
d
to
e
nsure
the
detect
ion
of
th
e
acci
den
t.
The
G
PS
prov
i
des
a
three
di
m
ension
al
posi
ti
on
as
well
as
current
ti
m
e
and
date
an
d
is
a
vaila
ble
eve
ry
wh
e
re.
[39
,
40
]
Using
GP
S
,
G
PRS
a
nd
GS
M
Tech
no
l
og
y
for
si
m
ple
acci
de
nt
de
te
ct
ion
but
in
our
case
,
our
syst
e
m
pro
po
se
d
A
D
RS
(
Accide
nt
Detect
ion
Re
portin
g
Syst
e
m
)
al
lowing
t
he
rob
ots
na
vig
at
ing
s
afely
,thin
g
that
rem
ai
ns
a
chall
eng
in
g
ta
sk
i
n
m
any
research
w
orks
.
Th
e
dev
el
op
e
d
s
yst
e
m
was
at
t
ached
t
o
the
t
op
of
th
e
r
obots
i
n
or
der
t
o
detect
a
cci
den
t
in
the
re
al
tim
e
,
reco
r
d
the
accu
rate
lo
cat
ion
of
t
he
a
cci
den
t
a
nd
s
e
nd
an
autom
at
ic
alert
m
essage to
the
em
erg
ency ce
nter
[41]
.
Figure
4
.
Ro
bot acc
ident
dete
ct
ion
repor
ti
ng syst
e
m
test
bed (ADRS
)
The
syst
em
was
powe
rin
g
up
an
d
dif
fe
ren
t
sens
ors
wer
e
te
ste
d.
Each
of
sens
or
com
m
and
s
the
Ra
spbe
rr
y
Pi3
to
gen
e
rat
e
a
s
pecific
S
MS
al
ert
with
the
ty
pe
of
the
acci
de
nt.
T
he
te
stbed
detect
s
tw
o
cases
of
re
ve
r
sal
:
Ri
gh
t
turno
ve
r,
Le
ft
turnov
e
r
as
show
n
in
F
ig
ure
5.
We
us
e
d
the
a
ccel
ero
m
et
er
data
to
cal
culat
e
the
incli
natio
n
of
the
ro
bot
al
ong
th
e
X
axis.
The
value
>
=
46
°
ind
ic
at
es
that
th
e
ro
bot
is
ov
e
rturne
d
on the
rig
ht.
If
the incli
natio
n al
ong
the
X axi
s is <=
-
70°
, thi
s i
m
plies t
hat the
rob
ot is ove
rturne
d
to
the
left.
We
us
e
d
a
li
ghte
r
to
li
ght
fire,
the
flam
e
sens
or
w
as
che
cked
us
in
g
norm
l
li
gh
te
r
an
d
the
se
ns
or
su
cces
fu
ll
dete
ct
ed
and
repo
r
te
d
the
fire
to
the
Ra
sp
be
rr
y
Pi3
Fig
ur
e
6.
The
la
tt
er
will
gen
e
rate
a
war
ni
ng
m
essage
in
dicat
ing
that
a
fire
has
occurre
d.
The
s
hock
sen
so
r
was
te
ste
d
by
creati
ng
sm
al
l
ob
sta
cl
e
to
report
the
colli
sio
n
(
Figure
7
)
.
T
he
sens
or
s
w
he
r
e
able
to
dete
ct
and
re
port
the
prob
le
m
s
in
ti
m
e
no
t
ex
ceed
ing
20
seco
nds
.
T
he
rob
ot
m
ast
er
P3
-
A
T
ca
rr
ie
s
a
plana
r
ob
j
ect
(
F
ig
ur
e
8
)
.
Aft
er
the
bl
ob
de
te
ct
ion
,
the
c
onto
ur
of
the
intere
ste
d
reg
i
on
is
ob
t
ai
ned
to
f
ur
t
he
r
acq
uire
the
centr
oid
,
wh
ic
h
will
be
con
du
ct
ed
in
each
con
t
ro
l
loop.
T
he
m
o
m
ents
are
c
om
pu
te
d
at
the
vide
o
rate
a
fter
bin
arizat
io
n
of
t
he
im
age,
the
resu
lt
s
ar
e
sho
wn
a
s
F
ig
ure
9,
we
s
uc
cess
fu
ll
y
detect
the
con
to
ur
.
The
im
age
ac
qu
i
red
at
the
de
sired
cam
er
a
po
sit
io
n
is
dis
play
e
d
in
Fig
ur
e
10
.
The
r
obot
tra
je
ct
or
y
is
sat
isf
act
or
y
us
i
ng
the
vis
uals
feat
ur
es
,
in
our
c
ontrib
utio
n
we
hav
e
t
o
est
i
m
at
e the d
epth
Z
for
eac
h
i
te
rati
on
.
Figure
5
.
Ri
gh
t
turn
ov
e
r
te
st, l
eft turn
ov
e
r
te
s
t
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Eff
ic
ie
nt a
nd s
ecure re
al
-
ti
me
mob
il
e r
obots
cooper
atio
n us
ing
vis
ual serv
oing
(
So
um
i
a
Bo
ud
r
a
)
3031
Figure
6
.
Flam
e senso
r
te
st
Figure
7
.
S
ho
c
k
se
nsor
te
st
Figure
8
.
N
on
sy
m
m
e
tric
al
p
la
nar o
bject
tra
ckin
g
Figure
9. Re
su
l
ts of detect
in
g t
he
bl
ob cont
our
Figure
10. Des
ired
im
age
The
ef
fecti
ve
ne
s
s
of
t
he
de
ve
lop
e
ds
yst
em
d
epends
on
t
he
pr
eci
sio
n
of
m
at
ching
betwee
n
the
cu
rrent
and
de
sired
posit
ion
of
the
vi
su
al
in
form
at
io
n,
a
s
s
how
n
in
F
igure
11(a
)
a
nd
F
ig
ure
12(a
)
wh
ic
h
c
onfir
m
that
the
syst
em
is
sta
ble
an
d
co
nv
erg
es
to
the
de
sired
val
ues
.T
he
cal
culat
ed
er
ror
betwee
n
th
e
cur
re
nt
inf
orm
at
ion
and
the
de
sire
d
i
nfor
m
at
ion
(fo
r
x
a
nd
y
)
is
relat
ively
unc
hangin
g
duri
ng
the
first
3m
s,
this
is
du
e
to
the
non
-
holo
no
m
ic
natur
e
of
t
he
r
obots
.
H
oweve
r,
at
32
-
37m
s
the
syst
e
m
con
ve
rg
e
d
to
the
rig
ht
p
os
it
ion
an
d
the
fi
nal
er
ror
was
0.0
1.
In
this
stu
dy,
we
discusse
d
t
he
m
ot
ion
c
on
t
ro
l
pro
blem
and
visua
l
ser
voing
of
non
-
holo
nom
ic
wh
eel
e
d
m
ob
il
es
ro
bots.
We
pr
ese
nted
kinem
at
ic
m
od
el
of
th
e
r
obot
in
ad
diti
on
t
o
their
con
t
ro
l
pr
op
e
rt
ie
s.
The
sim
ul
at
ion
res
ult
s
sh
ow
that
the
de
velo
ped
c
ontr
ol
la
w
co
ntr
ols
the
velocit
ie
s
of
al
l
us
e
d
r
obots
i
n
a
rob
us
t
an
d
preci
se
way
.
T
he
de
velo
ped
c
on
t
ro
ll
er
is
a
bl
e
to
dri
ve
a
se
t
of
wheel
ed
m
ob
il
e
rob
ots
to
a
ta
rget
obj
ect
un
ti
l
the
cam
eras
ob
s
erv
e
the
desire
d
vis
ual
char
ac
te
r
ist
ic
s
as
sh
own
in
Fig
ur
e
11(
b
)
-
Figure
12(
b)
.
Evaluation Warning : The document was created with Spire.PDF for Python.