A low cost 3D vision system for positioning welding mobile robots using a FPGA prototyping system
Abstract
The aim of this work is to explore some solutions for artificial vision systems applied to welding autonomous robots. We take advantage from the UNSHADES-1 system, developed in Filiation (blind for evaluation). This system can be used for building a powerful 3D vision system. The system concentrates and process the three images in parallel, producing an accurate value of the position, that can be improved if the relative position of the cameras are well defined.
Full text
A Low Cos 3D Vision Sys em o Posi ioning Welding Mobile Robo s
using a FPGA p o oyping sys em
I. Ibáñez, M.A. Agui e, A. To alba and L.G. F anquelo
Escuela Supe io de Ingenie os. Uni e sidad de Se illa
A da. Camino de los Descub imien os s/n,
41092 Se illa (SPAIN)
[email p o ec ed], o[email p o ec ed], leopold[email p o ec ed]
Abs ac – The aim o his wo k is o explo e some solu ions o
a i icial ision sys ems applied o welding au onomous obo s.
We ake ad an age om he UNSHADES-1 sys em, de eloped
in Filia ion (blind o e alua ion). This sys em can be used o
building a powe ul 3D ision sys em. The sys em concen a es
and p ocess he h ee images in pa allel, p oducing an accu a e
alue o he posi ion, ha can be imp o ed i he ela i e
posi ion o he came as a e well de ined.
I. INTRODUCTION
Au onomous Robo s a e comple e sys ems ha ope a e
inside complex en i onmen s wi hou being guided and
con olled by human ope a o s. They can es ablish
ela ionships, ake decisions in un- ime and adap hei
ac ion plans o di e en ex e nal ci cums ances. The ecei ed
in o ma ion make hem au o-p og ammables [1-2], al e ing
hei p og am in o de o sol e he ac ual si ua ion. Those
sys em ha e some AI in hei beha iou . Mos o cases, he
equi ed in o ma ions a e ela ed o posi ion, speed,
accele a ion, s eng h, o que, sizes and shapes om objec s
and empe a u e. Quan i ica ion o hose measu es a e made
by means o mechanical, op ical, e mical, elec ical,
ul asounds, ... senso s. Howe e , ad anced esea ch
specially akes in o accoun a i icial ision ansduce s. This
esea ch can ake ad an age om he newes mic oelec onic
sys ems, whe e design schedule ha e been sho ened using
he ad anced FPGA’s. These de ices can p ocess images in
“ eal ime”, p ocessing ame by ame on he ly.
Be ween he a i icial ision based solu ions, he mos
a ac i e ones o Robo ics applica ions a e objec
ecogni ion and classi ica ion, ensambling, welding, s ee ing
and synch onisa ion wi h o he obo s.
One o he mos complex applica ion is welding sys ems
based on au onomous obo s. An eme ging echnology ha
concen a es s a egies based on me alu gy, obo ics,
elec ical and elec onic enginee ing and mic oelec onics.
Welding obo s a e widesp ead demanded sys ems, specially
whe e he indus y has a high le el o au oma ion.
This pape ela es a new cus om sys em o de e mining
he exac posi ion o he welding poin . The me hod is based
on a sys em wi h h ee low cos came as ha a e placed and
ixed in he h ee space axes, and compu es he posi ion
h ough he simul aneous p ocessing o he images aken by
each came a. Each image is bina ised and he welding poin
is ob ained using a p og ammable bina y h eshold. So, o
each came a, he whole image is educed o a single poin . As
he h ee images a e p ocessed simul aneously, e e y
coo dina e is compu ed as pa o wo o hem and hen he
inal coo dina e is gi en as an imp o ed p ecision numbe .
Using a gene al pu pose sys em based on FPGA’s called
UNSHADES [4] we ha e de eloped and es ed he comple e
sys em. Typically his sys em would be made wi h a DSP
sys em ha would spend many ime p ocessing e e y image
in a se ial way and wouldn’ ake ad an age om he
pa allelism in e ms o accu acy.
This pape shows how o ake ad an age om he
UNSHADES-1 sys em. We ha e de eloped a ull con olled
sys em ha compu es he posi ion o he welding poin . In
sec ion II we’ll desc ibe he UNSHADES amewo k, in
sec ion III he Vision sys em is p esen ed, sec ion IV shows
he sys em design and inally he conclusions will be
p esen ed.
II. UNSHADES FRAMEWORK
A Ha dwa e Debugge is a comple e p og ammable
sys em ha can be cus omized o di e en applica ions.
Fig 1. UNSHADES-1 sys em
Hos Con olle
XC4010XL
In e ace &
Con olle
EPP
Da alink
P og ammable
Oscilla o
0 o 100MHz
VIRTEX De ice
XCV50 o XCV800
Low Speed Con ol
Signals
Gene al pu pose Signals
o Sys em In e ace
Main ad an age is ha he comple e sys em can be
moni o ized du ing execu ion ime. Main p ocesso s a e
subs i u ed by an ad anced FPGA and un embedded in o he
sys em i sel . I needed a o ally cus omized sys em can be
buil using he ac ual ne lis .
UNSHADES-1 gene al pu pose amewo k ha is based
on one Xilinx ad anced FPGA o emula ing he cus om
p ocesso o , in his case, image p ocessing. Main ad an age
o UNSHADES sys em is ha he use ha e ull con ol o
he execu ion o he FPGA due o he special cha ac e is ics
o his de ice. UNSHADES means Uni e sidad de Se illa
Ha dwa e Debugge Sys em, and allows a apid p o o yping
and he debugging o he p ocesso ci cui du ing execu ion.
Figu e 1 shows a gene al scheme o he sys em:
Main componen s o UNSHADES sys em a e lis ed below:
- An ad anced FPGA om Xilinx Vi ex amily [3].
This FPGA hos s he cus om design.
- PC so wa e, he human in e ace, ha p o ides a
comp ehensi e way o manipula ing he inne
in o ma ion o he FPGA.
- A p og ammable econoscilla o om Dallas
Semiconduc o DS-1075, ha gene a es he sys em
clock. This de ice can be uned om he PC.
- The e’s ano he FPGA ac ing as he in e ace
be ween he PC and he ad anced FPGA. All he
communica ions, p o ocol adap ing and signal
gene a ion a e suppo ed by de ice, which is smalle
han ad anced FPGA an has ixed i s p og am.
Communica ion be ween PC and UNSHADES ha dwa e is
pe o med using he EPP p o ocol ha can each 2 Mby es/s,
ha is speed enough o ou pu pose.
UNSHADES so wa e is a comple e sys em ha allows a
bi s eam managemen including some in e es ing ools:
downloade , in e nal egis e s inspec o , in e nal egis e
modi ie , oscilla o manage and design selec o , all ully
in eg a ed wi hin he Xilinx s anda d design low. Fo an
op imal speed in he bi s eam ans e , all he w i e and ead
ope a ions a e pe o med using pa ial econ igu a ion
echniques.
III. POSITIONING SYSTEM DESCRIPTION
The aim o his wo k is o build a simple, low cos and
e ec i e 3D digi al ision sys em o acking he posi ion
and p o ide a eedback in posi ioning au onomous obo s.
Sys ems like he one p esen ed can enla ge he e ec i e li e
o he au onomous obo s.
Main d awbacks o cons uc ing hese sys ems a e [5]:
1s . Vision sys ems a e as expensi e as he obo i sel .
2nd. In his kind o obo ics sys ems high h oughpu
echniques ha could gene a e he in o ma ion in eal- ime
a e needed.
3 d. The e’ e many p oblems ela ed wi h isual in o ma ion,
like shadows, illumina ions, uni, bi and h ee-dimensional,
hidden objec s,… e c.
The main objec i e is o o e come mos o hese di icul ies:
1s We p o ide wi h low cos CCD came as and ideo
decode s, he cos o he sys em s ongly depends on he
equi ed p ecision. Fo ou p o o ype we used small digi al
came as.
2nd The p ocessing echniques do no need o p ocess he
whole image o each came a. A simpli ied me hod allows a
esou ce sa ing ha makes a de ice like Vi ex V300 big
enough o he h ee came as. They a e p ocessed in pa allel
and he in o ma ion gene a ed simul aneously, so he esul is
p oduced using he same image ame aken in he same
ins an .
3 d All he imaging p oblems a e sol ed due o he use o
h ee came as placed a di e en posi ions. T inocula s e eo
ision inc eases he geome ic cons ain s and educes he
in luence o heu is ic cons ain s o s e eo ma ching [6-7].
Mos o cases we’ll ha e edundan in o ma ion, bu hidden
objec s will be o one came a, while he o he s will con inue
p oducing he in o ma ion. When he in o ma ion is
edundan , some p ecision imp o emen echniques can be
applied and be e weld acking will be ob ained.
4 h Ano he imp o emen comes om he image dis o ion
due o he lenses ha a e commonly cu e. Fo i s
compensa ion, a dis o ion ac o is in oduced o modi ying
he ypical linea model o he came a. The pa ame e s o
he compensa ion a e measu ed using pho og ame ic
me hods o calib a ion [8-9 ].
Ano he ad an ages a e in oduced when compa ed wi h
o he welding sys ems: The e’s no need o p oduce s imuli
o in e ac ion measu emen . Ou sys em can be conside ed a
passi e sys em, due o he e’s no need emi e s, because he
welding poin ligh ing is he signal aken o measu emen s.
The e’ e ano he echniques o p o iding he in o ma ion
ha ha e been ejec ed o ou pu pose: ul asound a e slow
and a objec s a e bad measu ed wi h many well known
p oblems. X ays need o use be ween he objec and he
senso an ampli ie o ans o ming X ays in o isible
spec um.
Finally, in a ed ha e many p oblems ela ed wi h
abso p ion, many dis o ions coming om he su ounding
emi e s.
In ou sys em he h ee images a e mixed, compensa ed,
s o ed and p ocessed in he same de ice.
IV. PROTOTYPE DESCRIPTION
Ou p o o ype has been designed and implemen ed in he
ela ed FPGA. Ou ex e nal ci cui only needs he ideo
acqui ing sys ems: TVCCD-30M Came as and he digi ize s.
Inside he FPGA, he compensa ions sys em and coo dina e
compu ing blocks a e p og ammed.
Figu e 2 p esen s a block diag am o he PCB boa d. I ’s
compound by he h ee digi ize s, he Phillips SAA7113H,
and ano he communica ion ea u es, like a RS232 line, a
VGA ou pu and a pa allel po .
came a 1
came a 2
came a 3
Video
Decode 3
F
P
G
A
In
e
a
ce
Pa allel
Po Se ial
Po
CDA
connec o
VGA
3D ision boa d
Video
Moni o
Compu e
UNSHADES
Sys em
Boa d
Video
Decode 2
Video
Decode 1
Fig. 2. Main PCB blocks.
Inside he FPGA, a ull digi al ci cui p ocess e e y image.
Figu e 3 shows he block diag am o he p ocesso . The i s
s age is o compensa e he lense dis o ion coming om he
came a lenses. Secondly a bina isa ion o he image sea ches
he welding poin , ead like a small a ea ob ained ha is a
small b igh ci cle. A e , his pa is s o ed in a memo y o
pos p ocessing.
P e iously a ci cui con igu es he digi ize s o an
adequa e image o ma .
The mos impo an block is he image p ocesso . The i s
s ep o he image p ocessing consis o inding he app opia e
h eshold ha would lead o a bina ised ound shape
co esponding o he welding poin p esen in he image.
E e y pixel is compa ed wi h he p e iously compu ed
h eshold, ha is ob ained om he image condi ion i sel .
We ha e selec ed he h eshold alue be ween he mean o
he luminance o he image and i s maximum alue. I he
luminance is below his alue he pixel has 0 alue (black)
and 1 alue (whi e) in ano he case. The line and pixel
numbe o whi e alues a e s o ed in he memo y.
Due o he ixed posi ion o e e y came a e e y decode
can p oduce wo alues (x,y), (y,z) o (z,y). The ideo
decode s p o ide in o ma ion wi h 625 lines and 1440 pixels
pe line. This in o ma ion is use ul o gene a ing e e y
coo dina e using simple coun e s.
Video
acqui e sys em
Dis o ion
compens
i
Dis o ion
compens
Dis o ion
compens
Bina isa ion Bina isa ion Bina isa ion
Memo y Memo yMemo y
P ocesso
Con igu e and
synch onise
decode s
Main
con ol
In e ace
VGA
In e ace
(X,Y,Z) (X,Y,Z)
Robo
Con ol Sys em
di i li ió
Main Blocks in
FPGA
PCB
PCB
Fig. 3. Inside FPGA p ocessing blocks.
The exac cen e o he welding poin is compu ed by
means o a g a i y cen e o each coo dina e, using hese
o mulas:
We’ll ob ain wo numbe s o each coo dina e because
hey’ e epea ed in wo memo ies. Some p ecision
imp o emen s can be p oduced i he came as a e placed
wi h a hal pixel o de ia ion o e e y coo dina e.
∑
∑
=
l
li
l
li
x
x
X
1
2
1
1∑
∑
=
m
mi
m
mi
y
y
Y
1
2
1
1∑
∑
=
n
ni
n
ni
z
z
Z
2
2
2
2
∑
∑
=
q
qi
q
qi
y
y
Y
2
2
2
2∑
∑
=
i
i
z
z
Z
3
2
3
3
∑
∑
=
p
pi
p
pi
x
x
X
3
2
3
3
2
31 XX
X+
=2
21 YY
Y+
=2
32 ZZ
Z+
=
The esul s a e ans e ed o he PC ha p o ides he
in o ma ion o he con olle s. The implemen a ion o his
sys em e esh he coo dina es e e y wo ames, due o he
pos -p ocessing da apa h. The comple e sys em has been
implemen ed in o a Vi ex 300, wi h a 60% o he
occupa ion. Sys em clock uns a 50MHz.
V. CONCLUSIONS
The pape p esen s a simple sys em o de e mine he
absolu e posi ion o a welding poin o an au onomous
obo s. Main ad an ages a e he cos , he p ecision and he
compu ing speed. In he inal e sion o his pape he
compu ing ime will be gi en. I ’s been es ima ed like a
coo dina e compu ing pe wo images, which can be
conside ed in e ms o obo speed eal- ime. This sys em
akes ad an age o he UNSHADES-1 sys em ha p o ides a
comple e, lexible and e y com o able pla o m o he
ision sys em.
V. REFERENCES
[1] D. Rdine Hundsdö e “Cos u as pe ec as. Aumen a el
empleo de la soldadu a po láse en el mecanizado de
chapa”. Re is a de Soldadu a y Tecnología de unión.
Año XII- Num 72, pp. 9-11 No iemb e/Dic 2001.
[2] P. Coi e , M. Chi ouze, “An In oduc ion o Robo
Technology”, Mc G aw Hill, 1983.
[3] Xilinx Da a Book. 2002
[4] M.A. Agui e, J. Tombs, A. To alba, L.G. F anquelo
“HADES-1: A apid p o o yping en i onmen based on
aad ances FPGA´s”. P oceedings o he Design o
Ci cui s and In eg a ed Sys ems, DCSI’01 Opo o 2001.
[5] J. M. Angulo Usa egui “Robó ica P ác ica. Tecnología y
aplicaciones”. Capí ulo 7: “Visión de Máquinas”,
Pa anin o, 5.ª ed., 2000.
[6] Ayache, “T inocula S e eo Vision o Robo ics”,IEEE
T ans. Pa e n Anal. & Machine In elligence, ol. 13,
no. 1, Jan. 1991.
[7] Yachida, “3D Da a Acquisi ion by Mul iple View”,
Robo ics Resea ch: Thi d In . Symp., Fauge as, O.D.
and Gi al , G., Eds. Camb idge, MA, MIT P ess, pp. 11-
18, 1986.
[8] B own D., 1971: Close-Range Came a Calib a ion.
Pho og amme ic Enginee ing, Vol. 37, No.8
[9] El-Hakim, S.F., 1986: Real-Time Image Me ology wi h
CCD Came as. Pho og amme ic Enginee ing and
Remo e Sensing, Vol. 52, No 11, pp. 1757-1766.