Upload
shanon-greer
View
227
Download
1
Embed Size (px)
Citation preview
Evolution-based least-squares fitEvolution-based least-squares fitting using Pythagorean hodograting using Pythagorean hodogra
ph spline curvesph spline curves
Speaker: Ying .LiuSpeaker: Ying .Liu
November 29. 2007November 29. 2007
Institute of Applied Geometry, JphaInstitute of Applied Geometry, Jphannes Kepler University ,Linz, Austirnnes Kepler University ,Linz, Austir
aawww.ag.jkwww.ag.jku.atu.at
Martin Aigner
Bert Juttler
Author:Author:
Martin Aigner:Martin Aigner:– Dr. Mag., research assistant Dr. Mag., research assistant – Email: martin. aigner @ jku .at Email: martin. aigner @ jku .at Zbynek Sir:Zbynek Sir:– Dr.; research assistant at Dr.; research assistant at FWFFWF-Projekt P1738-Projekt P1738
7-N12 7-N12 – Alumni Alumni
Author: Bert JuttlerAuthor: Bert JuttlerSelected scientific activities:Selected scientific activities:– Since 2003:associated editor Since 2003:associated editor
of CAGDof CAGD– Organizer of various Mini symposia Organizer of various Mini symposia – Member of program committeesMember of program committees
of numerous conferences of numerous conferences
Research interests:Research interests:– CAGD, Applied Geometry, CAGD, Applied Geometry,
Kinematics, Robotics, Kinematics, Robotics, Differential GeometryDifferential Geometry
IntroductionIntroduction
Using PH spline curves to evoluted fittinUsing PH spline curves to evoluted fitting a given set of data points or a curve g a given set of data points or a curve For example:For example:
Steps:Steps:
Introduce a general framework for Introduce a general framework for abstract curve fittingabstract curve fitting
Apply this framework to PH curvesApply this framework to PH curves
Discuss the relationship between this Discuss the relationship between this method and Gauss-Newton iterationmethod and Gauss-Newton iteration
An abstract framework for curve An abstract framework for curve fitting via evolutionfitting via evolution
Parameterized family of curves:Parameterized family of curves:(( s, us, u )) -> ->
– u is the curve parameteru is the curve parameter– s is the vector of shape parameterss is the vector of shape parameters
Let s depend smoothly on an evolution Let s depend smoothly on an evolution parameter t, s( t)=( )parameter t, s( t)=( )Approximately compute the limit Approximately compute the limit
)(uCs
)(),......,(),( 21 tststs n
)(lim tst
],[ bau nRs
An abstract framework for curve An abstract framework for curve fitting via evolutionfitting via evolution
Each point travels with the velocity:Each point travels with the velocity:
Normal velocity of the inner points:Normal velocity of the inner points:
)(|)(
)()( )(1
)()( tss
uCuCuv itss
n
i i
ststs
n
itsitss
i
stststs unts
s
uCunuvuv
1)()()()()( )())(|
)(()()()(
An abstract framework for curve An abstract framework for curve fitting via evolutionfitting via evolution
Assume a set of data points is Assume a set of data points is given.given.
Let and Let and
Expected to toward their associated Expected to toward their associated data points if data points if then then
NjjP ......1}{
)()( jtsj uCf ||)(||minarg )(],[
uCPu tsjbau
j
)()()( )( jtsjjjj unfPduv },{ bau j },{ bau j jjjj fPduv
)(
An abstract framework for curve An abstract framework for curve fitting via evolutionfitting via evolution
An abstract framework for curve An abstract framework for curve fitting via evolutionfitting via evolution
Time derivatives of the shape Time derivatives of the shape parameters satisfied the following parameters satisfied the following equation in least-squares senseequation in least-squares sense
N
bauj
N
baujRjjjtsvjjts
s j j
RwfPuvwduvws},{,1 },{,1
2)(
2)( ||)()(||))((minarg
Necessary condition for a minimum
))(()())(( tsrtstsM
))(())(())(()( 1 tsrtsMtsFts
An abstract framework for curve An abstract framework for curve fitting via evolutionfitting via evolution
Definition:Definition: – A given curve:A given curve:– a set of parameters U is said to be a set of parameters U is said to be
regular:regular:– A set parameters: that A set parameters: that
andand– Unit normal vectorsUnit normal vectors
That the matrix has That the matrix has a maximal ranka maximal rank
)(uCs
NjjuU .....1}{ Uba },{
0)()(
uC ts
)()( jtsj unn
)(, |)(
tssk
jsjkj s
uCnA
nN
An abstract framework for curve An abstract framework for curve fitting via evolutionfitting via evolution
LemmaLemma: in a regular case and if all : in a regular case and if all closet points are neither singular nor closet points are neither singular nor boundary points, then any solution of boundary points, then any solution of the usual least-squares fitting the usual least-squares fitting
is a stationary point of the is a stationary point of the differential equation derived from the differential equation derived from the evolution processevolution process
N
jjtsj
bausuCP
j1
2)(
],[||)(||minminarg
Evolution of PH splinesEvolution of PH splines
Ordinary PH curves c (u)=[x ( u) ,y (u)] satOrdinary PH curves c (u)=[x ( u) ,y (u)] satisfied the following conditions:isfied the following conditions:
Regular PH curves: let w=1.Regular PH curves: let w=1.The difference : gcd (x’ ( u ),y’ (u)) is a The difference : gcd (x’ ( u ),y’ (u)) is a square of a polynomialsquare of a polynomial called preimage curvecalled preimage curve
)(w'x 22 )2(w'y
)]u(),u([
Evolution of PH splinesEvolution of PH splines
Proposition: if a regular PH curve c Proposition: if a regular PH curve c (u) and then:(u) and then:– Smooth field of unit tangent vectors for Smooth field of unit tangent vectors for
all uall u– Parametric speed and arc-length are Parametric speed and arc-length are
polynomial functions polynomial functions – Its offsets are rational curvesIts offsets are rational curves
)0,0())u(),u((
Evolution of PH splinesEvolution of PH splines
Let an open integral B-spline curve,Let an open integral B-spline curve, and and
Let Let
)]u(),u([
km2m1mm1kk1k10u...uu,u,...,u,u,u...uu
m
0ik,ii
)u(N)u(
m
0ik,ii
)u(N)u(
)u(K2y
xud
)u()u(2
)u()u(
y
x)u(C
j,i
m
0i
m
0j
ji
jiji
0
0u
u
22
0
0
s1k
u
uk,jk,ij,i
1k
ud)u(N)u(N)u(K
Evolution of PH splinesEvolution of PH splines
In the evolution we fix the knot vector, sIn the evolution we fix the knot vector, so the shape parameters areo the shape parameters are
the velocitythe velocity
The unit normalsThe unit normals
],...,,...,y,x[sm0m000
m
0i
m
0jj,iij
ijij
ij0
0s
)u(K],[2]y,x[)u(v
m
0i
m
0j k,jk,ijiji
k,jk,i
m
0i
m
0j
jiji
ji
22
s
s )u(N)u(N)(
)u(N)u(N2
)u()u(
)u('C)u(n
Evolution of PH splinesEvolution of PH splinesThe length of PH spline:The length of PH spline:
The regularization term:The regularization term:
Which forces the length to converge tWhich forces the length to converge to some constant value o some constant value
1m
1k
u
u
m
0i
m
0j1mj,ijiji
22
s)u(K)(du))u()u((L
m
0jj,ij
i
s )u(K2L
m
0jj,ij
i
s )u(K2L
2
sse)LLL(:R
s
Le
L
Examples of PH splines evolutioExamples of PH splines evolutionn
Simple example: Simple example: – fitting two circular arcs with radius 1.fitting two circular arcs with radius 1.– Two cubic PH segments depending on 8 Two cubic PH segments depending on 8
shape parametersshape parameters– Initial position: straight lineInitial position: straight line
e
LRv
www
Examples of PH splines evolutioExamples of PH splines evolutionn
Examples of PH splinesExamples of PH splinesInitial: two Initial: two straight straight segments For the segments For the global shape global shape =8,=8,
Gradually raised Gradually raised length to 14length to 14Fix end pointsFix end points
Insert knotsInsert knots
eL
1 Rv www
1.0Rw 100vw
Examples of PH splinesExamples of PH splinesInitial: two straight Initial: two straight segments For the segments For the global shape =8,global shape =8,
Gradually raised Gradually raised length to 14length to 14
Fix end pointsFix end points
Insert knotsInsert knots
eL
1 Rv www
1.0Rw 100vw
Examples of PH splinesExamples of PH splinesInitial: two straight Initial: two straight segments For the glsegments For the global shape =8,obal shape =8,
Gradually raised lengGradually raised length to 14th to 14Fix end pointsFix end points
Insert knotsInsert knots
eL
1 Rv www
1.0Rw 100vw
Examples of PH splinesExamples of PH splinesInitial: two straight Initial: two straight segments For the segments For the global shape =8,global shape =8,
Gradually raised Gradually raised length to 14length to 14
Fix end pointsFix end points
Insert knotsInsert knots
eL
1 Rv www
1.0Rw 100vw
Examples of PH splinesExamples of PH splinesInitial: two straight Initial: two straight segments For the segments For the global shape =8,global shape =8,
Gradually raised Gradually raised length to 14length to 14
Fix end pointsFix end points
Insert knotsInsert knots
eL
1 Rv www
1.0Rw 100vw
Example of PH splinesExample of PH splinesInitial: two straight Initial: two straight segments For the segments For the global shape =8,global shape =8,
Gradually raised Gradually raised length to 14length to 14
Fix end pointsFix end points
Insert knotsInsert knots
eL
1 Rv www
1.0Rw 100vw
Examples of PH splinesExamples of PH splinesInitial: two straight Initial: two straight segments For the segments For the global shape =8,global shape =8,
Gradually raised Gradually raised length to 14length to 14
Fix end pointsFix end points
Insert knotsInsert knots
eL
1 Rv www
1.0Rw 100vw
Examples of PH splinesExamples of PH splinesInitial: two straight Initial: two straight segments For the segments For the global shape =8,global shape =8,
Gradually raised Gradually raised length to 14length to 14
Fix end pointsFix end points
Insert knotsInsert knots
eL
1 Rv www
1.0Rw 100vw
Examples of PH splinesExamples of PH splines
Initial value by Hermite interpolation Initial value by Hermite interpolation – Split data points at estimated inflectionsSplit data points at estimated inflections
Speed of convergenceSpeed of convergence
LemmaLemma: the Euler update of the : the Euler update of the shape parameters for the evolution shape parameters for the evolution with step h is equivalent to a Gauss-with step h is equivalent to a Gauss-Newton step with the same h of the Newton step with the same h of the problemproblem
Provided that Provided that
m
js
jsj uCP1
2 min||)(|| ||)(||minarg],[
uCPu sjbau
j
}...1|{},{ Njuba j 0Rw
Speed of convergenceSpeed of convergence
Speed of convergenceSpeed of convergence
Quadratic convergence of the Quadratic convergence of the methodmethod
Concluding remarksConcluding remarks
Least-squares fitting by PH spline cuves iLeast-squares fitting by PH spline cuves is not necessarily more complicated than s not necessarily more complicated than othersothersFuture work is devoted to using the apprFuture work is devoted to using the approximation procedure in order to obtain oximation procedure in order to obtain more compact representation of NC tool more compact representation of NC tool pathspaths
Q&AQ&A
Thanks!Thanks!