See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3193385

Orthogonal distance fitting of implicit curves and surfaces

Article  in  IEEE Transactions on Pattern Analysis and Machine Intelligence · June 2002

DOI: 10.1109/34.1000237 · Source: IEEE Xplore

CITATIONS READS
114 650

4 authors, including:

Sung Joon Ahn Wolfgang Rauh

CurvSurf, Inc. VITA Zahnfabrik H. Rauter GmbH&Co. KG
34 PUBLICATIONS   904 CITATIONS    29 PUBLICATIONS   623 CITATIONS   

SEE PROFILE SEE PROFILE

Hyungsuck Cho
Korea Advanced Institute of Science and Technology
141 PUBLICATIONS   2,317 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Point Cloud Engine View project

All content following this page was uploaded by Sung Joon Ahn on 03 September 2013.

The user has requested enhancement of the downloaded file.

620 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

Orthogonal Distance Fitting of

Implicit Curves and Surfaces
Sung Joon Ahn, Wolfgang Rauh, Hyung Suck Cho, Member, IEEE, and Hans-JuÈrgen Warnecke

AbstractÐDimensional model fitting finds its applications in various fields of science and engineering and is a relevant subject in
computer/machine vision and coordinate metrology. In this paper, we present two new fitting algorithms, distance-based and
coordinate-based algorithm, for implicit surfaces and plane curves, which minimize the square sum of the orthogonal error distances
between the model feature and the given data points. Each of the two algorithms has its own advantages and is to be purposefully
applied to a specific fitting task, considering the implementation and memory space cost, and possibilities of observation weighting. By
the new algorithms, the model feature parameters are grouped and simultaneously estimated in terms of form, position, and rotation
parameters. The form parameters determine the shape of the model feature and the position/rotation parameters describe the rigid
body motion of the model feature. The proposed algorithms are applicable to any kind of implicit surface and plane curve. In this paper,
we also describe algorithm implementation and show various examples of orthogonal distance fit.

Index TermsÐImplicit curve, implicit surface, curve fitting, surface fitting, orthogonal distance fitting, geometric distance, orthogonal
contacting, nonlinear least squares, parameter estimation, Gauss-Newton method, parameter constraint, parametric model recovery,
object segmentation, object classification, object reconstruction.

1 INTRODUCTION

W ITH image processing, pattern recognition, and compu-

ter/machine vision, dimensional model (curve and
surface) fitting to a set of given data points is a very common
shortest distance point on an implicit model feature from a
given point. Section 3 describes two new algorithms, the
distance-based and coordinate-based algorithms, for ortho-
task carried out during a working project, e.g., edge detection, gonal distance fitting of implicit model features. Various
information extraction from 2D-image or 3D-range image, fitting examples are given in Section 4 and Section 5,
and object reconstruction. For the purpose of dimensional including examples of real data fitting. This paper ends
with a summary in Section 6.
model fitting, we can consider three methods, namely,
moment method [15], [32], [35], [42], Hough transform [8], 1.1 Least-Squares Dimensional Model Fitting
[19], [28], and least-squares method (LSM) [23]. The moment The least-squares method [23] is one of the best known, and
method and Hough transform are efficient for fitting of most often applied, mathematical tools in various disci-
relatively simple models, while their application to a complex plines of science and engineering. With applications of LSM
object model or to an object model with a large number of to dimensional model fitting, the natural and best choice of
model parameters is not encouraged. In this paper, we the error distance is the shortest distance between the given
consider the LS-fitting algorithms for implicit model features. point and the model feature. This error definition is
By data modeling and analysis in various disciplines of prescribed in coordinate metrology guidelines [16], [30]. A
science and engineering, implicit features are very often used century ago, Pearson [32] elegantly solved the orthogonal
because of their compact description in form of f…a; x† ˆ 0 distance fitting problem of line and plane in space by using
and because of the possibility of a simple on-off and inside- the moment method (an art of LSM). However, except for
some simple model features, computing and minimizing
outside decision.
the square sum of the shortest error distances are not simple
We review the current LS-fitting algorithms for implicit
tasks for general features. For applications, an algorithm for
model features and introduce new approaches in this
dimensional model fitting would be very advantageous if
section. Section 2 contains two algorithms for finding the
estimation parameters are grouped and simultaneously
estimated in terms of form, position, and rotation para-
. S.J. Ahn and W. Rauh are with the Department of Information Processing, meters [24]. Furthermore, it would be helpful if the
Fraunhofer Institute for Manufacturing Engineering and Automation reliability (variance) of each parameter could be tested.
(IPA), Nobelstr. 12, 70569 Stuttgart, Germany. There are two main categories of LS-fitting problems of
E-mail: {sja, wor}@ipa.fhg.de.
. H.S. Cho is with the Department of Mechanical Engineering, Korea
implicit features, algebraic and geometric fitting, and these
Advanced Institute of Science and Technology (KAIST), Kusung-dong are differentiated by their respective definition of the error
373-1, Yusung-gu, 305-701 Taejeon, South Korea. distances involved [21], [36].
E-mail: [email protected]. Algebraic fitting is a procedure whereby model feature is
. H.-J. Warnecke is with the Fraunhofer Society, Leonrodstr. 54, 80636
described by implicit equation F …b; X† ˆ 0 with algebraic
Munich, Germany. E-mail: [email protected].
parameters b and the error distances are defined with the
Manuscript received 3 Apr. 2001; revised 25 Oct. 2001; accepted 29 Jan. 2002.
Recommended for acceptance by S. Sarkar.
deviations of functional values from the expected value
For information on obtaining reprints of this article, please send e-mail to: (i.e., zero) at each given point. If F …b; Xi † 6ˆ 0, the given
[email protected], and reference IEEECS Log Number 113920. point Xi does not lie on the model feature (i.e., there is
0162-8828/02/$17.00 ß 2002 IEEE
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 621

some error-of-fit). Most publications about LS-fitting of 1.2 New Approaches

implicit features have been concerned with the square sum We will now introduce two versatile and very efficient
of algebraic distances or their modifications [12], [36], [40] orthogonal distance fitting algorithms for implicit surfaces
X
m  m h
X i2 and plane curves. The general goal of orthogonal distance
2
20 ˆ F …b; Xi † or 20 ˆ F …b; Xi †=krF …b; Xi †k : fitting of a model feature to m given points is the estimation of
i i the feature parameters a minimizing the performance index
…1†
20 ˆ dT PT Pd …3†
In spite of advantages in implementation and computing
cost, the algebraic fitting has drawbacks in accuracy. The or
estimated parameters are not invariant to coordinate
20 ˆ …X X0 †T PT P…X X0 †; …4†
transformation (e.g., a simple parallel shift of the given
points set causes changes not only in position but also in where
form and rotation of the estimated model feature) and,
generally, we cannot find a physical interpretation of an . X: Coordinate column vector of the m given points,
algebraic error definition. Finally, resolving the algebraic XT ˆ …XT T T
1 ; . . . ; Xm †, Xi ˆ …Xi ; Yi ; Zi †;
0
parameters b into the physical parameters (e.g., in terms of . X : Coordinate column vector of the m correspond-
form, position, and rotation parameters) is not a simple task ing points on the model feature;
for general features [24]. . d: Distance column vector, dT ˆ …d1 ; . . . ; dm †,
In geometric fitting, also known as best fitting or orthogonal q
T
distance fitting, the error distance is defined as the shortest di ˆ kXi X0i k ˆ …Xi X0i † …Xi X0i †;
distance (geometric distance) of the given point to the model
feature. However, contrary to its clear and preferable
definition, the geometric distance has nonlinear nature with . PT P: Weighting matrix or error covariance matrix
the model parameters and, as said above, computing and (positive definitive);
minimizing the square sum of geometric distances are not . P: Nonsingular symmetric matrix [17].
easy tasks for general features. Thus, including the normal- If we assume independent identically distributed (i.i.d.)
ized algebraic distance F …b; Xi †=krF …b; Xi †k shown in the measurement errors, the two performance indexes (3) and (4)
second formula of (1) [36], [40], various approximating have the same value up to the a priori standard deviation of
measures of the geometric distance are used for dimensional the measurement errors.
model fitting [14], [29], [31]. Because using approximating We are calling the new algorithms minimizing the
measures of the geometric distance cannot satisfactorily fill performance indexes (3) and (4) distance-based algorithm
the gap between algebraic and geometric fitting, the devel- and coordinate-based algorithm, respectively. Our algorithms
opment of geometric fitting algorithms has been pursued in are the generalized extensions of an orthogonal distance
the literature. While there are geometric fitting algorithms for fitting algorithm for implicit plane curves [3], [4]. Because
explicit [11], and parametric features [5], [13], [21], [38], [41], no assumption is made about the mathematical handling of
we are considering in this paper only fitting algorithms for implicit features, the proposed algorithms can be applied to
implicit features. Sullivan et al. [39] presented a geometric any kind of implicit dimensional feature. Each of the two
fitting algorithm for implicit algebraic features algorithms consists of two nested iterations:


X
q
min min 20 fX0i …a†gm
iˆ1 : …5†
F …b; X† ˆ bj Xkj Y lj Z mj ˆ 0; 0 m
a fXi giˆ1 2F
j
The inner iteration of (5), to be carried out at each outer
minimizing the square sum of the geometric distances iteration cycle, finds the shortest distance points on the
d…b; Xi † ˆ kXi X0i k, where Xi is a given point and X0i is current model feature from each given point (Section 2) and
the shortest distance point on the model feature F from Xi . In the outer iteration cycle updates the model feature para-
order to locate X0i , they have iteratively minimized the meters (Section 3). Once appropriate break conditions are
Lagrangian function L…; X† ˆ kXi Xk2 ‡ F …b; X†. To satisfied, e.g., when no more significant improvement on
obtain the partial derivatives @d…b; Xi †=@b (these are neces- the performance indexes (3) and (4) could be gained
sary for the nonlinear iteration updating the algebraic through parameter updating, the outer iteration terminates.
parameters b), they have utilized the orthogonal contacting By the new algorithms, the estimation parameters a are
equations defined in machine coordinate system XYZ grouped in three categories and simultaneously estimated.
  First, the form parameters ag (e.g., radius r for a circle/
F cylinder/sphere, three axis lengths a, b, c for an ellipsoid)
F…b; Xi ; X† ˆ ˆ 0: …2†
rF  …Xi X† describe the shape of the standard model feature f defined
A weakness of Sullivan et al. algorithm, the same as in the case in model coordinate system xyz (Fig. 1)
of algebraic fitting, is that the physical parameters are
f…ag ; x† ˆ 0 with ag ˆ …a1 ; . . . ; al †T : …6†
combined into an algebraic parameters vector b. And, as we
will show in Section 3.1, they have unnecessarily expensively The form parameters are invariant to the rigid body motion
obtained the partial derivatives @d…b; Xi †=@b from (2). of the model feature. The second and the third parameters
Helfrich and Zwick described a similar algorithm in [27]. group, respectively, the position parameters ap and the
622 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

which usually has only a few form parameters. Functional

interpretations and treatments of the position/rotation
parameters concerning the coordinate transformation (7)
are identical for all implicit model features.

2 ORTHOGONAL CONTACTING POINT

The time consuming part of the proposed orthogonal
distance fitting algorithms in this paper is the inner iteration
of (5), finding the shortest distance points (orthogonal
contacting points) fX0i gmiˆ1 on a general model feature from
each given point fXi gm iˆ1 . For a given point xi in frame xyz

xi ˆ R…Xi Xo †; …9†
we determine the orthogonal contacting point x0i
on the
standard model feature (6). Then, the orthogonal contacting
point X0i in frame XYZ from the given point Xi will be
obtained through a backward transformation of x0i into XYZ.
For some implicit features (e.g., circle, sphere, cylinder, cone,
Fig. 1. Implicit features and the orthogonal contacting point x0i in frame and torus), the orthogonal contacting point x0i can be
xyz from the given point Xi in frame XYZ: (a) plane curve and (b) surface. determined in closed form at a low computing cost.
However, in general, we must iteratively find x0i satisfying
rotation parameters ar , describe the rigid body motion of the appropriate orthogonal contacting conditions. We describe
model feature f in machine coordinate system XYZ two algorithms for finding the orthogonal contacting point x0i
in frame xyz. The first algorithm is derived from the general
X ˆ R 1 x ‡ Xo or x ˆ R…X Xo †; where …7†
properties of the shortest distance point. The second
algorithm is based on a constrained minimization method,
R ˆ R R' R! ˆ …r1 r2 r3 †T ; R 1
ˆ RT ; the method of Lagrange multipliers [20]. Within the follow-
…8† ing sections, the orthogonal contacting point is marked with a
ap ˆ Xo ˆ …Xo ; Yo ; Zo †T ; and ar ˆ …!; '; †T :
prime, as X0i in frame XYZ and as x0i in frame xyz, once it has
We characterize the form parameters as intrinsic parameters been determined and is ready to be used. Otherwise, it is
and the position/rotation parameters as extrinsic parameters of notated with the general coordinate vector X or x.
the model feature according to their context. In this paper, we
intend to simultaneously estimate all these parameters 2.1 Direct Method
A necessary condition for the shortest distance point x on
aT ˆ …aT T T
g ; ap ; ar † the implicit model feature (6) from the given point xi is that
ˆ …a1 ; . . . ; al ; Xo ; Yo ; Zo ; !; '; † ˆ …a1 ; . . . ; aq †: the connecting line of xi with x should be parallel to the
feature normal rf at x (Fig. 1):
For plane curve fitting, we simply ignore all terms concerning
z, Z, !, and '. We may try to describe the model feature F with rf  …xi x† ˆ 0 ; where r ˆ …@=@x; @=@y; @=@z†T :
algebraic parameters b in machine coordinate system XYZ …10†

F …b…ag ; ap ; ar †; X† ˆ f…ag ; x…ap ; ar ; X†† ˆ 0; Equation (10) is equivalent to

although this has no interest for us from the viewpoint of rx f=…xi x† ˆ ry f=…yi y† ˆ rz f=…zi z†
our model fitting algorithms. and only two of three equation rows of (10) are indepen-
In comparison with other dimensional model fitting dent. Equation (10) describes a space curve as the
algorithms, our algorithms simultaneously estimate the intersection of two implicit surfaces defined in frame xyz
physical parameters a of the model feature in terms of
by two independent equation rows of (10). We can find that
form ag , position ap , and rotation parameters ar , thus
(10) generally satisfies the parallel condition of the vector
providing very useful algorithmic features for various
…xi x† to the feature normal rf of the iso-feature of (6)
applications, e.g., with robot vision, motion analysis, and
coordinate metrology. For such applications, algebraic para- f…ag ; x† const ˆ 0: …11†
meters b must be converted into physical parameters a even
though the conversion is not an easy task for general model As the constant in (11) is continuously varying, the
features. trajectories of the points lying on (11) and satisfying (10)
The novelty of our algorithms, as will be described in draw a curve described by (10). We name the curve (10)
Section 3, is that the necessary derivative values for nonlinear centrortho-curve about the point xi to the iso-features (11).
iteration are to be obtained from the standard model feature Our aim is to find the intersection of the centrortho-curve
equation (6) and from the coordinate transformation equa- (10) with the implicit model feature (6). In other words, we
tions (7). For application of our algorithms to a new model would like to find the point x which simultaneously
feature, we need merely to provide the first and the second satisfies (6) and (10) (orthogonal contacting condition, see
derivatives of the standard model feature equation (6) Fig. 2 for the case of a plane curve, an ellipse):
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 623

Fig. 3. Iterative search (direct method) for the orthogonal contacting

point x0i on f…ag ; x† ˆ 0 from the given point xi . The points xi;1 and xi;2
are, respectively, the first and the second approximation of x0i .

Fig. 2. Isocurves x2 =64 ‡ y2 =16 const ˆ 0 (const > 0) and the trajec- the isocurve tangent in xy-plane [3], [4]. From this fact and for
tories (centrortho-curves) …3x 8†…3y ‡ 2† ‡ 16 ˆ 0 (thick line) and
…3x 8†y ˆ 0 (broken line) of the orthogonal contacting points on the
the sake of a common program module for curves and
isocurves from the point (2, 2) and (2, 0), respectively. surfaces, we interchange the second and the fourth row of
  (12). Then, for the case of plane curves, we activate only the
f first two rows of the modified equation (12) in the program
f …ag ; xi ; x† ˆ ˆ 0 with xi ˆ R…Xi Xo †:
rf  …xi x†
module (see Appendix A and Fig. 2 for an ellipse).
…12†
2.2 Constrained Minimization
Orthogonal contacting equation (12) governs the variational
Alternatively, we can find the orthogonal contacting point
behavior of the orthogonal contacting point in frame xyz
minimizing the Lagrangian function [20]
relative to the differential change of the feature parameters a
and will play an important role with the coordinate-based L…; x† ˆ …xi x†T …xi x† ‡ f: …14†
algorithm to be described in Section 3.2.
We directly solve (12) for x by using a generalized The first order necessary condition for a minimum of the
Newton method starting from the initial point of x0 ˆ xi Lagrangian function (14) is
(Fig. 3) (how to compute the matrix @f =@x will be shown in    
rL 2…xi x† ‡ rf
Section 3.2): @L ˆ ˆ 0: …15†
@ f

@f   We iteratively solve (15) for x and  [27], [39]

x ˆ f …x†k ; xk‡1 ˆ xk ‡ 
x: …13†
@x    
k 2I ‡ H rf
x
If necessary, especially in order to prevent the point update rTf 0


x from heading for divergence zone with an implicit   …16†
2…xi x† rf @
feature having very high local curvature, we underweight ˆ ; where Hˆ rf:
f @x
the on-the-feature condition in (12) as below:
  In practice, we have observed that, once the second term of
wf the Lagrangian function (14) becomes negative (the first term
f …ag ; xi ; x† ˆ ˆ 0; with 0 < w  1:
rf  …xi x†
is always nonnegative), it is occasionally prone to become
In the first iteration cycle with the starting point of x0 ˆ xi , more negative (i.e., diverging) in the succeeding iteration
the linear system (13) and its solution (i.e., the first moving cycles of (16) to minimize (14), although the second term is
step) are equivalent to commanded to be zero. Consequently, the convergence of
    
f  iteration (16) is very affected by the selection of the initial
rf

x f 
ˆ and
x ˆ rf  : values. From this observation and with a hint at determining
rf 
x xi 0 xi krfk2 
x i the first moving step size taken from the direct method
It should be noted that the first moving step is the negative described in Section 2.1, we use the following initial values for
of the normalized algebraic error vector shown in the x and :
second formula of (1) which has been regarded in the 8 
> 2
literature [36], [40] as a good approximation of the < f=krfk  : x0 ˆ xi
xi
geometric error vector. 0 ˆ ‡1 : x0 6ˆ xi and f…ag ; x0 †  0
>
:
For plane curves on the xy-plane (i.e., z ˆ 0), we simply 1 : x0 6ˆ xi and f…ag ; x0 † < 0:
ignore the second and the third row of (12), which have terms
…17†
of coordinate z in the cross product and have been
automatically satisfied, then the fourth row of (12) without Iteration (16) with the initial values in (17) converges
terms of coordinate z in the cross product is nothing other relatively fast. Nevertheless, if the convergence of iteration
than the orthogonality condition of the error vector …xi x† to (16) fails, we catch it using the direct method.
624 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

2.3 Verification of the Orthogonal Contacting Point

If no closed form solution of the orthogonal contacting point
x0i is available, thus, if we have to find x0i through iteration,
we would like to start the inner iteration described in
Section 2.1 and Section 2.2, preferably from the initial point
of x0 ˆ xi because we need not provide a special initial
point (blind search). And, as another advantage, the initial
point of x0 ˆ xi almost always leads the iteration (13) and
(16) to the global minimum distance point that is just the
nearest point on the model feature from xi . However,
despite this advantageous initial point selection, the global
minimum distance of x0i from xi could not be generally
guaranteed. Conditions (12) and (15) merely constrain the
connecting line of point x on f…ag ; x† ˆ 0 with the given
point xi to be parallel to the feature normal rf at x. Thus,
the orthogonal contacting point x0i found by iterations (13)
or (16) is, in principle, only a local extreme (maximum or
minimum) distance point. For example, there is a maximum
of four local extreme distance points on an ellipse for a
given point (Fig. 2). While there is no general sufficient
Fig. 4. Verification of the orthogonal contacting point. The point x0i;1
condition for global minimum distance of x0i from xi , we
satisfies (18), while x0i;2 does not: (a) f…ag ; xi † > 0. (b) f…ag ; xi † < 0.
check the correctness of the point x0i by testing multiple
independent necessary conditions.
of the orthogonal distance fitting if we use the current
With many a well-known model feature having axis or
orthogonal contacting point X0i as a reasonable initial point
plane symmetry (e.g., ellipse, parabola, hyperbola, ellipsoid,
for the inner iteration inside the next outer iteration cycle
paraboloid, etc.), the two points x0i and xi must lie in the same
half-plane/space of the model coordinate frame xyz [4]. From x0;a‡
a ˆ Ra‡
a …X0i;a Xo;a‡
a †: …19†
this fact, we can simply check if x0i is not the global minimum
distance point from xi by comparing the coordinate signs However, a pitfall of using (19) must be mentioned. If the
between x0i and xi . If necessary, we selectively invert the current orthogonal contacting point X0i once becomes a local
coordinate signs of x0i and repeat iteration (13) or (16) starting (not global) minimum distance point after an update of the
from the modified x0i . parameters a, it can be inherited into the next outer iteration as
We can also check the correctness of point x0i by long as we use (19) (Fig. 5). In order to interrupt this
examining the direction of the feature normal rf at x0i . unintended development, we periodically (e.g., with every
Once iteration (13) or (16) is successfully terminated, we 5-10th outer iteration cycle) start the inner iteration from the
verify the direction of the feature normal by using another given point Xi during the second half-phase of the outer
orthogonal contacting condition (Fig. 4):
iteration

rf  xi x0i

sign…f…ag ; xi †† ˆ 0: …18† x0;a‡
a ˆ Ra‡
a …Xi Xo;a‡
a †: …20†
krfk x0 kxi x0i k
i
We summarize the strategy for an efficient and inexpen-
Rarely, if (18) could not yet be satisfied with a value sive finding of the orthogonal contacting point x0i :
(probably 2) other than zero, then the connecting line of
the two points xi and x0i pierces the model feature an odd . Use the closed form solution of x0i , if available;
number of times (i.e., at least, x0i is not the global minimum . Start the inner iteration from the initial point x0 of (20)
distance point). In this case, we repeat iteration (13) or (16) in the first half-phase of the outer iteration, during
which the parameter update
a is generally wild;
with an adjusted starting point (e.g., xi;1 on the connecting
. In the second half-phase of the outer iteration, start
line of the two points xi and x0i in Fig. 4).
the inner iteration from the initial point x0 of (19);
2.4 Acceleration of Finding the Orthogonal . Periodically, in the second half-phase of the outer
Contacting Point iteration, start the inner iteration from the initial
The use of the initial point of x0 ˆ xi for the iterative search of point x0 of (20).
the orthogonal contacting point x0i is very convenient and
preferable, however, it demands a relatively high computing 3 ORTHOGONAL DISTANCE FITTING
cost. Fortunately, we are aware of the fact that the parameter
update
a will generally have been moderate in the second With the algorithms described in Section 2 (inner iteration),
half-phase of the outer iteration (after the first 5-10 outer the shortest distance points fX0i gm
iˆ1 on the current model
iteration cycles with most fitting tasks) and the orthogonal feature from each given point fXi gm iˆ1 are found and now
contacting point X0i in frame XYZ changes its position very available for the parameter updating (outer iteration) in the
incrementally between two consecutive outer iteration cycles. nested iteration scheme (5). The inner iteration and the
Then, we can dramatically save the computing cost of the parameter updating are to be alternately repeated until
inner iteration and, consequently, the overall computing cost appropriate break conditions are satisfied. In this section, we
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 625

From (7) and (25), we get

T 
@di Xi X0i @X
Jdi ;a ˆ ˆ
@a kXi X0i k @a XˆX0
i


0 T 
Xi Xi 1 @x @R 1 @Xo 
ˆ R ‡ ‰xŠ ‡
kXi X0i k @a @a @a xˆx0
i


0 T 
xi xi @x 
ˆ
kxi x0i k @axˆx0


i
  !
Xi Xi 0 T   @R 1
 
0  I  ‰x0i Š ;
Fig. 5. Orthogonal contacting points X0i;1 and X0i;2 on the updated feature kXi X0i k   @ar
F …a ‡ 
a; X† ˆ 0 from the given point Xi . Iterative search of the
orthogonal contacting point may converge to X0i;1 , if the iteration started …26†
from the current orthogonal contacting point X0i on F …a; X† ˆ 0 (19),
while starting from the given point Xi converges to X0i;2 (20), being where
kXi X0i;1 k > kXi X0i;2 k.  
aT ˆ aT T T
g ; ap ; ar ;
present two algorithms for updating the model feature 0 1
  x 0
parameters, the distance-based algorithm and the coordi- @R @R @R @R B .. C
nate-based algorithm minimizing the performance index (3) ˆ ; and ‰xŠ ˆ @ . A:
@ar @! @' @
and (4), respectively. 0 x
3.1 Distance-Based Algorithm Furthermore, from the facts …xi x0i † == rfjxˆx0 and
i

The first order necessary condition for a minimum of the @x @f @x @f

performance index (3) as a function of the parameters a is rTf ˆ ˆ
@a @x
 @a @a 
  @f  T  T
@ 2 T @d ˆ  0  0 by f…a; x† ˆ f…ag ; x† ˆ 0;
0 ˆ 2JT PT Pd ˆ 0 ; where Jˆ : …21† @ag
@a @a
the Jacobian matrix Jdi ;a in (26), eventually becomes
We may iteratively solve (21) for a by using the Newton
0    1
method 0 T   
sign …x i x † rf @f   
B i  T  C
Jdi ;a ˆ@  0  0T A
T T


J P PJ ‡ HP Pd
a ˆ T T
J P Pd;T @2d
where H ˆ 2 ; krfk @ag   
@a xˆx0i

T    
…22† Xi X0i   @R 1 0
0  I  ‰xi Š :
or, more conveniently, using the Gauss-Newton method, kXi X0i k @ar
ignoring the second derivatives term in (22) …27†
  In comparison with Sullivan et al.'s algorithm [39], our
JT PT PJk
a ˆ JT PT Pdk ; ak‡1 ˆ ak ‡ 
a: …23†
algorithm, including (24) and (27), simultaneously esti-
In this paper, we iteratively estimate the parameters a by mates the parameters a in terms of form, position, and
using the direct form of the Gauss normal equation (23) rotation parameters. And, Sullivan et al.'s algorithm obtains
PJjk
a ˆ Pdjk ; ak‡1 ˆ ak ‡ 
a; …24† its Jacobian matrix unnecessarily expensively from (2). If
we apply a similar procedure of (26) and (27) to F …b; X† ˆ
with break conditions for iteration (24)
0 with algebraic parameters b defined in machine coordi-
 
kJT PT Pdk  0 or k
ak  0 or 20 k 20 k‡1  0: nate system XYZ, the Jacobian matrix for the Sullivan's
algorithm can be obtained as below at a lower implementa-
The first break condition denotes that the error distance
tion cost without using (2):
vector and its gradient vectors in the linear system (24) should
  
be orthogonal and is equivalent to condition (21). If this is
sign …Xi X0i †T rF @F 
satisfied, there will be practically no more parameter update Jdi ;b ˆ  ; with
(k
ak  0) and improvement on the performance index of 20 . krF k @b 
0
XˆXi
In order to complete the linear system (24), we have to T
r ˆ …@=@X; @=@Y ; @=@Z † :
provide the Jacobian matrix of each distance di between
the given point Xi and the orthogonal contacting point X0i
on the current model feature 3.2 Coordinate-Based Algorithm
In a similar way to the procedure of (21)-(24) carried out in
q
di ˆ kXi X0i k ˆ …Xi X0i †T …Xi X0i †: …25† Section 3.1, we construct the Gauss-Newton iteration
minimizing the performance index of (4)
626 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

@X0 generality, the second and the fourth row of (12). Then, for
Jˆ ; PJjk
a ˆ P…X X0 †jk ; ak‡1 ˆ ak ‡ 
a; …28†
@a plane curve fitting, we consider only the first two rows of the
with break conditions modified equation (12), in other words, we consider only the
  upper-left block of the FHG matrix in (31) (see Appendix A).
kJT PT P…X X0 †k  0 or k
ak  0 or 20 k 20 k‡1  0
3.3 Comparison of the Two Algorithms
and with the Jacobian matrices of each orthogonal contacting Each of the distance-based and coordinate-based algorithms
point X0i on the current model feature, directly derived for orthogonal distance fitting of implicit features is versatile
from (7), and very efficient from the viewpoint of application to a new
   model feature. We would like to stress that only the standard
@X 1 @x @R 1 @Xo  model feature equation (6), without involvement of the
JX0i ;a ˆ ˆ R ‡ ‰xŠ ‡
@a XˆX0 @a @a @a xˆx0 position/rotation parameters, is required in (31) for imple-
i
  
i
…29† mentationofeitherofthetwoalgorithmstoanewmodelfeature

1 @x @R 1 0 (the second derivatives @rf=@ag is required only for coordi-
ˆR ‡ 0 I ‰x Š :
@axˆx0 @ar i
nate-based algorithm). The overall structure of our algorithms
i
remains unchanged for all dimensional fitting problems of
The derivative matrix @x=@a at x ˆ x0i in (29) describes the implicit features. All that is necessary for orthogonal distance
variational behavior of the orthogonal contacting point x0i in fitting of a new implicit model feature is to derive the
frame xyz relative to the differential changes of the FHG matrix of (31) from (6) of the new model feature and to
parameters vector a. Purposefully, we obtain @x=@a from supply a proper set of initial parameter values a0 for iterations
the orthogonal contacting equation (12). Because (12) has an (24) and (28). This fact makes possible a realization of versatile
implicit form, its derivatives lead to and very efficient orthogonal distance fitting algorithms for
implicit surfaces and plane curves. An overall schematic
 
@f @x @f @xi @f @f @x @f @xi @f information flow is shown in Fig. 6 (coordinate-based
‡ ‡ ˆ 0 or ˆ ‡ ; algorithm). We compare the two algorithms from the view-
@x @a @xi @a @a @x @a @xi @a @a
point of implementation and computing cost as follows:
…30†
. Common features
where @xi =@a is, from (9),
- For implementation to a new model feature,
@xi @R @Xo
ˆ ‰Xi Xo Š R only the standard model feature equation (6) is
@a  @a @a  eventually required;
@R
ˆ 0 R ‰Xi Xo Š : - Computing time and memory space cost are
@ar
proportional to the number of the data points;
The other three matrices @f =@x, @f =@xi , and @f =@a in (13) and - The two algorithms demand about the same
(30) are to be directly derived from (12). The elements of these computing cost because the time-consuming
part of the overall procedure is the finding of
three matrices are composed of simple linear combinations of
the orthogonal contacting points.
components of the error vector …xi x† with elements of the
. Different features
following three vector/matrices rf, H, and G (FHG matrix):
    - The distance-based algorithm demands smaller
@f @f @f T @ @ f memory space (with plane curve fitting one half
rf ˆ ; ; ; Hˆ rf; G ˆ ; …31†
@x @y @z @x @ag rf and with surface fitting one third of what the
0 1 coordinate-based algorithm does);
0 0 0
- To a new model feature, the implementation of
@f B B i y y …x i x† 0 C
C the distance-based algorithm is relatively easy
ˆB CH
@x @ …zi z† 0 xi x A because it does not require the second deriva-
0 zi z …yi y† tives @rf=@ag of the model feature;
0 @f @f @f 1 0 1 - With the coordinate-based algorithm, each co-
0 0 0
B
@x @y @z
C ordinate Xi ; Yi ; Zi of a measurement point can
B B @f @f
0 C
B
@f
@y
@f
@x 0 CC @f B @y @x C be individually weighted, while only pointwise
‡B C; ˆB C
@f C; weighting is possible with the distance-
B @f
0 @f C @xi B
@
@f
0 @x A
@ @z @x A @z
based algorithm. This algorithmic feature of
@f @f
0 @f @f 0 @z @y the coordinate-based algorithm is practical
@z @y
0 1 for measuring equipment whose measuring
1 0 0 0
accuracy is not identical between measuring
@f B
B0 yi y …xi x† 0 C
C axes. From this fact, we can say the coordinate-
ˆB C… G 0 0 †:
@a @ 0 …zi z† 0 xi x A based algorithm is the more general one than
0 0 zi z …yi y† the distance-based algorithm;
- A comparison between (26) and (29) reveals
Now, (30) can be solved for @x=@a at x ˆ x0i and, conse-
quently, the Jacobian matrix (29) and the linear system (28) di ˆ NT
i …Xi X0i †; Jdi ;a ˆ NT
i JX0i ;a ;
can be completed and solved for parameter update
a. Xi X0i
For the sake of a common program module for surfaces where Ni ˆ :
kXi X0i k
and plane curves, we have interchanged, without loss of
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 627

Fig. 6. Information flow for orthogonal distance fitting of implicit features (coordinate-based algorithm).
 
3.4 Parameter Constraint PJ
ˆ UWVT
If any parameter constraint (e.g., on size, area, volume, Wc Jc
 
position, orientation of the model feature) with UT U ˆ VT V ˆ I; W ˆ diag…w1 ; . . . ; wq † ;
fcj …a† constj ˆ 0; j ˆ 1; . . . ; n; …32† and now the linear systems (34) and (35) can be solved for
or, in vector form,
a. After a successful termination of iterations (34) and (35),
along with the performance index 20 , the Jacobian matrix
f c …a† const ˆ 0 provides useful information about the quality of parameter
is to be additionally applied, we append any following estimations as follows:
n equations (33) with the large weighting values wc to the
linear systems (24) and (28): . Covariance matrix:
  " T  # 1
@fcj
PJ PJ
wcj
a ˆ wcj fcj …a† constj ; j ˆ 1; . . . ; n: …33† Cov…^a† ˆ ˆ VW 2 VT ;
@a Wc Jc Wc Jc
Considering parameter constraints (32), we can rewrite the
performance indexes (3) and (4) and the linear systems (24) . Parameter covariance:
and (28) as follows:
Xq  
 T   Vji Vki
Pd Pd aj ; a^k † ˆ
Cov…^ ; j; k ˆ 1; . . . ; q ; …36†
20 ˆ ; iˆ1
w2i
Wc …f c …a† const† Wc …f c …a† const†
   
PJ Pd …34†

a ˆ ; . Variance of parameters:
Wc Jc Wc …f c …a† const†
@d @f c 20
Jˆ ; Jc ˆ 2 …^
aj † ˆ aj ; a^j †;
Cov…^ j ˆ 1; . . . ; q ;
@a @a m‡n q
for the distance-based algorithm and
 T   . Correlation coefficients:
P…X X0 † P…X X0 †
20
ˆ ; Cov…^ aj ; a^k †
Wc …f c …a† const† Wc …f c …a† const† aj ; a^k † ˆ p ;
…^ j; k ˆ 1; . . . ; q:
    Cov…^ aj ; a^j †Cov…^ ak ; a^k †
PJ P…X X0 † …35†

a ˆ ;
Wc Jc Wc …f c …a† const†
Using the above information, we can test the reliability of
@X0 @f c
Jˆ ; Jc ˆ the estimated parameters ^ a and the propriety of the model
@a @a
selection (object classification). For example, we may try to
for the coordinate-based algorithm. fit an ellipsoid to a set of measurement points of a sphere-
3.5 Parameter Test and Object Classification like object surface. Then, besides a  b  c and the existence
The Jacobian matrix in (34) and (35) will be decomposed by of strong correlations between the rotation parameters and
SVD [25], [33], the other parameters, we get relatively large variances of
628 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

Fig. 7. Orthogonal distance fit to the points set in Table 1: (a) Circle fit. (b) Ellipse fit. (c) Superellipse fit. (d) Convergence of the fit. Iteration number 0-6:
circle, 7-21: ellipse, and 22-: superellipse fit.

the rotation parameters. In this case, the rotation parameters processing and pattern recognition is very desirable because
are redundant here (overparameterized) and, although the superellipse can represent various 2D-features, e.g., rectangle
fitting of an ellipsoid to the points set has better ("  1), ellipse (" ˆ 1), and star (" > 2). Many researchers
performance than a sphere fitting according to the have tried to fit a superellipse by using the moment method
performance index 20 , the proper (reliable) model feature [10], [43] or using the LSM with algebraic distances [34].
for the points set is a sphere. In order to demonstrate the outstanding performance of
The parameter covariance (36) with the distance-based our geometric fitting algorithms, we fit an extreme shape of a
algorithm is generally a little larger than that with the superellipse, a rectangle. We obtain the initial parameter
coordinate-based algorithm. If 20 is small enough, justifying values with " ˆ 1 from a circle fitting and an ellipse fitting,
ignoring of the second derivatives term in (24) and (28), successively. In detail, we use the gravitational center and rms
there is practically no difference of the parameter covar- central distance of the given points set as the initial parameter
iance (36) between the two algorithms. values for circle fitting, circle parameters for ellipse fitting [3],
[4], and, finally, ellipse parameters for superellipse fitting
(Fig. 7, Table 2). Superellipse fitting to the eight points in
4 FITTING EXAMPLES
Table 1 representing a rectangle terminated after seven
4.1 Superellipse Fitting iteration cycles for k
ak ˆ 7:8
10 9 (Fig. 7d, Table 2).
A superellipse [22] is a generalization of an ellipse and it is
described by
TABLE 1
f…a; b; "; x† ˆ …jxj=a†2=" ‡…jyj=b†2=" 1 ˆ 0; Eight Coordinate Pairs Representing a Rectangle

where a, b are the axis lengths and exponent " is the shape
coefficient. The use of superellipses in applications of image
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 629

TABLE 2
Results of the Orthogonal Distance Fitting to the Points Set in Table 1

4.2 Cone Fitting the constraint weighting value wc is large and, thus, locks the
The standard model feature of a cone in frame xyz can be parameters from changing. If we perturb this quasi-equili-
described below: brium state, similarly to the random walking technique,
through adding a small artifact error to the initial parameter
f… ; x† ˆ x2 ‡ y2 …z tan… =2††2 ˆ 0;
values, then we could expect a faster convergence. With a
where , the only form parameter, is the vertex angle of a nonzero initial vertex angle value of ˆ =10, the iteration
cone. The position Xo , the origin of frame xyz, is defined at terminates after only 13 cycles for k
ak ˆ 2:9
10 6 (Fig. 8d).
the vertex. The orientation of a cone is represented by the
direction cosines of the z-axis, r3 in (8). However, from the 4.3 Torus Fitting
viewpoint of coordinate metrology [30], a better parameter- A torus is a quartic surface, and is described by
ization of a cone is
f…r1 ; r2 ; x† ˆ x4 ‡ y4 ‡ z4 ‡ 2…x2 y2 ‡ y2 z2 ‡ z2 x2 †
2 2 2
f… ; r; x† ˆ x ‡ y …r z tan… =2†† ˆ 0; 2…r22 ‡ r21 †…x2 ‡ y2 † ‡ 2…r22 r21 †z2 ‡ …r22 r21 †2 ˆ 0;
with a constraint on the position and rotation parameters where r1 is the radius of the circular section of the tube and
(see (32)) r2 is the mean radius of the ring. We obtain the initial
parameter values from a 3D-circle fitting [5] (Fig. 9) with
fc …ap ; ar † ˆ …Xo X†T r3 …!; '† ˆ 0;
q
where the second form parameter r is the sectional radius of r1 ˆ 20;circle =m; r2 ˆ rcircle :
a cone cut by the xy-plane (z ˆ 0) and X is the gravitational
center of the given points set. Torus fitting to the 10 points in Table 5 representing a half
As one of the test data sets used in an authorized torus is terminated after nine iteration cycles for k
ak ˆ
1:4
10 7 (Fig. 9d, Table 6).
certification [18], [30] process of our algorithms, the 10 points
in Table 3 representing only a quarter of a cone slice were 4.4 Superquadric Fitting
prepared by the German federal authority PTB (Physikalisch- A superquadric (superellipsoid) [9] is a generalization of an
Technische Bundesanstalt). We have obtained the initial ellipsoid and is described by
parameter values with ˆ 0 from a 3D-circle fitting [5] and
a cylinder fitting, successively (Fig. 8, Table 4). The cone fitting f…a; b; c; "1 ; "2 ; x† ˆ
 "1 ="2
to the points set in Table 3 with the constraint weighting …jxj=a†2="1 ‡…jyj=b†2="1 ‡…jzj=c†2="2 1 ˆ 0;
value wc ˆ 1 is terminated after 60 iteration cycles for
k
ak ˆ 2:3
10 7 . The convergence is somehow slow because where a, b, c are the axis lengths and exponents "1 , "2 are the
the initial cone parameters (i.e., the cylinder parameters) seem shape coefficients. In comparison with the algebraic fitting
to already provide a local minimum and it is especially slow if algorithm [37], our algorithms can also fit extreme shapes of

TABLE 3
Ten Coordinate Triples Representing a Quarter of a Cone Slice
630 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

Fig. 8. Orthogonal distance fit to the points set in Table 3: (a) 3D-circle fit. (b) Circular cylinder fit. (c) Circular cone fit. (d) Convergence of the fit.
Iteration number 0-4: 3D-circle, 5-16: cylinder, and 17-:cone fit with the initial value of ˆ =10.

a superquadric (e.g., a box with "1;2  1 or a star with Superquadric fitting to the 30 points in Table 7 representing
"1;2 > 2). After a series of experiments with numerous sets a box is terminated after 18 iteration cycles for k
ak ˆ
of data points, we have concluded that our algorithms 7:9
10 7 (Fig. 10d). The intermediate ellipsoid fitting
safely converge with the parameter zone (a  b  c) showed a slow convergence in its second half-
phase (iteration number 10 to 42 in Fig. 10d) because of a
0:002  …"1 and "2 †  10 and "1 ="2  500:
relatively large variance of the estimated major axis length a^ of
Otherwise, there is a danger of data overflow destroying the the ellipsoid caused by the distribution of the given points. In
Jacobian matrices of linear systems (13), (16), (24), and (28).
other words, the performance index 20 is not being changed
Additionally, a sharp increase of the computing cost with
very much by the variation of a^ (compare the standard
inner iteration is unavoidable. We obtain the initial
parameter values with "1 ˆ "2 ˆ 1 from a sphere fitting deviation …^ a†ellip: ˆ 13:4393
a† of the intermediate ellipsoid …^
and an ellipsoid fitting, successively (Fig. 10, Table 8). with that of the superquadric …^ a†SQ ˆ 0:1034).

TABLE 4
Results of the Coordinate-Based Orthogonal Distance Fitting to the Points Set in Table 3
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 631

Fig. 9. Orthogonal distance fit to the points set in Table 5: (a) 3D-circle fit. (b) Initial torus. (c) Torus fit. (d) Convergence of the fit. Iteration number 0-6:
3D-circle, and 7-: torus fit.

5 REAL DATA FITTING because they distort the fitting results. Conceptually, our
5.1 Data Acquisition algorithms possess very advantageous algorithmic features
In order to show how our algorithms perform on real data, a for segmentation and outlier detection of dimensional
sphere and an ellipsoid will be fitted to the same set of measurement data, namely, the geometric error definition
measurement points of a sphere-like object surface (Fig. 11a). and the parameters grouping in terms of form, position, and
The points set is acquired by our stripe projecting rotation. In this paper, utilizing these algorithmic features, we
3D-measurement system with the Coded-Light-Approach
used the following interactive procedure between segmenta-
(CLA) [7], [44] using the Gray code [26]. The measuring
system consists of a standard CCD camera and an LCD stripe tion, outlier detection, and fitting of the measurement points:
projector of ABW [1] and is calibrated by photogrammetric
1. From the following procedure, exclude the measure-
bundle adjustment using the circular coded targets for
ment points which belong to the known objects, e.g.,
automation of the calibration procedure [2], [6]. Our measuring stage or already detected and identified
3D-measurement software determines the coordinates of objects.
the object surface points along the stripe edges. The number 2. Initialize the model feature parameters through a
of the measured surface points is about 29,400 (Fig. 11b), model fitting to a well-conditioned object portion or
from which every eighth point (subtotal 3,675 points) is used through a fitting of a simple model feature, e.g. a
for the sphere and the ellipsoid fitting in this section (Fig. 12). sphere instead of the target ellipsoid. Also, initialize
5.2 Segmentation, Outlier Detection, and the safety margin t of the point searching domain
Model Fitting volume to be used in the next step, e.g., t0 ˆ r=2 for a
sphere.
With real data fitting, we have to find an effective search
3. For the current model feature, determine the
(segmentation) method for the measurement points which
orthogonal error distances di of each measurement
potentially belong to the target model feature. And, outliers point lying inside the domain volume (see the next
should be detected and excluded from the model fitting section for a detailed description) and evaluate the
rms error distance
TABLE 5 s
Ten Coordinate Triples Representing a Half Torus 1X m
error ˆ d2 ;
m iˆ1 i

where m is the number of the measurement points

lying inside the domain volume.
632 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

TABLE 6
Results of the Coordinate-Based Orthogonal Distance Fitting to the Points Set in Table 5

Fig. 10. Orthogonal distance fit to the points set in Table 7: (a) Sphere fit. (b) Ellipsoid fit. (c) Superquadric fit. (d) Convergence of the fit. Iteration
number 0-42: ellipsoid and 43-: superquadric fit.

4. Determine the inliers set of measurement points accuracy of the measuring system, terminate the
lying inside the domain volume and having error procedure.
distances of di  tthreshold
error , where tthreshold is the 6. Update the safety margin t tgrowing
tthreshold

outlier threshold factor (we used tthreshold ˆ 2:5, i.e., fit error of the domain volume, where tgrowing is
we assumed 1.24 percent outlier probability in the the domain growing factor (we used tgrowing ˆ 1:0).
domain volume according to the Gaussian error 7. Repeat again from the third step.
distribution). Optionally, if there is no change in the
5.3 Domain Volume for Measurement Points
inliers set, terminate the procedure.
In the first step of the procedure described in the previous
5. Fit the model feature to the inliers set and save the
section, 256 of 3,675 measurement points were detected and
resulting rms error distance fit error . If fit error is removed from the points list as they belong to the measuring
smaller than a given error level concerning the stage (Z  0). In order to not only exclude the outliers from
the model fitting but also avoid unnecessary determination of
TABLE 7 the orthogonal contacting points in the third step of the
Thirty Coordinate Triples Representing a Box procedure, we applied a two-step domain volume criterion.
First, we restrict the measurement points domain within a
box with a safety margin t containing the interest portion of the
standard model feature (6) defined in frame xyz. In particular,
for an ellipsoid, the domain box can be defined below:

jxj  a ‡ t; jyj  b ‡ t; and jzj  c ‡ t:

With the domain box criterion, the potential measurement
points of the target object surface can be screened from a
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 633

TABLE 8
Results of the Coordinate-Based Orthogonal Distance Fitting to the Points Set in Table 7

large set of measurement points at minimal computing cost grouping in terms of form, position, and rotation, makes
(for each measurement point, we need only a few multi- possible a simple and inexpensive search for the measure-
plications for coordinate transformation). ment point candidates which potentially belong to the model
Next, of the measurement points lying inside the domain feature. The time-consuming determination of the orthogo-
box, we actually determine the orthogonal contacting point nal contacting points, which is necessary for the outlier
if the measurement point also lies between two iso-features detection, is to be carried out only for the measurement point
(11) with constinner < 0 and constouter > 0, respectively. In candidates lying inside the domain volume.
particular, for an ellipsoid, we use
5.4 Fitting Results
constinner ˆ f…ag ; xinner † with With the ellipsoid fitting, we have applied the parameter
8 T constraints of ! ˆ 0 and ' ˆ 0 with a weighting value of wc ˆ
< …max…a=2; a t†; 0; 0†
> : a ˆ min…a; b; c†
xinner ˆ …0; max…b=2; b t†; 0†T 1:0
108 (see Section 3.4) since the object surface seems to be
: b ˆ min…a; b; c†
>
: the top face of a horizontally lying ellipsoid. The computing
…0; 0; max…c=2; c t††T : c ˆ min…a; b; c†; time with a Pentium III 866 MHz PC was a total of 7.8 s for
and 17 rounds of sphere fitting with a final parameter update size
of k
ak ˆ 3:7
10 7 and a total of 94.1 s for 26 rounds of
constouter ˆ f…ag ; xouter † with ellipsoid fitting with k
ak ˆ 6:5
10 7 . The number of the
8 T
< …a ‡ t; 0; 0†
> : a ˆ min…a; b; c† outliers (including the out-of-domain points) was 727 for
xouter ˆ …0; b ‡ t; 0† T
: b ˆ min…a; b; c† error ˆ 0:0596 mm and 1,005 for error ˆ 0:0309 mm with the
>
:
…0; 0; c ‡ t† T
: c ˆ min…a; b; c†: sphere and the ellipsoid fitting, respectively (Fig. 12d). If we
compare Fig. 12b and Fig. 12c, an ellipsoid seems to be the
The above two-step criterion, defining the domain better model feature than a sphere for the measured object
volume at a low computing cost by utilizing the parameters surface. However, a look at Table 9 and Table 11 reveals the

Fig. 11. Sphere-like object surface: (a) Stripe projection. (b) Measured points cloud.
634 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

Fig. 12. Results of the model fitting to the points cloud (3; 419 ˆ 3; 675 256 points) from Fig. 11b. Distance error bars are 10 times elongated.
(a) Sphere fit without and (b) with outlier elimination. (c) Ellipsoid fit with outlier elimination under the constraints of ! ˆ 0 and ' ˆ 0. (d) Convergence
of the fit with outlier elimination. Iteration number 0-51: 17 rounds of sphere fitting and 52-162: 26 rounds of ellipsoid fitting. The initial sphere
parameters are taken from the gravitational center and the rms central distance of the data points. After the first round of the sphere fitting, the initial
number of the outliers (including the out-of-domain points) Noutlier ˆ 2; 515 is reduced to 177 and, thereafter, increased.

relatively large variances and the very strong correlations of 6 SUMMARY

the axis lengths of the estimated ellipsoid. From this fact, a
Dimensional model fitting finds its applications in various
sphere could be considered rather as the proper model
fields of science and engineering and is a relevant subject in
feature for the measured object surface. The maximal
computer/machine vision, coordinate metrology, and com-
anticorrelation between the position parameter Zo and the
puter-aided geometric design. In this paper, we have
sphere radius a (also the axis length c of the ellipsoid, Table 10
and Table 11) should be noted which is caused by the one- presented two new algorithms for orthogonal distance fitting
sided distribution of the data points on the top face of the of implicit surfaces and plane curves which minimize the
sphere/ellipsoid. On the other hand, there is no correlation square sum of the orthogonal error distances between
between the first two rotation parameters (! and ') and the the model feature and the given data points. Easily
other parameters of the ellipsoid, which is conditioned by the understandable optimization methods, but with a fresh idea,
parameter constraints of ! ˆ 0 and ' ˆ 0. are applied to the problem of geometric fitting, a widely

TABLE 9
Results of the Coordinate-Based Orthogonal Distance Fitting to the Points Cloud
of Fig. 12b (Sphere Fitting) and Fig. 12c (Ellipsoid Fitting)
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 635

TABLE 10 TABLE 11
Correlation Coefficients of the Sphere Parameters in Table 9 Correlation Coefficients of the Ellipsoid Parameters in Table 9

recognized difficult problem. The new algorithms

are versatile and very efficient from the viewpoint of
implementation and application to a new model feature.
. Rigid body motion (7) of the model in machine
Our algorithms converge very well and do not require a coordinate frame XY:
necessarily good initial parameter values set, which could also       
be internally supplied (e.g., gravitational center and rms X C S x Xo
ˆ ‡ ;
central distance of the given points set for sphere fitting, Y S C y Yo
sphere parameters for ellipsoid fitting, and ellipsoid para- with C ˆ cos…†; S ˆ sin…†:
meters for superquadric fitting, etc.). Memory space and
computing time cost are proportional to the number of the . FHG matrix (31):
given data points. The estimation parameters are grouped in  
x=a2
form/position/rotation parameters and are simultaneously rf ˆ ;
y=b2
estimated, thus providing useful algorithmic features for 0 1
  x2 =a3 y2 =b3
various applications. With the new algorithms, the quality of 1=a2 0 B C
parameter estimation (including the propriety of model Hˆ ; G ˆ @ 2x=a3 0 A:
0 1=b2
selection) can be simply tested by utilizing the parameter 0 2y=b3
covariance matrix. Together with other algorithms for
orthogonal distance fitting of parametric features [5], our . Orthogonal contacting equation (12):
algorithms are certified by the German federal authority PTB  
…x2 =a2 ‡ y2 =b2 1†=2
[18], [30], with a certification grade that the parameter f …a; b; xi ; x† ˆ ˆ 0:
…yi y†x=a2 …xi x†y=b2
estimation accuracy is higher than 0.1 m for length unit and
0.1 rad for angle unit for all parameters of all tested model . Linear system equation (13) of the direct method
features with all test data sets.
(Section 2.1):
  
APPENDIX A 0 0 1=a2 0
IMPLEMENTATION EXAMPLE WITH ELLIPSE FITTING yi y …xi x† 0 1=b2
 ! 
In this paper, the geometric fitting algorithms for implicit x=a2 y=b2
x
‡
curves and surfaces are quite generally described and, in y=b2 x=a2
y
practice, they are implemented in a highly modular manner.  
…x2 =a2 ‡ y2 =b2 1†=2
For application to a new curve/surface, the user needs merely ˆ :
…yi y†x=a2 …xi x†y=b2
to supply the FHG matrix (31) of the standard model feature
(6). Nevertheless, in order to help the interested reader realize . Linear system equation (16) of the constrained
his/her own implementation, we give, in this section, an
minimization (Section 2.2):
example of how some key intermediate equations really look 0 10 1
like with a particular fitting task, the ellipse fitting. 2 ‡ =a2 0 x=a2
x
B CB C
@ 0 2 ‡ =b2 y=b2 A@
y A
. Parameters vector of an ellipse in XY-plane:
x=a2 y=b2 0

a ˆ …a; b; Xo ; Yo ; †T :
0 1
2…xi x† x=a2
. Standard model (6) in model coordinate frame xy: B C
ˆ@ 2…yi y† y=b2 A:
  …x2 =a2 ‡ y2 =b2 1†=2
1 x2 y2
f …a; b; x† ˆ ‡ 1 ˆ 0:
2 a2 b2
636 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

. Linear system equation (24) for the distance-based APPENDIX B

algorithm (Section 3.1):
GRADIENT AND HESSIAN MATRIX OF SUPERELLIPSE
0 1 0 1 AND SUPERQUADRIC
Jd1 ;a d1
B .. C B . C Except for superellipse and superquadric, the derivation of
PB
@
C
. A
a ˆ P@ .. A;
B C
the FHG matrix (31) is straightforward with the curves and
Jdm ;a dm surfaces shown in this paper. Because the power function
q and the logarithm function can cause data overflow and
with di ˆ …Xi Xi0 †2 ‡ …Yi Yi0 †2 : domain error, a careful program source coding of the FHG
matrix is essential for the superellipse fitting and the
superquadric fitting.
. Jacobian matrix Jdi ;a in (27) for the distance-based
algorithm (Section 3.1): B.1 Superellipse
For the sake of a compact description and efficient source


Jdi ;a ˆ gi x0i 2 =a3 gi y0i 2 =b3 0 0 0 coding of the FHG matrix, we use the following super-
 T  ellipse function:
1 Xi Xi0 0 0 1 0 …Yi0 Yo †
;  2="  2=" !
di Yi Yi0 0 0 0 1 Xi0 Xo 1 jxj jyj
f…ag ; x† ˆ f…a; b; "; x† ˆ ‡ 1 ;
2 a b
with

then the FHG matrix of this superellipse function will be:
sign …xi x0i †x0i =a2 ‡ …yi y0i †y0i =b2
gi ˆ q :    
…x0i =a2 †2 ‡ …y0i =b2 †2 1 A=x 2 " A=x2 0
rf ˆ ; Hˆ 2 ;
" B=y " 0 B=y2

. Linear system equation (28) for the coordinate- 0 1

A=a B=b …A ln A ‡ B ln B†=2
based algorithm (Section 3.2): 1B C
G ˆ @ 2A="ax 0 A…1 ‡ ln A†="x A;
0 1 0 1 "
JX10 ;a X1 X10 0 2B="by B…1 ‡ ln B†="y
B JY10 ;a C B Y1 Y1 C
0
B C B C with A ˆ …jxj=a†2=" and B ˆ …jyj=b†2=" :
B .. C B .. C
PB . C
a ˆ PB . C:
B C B C From the fact limx!0 x ln x ˆ 0, we prevent the domain error
@ JX0 ;a A @ Xm X 0 A
m
m
of ln…0† by treating ‰0 ln…0†Š ˆ 0 as a group.
JYm0 ;a Ym Ym0
B.2 Superquadric
. Jacobian matrix JX0i ;a in (29) for the coordinate- For superquadric, we use the following function:
based algorithm (Section 3.2): f…ag ; x† ˆ f…a; b; c; "1 ; "2 ; x†
0 1
   1    2="1  2="1 !"1 ="2  2="2
C S @f @f @xi @f  1@ jxj jyj jzj
JX0i ;a ˆ ‡  ˆ ‡ ‡ 1A:
S C @x @xi @a @a  0 2 a b c
xˆxi
 
0 0 1 0 …Yi0 Yo † In order to reduce the danger of a data overflow by the
‡ ;
0 0 0 1 Xi0 Xo power function, we introduce some intermediate variables
cancelling similarly sized values below:
with
   A ˆ …jxj=a†2="1 =d; B ˆ …jyj=b†2="1 =d; C ˆ …jzj=c†2="2 ;
@f 0 0 1=a2 0
ˆ D ˆ d"1 = "2 ; with d ˆ …jxj=a†2="1 ‡…jyj=b†2="1 :
@x yi y …xi x† 0 1=b2
 2 
x=a 2
y=b Then, the FHG matrix of the above superquadric function
‡ 2
; will be:
y=b x=a2
   
@f 0 0 AD "1 "2
ˆ ; Hxx ˆ 2A ‡ 2 " 1 ;
@xi y=b x=a2
2 "1 "2 x2 "2
   
@xi 0 0 C S yi BD "1 "2
ˆ ; 0 1 Hyy ˆ 2B ‡ 2 " 1 ;
@a 0 0 S C xi AD=x "1 "2 y2 "2
  1@
@f 1 0 0 rf ˆ BD=y A; C
ˆ "2 Hzz ˆ 2 2 …2 "2 †;
@a C=z "2 z
0 yi y …xi x†  
0 1 2ABD "1 "2
x2 =a3 y2 =b3 0 0 0 Hxy ˆ Hyx ˆ ;
B C "1 "2 xy "2
@ 2x=a3 0 0 0 0 A:
Hyz ˆ Hzy ˆ Hzx ˆ Hxz ˆ 0;
0 2y=b3 0 0 0
AHN ET AL.: ORTHOGONAL DISTANCE FITTING OF IMPLICIT CURVES AND SURFACES 637
0 1
AD=a [12] F.L. Bookstein, ªFitting Conic Sections to Scattered Data,º Computer
B C Graphics and Image Processing, vol. 9, no. 1, pp. 56-71, 1979.
  B BD=b C
@f T 1BB
C
C;
[13] B.P. Butler, A.B. Forbes, and P.M. Harris, ªAlgorithms for
ˆ B C=c C Geometric Tolerance Assessment,º Report no. DITC 228/94, Nat'l
@ag "2 B C Physical Laboratory, Teddington, U.K., 1994.
@ D…A ln A ‡ B ln B†=2 A [14] X. Cao, N. Shrikhande, and G. Hu, ªApproximate Orthogonal
…C ln C ‡ D ln D†=2 Distance Regression Method for Fitting Quadric Surfaces to Range
Data,º Pattern Recognition Letters, vol. 15, no. 8, pp. 781-796, 1994.
[15] B.B. Chaudhuri and G.P. Samanta, ªElliptic Fit of Objects in Two
0   1 and Three Dimensions by Moment of Inertia Optimization,º
2
A "1"2"2 ‡ 1 Pattern Recognition Letters, vol. 12, no. 1, pp. 1-7, 1991.
B a
 C

B C [16] DIN 32880-1, Coordinate Metrology; Geometrical Fundamental Prin-
 T B 2
B "1 "2 C ciples, Terms and Definitions. Berlin: Beuth Verlag, German
@ @f AD B B b "2 C
C Standard, 1986.
ˆ B 0  C;
@ag @x "1 "2 x B  C [17] N.R. Draper and H. Smith, Applied Regression Analysis, third ed.
B "1 "2 C New York: John Wiley and Sons, 1998.
B …A ln A ‡ B ln B† "2 ‡ ln A C
@ A [18] R. Drieschner, B. Bittner, R. Elligsen, and F. WaÈldele, ªTesting
"1 Coordinate Measuring Machine Algorithms: Phase II,º BCR
"2 …1 ‡ ln D†
Report no. EUR 13417 EN, Commission of the European
Communities, Luxemburg, 1991.
0   1 [19] R.O. Duda and P.E. Hart, ªUse of the Hough Transformation to
2
A "1"2"2 Detect Lines and Curves in Pictures,º Comm. ACM, vol. 15, no. 1,
B 
a
 C pp. 11-15, 1972.
B C
 T B 2
B "1 "2
‡ 1 C [20] R. Fletcher, Practical Methods of Optimization. New York: John
@ @f BD B B b " C
2
C Wiley & Sons, 1987.
ˆ B 0  C; [21] W. Gander, G.H. Golub, and R. Strebel, ªLeast-Squares Fitting of
@ag @y "1 "2 y B  C
B "1 "2 C Circles and Ellipses,º BIT, vol. 34, no. 4, pp. 558-578, 1994.
B …A ln A ‡ B ln B† " ‡ ln B C [22] M. Gardiner, ªThe Superellipse: A Curve that Lies between the
@ 2 A
"1 Ellipse and the Rectangle,º Scientific Am., vol. 213, no. 3, pp. 222-
" …1 ‡ ln D† 234, 1965.
0 1 2
0 [23] C.F. Gauss, Theory of the Motion of the Heavenly Bodies Moving about
B C the Sun in Conic Sections (Theoria Motus Corporum Coelestium in
  B 0 C Sectionibus Conicis Solem Ambientum), first published in 1809,
@ @f T CB C
ˆ 2 B B 2=c C:
C
translation by C.H. Davis. New York: Dover, 1963.
@ag @z "2 z B C [24] R.N. Goldman, ªTwo Approaches to a Computer Model for
@ 0 A Quadric Surfaces,º IEEE Computer Graphics and Applications, vol. 3,
no. 9, pp. 21-24, 1983.
1 ‡ ln C [25] G.H. Golub and C. Reinsch, ªSingular Value Decomposition and
Least Squares Solutions,º Numerische Mathematik, vol. 14, no. 5,
pp. 403-420, 1970.
[26] F. Gray, ªPulse Code Communication,º US Patent 2,632,058, Mar.
REFERENCES 17, 1953.
[1] ABW GmbH, http://www.abw-3d.de/home_e.html, Feb. 2002. [27] H.-P. Helfrich and D. Zwick, ªA Trust Region Method for Implicit
[2] S.J. Ahn, ªCalibration of the Stripe Projecting 3D-Measurement Orthogonal Distance Regression,º Numerical Algorithms, vol. 5,
System,º Proc. 13th Korea Automatic Control Conf. (KACC '98), pp. 535-545, 1993.
pp. 1857-1862, 1998 (in Korean). [28] P.V.C. Hough, ªMethod and Means for Recognizing Complex
[3] S.J. Ahn and W. Rauh, ªGeometric Least Squares Fitting of Circle Patterns,º US Patent 3,069,654, Dec. 18, 1962.
and Ellipse,º Int'l J. Pattern Recognition and Artificial Intelligence, [29] G. Hu and N. Shrikhande, ªEstimation of Surface Parameters
vol. 13, no. 7, pp. 987-996, 1999. Using Orthogonal Distance Criterion,º Proc. Fifth Int'l Conf. Image
[4] S.J. Ahn, W. Rauh, and H.-J. Warnecke, ªLeast-Squares Orthogo- Processing and Its Applications, pp. 345-349, 1995.
nal Distances Fitting of Circle, Sphere, Ellipse, Hyperbola, and [30] ISO 10360-6:2001, Geometrical Product Specifications (GPS)ÐAc-
Parabola,º Pattern Recognition, vol. 34, no. 12, pp. 2283-2303, 2001. ceptance and Reverification Tests for Coordinate Measuring
[5] S.J. Ahn, E. WestkaÈmper, and W. Rauh, ªOrthogonal Distance Machines (CMM)ÐPart 6: Estimation of Errors in Computing
Fitting of Parametric Curves and Surfaces,º Proc. Int'l Symp. Gaussian Associated Features, Int'l Standard, ISO, Dec. 2001.
Algorithms for Approximation IV: Huddersfield 2001, I. Anderson and [31] D. Marshall, G. Lukacs, and R. Martin, ªRobust Segmentation of
J. Levesley, eds., 2002. Primitives from Range Data in the Presence of Geometric
[6] S.J. Ahn, W. Rauh, and S.I. Kim, ªCircular Coded Target for Degeneracy,º IEEE Trans. Pattern Analysis and Machine Intelligence,
Automation of Optical 3D-Measurement and Camera Calibra- vol. 23, no. 3, pp. 304-314, Mar. 2001.
tion,º Int'l J. Pattern Recognition and Artificial Intelligence, vol. 15, [32] K. Pearson, ªOn Lines and Planes of Closest Fit to Systems of
no. 6, pp. 905-919, 2001. Points in Space,º The Philosophical Magazine, Series 6, vol. 2, no. 11,
[7] M.D. Altschuler, B.R. Altschuler, and J. Taboada, ªMeasuring pp. 559-572, 1901.
Surfaces Space-Coded by a Laser-Projected Dot Matrix,º Proc. [33] W.H. Press, B.P. Flannery, S.A. Teukolsky, and W.T. Vetterling,
SPIE Conf. Imaging Applications for Automated Industrial Inspection Numerical Recipes in C: The Art of Scientific Computing. Cambridge,
and Assembly, vol. 182, pp. 187-191, 1979. U.K.: Cambridge Univ. Press, 1988.
[8] D.H. Ballard, ªGeneralizing the Hough Transform to Detect [34] P.L. Rosin, ªFitting Superellipses,º IEEE Trans. Pattern Analysis and
Arbitrary Shapes,º Pattern Recognition, vol. 13, no. 2, pp. 111-122, Machine Intelligence, vol. 22, no. 7, pp. 726-732, July 2000.
1981. [35] R. Safaee-Rad, K.C. Smith, B. Benhabib, and I. Tchoukanov,
[9] A.H. Barr, ªSuperquadrics and Angle-Preserving Transforma- ªApplication of Moment and Fourier Descriptors to the Accurate
tions,º IEEE Computer Graphics and Applications, vol. 1, no. 1, pp. 11- Estimation of Elliptical Shape Parameters,º Pattern Recognition
23, 1981. Letters, vol. 13, no. 7, pp. 497-508, 1992.
[10] M. Bennamoun and B. Boashash, ªA Structural-Description-Based [36] P.D. Sampson, ªFitting Conic Sections to `Very Scattered' Data: An
Vision System for Automatic Object Recognition,º IEEE Trans. Iterative Refinement of the Bookstein Algorithm,º Computer
Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 27, no. 6, Graphics and Image Processing, vol. 18, no. 1, pp. 97-108, 1982.
pp. 893-906, 1997. [37] F. Solina and R. Bajcsy, ªRecovery of Parametric Models from
[11] P.T. Boggs, R.H. Byrd, and R.B. Schnabel, ªA Stable and Efficient Range Images: The Case for Superquadrics with Global Deforma-
Algorithm for Nonlinear Orthogonal Distance Regression,º SIAM J. tions,º IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 12,
Scientific and Statistical Computing, vol. 8, no. 6, pp. 1052-1078, 1987. no. 2, pp. 131-147, Feb. 1990.
638 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 5, MAY 2002

[38] D. Sourlier, ªThree Dimensional Feature Independent Bestfit in Hyung Suck Cho received the BS degree from
Coordinate Metrology,º PhD Thesis no. 11319, ETH Zurich, Seoul National University, Korea in 1971, the MS
Switzerland, 1995. degree from Northwestern University Evanston,
[39] S. Sullivan, L. Sandford, and J. Ponce, ªUsing Geometric Distance Illinois, in 1973, and the PhD degree from the
Fits for 3-D Object Modeling and Recognition,º IEEE Trans. Pattern University of California at Berkeley in 1977. From
Analysis and Machine Intelligence, vol. 16, no. 12, pp. 1183-1196, 1977 to 1978, he was a postdoctoral fellow in the
Dec. 1994. Department of Mechanical Engineering, Univer-
[40] G. Taubin, ªEstimation of Planar Curves, Surfaces, and Nonplanar sity of California at Berkeley. Since 1978, he has
Space Curves Defined by Implicit Equations with Applications to been a professor in the Department of Production
Edge and Range Image Segmentation,º IEEE Trans. Pattern Engineering, Department of Automation and De-
Analysis and Machine Intelligence, vol. 13, no. 11, pp. 1115-1138, sign, Seoul Campus, and is currently with the Department of Mechanical
Nov. 1991. Engineering, Korea Advanced Institute of Science and Technology
[41] D.A. Turner, ªThe Approximation of Cartesian Coordinate Data by (KAIST), Korea. From 1984 to 1985, he was a visiting scholar at the
Parametric Orthogonal Distance Regression,º PhD thesis, School of Fraunhofer Institute for Manufacturing Engineering and Automation (IPA),
Computing and Math., Univ. of Huddersfield, U.K., 1999. Germany, where he carried out research on robot-based assembly. He
[42] K. Voss and H. SuÈûe, ºInvariant Fitting of Planar Objects by has been invited as a short-term visiting scholar to several universities:
Primitives,º IEEE Trans. Pattern Analysis and Machine Intelligence, Ritsumeikan University, Japan, the University of Paderborn, Germany,
vol. 19, no. 1, pp. 80-84, Jan. 1997. and the New Jersey Institute of Technology, in 1987, 1992, 1998,
[43] K. Voss and H. SuÈûe, ªA New One-Parametric Fitting Method for respectively. From 1995 to 1996, as a visiting professor, he participated in
Planar Objects,º IEEE Trans. Pattern Analysis and Machine graduate school education for the Advanced Manufacturing Program
Intelligence, vol. 21, no. 7, pp. 646-651, July 1999. (AMP) of the University of California, San Diego. Dr. Cho's research
[44] F.M. Wahl, ªA Coded-Light Approach for 3-Dimensional (3D) interests are focused on the area of environment perception and
Vision,º Research Report no. 1452, IBM Zurich Research Labora- recognition for mobile robots, machine vision and pattern classification,
tory, Switzerland, Sept. 1984. and application of artificial intelligence/machine intelligence. During his
research career, he has published six book chapters and more than 346
Sung Joon Ahn received the BS degree in research papers (276 papers in international journals and conferences
mechanical design and production engineering and 66 papers in Korean journals). He now serves on the editorial boards
from Seoul National University in South Korea in of five international journals, including Control Engineering Practice
1985. He received the MS degree in production (IFAC). In addition to his academic activities, he serves and/or has served
engineering from the Korea Advanced Institute of on a few technical committees (TC) of IFAC and IMEKO: TC on Robotics,
Science and Technology (KAIST) in 1987. He TC on Advanced Manufacturing (IFAC), and Measurements on Robotics
worked as a junior research scientist at the (IMEKO). He has been general chair and/or cochair of several interna-
research center of LG Electronics from 1987 to tional conferences, including SPIE-Opto Mechatronic Systems (2000,
1990. Since 1991, he has been working as a 2001) and IEEE/RSJ IROS (1999). For his achievements in research
research scientist for the Fraunhofer Institute for works in the fields of robotics and automation, Dr. Cho has been endowed
Manufacturing Engineering and Automation (IPA) in Stuttgart, Germany. with a POSCO professorship from Pohang Steel Company, Korea, since
His research interests include pattern recognition, optical 3D-measure- 1995. In 1998, he was awarded the Thatcher Brothers Prize from the
ment, close-range photogrammetry, camera calibration, and 3D-informa- Institution of Mechanical Engineers, United Kingdom, and he also
tion processing. received the fellowship of Alexander von Humboldt of Germany in 1984.
He is a member of the IEEE.

Wolfgang Rauh received the Doctorate degree Hans-JuÈrgen Warnecke studied mechanical
in 1993 from the University of Stuttgart. He engineering at the Technical University of
worked at the University of Stuttgart from 1984 Braunschweig, Germany. He was a research
until 1989 after having received the Diploma engineer at the Institute for Machine Tools and
degree in mechanical engineering from the Manufacturing afterward, from 1965 to 1970, a
University of Karlsruhe in 1984. He then became managing director for planning and manufactur-
head of the Industrial Image Processing Group ing at the Rollei-Werke, Braunschweig, a camera
at the Fraunhofer Institute for Manufacturing factory. Since 1971, he has been a full professor
Engineering and Automation (IPA). In 1991, he for industrial manufacturing and management,
was made head of the Information Processing University of Stuttgart, and the head of the
Department at the same institute. In this department, a wide variety of Fraunhofer Institute for Manufacturing Engineering and Automation
projects in the area of image processing are carried out. (IPA) of the Fraunhofer Society for Applied Research. His main fields of
research and development are management and organization of
industrial companies, computer integrated manufacturing, computer-
aided planning of material flow, simulation, production planning and
control, manufacturing processes, flexible manufacturing systems,
industrial robots, automation of handling and assembly, quality control,
industrial measurement, cleanroom manufacturing, semiconductor
manufacturing automation, and maintenance. He has published
10 books with many publications in the field of production, industrial
engineering, and automation. Since 1 October 1993, he has been the
president of the Fraunhofer Society, the largest institution for applied
research and development in Europe with a budget of EUR 0.9 billion, of
which EUR 300 million are proceeds from 3,000 private enterprises.
There are 11,000 employees in 56 institutes. In the years 1995 to1997,
he was the president of VDI, the Association of German Engineers, the
largest institution of this kind in Europe.

. For more information on this or any other computing topic,

please visit our Digital Library at http://computer.org/publications/dlib.

View publication stats