J. Indian Inst. Sci., July–Aug.

2005, 85, 161–182

© Indian Institute of Science.

A tracing method for parametric Bézier triangular surface/

plane intersection over triangular domain

R. SHARMA1* AND O. P. SHA2

1
Design Laboratory, Department of Ocean Engineering and Naval Architecture, Indian Institute of Technology,
Kharagpur 721 302, WB, India.
Fax no.: 1*+091-03222-282700, 2+091-03222-277190; Phone no.: 1*+091-03222-281601, 2+091-03222-283788.
email: 1*[email protected]; [email protected]

Received on October 3, 2003; Revised on October 3, 2004.

Abstract

Surface/plane intersection problem is a special case of surface/surface intersection, and is an active area of re-
search across many disciplines in computer-aided geometric design. This paper presents, in general the setting of
derivational continuities, (i.e. C0, C1, and C2), a surface/plane intersection algorithm for parametric Bézier trian-
gular surface over triangular domain. The present algorithm is a combinatorial algorithm that is based on a trac-
ing method, in which the intersection curves are traced out in the direction of tangent vectors at intersection
points. The intersection curves are represented by their piecewise polynomial approximations, given by ordered
pair of surface domain points. The aim is to obtain the smallest number of ordered points that correctly represent
the topology of the true intersection curves. Since the points on the intersection curves are obtained in an ordered
manner, no sorting is essential. The refinement of the intersection curve may include fitting an interpolatory cubic
spline, or adding and deleting points. The method has been used to compute different planar sections (i.e. inter-
section curves with X, Y, and Z planes) for surfaces having different derivational continuities (i.e. C0, C1, and C2).
Keywords: Bézier rational triangular patches, rational-triangular parametric surfaces, surface/plane intersection,
surface directional tangent vectors, topological subdivision.

1. Introduction
In general context, surface/plane intersection problem is a special case of surface/surface
intersection. This has various applications in CAD/CAM systems, computer graphics, sur-
face information systems, geographic information systems, etc. The computation of curve is
required for different objectives in many industries, i.e. surface slicing for laminated object
manufacturing and stereo lithography, NC tool path generation, intersection of hunting
planes to evaluate machining information, and contour lines for geographical and informa-
tion models. Though it is computationally expensive to treat surface/plane intersection
problem as a special case of surface/surface intersection rather than purely a planar cut
problem, theoretically it is better because some of the ideas of the algorithm may be utilized
to compute surface/surface intersection. The surface/surface intersection problem is consid-
ered to be one of the most basic but difficult problems in computer-aided geometric design
(CAGD). However, it has been dealt extensively to obtain exact, robust, and efficient solu-
tions in general setting for surfaces.

*Author for correspondence.

162 R. SHARMA AND O. P. SHA

Triangular surfaces are important because in areas where the geometry is not similar to
rectangular domain, the rectangular surface patch will collapse into a triangular patch. In
such a case, one boundary edge may collapse into a boundary vertex of the patch, e.g.
Wolter and Tuohy [1]. This may give rise to geometric dissimilarities, (e.g. shape parame-
ters, Gaussian curvature distribution, cross-boundary continuities, etc.) and topological in-
consistency. Furthermore, since a triangular patch (i.e. defined as a closed polygon) is a
basic 2D figure in algebraic topology, any fairly irregular complex geometry can be effi-
ciently modeled/designed/regenerated with a triangular patch, e.g. Farin [2]. However, tri-
angular surfaces over triangular domain remain relatively unexplored as compared to
rectangular surfaces over rectangular domain.
In the present work, we are interested in surface/plane intersection algorithm for paramet-
ric Bézier triangular surface over triangular domain. Since, in full generality, the curve trac-
ing in surface/surface intersection is a difficult problem, no infallible method is available
either in general context or in general setting of derivational continuities. Instead of work-
ing with one special type of derivationally continuous surface (i.e. C0, or C1, or C2 continu-
ous) or concentrating on a specialized area in the general problem of surface/plane
intersection, in this work we take a combinatorial approach that utilizes different ideas, and
concentrate on general settings of derivational continuities (i.e. C0, C1, and C2 continuity).
Our aim is to present a basic surface/plane intersection algorithm for parametric Bézier tri-
angular surface over triangular domain without any add-on feature.
We consider the non-self-intersecting surface/plane intersection problem as a special case
of non self-intersecting surface/surface intersection. Basically, we compute surface/surface
intersection, while defining the plane itself as a fixed domain surface. The planar surface is
designed/modeled/defined over the plane as a fixed domain surface, where the domain is
user-specified. This will ensure that within the bounds of the domain of the two surfaces,
one surface and another planar surface for the plane, an intersection does exist. The algo-
rithm used in this work is based on a tracing method, in which the intersection curves are
traced out in the direction of tangent vectors at intersection points. The intersection curves
are represented by their piecewise polynomial approximations, given by ordered pair of surface
domain points. The aim is to obtain the smallest number (i.e. in this work the number is be-
tween 2 and 23) of ordered points that correctly represent the topology of the true intersec-
tion curves. The algorithm presented is simple, but allows the correct computation of topo-
logy of the true intersection curves. This allows any complex and complicated feature (i.e.
the computation of intrinsic properties such as unit tangent vector, curvature vector, binor-
mal vector, curvature, torsion, and higher-order transversal and tangential derivatives), re-
quired to further analyze the intersection curve to be suitably added. Since the points on the
intersection curves are obtained in an ordered manner, no sorting is essential. We consider
the computation of multiple intersection curves at one plane with multiple solutions of start-
ing point and for each starting point the intersection curve is traced separately. However,
we do not include the computation of intersection curve with multiple branches (i.e. branch
points, and bifurcation points), or lower and higher-order singularities in our surface/plane
intersection algorithm. The surface/plane intersection algorithm presented in this work is a
structured algorithm that allows better insight into the problem, and stepwise computations.
This shall allow easy addition of any suitable add-on feature for future research.
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 163

This paper is organized as follows. In Section 1, a brief description of the work is pre-
sented. The current state of the art has been reviewed in Section 2. The theoretical back-
ground and implementation of the surface/plane intersection is presented in Section 3.
Numerical examples are discussed in Section 4. Section 5 concludes by identifying some
future applications and scope of research.

2. Brief review of the state of the art

Normally, the surface/surface intersection approaches are classified into four groups: ana-
lytical, lattice evaluation, subdivision, and tracing (i.e. including numerical methods)-based
methods. Of these four approaches, subdivision- and tracing-based methods are the most
widely used because of their generality. In this section we concentrate on the review of the
methods based upon subdivision and tracing.
Bajaj et al. [3] discuss a purely numerical approach that used a third-order Taylor ap-
proximation by taking steps of reliable lengths and the intersection curves are computed us-
ing the Newton iteration method. They also presented implemental techniques to address
the problem of singularity, but, because of the implicit nature of the method it is suitable to
only CAD systems in engineering where implicit surface definitions are used. Montaudouin
et al. [4] presented a method in limited context for approximation of intersections for alge-
braic curves and surfaces. They used power series in approximation. Asteasu [5], and Gar-
rity and Warren [6] also worked on similar lines.
Houghton et al. [7] discuss a subdivision-based method. It is dependent on C0 linear sub-
divided surface approximations, which is a limitation in general context. Filip et al. [8] also
worked on similar lines. Bürger and Schaback [9] presented a parallel algorithm using the
divide-and-conquer (i.e. subdivision based) method with the analysis of its complexity. The
method uses progressive refinement in interval search (e.g. Gregory [10]). Barnhill et al.
1
[11] discuss a general well-structured surface/surface intersection algorithm for C con-
tinuous surfaces in parametric rectangular domain. The algorithm computes the intersection
curve stepwise, and hence allows a better insight into the problem. This allows any ‘add-
on’ feature to be integrated with the algorithm if the geometric constraints so demand.
Later, Barnhill and Kersey [12] generalized the algorithm further by excluding explicit sur-
face definitions and including more geometric and topological definitions (i.e. evaluated
surface positions and tangents). The method uses oriented bounding boxes (e.g. common
with [7]) over the surface domain for selective subdivision and to get starting points. In
similar line, Cugini et al. [13] present the concept of shrinking bounding boxes over the
surface domain, and use conflicting bounding boxes to compute the starting points. Though
the concept of bounding boxes is elegant, simple, and effective in many cases, it is pure
geometric, and hence may miss the intersection points.
Abdel-Malek and Yeh [14] present a tracing method to compute the intersection curve for
surface/surface intersection. The authors deal with solid surfaces (i.e. closed surface de-
fined over a solid) and represent a solid by a number of parametric surfaces (i.e.
A
S Si (u , v) : U ⊂ R 2 → R 3 , A S Si = { A S S1 , A SS 2 , A S S3 ,...}, which is a differentiable map AS Si
from an open set U ⊂ R2 into R3). The authors define a solid surface over R3 scattered data,
onto different open surfaces over R2 parametric plane. The implicit condition in the algo-
164 R. SHARMA AND O. P. SHA

rithm presented by the authors is that the vectors ∂ A S Si / ∂u and ∂ A S Si / ∂v are linearly inde-
pendent. The linear independency of these partial derivatives is essential to have a nonzero
Jacobian (JC) of the surface SS(u, v). The requirement of a nonzero Jacobian (JC) of the sur-
face SS(u, v) is because if the Jacobian (JC) is zero then the computation of the inverse of
the Jacobian (JC) is computationally expensive, and numerically unstable too. This implicit
condition holds true only for a class of surfaces, which are higher-order continuous (i.e.
Gn/Cn continuous for n ≥ 1), and do not have a tangent crack (i.e. surface which has C0 con-
tinuity overall or in hybrid setting of derivational continuities). Because of this, the formu-
lation presented is restricted to surfaces in engineering sciences that are nonflat, nonhybrid,
and without a crack. Thus, the authors have only discussed the examples of analytical sur-
faces (i.e. nonflat higher-order continuous surfaces such as cylindrical and spherical sur-
faces). Additionally, the authors trace the curve with a purely numerical approach (i.e.
incorporating extra constraints representing local differential geometrical properties such as
tangent vector), and hence solve a system of highly nonlinear equations using a professional
software (PITCONTM*). The solution of nonlinear equations generally is a complex, com-
plicated, and computationally expensive process, and hence for robustness specialized
software is needed. Furthermore, the authors address the problem of singularity/bifurcation
(i.e. the intersection curve having multiple branches), by relying upon the change in the
sign of Jacobian (JC) of the surface, and the assumption is used in switching from one inter-
section curve branch to another of the same curve. This formulation holds true only for the
cases when Jacobian (JC) of the surface is nonzero and nonconstant. In engineering sci-
ences, this is valid only for a class of surfaces that have a fair distribution of curvatures (i.e.
cylindrical surfaces). For the objective of switching the authors use low rank deficiency in
the Jacobian (JC) (i.e. if only the first-order Taylor expansion of the surface is zero and not
the higher-order ones, then it is low rank). Numerically, rank deficiency of the Jacobian
(JC) can only be tackled efficiently with analytical surfaces, and not with piecewise poly-
nomial surfaces. This is because for analytical functions the degree in the series expansion
can be sufficiently large, while for piecewise polynomials the popular degrees are either 3
or 4. As in the present work, piecewise polynomial surfaces are investigated we avoid, in
our formulation, rank deficiency in the Jacobian (JC).
In recent years, research has been concentrated on the ‘loop detection’-based methods,
e.g. Cheng [15], Kriezis et al. [16], Sederberg et al. [17], Sederberg and Meyers [18], and
Sinha et al. [19]. In those techniques, one seeks the solution to sequential initial value prob-
lems. Each initial value problem terminates when the solution encounters one of the possi-
ble stopping points. Since it is impossible to estimate the parametric value at which such an
encounter might occur, a heuristic approach is taken into the curve tracking/tracing scheme.
This procedure in general depends upon prior setting of a few tolerances, a process that is
dependent on the geometrical features of the surfaces. Since the choices are not obvious, it
is not always easy to handle.
Grandine and Klein [20] approach this problem differently. Instead of working with ini-
tial value problem, the authors try to determine, in advance, which of the candidate stop-
ping points is the actual one, and set that prior to curve tracking/tracing. Hence, the authors

*Trademark and copyright are with Prof. W. C. Rheinboldt, Department of Mathematics, University of Pittsburg, USA.
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 165

work with boundary value problem in place of initial value problem. The approach is re-
stricted heavily to a class of surfaces (i.e. in the author’s case tensor product surfaces) to
which the computationally robust solution to a nonlinear system of equations exists. Again,
the surface has to be C1 continuous, and if not, it needs to be subdivided into subsurfaces
that are C1 continuous. The method can be extended to piecewise polynomial surfaces defined
over triangular domain (i.e. those used in the present work) and irregularly shaped regions,
and algebraic surfaces. Though the method computes excellent results in some cases, be-
cause of its restrictions and computational expenses, it cannot be applied in general context.
The ‘loop detection’ technique ensures that the different branches of the intersection
curve will be computed. However, this technique does not ensure the correct topological
computation of the intersection curve. Since, in interactive user environment in engineering,
the correct topology of the intersection curve is important for interactive design this is a se-
rious restriction. In general context, only the marching, tracing, and curve tracking makes
the computation of topological properties of the intersection curve an in-built feature in the
procedure. Because of this reason ‘loop detection’-based techniques are considered a suit-
able add-on, but not a sole methodology in themselves.
In general, numerical algorithms require the computation of a starting point on/or close
within specified tolerances to the intersection curve. Müllenheim [21] discusses an iterative
procedure to compute a starting point close within specified tolerances to the intersection
curve. Dokken [22], and Dokken et al. [23] present a surface/surface intersection based on
recursive subdivision. Abdel-Malek and Yeh [24] present two iterative numerical algo-
rithms to compute the starting points in surface/surface intersection with their complexity
analysis. The first numerical algorithm is an iterative, conjugate gradient-based optimiza-
tion technique. The second algorithm uses the Moore–Penrose pseudo-inverse of the con-
straint function to determine the starting point.
The recursive subdivision based on some type of global divide-and-conquer algorithms
can continue till the problem reduces to a simple optimization problem within specified tol-
erances, for example, Dokken et al. [23]. Since optimization moves from global to local via
subdivision, any error in subdivision at any level will continue and might affect the conver-
gence in the whole solution. Also, because of their global characteristics the subdivision-
based algorithms are slow and computationally expensive. The iterative methods consider
the intersection problem in local context of the parametric domain. Hence these are fast and
computationally efficient, and due to the same reason do not ensure the complete determi-
nation of multiple solutions. The Moore–Penrose pseudo-inverse, however, guarantees that
if a solution exists, it will be computed within/closest to user-specified tolerance (i.e. initial
guess). If multiple solutions exist, then the closest solution to the pre-specified initial guess
will be found.
The subdivision is important on surfaces that occur normally in engineering sciences
(discussed in Section 3.1.2.). The recursive subdivision methods are based on some type of
pure geometric primitives. And, with each step of subdivision, the method loses information
on the derivational continuities and subsequently the surface’s topological features too. It is
simple and computationally efficient to continue the subdivision till the problem reduces to
a simple optimization problem within specified tolerances. However, this restricts the use of
166 R. SHARMA AND O. P. SHA

recursive subdivision methods to only a class of surfaces where derivational continuities are
not of much importance (i.e. planar, flat, nearly flat surfaces, or surfaces having low Gaus-
sian or principal mean curvatures). Since these types of surfaces constitute only a small
type of general surfaces in engineering sciences, this is a serious restriction. The subdivi-
sion can be based upon topological features that maintain not only derivational continuities
but also convexity and other topological properties. This is desired in engineering sciences,
and allows for a wide variety of surfaces to be included in the surface/surface intersection
algorithm.
Almost all the methods mentioned above deal with rectangular surfaces defined over rec-
tangular domains. Both the methods (i.e. curve tracing and ‘loop detection’ technique) deal
with prior setting of a few tolerances, a process that is dependent on the geometrical fea-
tures of surfaces, and is always difficult to handle in semi-automatic or fully automatic
computation process. Because of this, in full generality, the surface/surface intersection
problem based on either curve tracing or ‘loop detection’ techniques is a difficult problem,
and hence no infallible method is available either in general context or in general settings of
derivational continuities. Any algorithm involving tolerances can be defeated by suffi-
ciently ill-chosen examples. Thus, the surface/surface intersection problem is user-
specified, tricky, and involves sufficient, correct and flexible user inputs. More comprehen-
sive review on iterative and subdivision-based methods can be found in [23, 24] and on sur-
face/surface intersection algorithms in [10–12, 20, 25, 26].

3. The tracing method

In this section, we present the surface/plane intersection algorithm considering it as a spe-
cial case of general surface/surface intersection problem. The algorithm used in this work is
based upon three basic ideas: computation of starting point to initiate the algorithm (e.g. [14
and 24]), topological subdivision of the surface to improve upon the aspect ratios of the
discretized triangular elements/patches (e.g. Goldman [27]), and tracing along intersection
curve in the direction of tangent vector in an ordered and structured manner (e.g. [12]). The
structured algorithm includes two processing steps for obtaining intersection approxima-
tions: getting starting points on intersection curve, and trace along intersection curve. The
compository structure of the algorithm used is the following:

1. Getting start points

¡ Subdivide the surface patches.
2. Trace along intersection curve
l Get step vector.
l Relax point to an intersection.
l Relax points on to boundaries if desired.
l Identify branch points and tangent points if desired.
l Tolerances.
l Repeat for all start points.

The implementation program of the algorithm is briefly presented in the next section.
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 167

3.1. Getting start points

Following [14, 24], let the equation of a surface parametrized in independent parameters u,
and v be
SS (u, v) (1)
with constrained parameters
u1 ≤ u ≤ u2 (2)
v1 ≤ v ≤ v 2 (3)
and let the equation of a plane defined as a fixed domain surface be
SPl (l, m) (4)
with constrained parameters
l1 ≤ l ≤ l2 (5)
m1 ≤ m ≤ m2. (6)
It may be noted here that SPl (l, m) is a planar and flat surface and has been defined as a C0
continuous surface. SPl (l, m) may be defined as C1 or C2 continuous surface, but this will
not change the nature of the numerical computations since a fixed domain surface for a
plane will remain planar irrespective of the continuities imposed. Therefore, a planar sur-
face as C1 or C2 continuous surface will only affect a few entries via the incorporation of a
constant valued column or row in the system matrices. This will not affect computations
per se, and only increase the computational space.
For numerical computation, let us reparametrize the above constraints by the introduction
of new generalized coordinate χi, in order that an inequality constraint in the form
Oimin ≤ O ≤ Oimax (7)
can be parametrized as

 O max + Oimin   Oimax − Oimin 

Oi =  i  +   * sin χ i (8)
 2   2 

 O max + Oimin   O max − Oimin 

where  i  is the midpoint and  i  is the half range of the inequality
 2   2 
constraint. The intersection problem is considered as solving a system of seven nonlinear
equations, in eight variables, and with the constraint function as

 S S − S Pl (l, m) 
u − [(u + u ) / 2 + sin( χ ) * (u − u ) / 2] 
 1 2 1 2 1 
ψ (Ρ ) = v − [(v1 + v2 ) / 2 + sin( χ 2 ) * (v2 − v1) / 2]  = 0 (9)
 
l − [(l1 + l2 ) / 2 + sin( χ 3 ) * (l2 − l1) / 2] 
m − [(m1 + m2 ) / 2 + sin( χ 4 ) * (m2 − m1) / 2]
 
168 R. SHARMA AND O. P. SHA

where P = [u v l m χ1 χ2 χ3 χ4] and the inequality constraints of eqns (2), (3), (5), and (6)
are parametrized as per eqn (8).

3.1.1. The Moore-Penrose pseudo-inverse method: The solution is desired for

ψP∆P = –ψ (10)
∂ψ
where ψP is defined as the Jacobian of the constraint, i.e. ψ Ρ = .
∂Ρ
There exist two different possibilities,
• If the constraint subJacobian matrix is square, then eqns (9) and (10) comprise the con-
ventional Newton–Raphson iteration method (i.e. with quadratic convergence properties,
e.g. Markot and Magedson [28]).
• If the constraint equation (9) has more rows than columns, and the constraint subJacobian
ψP has more columns than rows, and vice versa, then (10) will have more than one solu-
tion. Using the principle of minimum norm, (Haug et al. [29], and Noble and Daniel
[30]); i.e. to find the solution of the minimization problem;

min 1 ∆ΡT ∆Ρ (11)

2
ψP∆P = –ψ. (12)
Applying Lagrange multiplier approach (i.e. appending a multiplier vector number of times
the equation to be satisfied in the minimization function), we define

ς = 1 ∆ΡT ∇Ρ − ϕ T (ψ Ρ ∆z + ψ ) . (13)
2
For a solution to eqn (13) to exist, the gradient of the function in eqn (13) should be zero, i.e.
∆PT = ϕT.ψP. (14)
Transposing both sides in eqn (14),
∆P = ψ ΡT .ϕ , (15)
and substituting in eqn (12),
ψ Ρψ ΡT ϕ = −ψ . (16)

If the subJacobian ψP has full row or column rank, then the coefficient matrix on the left
hand side of eqn (10) is positive, definite, and nonsingular, and therefore
ϕ = (ψ Ρψ ΡT )−1(−ψ ) . (17)

Substituting in eqn (15)

∆Ρ = ψ ΡT (ψ Ρψ ΡT )−1(−ψ ) . (18)

The coefficients of the right-hand side in eqn (18) are called Moore–Penrose pseudo-
inverse of the subJacobian ([30]). The equation can also be written as
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 169

∆Ρ = ψ Ρ* (−ψ ) (19)

where ψ*P is the Moore–Penrose pseudo-inverse of the Jacobian ψP, and is defined as
ψ *P = ψ ΡT (ψ Ρψ ΡT )−1 . (20)

For numerical computation, let us define,

ρ ≡ (ψ Ρψ ΡT )−1(−ψ ) (21)

which is equivalent to the solution

(ψ Ρψ ΡT ) ρ = −ψ . (22)

Solving eqn (22) numerically for ρ, and substituting in eqn (18),

∆Ρ = ψ ΡT ρ . (23)

With the specification of initial Pi, the generalized set of coordinates is computed as

Pi + 1 = Pi + ∆P. (24)

The method described above has the same quadratic convergence as the standard Newton–
Raphson method (e.g. [21]). And, using this method, the starting point P0 is computed with
smaller number of iterations (i.e. 4–6), but it gives only one starting point that is nearest to
the user-specified initial point, Pi.

3.1.2. Subdivide the surface patches: In engineering sciences (i.e. aircraft, mechanical
and ship sciences), a surface is designed/modeled/generated with a set of discrete input
points called control points (i.e. control net or offset points). Derivational continuities do
not ensure that the surface will have a fair distribution of aspect ratios of the discretized
elements. Hence, there is no practical control over the aspect ratio of the elements gener-
ated in the surface. Furthermore, for surfaces of underwater or surface floating bodies, the
flow changes drastically in the normal direction to the surface and little in the tangential
longitudinal direction. This forces the discretized elements to have extreme side ratios, e.g.
Bertram [31]. Again, in the case of some surfaces the production features of the surface also
show drastic change from one end to another and that forces discretized elements to have
poor aspect ratios. This means that in some cases of surface generation process, the poor
aspect ratio of individual domain cannot be avoided.
The poor aspect ratios of long, narrow, and thin elements/patches often affect the stabil-
ity, efficiency, and robustness of the solution process. It is not only difficult to compute ei-
ther numerical or computational objective function within a domain or boundary of a long,
narrow, and thin element/patch, but even the simple geometric and algebraic function tends
to fail to converge for a long, narrow, and thin element/patch. Therefore, in the present
work the subdivision is used selectively, and for the following purposes: The long, narrow,
thin and patches with poor aspect ratio are subdivided explicitly before computing sur-
face/plane intersection.
170 R. SHARMA AND O. P. SHA

P2, 0
v P1, 0 P1, 2
P0,1
u
P0, 0
(b) Bézier surface control point P0, 2
FIG. 1. A simple triangular Bézier surface and corresponding control net.

Following Farin [2, 32], let the surface be parametrized by two parameters u and v and a
weight function Bid, j (u, v) describes the influence of a control point Pi,j on the shape of sur-
face, where d is the degree of the surface. The triangular rational Bézier surfaces are de-
fined as follows:
d d− j

∑∑ P
j =0 i = 0
d
i , j wi , j Bi , j (u, v)

P(u, v) = d d− j
(25)
∑∑ w
j = 0 i =0
d
i , j Bi , j (u, v)

with the corresponding Bézier weight functions:

Bid, j (u, v) = d! u i v j wk with w = 1 – u – v, k = d – i – j. (26)

i! j!k!

The each control point on the surface is assigned a weight wi,j for better control. In case all
the weights are constrained to be positive, then the two properties hold: the surface lies
completely in the convex hull of its control points, and the surface can be repeatedly subdi-
vided into subsurfaces of the same kind, e.g. [2] and [32]. Figure 1 shows a simple Bézier
surface and the corresponding control net.

The triangular surfaces can be subdivided with an extended form of the de Casteljau al-
gorithm, or any of its variant [33]. The triangular control point net is recursively subdivided
with trilinear interpolation for a parameter set, and the result being control net for the sur-
faces and a point on the surface. Furthermore, the geometric and topological properties are
computed as for the initial surface. The successive recursion to the resulting subsurfaces,
with de Casteljau algorithm leads to long, narrow, and thin triangular patches with poor as-
pect ratio, often causing numerical and computational problems [34]. Thus, the de Casteljau
algorithm is not practical for use in this work.

Goldman [27] presents a general subdivision algorithm, based on arbitrary subdivision

that allows explicit control over the aspect ratio. The idea is: each of the three corner points
Pj(uj, vj), j = 1, 2, 3 of an arbitrary triangular subsurface can be computed by evaluating a
sequence of d de Casteljau steps Cd ij (u j , v j ), i = 0, ... , d, where d is the degree of the sur-
face. The selection of an arbitrary step from among the three Cd ij (u j , v j ) for each of the d
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 171

P0 (u0 , v0 )
(u 0 , v 0 ) P2 (u2 , v2 )
(u 2 , v 2 )
P1 (u1 , v1 )

(u1 , v1 )

(a) Arbitrary parameter domain (b) Surface and control points for arbitrarily
subdivision of triangle. parametrically divided triangle.
FIG. 2. Arbitrary parameterization and corresponding subsurface and its control net.
steps of the sequence will compute a control point of the subtriangle. The computation for
all possible permutations of this step sequence will give all the control points of the triangu-
lar subsurface, and the geometric and topological properties are computed as for the initial
surface (Fig. 2).
This process is known as blossoming principle after Ramshaw [35], and was independ-
ently developed by de Casteljau [33]. In this work, we have used techniques described in
Goldman [27] to arbitrarily subdivide the parametric surface to improve upon the aspect ra-
tios of the discretized triangular elements/patches of the parametric surface SS(u, v), and
correspondingly the geometric and topological properties are computed.

3.2. Trace along intersection curve

3.2.1. Getting step vector: After getting start point P0 and selectively subdividing the sur-
face SS(u, v), the corresponding intersection curve is traced in the direction of tangent vec-
tor along intersection curve. For a step direction, it is necessary to determine a step distance
at which a new point (i.e. approximate) will be obtained. In general, step length is important
because a too small step length decreases the computational efficiency and a too large one
may miss the points on the intersection curve. In order to achieve balance in these two con-
flicting requirements, adaptive step length based on curvature estimates at intersection
points is used in this work.
In the case of nonparallel surface tangent planes, step vector direction is computed by the
crossproduct of the plane-surface normal. The tracing moves by stepping a distance in the
direction (i.e. +ve or –ve) along this vector (Fig. 3). Instead of crossproduct, another calcu-
lation that is computationally efficient is that for a given Jacobian matrix (JC) of the tangent
plane on one surface (i.e. SS), and the plane normal of the other surface (i.e. SPl) tangent
plane NPl, the intersection vector between these planes is represented by (JC∆(u,v)SS)⋅NPl = 0),
where ∆(u,v)SS is the domain gradient of the first surface (i.e. SS) corresponding to the inter-
section vector. The expansion will give the intersection vector;

S Sv .Ν Pl
I SV = S Sv − .S (27)
S Su .Ν Pl Su

where SSu and SSv are the partial derivatives of the surface SS at the previous point (i.e. P0).
172 R. SHARMA AND O. P. SHA

∆θ
Ρi +1

I SL

Ρi
ISL
ISV
Ρi Ρi+1

FIG. 3. The stepping vector and stepping length. FIG. 4. The approximation of stepping length.

In the case of parallel surface tangent planes (i.e. in the case of surfaces having tangent
cracks, or C 0 continuity), the computation of step vector is difficult. One possible approach
that has been utilized in the present work is to search coordinate points radially about the
starting point. It means that a new point is computed at a small distance ε from the start
point (i.e. P0) on the intersection. The search is done radially in each local parametric do-
main till a new point is computed. The new tangent vector then is computed as the differ-
ence between the new and the starting point (i.e. P0).

The next step is to compute step length. The method used in this work is adaptive and is
based upon approximate curvature and an angle tolerance with curve refinement tolerance
(CRT) as a minimum bound width for stepping distance, e.g. [11, 12]. As shown in Fig. 4,
for the computation of radius of curvature, an osculating circle is approximated at a given
point P.

The circle is defined by P, and two neighboring points at fairly small distances from P in
each direction along the tangent vector. The neighboring points are then relaxed to the in-
tersection using the methods described in Section 3.2.2. Within the computational limit,
these three points describe the true osculating circle. For these three points as computed, the
radius of their circumscribed circle that is assumed to be the approximate radius of curva-
ture is, e.g. [26];

Q – Ρ.R – Ρ. Q – Ρ − R – Ρ
f= . (28)
2. Q – Ρ × R – Ρ
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 173

In the case of surfaces with tangent plane parallelity, and possibly in some other cases,
when the three points are collinear, or if step length ISL > CRT, then CRT is chosen as the
step length. The step length (i.e. arc length ISL) is then approximated to ISL = f.∆θ. The
value of angle tolerance (i.e. ∆θ) is defined by the user.

3.2.2. Relax point to an intersection: The refinement (i.e. relaxing approximate points on
to intersections) is computed using linear iterative techniques as described in [12]. The ini-
tial approximate point is the tracing point with approximate domain and range values. For
an approximate intersection point P ∈ ℜ3, and coordinate pairs (u, v) and (l, m) in the do-
main of surfaces SS(u, v) and SPl(l, m) ∈ ℜ3 respectively, the aim is to relax P on to a true
intersection. The point P may not in general be lying on any surface. The one possible
method for refining P is to first compute surface ‘near point’, and then evaluating the lin-
earized surface equations at these points. The process will continue till convergence is
achieved, (i.e. ||SS(u, v) – SPl(l, m)||2 〈 SPT, where ||⋅|| is the Euclidean vector norm, and SPT
is the same point tolerance ([11]).
The surface near points is defined as points, one on surface, which is at a minimum dis-
tance from P. The point on surface SS(u, v) that is nearest to P is determined by minimizing
the norm of P – SS(u, v). The linearized form is computed as Taylor expansion, as:
 ∂S S ∂S S  ui +1 − ui 
   = [Ρ − S S (ui , vi )]. (29)
 ∂u u i , vi
∂v
 vi +1 − vi 
u i , vi 

The matrix (i.e. 3 × 2) on the left-hand side in eqn (29), which contains column vectors, is
the Jacobian transformation JC. Since the Jacobian matrix is not rank deficient, the least
square solution (i.e. 2-norm minimum) can be computed uniquely by inverting the Jacobian
using its pseudo-inverse;

[ui +1 − ui ] = [ JcT Jc]−1 JcT [Ρ − S S (ui , vi )]. (30)

The iterations (i.e. after 3–4 iterations in general) converge fast because of the quadratic
convergence of Newton methods. When the surfaces near points are not within the SPT
then the linearized forms of the surface equations are computed again. And, SS – SPl = 0, at
surface points (u, v) and (l, m) near to Pi, are tackled directly. The tangent planes at the sur-
face points are intersected with a third nonparallel plane to uniquely define a new point Pi+1.
There exist many methods to address this, e.g. [7, 11, 26]. The computationally efficient
method is to use the mid-point of the projection of the surface near points on to the inter-
section line of the tangent planes to describe the next iterate (Fig. 5).
If NS and NPl are the tangent plane normals at the surface near points, and NSPl = NS × NPl
is the third tangent plane normal, then the next iterate Pi+1 is computed from

 ΝTS   Ν S .S S 
 T   
 Ν Pl  [Ρi +1 ] =  Ν Pl .S Pl . (31)
 T   Ν SPl .( S S + S Pl )/2 
 Ν SPl 
174 R. SHARMA AND O. P. SHA

ΝS

Ν SPl
Ν Pl

Ρi+1
S Pl

SS
FIG. 5. The tangent planes on the surfaces, SS and SPl.

The convex combination of the two projected points is sufficient for convergence (e.g.
Mullenheim [36]). This means that the desired element (i.e. third) of the righthand side ma-
trix in eqn (31) can generally be taken as NSpl⋅(t⋅SS + (1 – t)SPl), where 0 ≤ t ≤ 1. In the pre-
sent work, as given in eqn (31), we have taken t = 1/2.

The methods used above may be problematic in the case of nearly or the same parallel
tangent planes at surface near points. If the planes are the same at two surfaces near points,
then there is an intersection, and computationally it should have been computed with SPT,
before refinement. If the tangent planes are parallel, then assuming twice differentiable sur-
face equation will lead to linear convergence. Though, theoretically in this case, the solu-
tion is not possible, but within SPT, a root shall always be computed; and that is sufficient
for computational purposes.

3.2.3. Relax points on to boundaries: Following [11], let the initial domain values be v0,
l0, m0, and the constant isoparametric u = u on the surface SS (i.e. the u = 0, or u = 1
boundary of the triangular domain), then a new iterate can be computed by minimizing the
norm of SPl(l, m) – SS(u, v). Computing the linearized form will result in 3 × 3 Newton itera-
tion on v, l, m, meaning;

v − v 
 ∂S S ∂S Pl ∂S Pl   i +1 i 
−  li +1 − li  = [S S (u , vi ) − S Pl (li , mi )] (32)
 ∂v u , vi
∂l l i , mi
∂m li , mi 
 m − m 
 i +1 i

with the termination of iteration at || S S (u , vi +1) − S Pl (li +1, mi +1) ||2 〈 SPT. Alternatively, in a
more computationally efficient manner, it is possible to incorporate u = constant constraint,
directly into the system, and this results in a 4 × 4 system matrix, in four unknowns. Let
∆u = u2 – u1, and constraint u2, as u2 = ∆u + u1 = a, where a is a constant, for a boundary
curve with a = 0 or a = 1, then ∆u = a – u1, and the system matrix is;

 ∆u 
 ∂S S ∂S S ∂S Pl ∂S Pl   ∆v 
− −  S (u , v ) − S Pl (li , mi ) 
 = S i i
 ∂u ui ,vi ∂v ui , vi ∂l li , mi ∂m li , mi 
 ∆l    (33)

a – ui 
 1 0 0 0   
 ∆m 
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 175

Similarly, for u + v = 1 boundary of the triangular domain of surface SS, the system is al-
tered slightly for computational easiness. In that case, the sum of equations u2 = ∆u + u1 and
v2 = ∆v + v1, is used to get ∆u + u1 + ∆v + v1 = 1. In other words, ∆u + ∆v = 1 – u1 – v1, and
the corresponding system matrix is;

∆u 
 ∂S S ∂S S ∂S Pl ∂S Pl  
− ∂u −
∂v ∂l ∂m  ∆v  S S (ui , vi ) − S Pl (li , mi )
=
∆l   .
 u i , vi u i , vi li , mi li , mi  (34)
1 − ui − vi 
 1 1 0 0   
∆
 
m

The methods used above may be problematic in the case of nearly or the same parallel tan-
gent planes at surface near points. In that case, similar techniques as mentioned before are
used.

3.2.4. Termination of intersection curve: In this work, the intersection curve terminates,
when,

a) The intersection curve being traced meets a boundary. In this case, the methods de-
scribed in Section 3.2.3 are applied.
b) The intersection curve being traced meets an earlier intersection curve.

The second situation in general points to the existence of a ‘branch point’, where more than
one intersection curve meet. In general, branch points are the points where surfaces share
tangent plane (i.e. tangent points) [12]. In the present work, we only detect this problem,
but have not addressed this. Hence, the intersection curve terminates even when it meets a
branch point.

3.2.5. Tolerances: The basic tolerance used in this work is SPT. Barnhill and Kersey [12]
examine whether the two points (i.e. P1 and P2) are the same or not and base the answer on
the Euclidean distance between the points. In this work, SPT has been chosen to be
SPT = 1.0e – 8. However, for engineering applications, SPT can be chosen anywhere be-
tween 1.0e – 4 and 1.0e – 8. Another tolerance used in this work is CRT, which is used as a
prescribed stepping distance. This distance measurement is dependent on the relative mag-
nitude of surfaces. As mentioned in Section 3.2.1, angle tolerance has been used, along with
the estimation of curvature of intersection curves, and is incorporated in adaptive step
length estimations. An adaptive step length provides better control over efficiency and ro-
bustness, and the angular tolerance positively affects the dependency on relative surface
magnitudes and units of measurements. Hence, this combination is better suited for general
surface/plane intersection problems.

3.2.6. Repeat for all start points: Once the intersection curve corresponding to the starting
point (i.e. P0) has been traced exhaustively (using steps from Section 3.1–3.2.1–3.2.4), then
within different settings, another starting point is computed as mentioned in Section 3.1,
and the whole process is run again.
176 R. SHARMA AND O. P. SHA

In this work, the intersection curves have been computed at different X = constant,
Y = constant, and Z = constant planes for different surfaces (i.e. surfaces having C0, C1, and
C2 continuities). The aim is to obtain the smallest number of ordered points that correctly
represent the topology of the true intersection curve. In this work, we compute two points
(minimum) that are desired even for an intersection curve that is linear. Since we have used
cubic surfaces (i.e. d = 3 the most popular degree of surfaces in engineering sciences), the
generated surfaces consist of 100 elements within a patch (i.e. clearly shown in Fig. 7(a),
10(a), and 12(a)). So, an arbitrary plane may pass through all the ten triangular elements in-
side one patch, though ideally one must compute three points on each triangle (i.e. two at
the local boundary and one inside/at the triangle). This is computationally expensive, and
reduces the speed of computation, and in practice really does not improve the quality or to-
pology of the intersection curve. Since in this work the intersection curves are represented
by their piecewise polynomial approximations, the computation of one point inside/at the
triangle is sufficient for applications in engineering sciences. Therefore, we compute one
intersection point on each triangle of each patch. Again, since the exact computation of the
interior edge or line of derivational continuity, where it is imposed, is essential in engineer-
ing sciences for surface-related analysis, we compute one point on the interior edge, or line
of derivational continuity and re-check via backtracking. Hence, for a two-patch surface we
compute total 23 numbers of points (i.e. one starting point on the surface’s exterior bound-
ary + 2*10 + one point on the interior edge/line of derivational continuity + one end point
of the surface at the surface’s exterior boundary = 23) for an intersection curve.
The algorithm has been implemented in C++ using its object-oriented features on a Sili-
con GraphicsTM** OriginTM** 200 workstation. The surface views and intersection curves
are represented explicitly in AutoCADTM*** R2000.

4. Numerical examples and discussion

To illustrate the theoretical model of this work, we have considered different problems. The
first problem deals with the intersection curve of C0 – C0 continuities combination. The
second, under the cases (1) and (2), deals with the intersection curve of C1 – C0 and C2 – C0
continuities combinations. For the sake of reproducibility, we give the explicit coordinates
and geometric definition with continuities of the point set.

Problem 1. This is basically a Y = 5 plane, modeled as a flat plate with five points and four
patches. The point set consists of P1 = (5, 5, 5), P2 = (–5, 5, –5), P3 = (–15, 5, 5), P4 = (–
5, 5, 15), and P5 = (–5, 5 ,5). The indices of the vertices of the triangles in geometry defini-
tion F are given by (1, 2, 5), (2, 3, 5), (3, 4, 5), and (4, 1, 5).

Intersection curves: The triangular domain is defined as in Fig. 6. Here C0 continuity has
been imposed across all the four patches (i.e. pa 1–pa 4). In this case, Fig. 7(a) shows the
front view in which the connectivity and triangulation pattern is clear. Figure 7(b) shows
the isometric view.

**Trademark and copyright with Silicon Graphics Corporation, USA.

***Trademark and copyright with AutoDesk Corporation, USA.
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 177

P4 P4

P4 interior edges or lines P1

of derivational
continuities as
imposed P1 P3
pa 4 pa 3
P1 P5 P3
pa 1 pa 2
P3

(a) (b)
P2 P2 P2
FIG. 6. Definition of the triangular domain
FIG. 7. Surface for a flat rectangular plate, C0 continuous.
for flat rectangular plate.

Since the surface is C0 continuous, it is purely flat. We plot the intersection curves of the
surface with constant X planes at the interval of 1.0, from (X = –14.0 to X = 4.0). Figure 8
shows the intersection curves at X planes. The surface is purely flat, so the intersection
curves are constantly linear (i.e. this follows the obvious; any continuous surface for a flat
plate will be a flat surface with linear curves of intersection).

Problem 2. This is a simple problem of a nonflat plate where one corner has been raised
with respect to the plane passing through the three points; and it has been modeled with two
patches. The point set consist of P1 = (0, 1, 0), P2 = (1, 0, 0), P3 = (1, 1, 1) and
P4 = (0, 0, 1). The indices of the vertices of the triangles in geometry definition F are given
by (1, 2, 4) and (2, 3, 4).

Case 1–Intersection curves: The triangular domain is defined as in Fig. 9. Here, C1 conti-
nuity has been imposed across the patches (i.e. pa 1–pa 2). Figure 10(a) shows the front
view in which the connectivity and triangulation pattern is clear. Figure 10(b) shows a typi-
cal isometric view. We have computed the intersection curves of the surface with constant Y
planes at the interval of 0.1, from (Y = 0.1 to Y = 0.9). Figure 11 shows the intersection
curves at constant Y planes. As stated earlier, since the surface is nonflat but is C1 continu-

P4 X = –14.0

P3
X = 4.0
P4 P3
pa 2
X = –14.0 interior edge or line of
derivational continuity as
P1 imposed
pa 1

P2 P1 P2
X = 4.0
FIG. 8. Intersection curves of constant X planes with FIG. 9. Definition of the triangular domain for a nonflat
surface for flat rectangular plate, C0 continuous. rectangular plate.
178 R. SHARMA AND O. P. SHA

P4 P3 P3 P4

P2

(a)
(b)

P1 P2 P1
FIG. 10. Surface for a nonflat rectangular plate with one corner raised, C1 continuous.
Y = 0 .5
P4 P3
Y = 0 .1 Y = 0 .9

Y = 0 .6
Y = 0 .4
Y = 0 .5
Y = 0.6 Y = 0 .4

Y = 0 .1

P1 P2 Fig. 11. Intersection curves of constant Y planes with

surface for a nonflat rectangular plate with one corner
Y = 0.9 Y = 0 .5 raised, C1 continuous.

ous, the intersection curves are continuous (i.e. nonlinear variation) without any break in
continuity.

Case 2–Intersection curves: Here C2 continuity has been imposed across all the patches
(geometric definition as in Fig. 9). Figure 12(a) shows the front view in which the slightly
curved sides, twist in space, higher curvatures, connectivity and triangulation pattern are
visible and clear. Figure 12(b) shows a typical isometric view. As previously, we have
computed the intersection curves of the surface with constant Y planes at the interval of 0.1,
from (Y = 0.1 to Y = 0.9). Figure 13 shows the intersection curves at Y planes. The surface is
nonflat but is C2 continuous, so the intersection curves are highly nonlinear but continuous.

P4 P3 P3 P4

P2

(a)
(b)
P1 P2 P1
Fig. 12. Surface for a nonflat rectangular plate with one corner raised, C2 continuous.
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 179

Y = 0 .5
P4 P3
Y = 0 .1 Y = 0 .9

Y = 0 .6
Y = 0 .4
Y = 0 .5
Y = 0.6
Y = 0 .4

Y = 0 .1
P1 P2 Fig. 13. Intersection curves of constant Y planes with
surface for a nonflat rectangular plate with one corner
Y = 0.9 Y = 0 .5 raised, C2 continuous.

5. Conclusions
This work has presented in general setting of derivational continuities a surface/plane inter-
section algorithm for non self-intersecting rational triangular Bezier surfaces defined over
triangular domain using combination of ideas (i.e. [11, 12, 14, 24, 27]). The algorithm pre-
sented is a stepwise, structured, and combinatorial one. Hence, it allows better insight into
the problem of the computation of surface/plane intersection curve. The method has been
used to compute different planar sections (i.e. intersection curves with X, Y, and Z planes)
for surfaces having different combination of continuities (i.e. C0–C0, C1–C0, and C2–C0 con-
tinuous surface/plane). The algorithm includes the computation of multiple intersection
curves at one plane but does not include the computation of intersection curve with multiple
branches (i.e. branch points), bifurcation points, and lower and higher-order singularities. A
more complex algorithm in general settings of surfaces in engineering sciences will pref-
erably include these features. Again, the detailed analysis of efficiency and complexity of
the present algorithm has not been addressed. Though we have also computed the intersec-
tion curves for parallel planes, the spatial coherency, (e.g. Kulkarni and Dutta [37], Jun et
al. [38]), among the planes, have not been incorporated in the algorithm. Again, the surface
subdivision used in the present work is explicit, and theoretically based upon Goldman [27]
that uses a 5-simplex of polar value numerical computations. However, if the explicit con-
trol over the aspect ratio is not desired, then the same can be achieved in four calls (e.g.
Prautzsch [39] and DeCarlo and Gallier [40]). The incorporation of the four-call procedure
in the surface subdivision algorithm would mean better robustness and computational effi-
ciency. Since in the present algorithm the surface/plane intersection problem has been con-
sidered as a special case of the surface/surface intersection problem in general setting of
derivational continuities, theoretically some ideas of the algorithm may be utilized to com-
pute intersection curves in surface/surface intersection. Furthermore, in the present algo-
rithm, the correct computation of the topological properties of the intersection curve is an
in-built feature, hence a suitable ‘loop detection’ technique can be implemented as an add-
on feature. This is under investigation.

Acknowledgements
The authors thank Prof. R. P. Gokarn, formerly faculty at the Department of Ocean Engi-
neering and Naval Architecture, Indian Institute of Technology, Kharagpur, India, for his
180 R. SHARMA AND O. P. SHA

critical comments and suggestions on the manuscript, which have helped us immensely in
revising the manuscript. They also thank Prof. S. P. Pal, Department of Computer Science
and Engineering, Indian Institute of Technology, Kharagpur, India, for discussions, which
have contributed to improving the clarity of the paper. The authors gratefully acknowledge
their support and encouragement.

References
1. F. E. Wolter and S. T. Tuohy, Curvature computations for degenerate surface patches, Computer Aided
Geometric Des., 9, 241–270 (1992).
2. G. Farin, Curves and surfaces for computer aided geometric design. A practical guide, 3rd edn, Academic
Press (1993).
3. C. L. Bajaj, C. M. Hoffmann, R. E. Lynch and J. E. H. Hopcroft, Tracing surface intersections, Computer
Aided Geometric Des., 5, 285–307 (1988).
4. W. T. Montaudouin, W. Tiller and H. Vold, Applications of power series in computational geometry, Com-
puter-Aided Des., 18, 514–524 (1986).
5. C. Asteasu, Intersection of arbitrary surfaces, Computer-Aided Des., 20, 533–538 (1988).
6. T. Garrity and J. Warren, On computing the intersection of a pair of algebraic surfaces, Computer Aided
Geometric Des., 6, 137–153 (1989).
7. E. G. Houghton, R. F. Emnett, J. D. Factor, and C. L. Sabharwal, Implementation of a divide-and-conquer
method for intersection of parametric surfaces, Computer Aided Geometric Des., 2, 173–183 (1985).
8. D. Filip, R. L. Magedson, and R. P. Markot, Surface algorithms using bounds on derivatives, Computer
Aided Geometric Des., 3, 295–311 (1986).
9. H. Bürger and R. Schaback, A parallel multistage method for surface/surface intersection, Computer Aided
Geometric Des., 10, 277–291 (1993).
10. J. A. Gregory, The mathematics of surfaces (J. A. Gregory, ed.), Clarendon Press (1986).
11. R. E. Barnhill, G. Farin, M. Jordan, and B. R. Piper, Surface/surface intersection, Computer Aided Geomet-
ric Des., 4, 3–16 (1987).
12. R. E. Barnhill and S. N. Kersey, A marching method for parametric surface/surface intersection, Computer
Aided Geometric Des., 7, 257–280 (1990).
13. A. Cugini, S. Radi, and C. Rizzi, A system of parametric surface intersection, in Computer-aided surface
geometry and design. The mathematics of surfaces IV (A. Bowyer, ed.), Oxford University Press, pp. 231–
242 (1994).
14. K. Abdel-Malek and H. J. Yeh, Determining intersection curves between surfaces of two solids, Computer-
Aided Des., 28, 539–549 (1996).
15. K.-P. Cheng, Using plane vector fields to obtain all the intersection curves of two general surfaces, in The-
ory and practice of geometric modeling (W. Strasser and H.-P. Seidel, eds), Springer-Verlag, pp. 187–204
(1989).
16. G. A. Kriezis, N. M. Patrikalakis, and F. E. Wolter, Topological and differential equation methods for sur-
face intersections, Computer-Aided Des., 24, 41–55 (1992).
17. T. W. Sederberg, H. N. Christriansen, and S. Katz, An improved test for closed loops in surface intersec-
tions, Computer-Aided Des., 21, 505–508 (1989).
18. T. W. Sederberg and R. J. Meyers, Loop detection in surface patch intersections, Computer-Aided Des., 5,
161–171 (1988).
19. P. Sinha, E. Klassen, and K. K. Wang, Exploiting topological and geometric properties for selective subdivi-
sion, in Proc. ACM Symp. on Computational Geometry, pp. 39–45 (1985).
20. T. A. Grandine and F. W.-Klein, IV A new approach to the surface intersection problem, Computer Aided
Geometric Des., 14, 111–134 (1997).
 PARAMTERIC BÉZIER TRIANGULAR SURFACE/PLANE INTERSECTION 181

21. G. Müllenheim, On determining start points for a surface/surface intersection algorithm, Computer Aided
Geometric Des., 8, 401–408 (1991).
22. T. Dokken, Finding intersections of B-spline represented geometries using recursive subdivision techniques,
Computer Aided Geometric Des., 2, 189–195 (1985).
23. T. Dokken, V. Skytt, and A. M. Ytrehus, Recursive subdivision and iteration in intersections and related
problems, in Mathematical methods in computer aided geometric design (T. Lyche and L. L. Schumaker,
eds), Academic Press, pp. 207–214 (1989).
24. K. Abdel-Malek and H. J. Yeh, On the determination of starting points for parametric surface intersections,
Computer-Aided Des., 29, 21–35 (1997).
25. G. Farin, An SSI bibliography, in Geometry processing for design and manufacturing (R. E. Barnhill, ed.),
SIAM, pp. 205–207 (1992).
26. I. D. Faux and M. J. Pratt, Computational geometry for design and manufacture, Ellis Horwood (1985).
27. R. N. Goldman, Subdivision algorithms for Bézier triangles, Computer-Aided Des., 15, 159–166 (1983).
28. R. P. Markot and R. L. Magedson, Procedural method for evaluating the intersection curves of two paramet-
ric surfaces, Computer-Aided Des., 23, 395–404 (1991).
29. E. J. Haug, C. M. Luh, F. Adkins, and J. Y. Wang, Numerical algorithms for mapping boundaries of manipu-
lator work-spaces. Proc. 24th ASME Mechanisms Conf., USA (1994).
30. B. Noble and J. W. Daniel, Applied linear algebra, Prentice-Hall (1988).
31. V. Bertram, Practical ship hydrodynamics, 2nd edn, Butterworth-Heineman, pp. 34–37 (2000).
32. G. Farin, Triangular Bernstein-Bézier patches, Computer Aided Geometric Des., 3, 83–127 (1986).
33. P. de Casteljau, Shape mathematics and CAD, Kogan Page (1986).
34. W. Stürzlinger, Ray-tracing triangular trimmed free-form surfaces, IEEE Trans. Visualization Computer
Graphics, 4, 202–214 (1998). web site address: http://citeseer.nj.nec.com/92665.html (2001).
35. L. Ramshaw, Bézier and B-splines as multiaffine maps, in Theoretical foundations of computer graphics
and CAD (R. Earnshaw, ed.), Springer-Verlag, pp. 757–776 (1988).
36. G. Müllenheim, Convergence of a surface/surface intersection algorithm, Computer Aided Geometric Des.,
7, 415–423 (1990).
37. P. Kulkarni and D. Dutta, Adaptive slicing of parameterizable algebraic surfaces for layered manufacturing,
Proc. ASME Design Technical Conf., Boston, MA, September 1995 (1995).
38. Cha-Soo Jun, Dong-Soo Kim, Deok-Soo Kim, Hyun-Chan Lee, Jiseon Hwang, and Tien-Chien Chang, Sur-
face slicing algorithm based on topology transition, Computer-Aided Des., 33, 825–838 (2001).
39. Helmut Prautzsch, Unterteilungsalgorithmen für multivariate splines, Ph D thesis, TU, Braunschweig,
Braunschweig, Germany (in German), (1984).
40. D. DeCarlo and J. Gallier, On the efficiency of strategies for subdividing polynomial triangular surface
patches, Department of Computer and Information Science, University of Pennsylvania, Philadelphia, USA,
pp. 1–19, web site address: http://www.cis.upenn.edu/~jean/home.html (1999).

Nomenclature
JC = Jacobian matrix
ε = small distance in Euclidean or parametric space
SS(u, v) = surface parametrized in independent parameters (u, v)
SPl(l, m) = plane defined as a fixed domain surface parametrized in parameters (l, m)
P0 = starting point on the intersection curve of surface/plane intersection
Pi = point on the intersection curve of surface/plane intersection
182 R. SHARMA AND O. P. SHA

ISV = intersection stepping vector

ISL = intersection stepping length
∆θ = user-specified angle tolerance
ψ* = Moore-Penrose pseudo-inverse of the Jacobian
NPl = tangent plane normal of surface SPl(l, m)
NS = tangent plane normal of surface SS(u, v)
NSi = surface normal for surface SSl(u, v) at point Pi
NPl
i = surface normal for surface SPl(l, m) at point Pi
SPT = user-specified same point tolerance
CRT = user-specified curve refinement tolerance