You are on page 1of 4

Seismic Theory 10: 3-D Model Building and Ray Tracing

Tuesday p.m., Nov. 4


ST 10.1
Automatic Reconstruction of 3D Complex Geologic Model from a Set of Polylines
Toshi Chang, Luis Canales and Chung-Chi Shih, CogniSeis Development, Inc.

Summary Delaunay triangulation on the scattered data points. The

Downloaded from http://onepetro.org/SEGAM/proceedings-pdf/SEG97/All-SEG97/1947224/seg-1997-1707.pdf by University of Baghdad, Hussein Y. Ali on 01 July 2021


second step is to successively remove the triangles or
tetrahedra ( to sculpt ) from the convex hull to the objective
We introduce a new approach to automatically construct a 3D boundary of the 3D model. In both methods, described
geologic model from a series of polylines, The main above, the generated 3D model will cause an error when the
contribution of this work is the rapid and precise computation data points representing the surface are not dense. However,
of the 3D model, which might be given by a set of digitized a 2D model can be triangulated and shrunk to an exact 2D
data from a seismic section or wells with a multi-valued shape, if it is constrained by a set of edges (or segments) as
surface. Our motivation comes from the major problem in shown in figures, [1], [2], [3], [4] and [5].
most of the methods of CAD (computer aided design)
models based on an unorganized collection of points in
space. The generated 3D model can not be corrected if the
digitized data are extremely complex. Even if the generated
model is corrected, such meshes are notoriously expensive to
store, transmit, and render. On the contrary, our algorithm to
generate a 3D model, does not need the digitized data to be
fine enough to represent the 3D surface. Compared to other
methods, ours has the advantages of simplicity, efficiency
Figure [1] Figure [2]
and uniformity. The first step of our approach is to generate a
convex hull by applying 2D Delaunay triangulation. The
second step is to shrink the convex hull to a 3D object by the
implementation of 2D local triangulation, constrained by the
given set of polylines. The algorithms are illustrated step by
step by a mushroom model with a closed surface, and then
are tested with two complex geologic models with an open
surface.

1.1 Introduction
Figure [3] Figure [4]
The reconstruction of a 3D complex model from interpreted
seismic data has been an intriguing problem during the last
decade. The main motivations for this problem come from
the applications such as the study of salt and salt tectonics,
which will make themselves more effective in oil
exploration. Data obtained by the digitization of 3D seismic
section can be triangulated in order to represent a 3D model.
When the model has a simple shape, the triangulated meshes
can be easily generated. However, if a set of digitized data
corresponds to a multi-valued surface such as a salt dome, Figure [5]
no method relying on projection can be applied.
Edelsbrunner and Mucke [1] tried to define the geometrical The 3D case is more complicated. Each set of points can be
data structure by the notion of “shape” for a set of points. triangulated, but there are polyhedra that can not be filled
Starting from the Delaunay triangulation, an alpha coefficient with given tetrahedra or triangles. It is very difficult to build
is associated with each point, edge, triangle and tetrahedron. a 3D model based on the given shape of triangulated meshes.
One can imagine of a ball-shaped eraser, whose radius is Therefore, to tetrahedralize a geometric structure involving
moved in the space. For alpha equals infinite, the shape of a vertices, edges, and triangles, Steiner points must be added in
3D object is identical to the convex hull. However, as alpha the general case [3]. Their number can be very large,
decreases, the shape shrinks by gradually developing cavities. resulting in a far more complex tetrahedralization.
Unfortunately, based on our experience, it is difficult to Boissonnat proposes another surface-based approach to build
define a continuous triangulated boundary from a set of a 3D model by means of a local procedure [3]. However, the
points. proposed method does not guarantee a satisfactory
Boissonnat [2] proposed a volume-based approach to build a triangulation of a 3D surface even if the discretization is fine
3D object. The first step of this approach is to implement enough.
1707
Automatic Reconstruction of 3D Model
In the following description, we propose a method to After all the points in 3D space have been checked and
circumvent this problem by applying constrained triangulated, we are able to generate a convex hull with all
triangulation on a 3D surface. Rather than implementing 3D the points either inside or on the surface of the convex hull,
triangulation and removing the tetrahedra step by step, we as shown in figure [10].
implement 2D Delaunay triangulation to generate the convex
hull. We then implement local trigulation, constrained by the
given polylines,* which will guarantee an efficient and
accurate result. The algorithm is illustrated step by step by a

Downloaded from http://onepetro.org/SEGAM/proceedings-pdf/SEG97/All-SEG97/1947224/seg-1997-1707.pdf by University of Baghdad, Hussein Y. Ali on 01 July 2021


3D mushroom model. Finally, the algorithm is tested with
two complex geologic models, one with an extremely
complex fault, and the others with recumbent salt dome.

2.1 Convex Hull Generation with Delaunay Triangulation

By providing a set of points in space, as shown in figure [6],


one can build a convex hull filled with triangulated meshes. Figure[10]
At the beginning, four triangles is built as a single
tetrahedron, which contains four vertices but not any three 2.2 Local Triangulation Constrained by Polylines
points on a line, as shown in figure [7]. For any point outside
the existing polyhedron, we add and locally triangulate on the A polyline is generated, if a set of points in 3D space is in a
3D surface, as shown in figures [8] and [9]. The triangulation sequential order as shown in Figure[11]. We are able to use
scheme we use satisfies the Delaunay criterion; that is, there sequentiat points as a guide to ensure the proper triangulation
is no point in space defined by each circumscribing sphere in the 3D model.
associated with each triangle.

Figure [6]
Figure [11]

The first step is to compute the normal vector from the given
polylines. For each point in 3D space, we assume the vector
is orthogonal to the 3D object and intersected with a specific
triangle on the convex hull. The normal vector gives us the
direction to apply local triangulation. Note that the lines
projected from a normal vector should not intersect with its
polyline. Therefore, our algorithm relies on checking of the
Figure [7] Figure [8] minimum distance of a query point inside the convex hull.
Each time, we select a point with the minimum distance to
the convex hull, and locally triangulate to the 3D surface.
Rather than building a Delaunay triangulated surface, we
implement 2D local triangulation, constrained by polylines.
This is to ensure that the edges of triangular meshes are
constrained by the polylines. Figures [12], [13], [14] and [15]
illustrate the shrink processing to obtain a 3D model from a
convex hull.
Figure [9]

1708
Automatic Reconstruction of 3D Model
Implementing the same algorithm as mentioned above, we
are able to generate a convex hull, as shown in figure [17]
The normal vector, provided by a set of polylines are shown
in figure [18]. The convex hull is then shrunk by applying
2D triangulation. Finally, we generate a model with an open
surface, as shown in figure [19]. The open surface is
generated by elliminating the unrelated triangles, based on
the pointing vectors provided by the polylines.

Downloaded from http://onepetro.org/SEGAM/proceedings-pdf/SEG97/All-SEG97/1947224/seg-1997-1707.pdf by University of Baghdad, Hussein Y. Ali on 01 July 2021


Figure [12] Figure [13]

Figure [18] Figure [19]

The second example is a recumbent 3D mode. Figure [20]


displays a set of points in 3D space. We apply the same
procedure, as described in previous example. First, we
generate a convex hull, as shown in figure [21]. The convex
hull is then shrunk to a 3D model, as shown in figure [23],
constrained by a set of polylines, as shown in figure [22]
Figure [14] Figure [15]

2.3 Examples

Our algorithm not only can be used for generating a 3D


model with a closed surface, but also can be used for
generating any model with a multi-valued, open surface.
Traditionally, people perfer to partition several patches
extracted from a given set of points. All the patches are
triangulated separately and then merged together to build a Figure [20] Figure [21]
3D model. This is cumbersome and not easy to handle
correctly on the boundaries among the patches. The
following two examples illustrate how the same algorithm
can be used to simply and efficiently construct a 3D model
with an open surface. Two set of points in 3D space, are
shown in both figure [16] and figure [20] with a multi-
valued surface. There is no direction from which either set of
points could be projected onto a 2D plane.

Figure [22] Figure [23]

Figure [16] Figure [17]

1709
Automatic Reconstruction of 3D Model
4. Conclusion

A 3D model construction constrained by a set of polylines


provides a simple and accurate result. In particular we
provide a boundary-based model without using an irregular
tetrahedra mesh, which will guarantee to generate the
triangular facets to match the surface of the 3D model. Our
algorithm to generate a 3D model does not need a fine

Downloaded from http://onepetro.org/SEGAM/proceedings-pdf/SEG97/All-SEG97/1947224/seg-1997-1707.pdf by University of Baghdad, Hussein Y. Ali on 01 July 2021


discretization on the 3D surface. However, if the model needs
to be finer, recursive subdivision on the triangulated meshes
can be implemented after the shrink.

Acknowledgment

We gratefully appreciate the support and encouragement of


CogniSeis Development, Inc. in providing the resources that
enable the investigation of this methods. In particular we are
grateful to the significant contributions of Mike McCabe and
Chris Todd for there assistance and guidance to provide the
data to demostrate this method.

References

[1] Edelsbrunner, H., and Mucke, E. P. Three-dimensional


alpha shape. ACM Transactions on Graphics 13, 1 (Jan.
1994), 43-72.
[2] Boissonat, J. D. Geometric structures for three-
dimensional shape representation. ACM Transactions on
Graphics 3,4 (Oct. 1984), 266-286.
[3] Nackman, L. R., and Srinivasan,V, Point placement for
Delaunay triangulation of polygonal domains. In Proc. 3rd
Canad. Conf. Comput. Geom, 1991,37-40.

1710

You might also like