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

net/publication/222831890

Topological models and frameworks for 3D spatial objects

Article  in  Computers & Geosciences · May 2004

DOI: 10.1016/j.cageo.2003.06.004

CITATIONS READS
159 858

3 authors:

Sisi Zlatanova Alias Abdul Rahman

UNSW Sydney Universiti Teknologi Malaysia
375 PUBLICATIONS   5,375 CITATIONS    154 PUBLICATIONS   1,187 CITATIONS   

SEE PROFILE SEE PROFILE

Wenzhong Shi
The Hong Kong Polytechnic University
206 PUBLICATIONS   4,805 CITATIONS   

SEE PROFILE

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

4TU.Bouw ExcaSafeZone View project

A Project of Shandong Province Higher Educational Science and Technology Program (J17KA064) View project

All content following this page was uploaded by Sisi Zlatanova on 01 November 2018.

The user has requested enhancement of the downloaded file.

 ARTICLE IN PRESS

Computers & Geosciences 30 (2004) 419–428

Topological models and frameworks for 3D spatial objects

Siyka Zlatanovaa,*, Alias Abdul Rahmanb, Wenzhong Shic
a
GIS Technology Section, Delft University of Technology, Thijssweg 11, 2629 JA, Delft, The Netherlands
b
Center for Geographic Information and Analysis (CGIA), Universiti Teknologi Malaysia, 81310 UTM Skudai, Johor, Malaysia
c
Advanced Research Centre for Spatial Information Technology, Department of Land Surveying and Geo-Informatics, The Hong Kong
Polytechnic University, Hung Hom, Kowloon, Hong Kong
Received 9 December 2002; accepted 2 June 2003

Abstract

Topology is one of the mechanisms to describe relationships between spatial objects. Thus, it is the basis for many
spatial operations. Models utilizing the topological properties of spatial objects are usually called topological models,
and are considered by many researchers as the best suited for complex spatial analysis (i.e., the shortest path search). A
number of topological models for two-dimensional and 2.5D spatial objects have been implemented (or are under
consideration) by GIS and DBMS vendors. However, when we move to one more dimension (i.e., three-dimensions),
the complexity of the relationships increases, and this requires new approaches, rules and representations. This paper
aims to give an overview of the 3D topological models presented in the literature, and to discuss generic issues related to
3D modeling. The paper also considers models in object-oriented (OO) environments. Finally, future trends for
research and development in this area are highlighted.
r 2004 Elsevier Ltd. All rights reserved.

Keywords: 3D topological models; 3D data structures; Object-oriented models; 3D analysis; Spatial operations

1. Introduction Two types of models (geometrical and topological) have

been examined in many studies. The geometrical models
Currently, geographic information system GIS pro- are more intuitive and easier to implement. Several
fessionals and users are content with the capability of mainstream DBMSs (Oracle, Ingres, Informix, IBM and
existing GISs (i.e., two-dimensional (2D) GISs). These DB2) support spatial objects organized in geometrical
systems can perform numerous 2D spatial analyses and models. Some of them even follow the Open GIS
applications. The Open GIS Consortium has agreed on standards. Many GIS and CAD packages (MapInfo,
Simple Feature Specifications (geometry) and Complex ArcGIS, MicroStation, AutoCAD) use geometrical mod-
Feature Specifications (topology). The first implementa- els of DBMS (Zlatanova et al. 2002). Most geometric types
tions of the OpenGIS, SQL/SFS (which became avail- supported by DBMS can display 3D spatial objects as 2D
able in 1999), marked an important step forward in the objects with 3D coordinates, but their spatial operations
development of GIS, and OpenGIS became a part of the are still 2D. Real 3D objects and their corresponding
mainstream ICT. As the world we are living in has three validation functions remain to be implemented.
or more dimensions, we have to manipulate three- The evolution of topological models into the third
dimensional (3D) spatial data instead of 2D spatial data. dimension is rather complex when compared to the
geometrical models. Many GIS packages construct 2D
*Corresponding author. topological models, and some CAD packages provide
E-mail addresses: [email protected] (S. Zlatanova), tools to check topological consistency (e.g., GeoParcel,
[email protected] (A.A. Rahman), MicroStation), and some mainstream DBMSs have
[email protected] (W. Shi). 2D topological implementations in their development

0098-3004/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cageo.2003.06.004
 ARTICLE IN PRESS
420 S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428

agendas. 3D topology is still being researched. The third 2.1. 3D topological models with explicit representation of
dimension introduces a number of new issues in objects
representing the objects (primitives, rules and con-
straints) and in detecting their relationships (topology, 2.1.1. 3D FDS
order, etc.). The suitability of the topological models in The formal data structure (FDS) was the first data
3D for different applications also varies. structure to consider spatial objects as an integration of
This paper addresses difficulties of designing 3D geometric and thematic properties (Fig. 1). A conceptual
topological models and representing the relationships model and 12 conventions (rules for the partitioning of
between them. The paper is divided into three sections. physical objects) define the structure (Molenaar, 1990).
The first reviews various models and discusses the Rikkers et al. (1993) proposed mapping the model into a
concepts behind them. The second concentrates on relational database. The model assumes the full partition
frameworks for detecting relationships. The final section of space (similar to the planar partition in 2D space).
concludes the discussion and suggests avenues for Besides the feature related to a thematic class, four
further research. elementary objects (point, line, surface and body) and
four primitives (node, arc, face and edge) can be
distinguished. Arcs and faces cannot intersect by
convention unless a node and an arc are created.
2. Review of 3D topological models Singularities are permitted in such a way that arcs and
nodes can exist inside faces or bodies. The role of an edge
One general question raised when referring to is dual; i.e., to define the border of a face (relationship
topological models is whether it is possible to have one face-arc) and to establish an orientation of a face, which
3D topological model, that is suitable for all types of is needed to specify the left and right body. The number
applications. The answer is negative. The design of 3D of arcs constituting an edge is not restricted. Arcs must
topological models is always closely related to the be straight lines and faces must be planar. The surface
specific requirements of a particular category of has one outer boundary and may have several non-
application. For example, space partitioning (full, nested boundaries; i.e., may have holes or islands. The
embedding) depends on the types of queries that have body has one outer surface and can have several non-
to be represented. In case of many neighborhood nested bodies or holes.
operations between 3D objects (such as for geological The fundamental rule of 3D FDS is the concept of a
bodies), full partitioning is recommended. However, if single-valued map; i.e., the node, arc, face or edge can
the objects are surrounded by ‘free space’ (e.g., appear in the description of only one geometric object of
buildings), the embedding approach is more appropri- the same dimension (Molenaar, 1998). A single-valued
ate. Another aspect relating to application are the types approach can partition the space into non-overlapping
of simple objects (e.g., point, line, surface, body) and objects, thus ensuring 1:1 relationships between primi-
primitives used to describe the objects, in 0-dimension tives and objects of the same dimensions, like surfaces
(0D), one-dimension (1D), 2D and 3D. In many cases, and faces. Primitives of different dimensions, however,
0D and 2D primitives are sufficient; e.g., for describing can overlap; for instance, node-on-face, arc-on-face,
buildings. The last aspect is related to the rules of node-in-body and arc-in-body relationships are stored
construction: the types of interrelationships allowed explicitly.
between objects, planarity and convexity rules, and so 3D FDS are used by many to incorporate 3D objects.
forth. For example, if only triangles are allowed, many For example, Shibasaki and Shaobo (1992) implemented
redundant subdivisions of original polygons will be
performed, such as windows on a building wall. Class Class
Class Class
In the following, we will give a brief overview of
several 3D topological models focusing on the three Belongs to Belongs to Belongs to Belongs to

aspects as mentioned above: space partitioning, sup- Surface Body Line Point
ported objects and primitives, and constructive rules.
Left
Two main groups of data structures were found in Right Part of
Part of
previous studies: those that maintain objects and those Face
Is in
Represents
Is in
that maintain relationships. In the first group (object- Is on
Is on
Border
oriented, OO), it is mostly the relationships between
Forward Begin
objects that have to be derived; in the second (topology- Edge Arc
End
Node
Backword
oriented), it is the representation of the objects. Many
data structures, for example, that maintain an explicit XYZ
storage of objects, also maintain an explicit storage of
relationships; i.e., singularities. Fig. 1. 3D Formal Data Structure (3DFDS) (Molenaar, 1990).
 ARTICLE IN PRESS
S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428 421

the model for the maintenance and visualization of 3D oriented applications where spatial queries need to be
city models. De Hoop et al. (1993) investigated possible visualized on the screen as 3D models (Zlatanova, 2000).
relationships (based on the nine-intersection model) for The model does not require the full partition of space;
3D FDS. The CC-modeler presented by Grun . and Wang i.e., all of the objects are embedded in 3D. The simple
(1998) records 3D reconstructed objects in a schema objects are four, but the primitives used are only two;
similar to that of 3D FDS but extended to incorporate i.e., node and face. This is the first representation that
textures per face. avoids the storage of a 1D primitive. The model removes
the uniqueness of the relationship arc/face in 3D space;
i.e., one arc can be part of more than two faces.
2.1.2. TEN
However, two successive nodes can implicitly define this
The next model, the TEtrahedral Network (TEN)
primitive. A 3D primitive is not maintained, and faces
(Fig. 2), was introduced by Pilouk (1996) to overcome
represent the 3D objects. The rules used to describe
some difficulties encountered by 3D FDS in modeling
objects are slightly different from the first two. Faces
objects with indiscernible boundaries, such as geological
must be planar and convex. Singularities are allowed,
formations, pollution clouds, and so forth. TEN,
with node-in-face and face-in-body stored explicitly. The
employing a simplex-oriented approach, was proposed
orientation of the faces is also stored explicitly, and the
to represent 3D objects in the real world (Carlson, 1987).
order of the nodes describes a face.
Like3D FDS, it has four primitives (tetrahedron,
triangle, arc and node) and the subdivision of the space
is full. It should be noted that this model has a real 3D
primitive. In the relational implementation, the arc-node
relationship is stated in the ARC table; the TRIANGLE body surf line point
table contains the tetrahedron-triangle-edge link. A body
is composed of tetrahedrons, a surface of triangles, a line
BID SID LID PID
of arcs and a point of nodes. The general rule for creating
the model is based on the fact that each node is part of Node on
an arc, each arc is part of a triangle and each triangle is face
FID

part of a tetrahedron. The constraints are simple and face

very strict: everything is classified into arc, triangles and
tetrahedrons. Singularities are not permitted.
node
Node in
2.1.3. SSM body

The Simplified Spatial Model (Fig. 3) was the first

NID X Y Z
topological structure that focused on visualization
aspects of the queries. It was designed to serve web- Fig. 3. Simplified Spatial Model (SSM) (Zlatanova, 2000).

SURFACE BODY LINE POINT

SID sclass BID bclass LID lclass PID pclass pnode

TRISURF TETRA ARCLINE

TRINR tsid TETNR tbid ARCNR alid

TRIANGLE

TRINR tet1 tet2 edge 1 edge 2 edge 3

ARC NODE

ARCNR node1 node2 NODENR x y z

Fig. 2. TEtrahedral Network (TEN): relational implementation for 3D (Pilouk, 1996).

 ARTICLE IN PRESS
422 S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428

2.1.4. UDM
The Urban Data Model (Fig. 4) is built on a full
partition of space and represents the geometry of a body
or a surface by planar convex faces (Coors, 2003). Each
face is defined by a set of nodes. Similar to SSM, a 1D
primitive is not supported. Two convex planar faces are
adjacent to each other if they share at least two nodes.
The orientation of a face is stored implicitly. The
constructing rules are stricter than SSM. In the
relational representation of the model, every face having
more than three nodes is decomposed into triangles, and
the FACE table contains only three columns; i.e., the
IDs of the three triangle nodes. As with 3D FDS, face- Fig. 5. 3D TIN-based OO model (Abdul–Rahman, 2000).
body relationships are explicitly stored in the FACE
table. The partition of the objects is higher and all of the
surfaces have to be triangulated. Depending on the
complexity of the surfaces (e.g., the number of windows
on a wall), this triangulation may increase the number of
databases. However, in the case of simple fa@ades (e.g.,
without windows), the constant number of columns in
the FACE table compensates for the number of elements
increased for maintenance. Singularities are reduced
relatively; i.e., the node-on-face and arc-on-face relation-
ships are resolved.

2.2. Object-oriented models

The models mentioned are mapped in a relational

DBMS, which is often considered less appropriate for Fig. 6. SOMAS (Pfund, 2001).
describing real-world objects. Abdul-Rahman (2000)
utilized the FDS model (Molenaar, 1998) to construct a
3D TIN based on spatial objects in an OO environment; objects. The model works with four spatial primitives
i.e., by using the commercial OO DBMS, also known as (nodes, lines, surfaces, and solids).
the Persistent Object and Extended Technology (POET Simple topological relationships between the primi-
OO DBMS). The schema of the model is illustrated in tives of the TIN-based objects can be established, such
Fig. 5, where 3D objects (such as boreholes) are as point–line, point–surface, point–solid, line–surface,
represented by a series of 3D TINs primitives (i.e., line–solid and surface–solid. All of the constructed
tetrahedral). TIN nodes represent point objects, TIN classes of the model are then mapped according to the
edges represent area objects, TIN surfaces (triangles) schema of the POET OO DMBS database.
represent area objects and 3D TINs represent solid Other solutions of explicitly structures-maintaining
objects are presented by the Solid Object Management
System (SOMAS) (Pfund, 2001) or by the model of de la
Losa and Cervelle (1999). Figs. 6 and 7 show the
Base Geometry conceptual models. The structures, however, are not
implemented in a DBMS.
The authors of the OO model proposed the order of
the faces with respect to a common edge to be explicitly
Body Surface Line Point maintained in the model. Thus, the normal vector of
0..2 0..* 0..* each face is determined by the direction of the edge and
0..*
part o f part of
may not always be directed outside of the 3D object.
left/right represent
2.2.1. OO3D
4..* 1..* 2..* 1
1..1
bounds Shi et al. (2003) developed an OO data model to
Face Node handle complex 3D objects in GIS (OO3D) (Fig. 8). The
0..* 3..n
conceptual OO3D data model is developed based on the
Fig. 4. Urban Data Model (UDM) (Coors, 2003). principle of OO data modeling. This model is founded
 ARTICLE IN PRESS
S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428 423

on the following three basic geometric elements: node, The proposed model can handle complex objects,
segment and triangle. The abstract geometric objects are such as complex buildings and TV towers, which is an
defined accordingly. These include points, lines, surfaces essential function for the building of large-scale cyber
and volumes. Second, the corresponding 3D logical cities. The proposed data model is proving to be very
model is designed based on the defined abstract objects efficient, particularly in visualization and rendering. The
and the relationships between them. Third, a formal experimental results of the model demonstrate more
representation of the 3D spatial objects is provided in compacted data volumes and improved visualization
detail. The model is applied in a piece of 3D GIS- speeds for 3D objects than the existing models.
developed software—SpaceInfo.
2.3. 3D structures with explicit representations of
relationships

The second type of topological model has only one

representative; i.e., the spatial model introduced by
Brisson (1990) and extended by Pigot (1995). It is viewed
as the tuple model. It defines cells and cell complexes by
the fundamental properties of a manifold. The subdivi-
sion of the space is full. A clear separation between
objects and primitives does not exist. The model is based
on four primitives, called cells. The description of each
cell gives the reference to all of the neighboring cells
from all dimensions. The relationships (0: non-existent,
1: existent) can be organized in a table with four
columns representing the four cells. For example, to
describe a 3-cell cube, the table will contain 16 records.
The initial works do not allow singularities, but in some
further extensions of the model (Mesgari, 2000),
singularities are permitted; for example, a 0-cell inside
a 2-cell, a 2-cell inside a 2-cell (holes), a 3-cell inside a
3-cell (tunnels). To classify a spatial object, one should
keep track of information on which cells belong to
which objects. Under these circumstances, spatial
objects can be described as a set of 3-cell, 2-cell, 1-cell
and 0-cell tuples.

2.4. Comparison of different models

The advantages and disadvantages of a model change

subject to application. For example, the arbitrary
Fig. 7. OO-model of de la Losa and Cervelle (1999). number of nodes per face can be seen as an advantage

Fig. 8. OO3D model of Shi et al. (2003).

 ARTICLE IN PRESS
424 S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428

or disadvantage, depending on the applications. The for a simple box occupy twice as much space as in
modeling of complex 3D objects (e.g., buildings) is 3D FDS.
convenient, since an inappropriate partitioning (from One of the major advantages of OO3D models, such
the user’s point of view) is not necessary and the faces on as OO3D, is that they are capable of handling complex
the boundary can represent a 3D object. However, the 3D objects. This further improvement is crucial,
same freedom in face description may lead to problems particularly for developing a cyber city for large cities
in visualization, as the rendering engines can only handle where many complex objects exist, such as numerous
triangles. Furthermore, the operators for consistency buildings.
check become very complex. Another example is the Some of the OO models are designed with compact
face–body relationship. Navigating through 3D objects characteristics, for example, the OO3D model has the
is easy, but in some cases (e.g., in urban areas) non- basic elements of node, segment and triangle. This
significant data may be stored (i.e., ‘‘open air’’ also has design differs from the TEN and 3D FDS models, which
to be stored as a right body). do not contain any arc elements, reducing data storage
The major problem with TEN occurs at the stage of when spatial objects are constructed. However, the
modeling. Since the space is completely subdivided into topological relationships are not stored explicitly. The
tetrahedrons, the interiors of objects (e.g., buildings), as performance of some of the spatial analysis-related
well as open spaces, are also decomposed into tetra- applications may not be as efficient as that of other 3D
hedrons. Such subdividing hinders the formation of 3D models.
man-made objects. Pilouk (1996) suggests that these It is rather difficult to compare models using only the
objects be represented by 3D FDS features in TEN. references in the literature. The topological models are
However, the subdivision of triangles furnishes the data implemented under different conditions, for example,
needed for displaying graphic information. In this different DBMSs and different server and client config-
respect, TEN and UDM are perhaps the optimal models urations, and tested with different data sets. Zlatanova
for the visualization of surfaces. The maintenance of (2000) presented performance tests for SSM and 3D
triangles solves other modeling problems such as holes FDS with respect to visualization queries. The queries
or the explicit storage of relationships like arc-on-face can be ‘translated’ as ‘extract and visualize all the
and node-on-face. An additional disadvantage of TEN is objects according to a given condition (e.g., IDo100).’
its much larger database size compared with other Tests are performed under the same computer config-
representations, and the need to process tetrahedrons, urations, DBMSs and data sets. The results of the tests
which is not required for visualization. showed that SSM gives significantly a better perfor-
The omission of arcs enables data structures (SSM, mance than FDS.
UDM) to benefit form the significantly faster data
traverse. However, the navigation of rough surfaces
(e.g., ‘‘following the shortest path’’) may become time-
consuming. The representation of bodies as a set of faces 3. Spatial relationhsips frameworks
(e.g., SSM) can extract the geometries of the objects, but
navigational queries may be disturbed since the co- 3.1. Frameworks for representing spatial relationships
boundary of face–body relationships is not explicitly
maintained (i.e., it has to be derived). Three different approaches to encoding spatial
The cell tuple data structure provides the largest relationships are discussed in the literature; i.e., metric,
spectrum of topological relations between cells and topology and order. The metric is a pure computational
complex cells. The model is built on solid mathematical approach based on the comparison of numerical values
foundations, thus promising easy maintenance. With related to the location of the objects in space. For
respect to visualization, the extraction of faces and example, the spatial relationship between a house and a
points is a simple operation, as the links between cells parcel (e.g., inside, outside, to the south) can be clarified
are stored explicitly. Data obtained from the tuple by a point-in-polygon metric operation performed for
representation, however, lacks an indication of order. each point constituting the footprint of the building. The
Supplementary records are needed to establish order establishes a preference based on the mathematical
the clockwise or anti-clockwise order of cells (note relation ‘‘o’’ (strict order) or ‘‘p‘‘(partial order), which
that the cyclic order is ensured). Assuming a relational allows for a tree-like organization of objects. For
implementation, the entire body of tuple information example, if a building is inside a parcel, the spatial
is available in one relational table. On the one relationship is represented as ‘‘building o parcel.’’ The
hand, there is no need to perform JOIN operations to applicability of representing spatial relationships has
select any data. On the other hand, the size of the table been investigated by Kainz (1989). Kainz argued that it
grows tremendously, which slows down the speed of has advantages in the expression of inside–outside
SELECT operations. For example, the records relationships.
 ARTICLE IN PRESS
S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428 425

Topology allows the encoding of spatial relationships reality, the intersections have not been further investi-
to be constructed on the neighborhoods of objects, gated and that many object intersections are topologi-
regardless of the distance between them. The main cally equivalent), the framework provides a systematic,
property of topology, the invariance under topological easy-to-implement method of detecting spatial relations.
transformations (i.e., rotation, scaling and translation),
makes the computer maintenance of spatial relation- 3.1.2. Voronoi-based spatial algebra for spatial
ships appropriate. The following section discusses the relationships
general framework of topology. Li et al. (2002) suggest a voronoi-based spatial algebra
for spatial relationships. Appropriate operators from set
3.1.1. The 9-intersection model operators are used in the solution to distinguish the
The framework of the model (Egenhofer and Herr- spatial relationships between neighboring spatial ob-
ring, 1990) utilizes the fundamental notions of general jects. Three values (contents, dimensions and number of
topology so that the topological primitives can investi- connected components) are employed as the computa-
gate the interactions of the spatial objects. The tional results of the operation of the sets. The voronoi
topological primitives of a spatial object can be defined region of an object enhances the interaction of an object
in each spatial model; hence, the framework can be with its neighbors.
applied to any spatial model. The basic criterion for
distinguishing different relationships is the detection of 3.1.3. Uncertain topological relationships modeling
empty and non-empty intersections between topological Shi and Guo (2002) presented a study of a formal
primitives. Depending on the number of topological representation of the topological relationships between
primitives considered, two intersection models were uncertain spatial objects. First, after reviewing the
presented in the literature. The first investigates the related definitions and representation concerning un-
intersection of the interiors and boundaries of two certain spatial objects, they proposed a unified structure
objects. This results in a 24=16 relationship between for representing certain or uncertain spatial objects.
two objects. Exterior evaluation is adopted when two Second, they presented a framework to formally
topological primitives are inadequate for differentiating represent topological relationships between uncertain
many relations. In this case, the number of detectable spatial objects. Third, they provided the algorithm to
relations between two objects increases to 29=512. determine topological relationship.
Eight relationships are possible between 3D and 3D
objects, and they are given the following names: disjoint, 3.1.4. Extended topological relationships in GIS
meet, contains, covers, inside, covered By, equal and Liu and Shi (2003) proposed a further development on
overlap (Fig. 9). For example, if the boundaries of the topological relations between any two objects in GIS.
two objects intersect but the interiors do not, the First, they adopted a new definition of the topological
conclusion is that the objects meet. Despite the criticism relations between two objects. Based on this new
(i.e., that not all of the relationships are possible in definition, the topology of the object itself and several
topological properties (such as compactness, connectiv-
ity, first fundamental group, subspace topology, etc.), a
sequence of topological relations between any two hole-
less objects is discovered. Based on the proposed
A B R031 disjoint A B R287 meet extended topological relationships models, the number
of topological relations between two infinites is found.
These can then be approximated by a sequence of
matrices. Furthermore, the topological relations be-
A A
B R179 contains R435 covers tween two convex sets can be approximated by a
B
sequence of 4  4 matrices, which are the topological
properties of A -B ; A \B; B \A; qA-qB:
B
A R220 inside B 3.1.5. The dimensional model (DM)
A
R476 coveredBy
The DM is another framework utilizing the order of
points, which is related to the study of affine space
(a subspace of the topological space) and convex shapes.
R400 equal A The formal definition of the model can be found in
B R511 overlap
A B Billen et al. (2002). Here, we will use a simple example
for illustration. If one looks at a triangle in R2 ; the
Fig. 9. 9-intersection model: possible relationships between 3D points of order 0 are the vertices. The points on the
and 3D objects (Egenhofer and Herrring, 1990). edge have an order of 1 and the points of order 2 are all
 ARTICLE IN PRESS
426 S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428

of the points that are ‘‘inside’’ the triangle. Applying this a consistent data structure and to update it are
formalism, spatial objects can be described and their investigated and developed. For example:
spatial relationships can be decoded. In the 3D
Euclidean space (R3 ), four types of dimensional * operations to organize data according to the data
elements are allowed; i.e., 0D, 1D, 2D and 3D elements. structure; i.e., operations for planarity, convexity and
For example, a polygon has a 2D-element, a 1D-element discontinuity, as defined in the model;
and a 0D-element. The 2D-element coincides with the * operations to check for consistency: the validation of
spatial object (i.e., the polygon). To represent the the objects (e.g., polygon closed, body closed), node-
dimensional relationships between two objects, one has on-line, node-on-face, node-in-body, line-on-face,
to consider all of the dimensional properties of these line-in-body, intersection of lines, face-on-face, inter-
elements. For example, the dimensional relationships section of faces, face-in-body;
between two simple spatial objects of two dimension * 3D overlay, which is based on the same operation of
(i.e., polygons A and B) can be defined in the following consistency check and 3D editing;
order: first, check the dimensional relationship between * operation for 3D editing: the adding, deleting and
the 2D element of A and all of the dimensional elements updating of cells.
of spatial object B; then, check the dimensional relation-
ship between the 1D element of A and all of the Apart from the constructing operators that have been
dimensional elements of spatial object B; etc. mentioned, GISs facilitate a number of specialized
The dimensional relationship can be partial, total or operations such as selection, navigation and specializa-
non-existent, depending on the interaction between the tion. Molenaar (1998) described the GIS query as a
interiors of the objects (Fig. 10). The benefit gained from selection operation with three components: data type
using these frameworks is flexibility while deciding on specification, conditions and operations that have to be
which dimensional elements are to be used. In general, a performed on the data. The selection can then be
larger number of relationships can be distinguished performed on semantics, geometry or topology. For
compared to the 9-intersection model (see Fig. 11). example, ‘‘select the buildings (data type) higher than
15 m (condition) and show their ID (operation).’’
Sophisticated operations on data may obscure the
3.2. Spatial operators boundary between query and analysis. Theoretically,
an original operation and further processing can be
Having specified the data structure and the frame- encapsulated in a new operation. Many classifications of
work for representing relationships, the next step is to operations are included in the literature (Aronoff, 1995;
define the operations to be supplied by the system. The Goodchild, 1987). In general, the operations can be
operations describe all of the actions that can be subdivided in three large groups with respect to
performed on the data. First of all, operations to build geometric and semantic characteristics and spatial
relationships. Most interesting are the operations related
to the geometry and to the spatial relationships. These
operations can be classified as follows:

Fig. 10. Dimensional relationships: non-existent, partial and

* metric operations are selection operations based on
total. the shape and size of objects and on further
computations; e.g., of distance, volume, area, length,
center of gravity, intersect;
* position operations are selection operations based on
position (without further processing); e.g., objects in
a certain area;
* proximity operations are selection operations based
on geometric characteristics and on the creation of
new objects; e.g., a buffer, convex hull, union of
objects;
* relationship operations are selection operations
based on spatial relationships (without further
processing); e.g., neighboring operations, overlay;
* network operations are selection operations based on
spatial relationships and geometries, and on further
Fig. 11. Dimensional model: possible relationships between 3D processing with different levels of complexity; e.g.,
and 3D objects. route planning;
 ARTICLE IN PRESS
S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428 427

* visibility operations are selections based on geometric operations such as ‘‘meet,’’ and ‘‘overlap’’ will be
characteristics and further processing; e.g., sign of performed on the topological model.
view;
* semantic operations are selections based on semantic
characteristics;
* mixed operations are selections founded on geo- References
metric and semantic characteristics.
Abdul-Rahman, A., 2000. The design and implementation of
Apparently, operations relating to the spatial relation- two and three-dimensional triangular irregular network
ships of the objects are highly influenced by data (TIN) based GIS, Ph.D. Dissertation, University of
structure. As mentioned in the previous section, some Glasgow, Glasgow, United Kingdom, 250p.
of the structures may perform certain queries better than Aronoff, S., 1995. Geographic Information Systems: A Manage-
others. Moreover, it should be noted that spatial ment Perspective. WDL Publications, Ottawa, 293p.
Billen, R., Zlatanova, S., Mathonet, P., Boniver, F., 2002. The
analysis can be performed on geometric models, as well.
dimensional model: a framework to distinguish spatial
Many relational DBMSs offer support to spatial objects, relationships. In: Richardson, D., van Oosterom, P. (Eds.),
especially geometric models; and supply a number of Advances in Spatial Data handling, 10th International
spatial operations, for example, validation, point-in- Symposium on Spatial Data Handling. Springer, Berlin,
polygon, objects-within-distance, area, length, etc. The pp. 285–298.
operations only make use of X and Y coordinates, Brisson, E., 1990. Representation of d-dimensional geometric
although some of them accept 3D faces. objects, Ph.D. Dissertation, University of Washington,
Seattle, WA, 165p.
Carlson, E., 1987. Three dimensional conceptual modelling of
subsurface structures. Technical Papers of American Society
4. Conclusions for Photogrammetry and Remote Sensing/American Con-
gress on Surveying and Mapping Annual Convention,
In this paper, we have given a brief overview on Baltimore, Vol. 4., pp. 188–200 (Cartography).
relational or OO topological models and discussed two Coors, V., 2003. 3D GIS in Networking environments’,
frameworks for detecting spatial relationships between Computer, Environment and Urbaqn Systems, Vol. 27 (4),
objects. With reference to the discussion on the advan- pp. 345–357.
tages and disadvantages of the different models, we de Hoop, S., van de Meij, L., Molenaar, M., 1993. Topological
conclude that selecting an appropriate structure is a relations in 3D vector maps, In: Proceedings of EGIS’93,
Genoa, Italy, pp. 448–455.
complex process and that the characteristics of the
de la Losa, A., Cervelle, B., 1999. 3D Topological modelling
applications, for instance, objects of interest, resolution, and visualisation for 3D GIS’. Computers & Graphics 23
required spatial analysis, etc., should be examined. A (4), 469–478.
model that is good for 3D spatial analysis may exhibit a Egenhofer, M.J., Herring, J.R., 1990. A mathematical frame-
dissatisfactory performance on 3D visualization and work for the definition of topological relationships, In:
navigation. Moreover, the relational or OO implementa- Proceedings of Fourth International Symposium on Spatial
tion of the model also has an impact on its performance. Data Handling, Zurich, Switzerland, pp. 803–813.
Following the current trends for the integrated Goodchild, M., 1987. A spatial analytical perspective on
maintenance of spatial and non-spatial data, many geographical information systems. International Journal of
DBMSs have already provided support to spatial Geographical Information Science 1 (4), 327–334.
. A., Wang, X., 1998. CC-modeller: a topology generator
Grun,
objects. According to the abstract specifications of
for 3D city models. ISPRS Journal for Photogrammtry and
OpenGIS (Open GIS consortium Inc., 1999), spatial Remote Sensing 53 (5), 286–295.
objects are stored in the database with their geometric Kainz, W., 1989, Order, topology and metric in GIS, American
and topologic representations to ensure consistency Society for Photogrammetry and Remote Sensing/American
between the two models after conversion operations. Congress on Surveying and Mapping Annual Convention,
This does not imply that all vendors have to accept one Vol. 4, Baltimore, pp. 154–160.
3D topological model for implementation. As discussed Li, Z., Zhao, R., Chen, J., 2002. A Vornori-based spatial
above, different models may be suitable for the algebra for spatial relations in GIS. Progress in Natural
execution of specific tasks. Oosterom et al. (2002) Science 12 (7), 528–536.
proposed the maintaining of multiple topological Liu, K.F., Shi, W.Z., 2003. Analysis of topological relation-
ships between two sets. In: Shi, W.Z., Fisher, P.F.,
models in one database by describing the objects, rules
Goodchild, M. F. (Eds.), Proceedings of the Second
and constraints of each model in a metadata table. Such International Symposium on Spatial Data Quality, March
an approach will maximize efficiency and effectiveness in 19th–20th, pp. 61–71.
the provision of operations. Metric and position Mesgari, S.M., 2000. Topological cell-tuple structures for three-
operations such as area or volume computations will dimensional spatial data, ITC Dissertation number 74, ITC,
be presented on the geometric model, while relationship The Netherlands, 200p.
 ARTICLE IN PRESS
428 S. Zlatanova et al. / Computers & Geosciences 30 (2004) 419–428

Molenaar, M., 1990. A formal data structure for 3D vector Rikkers, R., Molenaar, M., Stuiver, J., 1993. A query oriented
maps. In: Proceedings of EGIS’90, Vol. 2. Amsterdam, The implementation of a 3D topologic data structure.
Netherlands, pp. 770–781. In: Proceedings of EGIS’93, Vol.2. Genoa, Italy,
Molenaar, M., 1998. An Introduction to the Theory of Spatial pp. 1411–1420.
Objects Modelling. Taylor & Francis, London. Shi, W.Z., Guo, W., 2002. Topological relationships between
Oosterom, P.J.M.van, Stoter, J.E., Zlatanova, S., Quak, W.C., spatial objects with uncertainty. In: Shi, W.Z., Fisher, P.F.,
2002. The balance between geometry and topology. In: Goodchild, M.F. (Eds.), Spatial Data Quality. Taylor and
Richardson, D., van Oosterom, P. (Eds.), Advances in Francs, London, pp. 50–61.
Spatial Data Handling, 10th International Symposium on Shi, W.Z., Yang, B.S., Li, Q.Q., 2003. An object-oriented data
Spatial Data Handling. Springer, Berlin, pp. 209–224. model for complex objects in three-dimensional geographic
Open GIS Consortium, Inc., 1999. The OpenGIS abstract information systems. International Journal of Geographic
specification, topic 1: Feature geometry, Technical Report Information Science, Vol. 17 (5), pp. 411–430.
Version 4 (99-101.doc), Open GIS Consortium, /http:// Shibasaki, R., Shaobo, H., 1992. A digital urban space model—
www.opengis.org/techno/abstract.htmS. a three dimensional modelling technique of urban space in a
Pfund, M., 2001. Topologic data structure for a 3D GIS. In: GIS environment. International Archives for Photogram-
Proceedings of International Society for Photogrammetry metry and Remote Sensing, Vol. XXIX. Part B4, Commis-
and Remote Sensing Journal, Vol. 34. Part 2W2, May, sion IV, Washington, DC, USA, pp. 257–264.
Bangkok, Thailand, pp. 233–237. Zlatanova, S., 2000. 3D GIS for urban development. Ph.D.
Pigot, S., 1995. A topological model for a 3-dimensional Spatial Dissertation, ITC, The Netherlands, 222pp.
Information System, Ph.D. Dissertation, University of Zlatanova, S., Abdul Rahman, A., Pilouk, M., 2002. 3D GIS:
Tasmania, Australia. current status and perspectives. In: Proceedings of the Joint
Pilouk, M., 1996. Integrated modelling for 3D GIS, Ph.D. Conference on Geo-spatial theory, Processing and Applica-
Dissertation, ITC, The Netherlands. tions, 8–12 July, Ottawa, 6p. CDROM.

View publication stats