You are on page 1of 12

ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145

Contents lists available at SciVerse ScienceDirect

ISPRS Journal of Photogrammetry and Remote Sensing


journal homepage: www.elsevier.com/locate/isprsjprs

Transaction rules for updating surfaces in 3D GIS


Gerhard Gröger ⇑, Lutz Plümer 1
Institute for Geodesy and Geoinformation, University of Bonn, Meckenheimer Allee 172, 53115 Bonn, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Three-dimensional surface models representing the terrain and the outer hull of objects such as buildings
Received 30 January 2011 and bridges support important 3D GIS applications, for example telecommunication planning and noise
Received in revised form 27 January 2012 emission simulation. Updates of surface models often introduce errors which violate basic assumptions of
Accepted 1 March 2012
users and their applications. The notion of geometric-topological consistency covers many of these assump-
Available online 10 April 2012
tions. It guarantees that objects do not penetrate mutually or that objects completely cover other objects.
Assuring that updates do not violate geometric-topological consistency constitutes a major challenge for
Keywords:
3D GIS which has not been satisfactorily met so far. This article presents a solution which is based on effi-
Updating
GIS
cient transaction rules for updating 3D surface models. We show that these rules are safe (consistency is
Surface preserved by any rule application) and complete (any consistent surface model can be generated by suc-
Three-dimensional cessive rule applications). For both properties rigorous mathematic proofs are given.
Quality Ó 2012 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier
B.V. All rights reserved.

1. Introduction 2.8D map, which extends digital terrain models (Okabe et al., 2000).
They are often called 2.5D and are restricted because (x,y)-loca-
There is a rapidly increasing demand for three-dimensional tions have exactly one height value. This restriction is omitted in
applications of geographical information systems (GIS). Prominent 2.8D maps, allowing for the representation of vertical walls and
examples are noise emission simulation and mapping (Czerwinski protrusions like balconies, ledges or roof overhangs. Hence, the
et al., 2007), urban and telecommunication planning (Köninger and outer hull of most urban objects can be represented. The geometry
Bartel, 1998), or disaster management (Zlatanova and Holweg, is three-dimensional, whereas topology remains two-dimensional.
2004; Kolbe et al., 2008; Gröger and Plümer, 2010). Most commer- From a mathematical point of view, a 2.8D map basically is a 2-
cial GIS currently available are, however, limited to two or two and manifold (for details, see Mäntylä, 1988 and Hatcher, 2001). In
a half dimensions, and are not able to cope with these applications addition, 2.8D maps allow for the modeling of overpasses like
adequately. Three-dimensional systems originating from computer bridges, viaducts, tunnels and pedestrian underpasses, which are
aided design (CAD) (Mäntylä, 1988) or computer graphics (Foley captured by the topological notion of a handle. The term ‘2.8D
et al., 1995) do not handle semantic properties of objects map’ wants to express that the model exceeds 2.5D, but is less than
adequately and do in most cases not provide the required GIS three-dimensional.
functionality. The corresponding models, boundary representation In many cases data sets which claim to be 2.8D maps in fact do
(B-Rep) or constructive solid geometry (CSG) (Mäntylä, 1988; Foley not fulfill these requirements. Well-known errors are mutually
et al., 1995) are complex and difficult to handle from an algorith- intersecting, self-intersecting or self-penetrating boundaries of ob-
mic perspective. jects, gaps in boundaries of objects, or objects like buildings float-
Three-dimensional surface models are sufficient for many impor- ing above the terrain. Hence, there is an urgent need for methods
tant 3D applications while preserving to a large degree the algo- and tools for checking effectively and efficiently whether data sets
rithmic and conceptual simplicity of the 2D. In those surface are geometric-topologically consistent, i.e. are 2.8D maps. The
models, the outer hull of man-made objects like buildings, bridges, declarative mathematical definition of a 2.8D map in terms of a
or tunnels are integrated into the terrain (see Fig. 1). In Gröger and 2-manifold, however, is not suitable to be checked effectively by
Plümer (2003, 2005, 2011a, 2011b), we introduced the concept of a automatic procedures. Hence, we have provided an alternative axi-
omatic characterization of 2.8D maps (Gröger and Plümer, 2003,
2005, 2011b). These axioms can readily be implemented in avail-
⇑ Corresponding author. Tel.: +49 (0) 228 73 1764.
able database systems and are equivalent to the declarative math-
E-mail addresses: groeger@igg.uni-bonn.de (G. Gröger), pluemer@igg.
uni-bonn.de (L. Plümer).
ematical definition of a 2.8D map: axioms have been proven to be
1
Tel.: +49 (0) 228 73 1750. complete (each violation of the declarative mathematical definition

0924-2716/$ - see front matter Ó 2012 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.isprsjprs.2012.03.004
G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145 135

Fig. 1. Three-dimensional surface model representing the terrain and a bridge integrated in the terrain.

of a 2.8D map is detected by the axioms) and correct (each violation is proven mathematically, guaranteeing the reliability and the flex-
of an axiom is in fact a violation of the definition). ibility of our method. Some of these rules have already been men-
The axioms for 2.8D maps are suitable for checking the consis- tioned in Gröger and Plümer (2009), but no mathematical
tency of a data set as whole. Assuring that an update of data does derivation has been given so far.
not affect consistency is a different problem. Since our natural The topological concept of a handle, which allows representing
and man-made environment changes frequently, data representing objects like bridges, tunnels or arcades in 3D surface models (c.f.
this environment have to be updated accordingly in order to reflect Fig. 1), plays a crucial role for surface models and transaction rules.
those changes. Of course, up-to-dateness is an important quality In Gröger and Plümer (2012) we have characterized handle objects
element of spatial data, and the usage of out-dated data sets may in terms of different types of cycles of the surface. In the paper at
lead to erroneous results that do not correspond to reality. More lo- hand, we transfer that static characterization of handles to the dy-
cal and more specific concepts are required to assure that updates, namic case of the insertion and deletion of handles.
which typically have local impacts, maintain consistency. Globally Identifying the adequate level of granularity poses a major chal-
checking consistency of a large dataset after each local update is lenge when creating transaction rules. If the rules are too fine-
not feasible from an efficiency point of view. Instead, we propose grained, the proof of the safety is simple, but not all operations
to employ transaction rules to obtain consistent updates. Transac- preserving consistency are permitted. Hence, those rules are not
tion rules, which initially emerged from the field of active dat- complete. The rules defined by Gold (2003) and Tse and Gold
abases (Widom and Ceri, 1996), locally check local updates by (2002), for example, are restricted to inserting prism-shaped,
restricting the axioms to that part of the data set which is actually straight tunnels (see Section 5). If due to an obstacle a non-straight
affected by the changes. Furthermore, transaction rules complete tunnel, i.e. a tunnel with an angle, has to be constructed, the rules
imperfect update requests in a consistent way and reject inconsis- fail and completeness is affected. On the other hand, if rules are too
tent updates. Hence, a crucial property of transaction rules is that coarse, they will be too complex to handle and will not permit an
consistency is preserved for arbitrary applications of the rule. Such efficient safety check. We define the right level of granularity by
rules are called safe. Another important property is the usability the notion of a composite surface, which is flexible with regard
and flexibility of rules in the sense that each (consistent) 3D city to geometry, but restricted topologically. In this approach, the
model can be generated by rule applications. Such a set of rules occurrence of nested structures, and particularly of nested handles,
is called complete. is a major source of complexity. The proofs of completeness in Sec-
Safe and complete transaction rules for updates of 2D datasets tion 4.5 cope with this complexity.
have been introduced in Gröger and Plümer (1997). Likewise, rules The focus of the transaction rules presented in this paper is set
covering the case of updating three-dimensional solid models (Grö- on topology and geometry. Concepts for the definition of semanti-
ger and Plümer, 2011a) have already been developed (Gröger and cal objects based on surface models have been presented for non-
Plümer, 2012). Solid models extend surface models since they en- handle objects like buildings in Gröger and Plümer (2005), and for
able the modeling of interior structures (e.g., stories, rooms). How- handle objects (tunnels, bridges) in Gröger and Plümer (2011b).
ever, surfaces are an important component for solid models, since a The contents of the semantical model, e.g., the object type (resi-
solid is bounded by a surface and since solid models require the dential building or industrial building for example), or attributes
representation of the terrain as a surface. The transaction rules (e.g., the year of construction or the building function) are defined
for solids (Gröger and Plümer, 2012), however, do not cover the in semantic 3D city models such as CityGML (Gröger et al., 2008;
case of 3D surfaces. These rules split a solid by the insertion of a Kolbe et al., 2008).
3D surface into two solids. An example for such a rule application The rest of this paper is organized as follows: The second sec-
is the splitting of a room into two rooms. An inverse rule merges tion will define basic mathematical notions which are relevant
two neighboring solids by deleting a 3D surface. In contrast, the for surface modeling. The third section will specify the surface
rules for 3D surfaces presented in this paper maintain a single, 2- model (termed 2.8D maps) and provide an axiomatic characteriza-
manifold 3D surface, which does not delimit a solid. In the rules tion of such models, which provide the base of the transaction
for solids, the question how to obtain a 3D surface for splitting a rules. The rules for 2.8D maps will be introduced in Section 4. First
solid is explicitly left open. Likewise, the completeness of the rules the rules for the 2D case (which have been published earlier) will
for solids (Proposition 7 in Gröger and Plümer, 2012) is based on be briefly recalled. Second we will outline the 2.8D rules and intro-
the completeness result for 3D surfaces, which has not been pro- duce two rules that generate and modify 2.8D maps, at the same
ven yet. Hence, the issue of safe and complete transaction rules time retaining the number of handles. Afterwards, rules for gener-
for surfaces in 3D has so far remained unsolved. This solution is ating and removing handles will be developed. The safety of each
presented by the paper at hand. The main contribution is a rigorous rule as well as the completeness of the whole set of rules will be
mathematical derivation of a set of transaction rules for the gener- formally proven. After having compared our approach to related
ation of arbitrary 3D surfaces. Safety and completeness of all rules work on checking consistency of updates in GIS in Section 6, this
136 G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145

Fig. 3. Cuboid with two opposite open sides as example for a surface with two
boundaries (depicted dotted).

used in geometrical modeling to represent the boundary of solids


(Mäntylä, 1988; Mortenson, 1997, c.f. Fig. 2b).
The number of handles is another topological invariant of sur-
faces. Handles in surfaces are used to model bridges, tunnels, ar-
cades or similar features. The number of handles of a surface is
given by its genus.3 The genus is defined as the maximal number
of closed, continuous, non self intersecting and mutually non-inter-
secting curves, the cutting of which preserves the connectivity of the
surface. A sphere or a cuboid surface, for example, has genus zero
and hence no handles, whereas the genus of a torus is one, i.e. it
has one handle. The 2-torus has two handles (see Fig. 4).
Fig. 2. Examples for 2-manifold surfaces (a and b) and non-2-manifold surfaces A surface is orientable, if two opposite sides of the surface can be
(c–e). distinguished. For the general case a more formal definition of ori-
entability is given in Armstrong (1997). Orientability of surfaces
which are composed of faces bounded by edges can be checked
paper will end with concluding remarks and a discussion of open
by a simple algorithm (Mortenson, 1997), as follows: First, the
questions and further work.
set of edges is duplicated in a manner that gives each face its
own edges within its boundary. Then, for each face, the edges are
2. Basic mathematical notions orientated consistently, either clockwise or counter-clockwise. If
there is an orientation for each face, giving opposite orientation
In this section, we recall notions from mathematical topology to all pairs of coincident edges, the surface is orientable, otherwise
(Alexandroff, 1961; Mäntylä, 1988; Armstrong, 1997; Hatcher, it is not orientable.
2001) to fix terminology. Particularly used are standard notions
from graph theory (Harary, 1969). We consider graphs that are
2.2. The Euler–Poincaré formula
embedded in Euclidean space R3 by assigning a 3D position to each
vertex. Such an embedding defines faces as atomic areal entities,
If a surface is represented by a graph embedded in R3 , the num-
which in our case are planar, i.e. are located in the same plane. In
bers E of edges, V of vertices, F of faces, H of handles and B of
our case, faces are bounded by simple cycles.
boundaries are related by the Euler–Poincaré formula (we assume
that the graph is connected):
2.1. Topological concepts for surfaces
V  E þ F þ B ¼ 2H þ 2 ð1Þ
A 2-manifold is a topological space where each point has a
This formula is also valid for surfaces which have an unbounded
neighborhood which is topologically equivalent to an open two-
outer face (often called Out) instead of an outer boundary.
dimensional disk (Mäntylä, 1988; Molenaar, 1992; Foley et al.,
1995; Mortenson, 1997; Hatcher, 2001; Herring, 2001). Examples
for 2-manifolds are the open disk, the open cylinder surface, the 2.3. Cell complexes
hull of a cuboid, or the sphere. Fig. 2 depicts the open disk (a)
and the hull of a saddle roof building (b) as examples for 2-mani- The theory of cell complexes (Hatcher, 2001), a branch of Alge-
folds, as well as three counterexamples: three rectangles meeting braic Topology, defines concepts for aggregating primitive objects
in a common edge (c), the hull of two cuboids meeting in a com- to more complex ones in a topologically clean manner. Hence, cell
mon edge (d), and the hull of two saddle roof buildings, where complexes are the base of many topological data models in GIS,
the roof of one building penetrates the wall of the other (e). The CAD and Computer Graphics (Molenaar, 1992; Foley et al., 1995;
penetration is depicted hatched. Herring, 2001; Lake et al., 2004; Boguslawski and Gold, 2008,
Another important property of surfaces, which is also a topolog- 2009). Primitives are nodes, edges, and faces, which are also called
ical invariant, is the number of boundaries. A sphere or a surface 0-cells, 1-cells, and 2-cells, respectively. Each n-cell is bounded by
representing the boundary of a pitched roof building (Fig. 2b) has (n-1)-cells, which are called the boundary of the n-cell (that notion
zero boundaries, an open disk (Fig. 2a) has one boundary, and an of a boundary generalizes the notion of the boundary of a surface in
open cylinder surface or a cuboid with two opposite open sides Section 2.1). A cell complex is an aggregation of cells where the fol-
(Fig. 3) has two boundaries. Surfaces without boundaries are lowing condition holds true: The intersection of two cells in the
closed,2 since they enclose a volume completely. Such surfaces are cell complex is either empty or is a cell which is part of the bound-
aries of both cells. Thus, cell complexes avoid overlapping cells as
2
Note that the term closed is overloaded: it may denote a surface with boundary
3
where the boundary points belong to the set (in contrast to open sets) or a surface This proposition is valid for connected, orientable surfaces. We assume that all
which has no boundary at all. In this paper the second designation is meant. surfaces discussed in this paper have both properties.
G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145 137

Fig. 4. (a) The cuboid surface has no handle (genus zero). (b) The torus has one handle (genus one). (c) The 2-torus has two handles (genus two). Each cycle on the surface in
(a) separates the surface, whereas the cycle on the surface in (b) as well as both cycles in (c) do not do so. Each additional, non-intersecting cycle separates the surfaces.

well as penetrations of cells, and touching of cells is defined explic-


itly by commonly shared boundaries. To exemplify the concept of
cell complexes, Fig. 5 depicts a cell complex consisting of two 2-
cells. The touching of both cells is represented explicitly by a 1-cell
and two 0-cells.

3. Consistency of surfaces in 3D

In Gröger and Plümer (2003, 2005), we have introduced the


Fig. 6. Simple example of a 2.8D map which represents the outer hull of a saddle
concept of 2.8D maps for the representation of the terrain and
roof building and the surrounding terrain (depicted from below).
the outer hull of buildings, including vertical walls and overhangs.
In Gröger and Plümer (2011b) the model is extended to handle ob- locality of updates, identify the region where violations may occur
jects like tunnels and bridges. Intuitively, a 2.8D map can be imag- and restrict the consistency checks to that region. In that sense, a
ined as a cloth draped over the terrain and man made structures. transaction rule is a specific, regionally restricted version of the
More formally, a 2.8D map is a 2-manifold, connected cell complex axioms, tailored for a specific update operation. All rules presented
embedded in 3D space, which consists of vertices, edges and faces. in this section are safe and complete for 2.8D maps. Hence, consis-
Each face is bounded by a simple cycle of edges. A special, un- tency is preserved and all permitted modifications are realizable.
bounded face Out surrounds the bounded faces of the map. Fig. 6 We first describe the structure of the rules.
depicts the hull of a saddle roof building as an example for a A transaction rule consists of a transaction, a condition, and an
2.8D map. It contains vertical walls and overhangs (roof over- action part. The transaction part specifies the input and the effect
hangs). The face Out is depicted hatched. of the rule. The condition part is the guard of the rule. It defines
The definition of a 2.8D map cannot be implemented immedi- the conditions under which the rule is applicable. The action part
ately and therefore does not provide a base for tools for consis- is a sequence of statements being executed in order to consistently
tency checking. Hence, we have introduced so-called axioms for complete the user’s request. In some sense, our transaction rules
2.8D maps (Gröger and Plümer, 2005, 2011b), which can immedi- are comparable to Event-Condition-Action-rules (ECA-rules: On E if
ately be implemented in effective and efficient tools for checking C do A), which are one of the major mechanisms of active dat-
the 2.8D map property (geometrical-topological consistency). The abases (Widom and Ceri, 1996).
reliability of such a procedure is assured by two properties of the
axioms, which guarantee that both formalizations are equivalent:
4.1. The 2D case
the axioms are correct, if each violation of the axioms is in fact
an error (violation of the 2.8D map definition), and the axioms
2D maps are a special case of 2.8D maps, where the coordinates
are complete if each error is detected by an axiom. Both correctness
of all vertices are two-dimensional. The axioms for 2.8D maps can
and completeness formally have been proven (Gröger and Plümer,
be simplified for that special case: axioms 3, 9, 10, 12 and 13 can be
2005, 2011b).
omitted since they are implied by the other axioms (see Plümer
and Gröger, 1997).
4. Transaction rules for 2.8D surfaces We shortly recall transaction rules for 2D maps (Gröger and
Plümer, 1997):
This section introduces the transaction rules which enable con-
sistent updates of 2.8D surfaces, e.g., the deletion or insertion of a  TR1: Initial insertion of a surface.
building, the addition of a balcony to a building or the modification  TR2: Final deletion of a surface.
of the roof shape of a building. The transaction rules exploit the  TR3: Splitting of a face by inserting a polyline.
 TR4: Merging of two faces by deleting a polyline.

The first and the second rule are used to create or delete an initial
model, which is a 2D surface that consists of one bounded face only.
Based on that model, more complex 2D maps can be generated by
repeatedly splitting or merging faces. Fig. 7 illustrates the four rules.
Note that the preconditions of the particular rules avoid the genera-
tion of inconsistent maps. For example, when merging two faces by
rule TR4, it must be assured that the boundary of the resulting new
face is a simple cycle. Fig. 8 depicts an error case where the cycle of
the resulting face Out is not delimited by a simple cycle. For more de-
Fig. 5. A cell complex consisting of two 2-cells A and B sharing a common edge e2 tails, particularly for a description of the preconditions of the indi-
and common vertices v1 and v2. vidual rules, see Gröger and Plümer (1997).
138 G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145

Fig. 7. Illustration of the transaction rules TR1–TR4 for 2D maps. (a and b) Initial generation of a single bounded face f. (c and d) Splitting/merging of faces by inserting a
polyline p (depicted hatched).

The following propositions hold:

Proposition 1. The transaction rules TR1–TR4 are safe.

Proposition 2. The transaction rules TR1–TR4 are complete for 2D


maps.
The proofs of both propositions can be found in Gröger and Plü-
mer (1997).
Fig. 8. Error case for TR4: merging of faces f1 and Out in (a) by deleting polyline p
results in a face Out which is not bounded by a simple cycle (b).
4.2. Rules for 2.8D maps: overview

The transaction rules for 2.8D maps extend the rules TR1–TR4.
They are as follows (see Fig. 9 for examples):

 TR5: Changing the geometry of vertices.


 TR6: Replacing a composite surface.
 TR7: Insertion of a handle.
 TR8: Deletion of a handle.

The rule TR5 is used to generate a 2.8D map from a 2D map by


changing the geometry of vertices. The geometries of the incident
edges and faces are adapted automatically. Since all incidence rela-
tions between vertices, edges and faces are maintained by the rule,
topology is preserved.4 In addition, the rule is useful for local
changes of the map’s geometry. An example is the modification of
the height of a saddle roof building (c.f. Fig. 9a). The vertices that
are changed geometrically are accentuated by an additional circle.
The next rule, TR6, replaces a composite surface (bounded by a sim-
ple cycle c) by another surface that is also bounded by c. This rule
may be used to add a balcony to a building, or to add a dormer or
a chimney to a roof. Fig. 9b gives an example, where the footprint
area of a building (depicted hatched) is replaced by the outer hull
of the corresponding saddle roof building (depicted in dark grey col- Fig. 9. Examples for applications of the transaction rules for 2.8D maps: (a) TR5:
or). The rule TR7 adds a handle to a surface. This is illustrated in changing geometry. (b) TR6: replacing a composite surface (hatched) by another
(dark gray). (c) TR7: incrementing the number of handles by replacing two
Fig. 9c, where two rectangles (depicted hatched) are replaced by a
composite surfaces (hatched) by a single surface with two boundaries and vice
single surface. In contrast to TR6, this surface has two boundaries versa decrementing the number of handles by TR8.
and touches the terrain surface in two simple cycles. TR8 is the in-
verse rule of TR7, since it removes a handle from a surface.
For the detailed presentation of the rules TR5–TR8, two notions 1989) and its 3D extension (Zlatanova, 2000), touch means that
are introduced. First, the concept of a composite surface (Herring, both structures meet only in the simple cycle and are disjoint else-
2001) with boundary is required to specify the input of transaction where. In the following we take for granted that both structures
rules. In contrast to a 2.8D map, the unbounded face Out is omitted have been preprocessed and hence, are identical with regard to
in composite surfaces. Hence, a composite surface has one bound- topology and geometry. Furthermore, we assume that after the ac-
ary which is a simple cycle. Unless stated otherwise, a composite tion part of the corresponding rule has been executed, both struc-
surface has exactly one boundary. For dealing with handles, com- tures are merged, duplicates are removed and corresponding
posite surfaces with two or more boundaries are used later in this vertices and edges are connected to yield a cell complex. This can
paper. However, each boundary is still a simple cycle and all be accomplished by using standard geometry and topology
boundaries are disjoint pairwise. procedures.
Second, in some of the rules presented this paper, two struc-
tures are considered which touch in one or two simple cycles. As 4.3. Updating surfaces retaining the number of handles
defined in the well-known 4-intersection-model (Egenhofer,
This section introduces the rules TR5–TR6, which do not change
4
In the case of surfaces with handles, topology may be changed. See remark in the number of handles, in detail. Note that, however, the surfaces
Section 4.3. which are modified by the rules may contain handles.
G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145 139

Fig. 10. Error case for rule TR5: if the (vertices of the) building in the right of (a) are moved towards the left building, penetrations of faces (depicted hatched) occur (b).

The first rule, TR5, changes the geometry of a subset Vc of the set
Table 1
of vertices of a 2.8D map. An example has already been presented
Complete and correct axioms for 2.8D maps (Gröger and Plümer, 2005, 2011b).
in Fig. 9a. Such a translation of vertices can violate the 2.8D map
properties since penetrations of faces, intersections of edges and Vertices
1 Different vertices have different coordinates
coinciding vertices may be introduced (see Fig. 10 for an example).
2 Each vertex has at least two incident edges
In addition, the planarity of faces may be affected. However, it is 3 Each vertex is surrounded by exactly one alternating sequence of edges
not necessary to check the axioms for 2.8D maps (Table 1) for all and faces. These faces do not penetrate pairwise
vertices, edges, and faces of the map. Instead, we can safely restrict Edges
checking to a subset which is defined as follows: Let the set of ver- 4 Each edge has exactly two distinct vertices as end points
5 Edges are straight line segments geometrically
tices Vc the geometry of which is modified be the modified vertices.
6 Edges intersect only at common vertices
A modified edge is an edge of the map which is incident to at least 7 Each edge has exactly two distinct incident faces
one modified vertex, and a modified face is a face which is incident Faces
to at least one modified edge. If we restrict the axioms that can be 8 Each face is bounded by exactly one simple cycle of edges
9 Bounded faces are planar
violated by changing geometry (axioms 1, 6, 9, 10) to the modified
10 No point of an edge touches the interior of a face
graph elements, we obtain the precondition of TR5: 11 There is exactly one unbounded face OUT
Graph/map
Transaction rule TR5 12 The underlying graph is connected
Transaction 13 The map is orientable
Change of geometry of a subset VC = {v1,. . .,vn} of vertices
(modified vertices) of a 2.8D map.
Condition
(1) The coordinates of all modified vertices are pairwise dences are relevant (c.f. Section 2.1). Since those incidences are not
distinct, and the coordinates of all modified vertices affected by TR5, orientability (axiom 13) is assured. Since all axi-
are distinct to all non-modified vertices. oms are valid after rule application, TR5 is safe. h
(2) The modified edges are pairwise non-intersecting and Note that an application of TR5 guarantees a topologically
the modified edges do not intersect non-modified equivalent 2.8D map only in the case that the map has no handles.
edges. If handles are present, a handle can ‘switch’ from one side of a sur-
(3) No edge touches the interior of a modified face. face to the other by a rule application. Hence, there is no continu-
(4) No modified edge touches the interior of a face. ous topological deformation from the original map to the resulting
(5) Each modified bounded face is planar. one. However, all incidences between vertices, edges and faces re-
Action main unchanged.
Change the geometry of each vi in VC accordingly. The rule TR5 is well suited for geometrical modifications. To
The application domain of rule TR5 is very general. In addition perform a topological change, e.g., to add vertices, edges, or faces
to local modifications of geometry (c.f. Fig. 9a), it can also be used it can be used in principle as follows: The 2.8D map is translated
to completely derive 2.8D maps from 2D maps. This is due to the to a topological equivalent 2D version, the topological changes
fact that each 2.8D map is topologically equivalent to a 2D map. are accomplished by subsequent applications of the 2D rules TR4
Fig. 11 illustrates that process: By translating each vertex in the flat or TR3, and the result is transformed back to 3D space, yielding
map (a) to a corresponding 3D position, a 3D surface representing a the desired result. However, such an approach is very inefficient
saddle roof building with a chimney is generated. and virtually not feasible. Instead, the rule TR6 is introduced for
The following proposition states that arbitrary applications of topological changes. It replaces a surface bounded by a simple cy-
TR5 maintain the consistency of 2.8D maps: cle by another surface. Both surfaces must be identical with regard
to their boundaries, but can differ in geometry and topology else-
Proposition 3. Transaction rule TR5 is safe. where. Examples for applications of TR6 are given in Fig. 12, where
a building is added to a surface. In Fig. 13, a balcony is added to a
Proof. We have to show that the validity of the axioms (Table 1) is building by rule TR6. In both cases, a simple cycle is selected as
preserved by arbitrary applications of TR5. Assume the axioms are boundary for both surfaces.
valid before TR5 is applied. The axioms 2, 3, 4, 5, 7, 8, 11 and 12 A subtle error case may occur when TR6 is applied (Gröger and
cannot be violated by the application of TR5 and hence are valid Plümer, 2009, 2011b): When a surface to be updated has handles,
afterwards. Axioms 1, 6 and 10 can be violated only for pairs of and when the cycle selected for replacing one surface by another is
graph elements where at least one is modified. If both are non- involved in that handle, the rule application may fail. This is due to
modified, the corresponding axiom cannot be violated. These the fact that on surfaces with handles, Jordan’s curve theorem
checks are accomplished in Cond. 1 (axiom 1), Cond. 2 (axiom 6) (each simple cycle separates a surface in a bounded and an un-
and Cond. 3 and 4 (axiom 10). Likewise, planarity (axiom 9) can bounded part) is not valid. The differentiation in handle and Jordan
only be violated for modified faces and has only to be checked cycles (Gröger and Plümer, 2009, 2011b) solves that problem: only
for those (Cond. 5). For orientability, only the face–edge–face inci- Jordan Cycles (that partition the surface in one bounded and one
140 G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145

Fig. 11. Generation of a 2.8D map representing a saddle roof building (b) by an application of TR5 to a 2D map (a).

Fig. 12. Application of TR3: example.

Proposition 4. TR6 is safe.

Proof. We have to show that the result m0 of any rule application


to a 2.8D map is again a 2.8D map. Since the axioms (Table 1) are
complete, it is sufficient to confirm that the axioms are valid for m0 .
The crucial observation is that the part of m inside the cycle c is
well-defined, since c is a Jordan-cycle (Cond. 4) that separates
the surface m into an unbounded part u and a bounded part b.
Hence, b can be deleted safely (action 1). Validity of axioms 1–10
Fig. 13. Another example for an application of TR3: A balcony is added to a is inherited from m and cs for m/c and cs/c (the symbol / has the
building. usual set-theoretical meaning ‘without’). This is due to the fact that
m and cs touch, i.e. are disjoint apart from c. To show that axioms 1
to 10 are valid for c as well, we only have to check axiom 7: By con-
struction, each edge in c is incident to a face in m and one in cs,
unbounded part) are suitable for aggregating a surface. An efficient hence, axiom 7 is valid. It remains to proof axioms 11, 12 and 13
algorithm to decide whether a given cycle is a handle or a Jordan for m0 . Since no additional unbounded faces are introduced by
cycle has been given in Gröger and Plümer (2011b). This differen- the action part, axiom 11 holds. Connectedness is implied by the
tiation is reflected in the precondition of TR6: connectedness of m and of cs, and by the fact that both touch. Ori-
Transaction rule TR6: (Replacing a surface) entability (axiom 13) is more difficult to show. The map m00 inherits
Transaction orientability from m, and the surface cs is orientable. Since both
Replacing a composite surface by another composite sur- touch in a single, simple cycle c, there is an orientation of the faces
face cs in a 2.8D map m. in m00 and in cs, making both coincide at each edge in c. Hence, m0 is
Condition orientable. h
(1) m is a 2.8D map. A problem which has not been solved yet is how to obtain a
(2) cs is a composite surface of genus zero. composite surface cs which can be replaced in a 2.8D map by
(3) m and cs touch in a simple cycle c. TR6. Since a composite surface is a special 2.8D map where the face
(4) c is a Jordan cycle in m. Out is omitted, it can be generated by rules TR1–TR8. Due to the
Action completeness result for the transaction rules (Proposition 10, see
(1) Delete all vertices, edges and faces inside c from m, Section 4.5), each consistent composite surface can be derived.
yielding m00 . Although there is an overlap of the application areas of the rules
(2) Merge m00 and cs, yielding m0 . TR5 and TR6, no one of them can be omitted. In the proof of the
completeness of the rules in Section 4.5, both rules are required.

Note that according to the remarks at the beginning of that sec- 4.4. Updating surfaces changing the number of handles
tion, touch in Cond. 3 means that m and cs meet only in a single cy-
cle and are disjoint elsewhere. Rule TR3 aggregates a composite surface with a single boundary
The following proposition states that each application of TR6 (which is a simple cycle) to a 2.8D map. This composite surface
maintains the consistency of a 2.8D map: must have genus zero. Hence, the rule cannot be used to introduce
G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145 141

handles which represent tunnels, bridges, arcades or similar han-


dle objects. To insert such objects, the surface to be replaced has
to touch the map in at least two disjoint cycles. This case is covered
by the next rule, which is illustrated in Fig. 14: A composite surface
csh with two boundaries – topologically equivalent to a cylinder
surface, see Fig. 3 for an example – is added to a 2.8D map. The sur-
face csh touches the 2.8D map m in two simple cycles c1 and c2,
which both are Jordan cycles. This condition, which assures that
the vertices, edges and faces inside the cycles can be deleted safely,
is similar to the Cond. 4 of TR6.
Even more complex bridges or tunnels can be inserted by TR6.
Consider, for example, the bridge in Fig. 1. The outer bridge hull Fig. 15. Nested cycles (c1 and c2, hatched lines): an atrium building instead of a
and the surrounding terrain surface together are a surface of genus handle is added.

three. The bridge can be inserted into the terrain surface in three
subsequent applications of TR6: First, the main part of the bridge
is inserted, then one pillar and eventually the other pillar. In gen- b1,...,bn where each bi is a handle cycle. If before the insertion of a
eral, each bridge or tunnel corresponding to a surface of genus n handle, the b1,...,bn are not nested and Jordan-cycles, this static con-
can be inserted by n applications of TR6. dition is valid afterwards and vice versa.
The condition that the surface to be inserted touches the 2.8D Now all prerequisites for a detailed description of the rule for
map in two cycles in not sufficient to confirm that in fact a handle inserting handles into 2.8D maps are available. The rule is as
is introduced. The surface cs (depicted dark grey) in Fig. 15, for follows:
example, touches the 2.8D map in two cycles c1 and c2, but there
is no handle involved. Instead, an atrium building, which is inte- Transaction rule TR7: (Insertion of a handle)
grated in the terrain, is represented: one cycle (c1) is the outer cy- Transaction
cle of the building, and the other (c2) delimits the inner courtyard. Adding a surface csh with two boundaries to a 2.8D map.
To differentiate the atrium building case from the handle case, an Condition
additional condition is required: if the cycles c1 and c2 are both dis- (1) m is a 2.8D map.
joint and not nested, a handle is introduced. In Fig. 15, c1 and c2 are (2) csh is a composite surface with two boundaries c1 and
nested (c2 is inside c1), but in Fig. 14, both cycles are not nested. In c2.
the general case of cycles on surfaces, it is not possible to decide (3) m and csh touch in two simple cycles c1 and c2, which
whether two cycles are nested (i.e. whether one cycle is inside an- are disjoint and not nested in m.
other cycle). However, in our case both cycles are Jordan cycles, (4) c1 and c2 are Jordan cycles in m.
which uniquely define a bounded 2-manifold area. Hence, the no- Action
tion of ‘nested’ is well defined for such cycles. Two cycles are not (1) Delete all vertices, edges and faces inside c1 or c2 from
nested, if the bounded areas defined by each of them are disjoint. m, yielding m00 .
This condition is equivalent to the static condition for handle ob- (2) Merge m00 and csh.
jects introduced in Gröger and Plümer (2011b), which states that
a surface defining a handle object has at least n P 2 boundaries A surface csh which is suitable as input of TR7 can be generated
in a similar way as an input surface for TR6. First, a 2.8D map (pos-
sibly with handles) is generated by rules TR1–TR8. Afterwards, the
unbounded outer face and one bounded face are removed (the
boundary of the outer face and of the face to be removed must
be disjoint). The result is a surface with two boundaries.
The following proposition states that TR7 maintains
consistency:

Proposition 5. Transaction rule TR7 is safe.

Proof. The proof is similar to the safety proof of TR6 (Proposition


4). The argument applied to one single cycle c can directly be trans-
ferred to the two cycles c1 and c2, since the conditions are identical.
Only the proof of the orientability (axiom 13) differs: In Proposi-
tion 4 it was shown that the orientation coincides with regard to
one cycle. However, here we have two disjoint cycles, the orienta-
tions of which have to be compatible. We have to avoid a case
where one touch in c1 between m and csh is from one side of m,
and the other, c2, is from the other (c.f. Fig. 16). But in that case,
the handle csh must have changed the side of m and hence must
penetrate m. This is prevented by Cond. 3. Hence, m0 is
orientable. h
It remains to be shown that TR7 in fact inserts a handle:

Proposition 6. An application of TR7 to a 2.8D map m increases the


number of handles by one, plus the number Hh of handles in the
Fig. 14. Insertion of a handle by TR7: example. surface csh which is inserted.
142 G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145

Fig. 16. Profile of a surface with handle which is not orientable.

Proof. We consider the Euler–Poincaré equation for m (zero


boundaries, one unbounded face), c.f. Eq. (1) in Section 2.2: Fig. 17. Removal of a handle – error case: deletion of surface cs (dark grey color)
with two boundaries b1 and b2 (hatched) yields an unconnected surface.

V m  Em þ F m ¼ 2 Hm þ 2 ð2Þ
Transaction rule TR8: (Removal of a handle)
Transaction
and for csh (two boundaries, i.e. Bh = 2) Deleting a surface csh with two boundaries b1 and b2 from a
2.8D map m, inserting two composite surfaces cs1and cs2.
V h  Eh þ F h þ 2 ¼ 2 Hh þ 2 ð3Þ Condition
(1) m is a 2.8D map.
If the vertices, edges and faces inside c1 and c2 are deleted, we (2) csh is a composite surface with two disjoint boundaries
obtain from Eq. (2) for m00 : b1 and b2.
(3) csh is part of m.
V m00  Em00 þ F m00 ¼ 2 Hm þ 2  2 ð4Þ (4) The composite surfaces cs1 and cs2 to be inserted touch
00 m in b1 and b2.
If we merge m and csh, we have to add Eqs. (3) and (4) to obtain
(5) b1 is a handle cycle.
the equation for m0 (note that deleting the inside of c1 and c2 does
Action
not change the formula, since V and E are equal):
(1) Delete all vertices, edges and faces which are in csh, but
ðV m00 þ V h Þ  ðEm00 þ Eh Þ þ ðF m00 þ F h Þ ¼ 2ðHm þ Hh Þ ð5Þ not in b1 or b2, yielding m00 .
(2) Merge m00 with cs1 and cs2.
0
Since m has zero boundaries and one unbounded face, the
Euler–Poincaré formula for m0 has to be
As in the case of TR7, we show the safety of the rule:
ðV m00 þ V h Þ þ V h Þ  ðEm00 þ V h Þ þ Eh Þ þ ðF m00 þ V h Þ þ F h Þ
¼ 2 Hm0 þ 2 ð6Þ Proposition 7. Transaction rule TR8 is safe.

which implies
Proof. Since csh is connected to the surface only by b1 and b2, and
Hm0 ¼ Hh þ Hm þ 1  ð7Þ since b1 and b2 are disjoint (Cond. 2), the deletion of csh yields a
The next rule we present, TR8, is the inverse rule of TR7. It de- surface with two disjoint holes. These holes are sealed by cs1 and
creases the number of handles in a 2.8D surface by deleting a sur- cs2. Since the preconditions are exactly the same as in TR6, the
face with two boundaries (see Fig. 14 for an example). To restore safety proof for TR6 (Proposition 4) can be applied to cs1 and cs2.
the 2-manifold-property of the surface, both ‘holes’ resulting from We only have to show that the result m00 of a rule application is
the deletion have to be sealed. At first glance this operation seems connected (axiom 12). Suppose m0 is divided in two non-connected
to be simple, but in general the problem of triangulating a 3D poly- components A and B, where b1 is part of A and b2 of B. Since b1 is a
gon is very complex due to penetrations and nested structures handle cycle (Cond. 5), there exists a path p connecting A with csh
(Barequet et al., 1998). To derive such a triangulation in the general not crossing b1. Since csh is connected to the remaining surface
case is extremely time-consuming: this problem is NP-Complete only by b1 and b2 (Cond. 4), p must cross b2. Hence, p must connect
(Garey and Johnson, 1979), i.e. up to now there is no algorithm A and B, a contradiction. Thus m’ is connected. h
known with polynomial running time. However, with the proof Finally, we show that TR8 decreases the number of handles by
of Proposition 9 we show the existence of sealing triangulations one:
for our case. To avoid such problems in the transaction rules, we
assume that the two sealing triangulations are part of the input Proposition 8. An application of TR8 decreases the number of
and are not derived automatically in the action part of the rule. handles by one, minus the number Hh of handles in csh.
An error case of the rule decrementing the number of handles is This proposition is implied immediately by Proposition 6, since
depicted in Fig. 17: the deletion of a surface with two boundaries the surfaces (and the corresponding Euler–Poincaré equations) in-
b1 and b2 yields an unconnected map. The difference between volved in TR7 and TR8 correspond to each other.
the boundaries in that error case and boundaries in a handle case
(c.f. Fig. 14) is the type of cycle: in Fig. 17, b1 and b2 are Jordan cy- 4.5. Completeness
cles,5 whereas in Fig. 14, cycles c1 and c2 both are handle cycles.
Hence, this criterion is used in the condition part of the following We now prove that the rules that were presented in Sections 4.3
rule to avoid that error case: and 4.4 are complete, i.e. that each arbitrary 2.8D map can be con-
structed by subsequent rule applications. The proof proceeds in
5
Note that the surfaces cs (depicted dark grey) in Figs. 15 and 17 are topologically two steps: First we show the result for 2.8D maps without handles.
equivalent. This proof refers to the completeness result for the 2D case in Prop-
G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145 143

osition 2. Based on the non-handle case, the completeness for arbi- (b) There are some vertices, edges or faces (a structure s) inside
trary 2.8D maps is proven (Proposition 10). h, which may penetrate simple triangulations cs1 and cs2
that are constructed according to case (a). Here we have
Proposition 9. The transaction rules TR1–TR6 are complete for 2.8D two sub-cases:
maps without handles. (i) The structure s is not connected on both sides of b1 and
b2 to mn+1, i.e. s is not part of a handle inside the handle
h. In that case, h can be removed similar to case a) by
Proof. Let m be an arbitrary 2.8D map with genus zero, i.e. without
sealing the two openings resulting from unraveling by
handles. Since such a map is topologically equivalent to a 2D map,
triangulations cs1 and cs2 which ‘embrace’ s (see the dot-
there exists a translation t of all vertices of m to a topologically
ted area in Fig. 19d).
equivalent 2D map m0 . This translation can effectively be con-
(ii) The structure s is connected on both sides of b1 and b2 to
structed. Due to the completeness result for 2D maps (Proposition
mn+1 and therefore cannot be embraced by cs1 and cs2
2), m0 can be generated by the rules TR1–TR4. By applying TR5
which seal both openings. Since the surface is a 2-man-
(using the translation that is inverse to t) to m0 we obtain m. Hence,
ifold, s must be (part of a) a handle h0 inside h, i.e. there
m can be generated by the rules TR1–TR6. h
are nested handles. Now we chose the handle h0 for
The idea of the proof for the handle case (which is very techni-
unraveling instead of h, calling the procedure (three
cal) is to proceed by induction on the number of handles. First, an
cases) for h0 recursively. Since the number of handles is
arbitrary handle is removed by TR8. Then the induction assump-
finite, eventually case (a) or (b)-(i) must apply. Hence,
tion is applied, and the same handle is re-added by the inverse rule
there exist penetration-free triangulations cs1 and cs2
TR7. The difficult part of the proof is to show that after removal of
(see Fig. 19e and f), and one handle has been removed.
the handle by TR8, the two resulting stubs can be sealed consis-
tently. Problems occur due to the penetration with other parts of
Note that the cases (a), (b)-(i) and -(ii) define a procedure which
the surface inside the handle. These structures may either be by-
removes a handle. For any cycles b1 and b2 that embraces handle h,
passed, or in case of nested handles, an inner handle is chosen
one of the three cases (a), (b)-(i) or -(ii) applies. In cases (a) and (b)-
for removal. However as the proof shows in detail, this procedure
(i), the sealing triangulations can be constructed immediately. In
eventually terminates:
case (b)-(ii), the procedure is recursively called for the handle h0
which is contained in h. Since the number of nested handles is
Proposition 10. The transaction rulesTR1–TR8 are complete for 2.8D
finite, case (b)-(ii) must apply eventually. Hence, the procedure
maps.
terminates.
Note also that this procedure is the application of rules TR6 and
Proof. The proof proceeds by induction on the number n of han- TR8 to mn+1. Since the resulting map mn has n handles, the
dles of the map. For the base case n = 0, the claim is implied by inductive hypothesis applies to mn and the proposition is valid for
Proposition 9. As the inductive hypothesis we assume that the mn. Now we apply the inverse rule TR7 to mn, followed by inverse
proposition holds for maps mn with n handles. Let mn+1 be a map applications of the rule TR6, yielding mn+1. Hence, mn+1 can be
with n+1 handles. We choose an arbitrary handle h in mn+1 and constructed by applying TR1–TR8. h
show that h can be removed. Let the resulting structure be mn.
Since h is a handle, there must be two handle cycles b1 and b2 5. Related work
(the edges of which must be inserted by TR6 if necessary, see the
three cases below) which embrace h. We have to consider three Transaction rules for the 2D case have been developed in Gröger
cases (two main cases, where the second one has two sub-cases): and Plümer (1997), Kufoniyi (1995) and Kufoniyi et al. (1995), and
for the special case of simplicial complexes (i.e. cell complexes con-
(a) The handle h can immediately be removed by cutting h at sisting of triangles) in Egenhofer et al. (1989). For constructing 3D
cycles b1 and b2. The cycles b1 and b2 can be constructed surface models stepwise, the well-known Euler Operators (Mäntylä,
by applying TR6, such that the distance between b1 and b2 1988; Mortenson, 1997) have been introduced. However, Euler
has an infinitesimal small size (see Fig. 18b). This operation Operators only preserve topological consistency and do not focus
yields two disconnected stubs with an infinitesimal small on geometric-topological consistency: the geometry of objects is
distance. Both stubs are sealed by triangulating the two not considered and penetrations are not explicitly avoided. The
openings resulting from cutting. Let the two triangulations granularity of Euler operators is very fine in contrast to the rules
be cs1 and cs2 (see Fig. 18c). Since in case (a), ‘inside’ the han- presented in this paper. Only single surfaces can be subject to an
dle h there are no other vertices edges or faces penetrating operation, not composite surfaces as in our case. Hence, an error
cs1 or cs2 (see Fig. 19a and b), the constructions of the trian- case involving handles which is prevented by Cond. 4 of TR6, can
gulations cs1 and cs2 are simple. We do not need to effec- not occur if Euler operators are applied.
tively construct cs1 and cs2, it is sufficient to assume the In Gold (2003) and Tse and Gold (2002), Euler Operators
existence of both triangulations. and GIS are coupled by applying the operators to triangula-

Fig. 18. Detailed illustration of the removal of handle h in Proposition 10. (a) Handle h. (b) Two handle cycles b1 and b2 embracing h are constructed. Both cycles define the
composite surface csh (hatched). (c) Removal of csh (and hence, h) by applying TR8 to b1, b2 and csh. The composite surfaces cs1 and cs2 which have been inserted in order to
seal both stubs are depicted hatched.
144 G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145

Fig. 19. Diagrammatic illustrations (profiles) of the procedure to remove handles in the completeness proof. (a and b) Simple case where both stubs of handle h (hatched) can
be sealed by triangulations cs1 and cs2 (thick lines). (c and d) Non-handle objects (light grey) have to be by-passed by both triangulations. (e and f) Nested handles, interior
handle h0 (light grey) chosen for removal by application of the first case (a) and (b). The two triangulations cs1 and cs2 are depicted in thick lines.

tions. The approach is restricted to triangulations, while our of non-handle objects like buildings, balconies and roofs, as well as
rules deal with planar faces of arbitrary shape. For some oper- of handle objects such as bridges, tunnels or arcades. The rules can
ators, the granularity becomes much coarser than the corre- be implemented based on available database technology. They pro-
sponding Euler operator. Particularly, there is one operator vide a topological foundation for automatic procedures which reli-
which inserts a handle. However, the handle is restricted to ably detect errors possibly occurring whenever data sets are
a prism geometrically. This is in contrast to our rule TR7 updated. To the best of our knowledge, this is the first time that
which inserts an arbitrary composite surface with two bound- rules, which are both safe and complete, are given for 3D surfaces
aries. Hence, this rule does not provide the flexibility of our in GIS.
rules, and the rules are not complete, because non-straight Besides the application for updating surfaces, the transaction
handles are outside the scope of the approach. In Boguslawski rules play a crucial role for updating solids, which model the com-
and Gold (2008) and Boguslawski and Gold (2009), Euler oper- plete hull of buildings and their interior structures composed of
ators are used to generate a single solid model, whereas ex- rooms and staircases. Due to its modular structure, the transaction
tended Euler operators are mentioned, which are capable of rules TR1–TR8 can be integrated in a system of rules for solid mod-
constructing models consisting of multiple solids. However, as els (Gröger and Plümer, 2012) and enable updates that affects only
it is the case with general Euler operators, the focus is on a single surface and allow for the generation of surfaces that are
topological consistency only, and no proofs of completeness used to construct solids and aggregations of solids.
or correctness of the rules are delivered. In this paper, we are providing mathematical proofs of the
safety and the completeness of the transaction rules, guaranteeing
the reliability of our method. The completeness proof (Proposition
6. Conclusions 10) turned out to be very difficult, due to the occurrence of handles
in surfaces. For completeness, it is shown that handles can be re-
This paper provides a rigorous mathematical treatment of a set moved without violating consistency. To maintain a 2-manifold
of transaction rules that enables updates of 3D surface models surface, the two stubs resulting from the removal of a handle have
which reflect changes of our urban and rural environment. These to be sealed. This turned out to be demanding since handles may
rules maintain consistency of the surface model once given (safety) contain other structures or even other handles (nested handles)
and provide the flexibility to generate arbitrary consistent surface in its interior. However, we proved that consistent sealing is possi-
models (completeness). They aim at constructing and updating ble in any case.
surfaces embedded in 3D space and modeling the terrain and the The given rules deal with faces having exactly one exterior cycle
outer hull of objects. They allow for the construction and removal as boundary. More general are cases where faces are additionally
G. Gröger, L. Plümer / ISPRS Journal of Photogrammetry and Remote Sensing 69 (2012) 134–145 145

delimited by interior cycles. Such cycles form holes or enclaves in Gold, C., 2003. But is it GIS? Journal of Geospatial Engineering 5 (2), 11–26.
Gröger, G., Kolbe, T.H., Czerwinski, A., Nagel, C. (Eds.), 2008. OpenGISÒ City
the face. That generalization alleviates the representation of, for
Geography Markup Language (CityGML) Encoding Standard. Version 1.0.0.,
example, free-standing buildings or windows in walls. In our ap- Open Geospatial Consortium, OGC Doc. No. 08–007r1.
proach, interior cycles are represented by using so-called pseudo Gröger, G., Kolbe, T.H., Schmittwilken, J., Stroh, V., Plümer, L., 2005. Integrating
edges, which add artificial edges with no counterpart in reality. versions, history and levels-of-detail within a 3D geodatabase. In: Gröger, G.,
Kolbe, T.H. (Eds.), Proceedings of the 1st international ISPRS/EuroSDR/DGPF-
Examples are the two edges connecting the outer boundary of Workshop on Next Generation 3D City Models, Bonn, June 21–22. EuroSDR,
the structure in Fig. 12 with the building footprint (depicted Publication No.49, pp. 35–40.
hatched). If interior cycles are considered, we have to cope with Gröger, G., Plümer, L., 1997. Provably Correct and Complete Transaction Rules for
GIS. In: Laurini, R., Bergougnoux, P. (Eds.), Proceedings of the 5th International
disconnected graphs, and some transaction rules have to be mod- Workshop on Advances in Geographic Information Systems (ACM-GIS’97). ACM
ified. In addition, the algorithm to discriminate between handle Press, Las Vegas, Nevada, pp. 40–43.
and Jordan cycles has to be adjusted accordingly. Gröger, G., Plümer, L., 2003. Exploiting 2D Concepts to Achieve Consistency in 3D
GIS Applications. In: Hoel, E., Rigaux, P. (Eds.), Proc. of the 11th International
The combination of the rules TR5–TR8 contributed by this paper Symposium on Advances in Geographic Information Systems (ACM-GIS’03).
with the rules for the 2D case (TR1–TR4, see Gröger and Plümer, ACM Press, New Orleans, Louisiana, pp. 78–85.
1997) and the rules for solids and aggregations of solids (TR9– Gröger, G., Plümer, L., 2005. How to get 3-D for the price of 2-D-Topology and
Consistency of 3-D Urban GIS. Geoinformatica 9 (2), 139–158.
TR12, see Gröger and Plümer, 2012) constitutes a powerful concept Gröger, G., Plümer, L., 2009. Updating 3D City Models – how to preserve geometric-
for maintaining updates for general 3D city models in a reliable, topological consistency. In: Agrawal, D., Aref, W.G., Lu, C.T., Mokbel, M.F.,
flexible way. This concept will be implemented as part of a sustain- Scheuermann, P., Shahabi, C., Wolfsohn, O. (Eds.), Proceedings of the 17th ACM
SIGSPATIAL International Conference on Advances in Geographic Information
able spatial data infrastructure (SDI). The Aristoteles 3D, a Java-based
Systems (ACM SIGSPATIAL GIS 2009), Seattle, Washington, November 4–6,
3D GML viewer which has been developed at the Institute for 2009. ACM Press, New York, pp. 536–539.
Geodesy and Geoinformation at the University of Bonn, already Gröger, G., Plümer, L., 2010. Derivation of 3D indoor models by grammars for route
provides functionality for semantical editing and a powerful planning. Photogrammetrie, Fernerkundung, Geoinformation - PFG 3, 193–210.
Gröger, G., Plümer, L., 2011a. How to achieve consistency for 3D city models.
plug-in concept. The Aristoteles will be augmented by a compo- Geoinformatica 15 (1), 137–165.
nent which enables the interactive geometrical editing of 3D city Gröger, G., Plümer, L., 2011b. Topology of surfaces modelling bridges and tunnels in
models and by a component which checks those updates by imple- 3D-GIS. Computers, Environment and Urban Systems 35 (3), 208–216.
Gröger, G., Plümer, L., 2012. Provably correct and complete transaction rules for
menting the transaction rules. Since the Aristoteles plays the role updating 3D city models. Geoinformatica 16 (1), 131–164.
of a client for transactional Web Feature Services (WFS-T) (Vretanos, Gröger, G., Reuter, M., Plümer, L., 2004. Shall 3-D city models be managed in a
2004) in a SDI, the application of our tool will assure the consis- commercial database? Geo-Informationssysteme (GIS) 17 (9), 9–15.
Harary, F., 1969. Graph Theory. Addison Wesley Publishing Company.
tency of 3D city models being stored in a CityGML database (Gröger Hatcher, A., 2001. Algebraic Topology, 10th ed. Cambridge University Press,
et al., 2004, 2005) and being accessible by a standardized WFS-T Cambridge.
interface. Herring, J.R., 2001. The OpenGIS Abstract Specification, Topic 1: Feature Geometry
(ISO 19107 Spatial Schema). Open Geospatial Consortium, OGC Document
Number 01–101, Version 5.
Acknowledgements Kolbe, T.H., Gröger, G., Plümer, L., 2008. CityGML – 3D City Models for Emergency
Response. In: Zlatanova, S., Li, J. (Eds.), Geospatial Information Technology for
Emergency Response. Taylor & Francis, London, pp. 257–274.
We would like to thank Michael Kneuper for his assistance in
Köninger, A., Bartel, S., 1998. 3D-GIS for Urban Purposes. Geoinformatica 2 (2), 79–
preparing the illustrations. 103.
Kufoniyi, O., 1995. Spatial coincidence modelling, automated database updating and
References data consistency in vector GIS. PhD thesis. ITC Publication 28, ITC Dissertation
20, International Institute for Aerospace Survey and Earth Sciences (ITC),
Enschede, The Netherlands.
Alexandroff, P., 1961. Elementary concepts of topology. Dover, New York. Kufoniyi, O., Molenaar, M., Bouloucos, T., 1995. Topologic consistency operations in
Armstrong, M.A., 1997. Basic Topology, Corr. 5. printing. Springer, New York. vector maps. Geo-Informations systeme (GIS) 8 (4), 7–13.
Barequet, G., Dickerson, M., Eppstein, D., 1998. On triangulating three-dimensional Lake, R., Burggraf, D., Trninic, M., Rae, L., 2004. Geography Mark-Up Language:
polygons. Computational Geometry: Theory and Applications (CGTA) 10 (3), Foundation for the Geo Web. John Wiley & Sons, Chichester.
155–170. Mäntylä, M., 1988. An Introduction to Solid Modeling. Principles of Computer
Boguslawski, P., Gold, C., 2008. Construction Operators for Modelling 3D Objects Science. Computer Science Press.
and Dual Navigation Structures. In: Lee, J., Zlatanova, S. (Eds.), 3D Geo- Molenaar, M., 1992. A topology for 3D vector maps. ITC Journal 1992–1 (1), 25–33.
Information Sciences. Springer, Berlin, Heidelberg, pp. 47–59. Mortenson, M.E., 1997. Geometric Modelling. John Wiley & Sons.
Boguslawski, P., Gold, C., 2009. Euler Operators and Navigation of Multi-shell Okabe, A., Boots, B., Sugihara, K., Chiu, S.N., 2000. Spatial Tessellations. Concepts and
Building Models. In: Neutens, T., Maeyer, P. (Eds.), Developments in 3D Geo- applications of Voronoi diagrams, 2. ed. Wiley, Chichester.
Information Sciences. Springer, Berlin, Heidelberg, pp. 1–16. Plümer, L., Gröger, G., 1997. Achieving integrity in geographic information systems-
Czerwinski, A., Sandmann, S., Stöcker-Meier, E., Plümer, L., 2007. Sustainable SDI for maps and nested maps. Geoinformatica 1 (4), 345–367.
EU noise mapping in NRW – best practice for INSPIRE. International Journal for Tse, R.O.C., Gold, C., 2002. TIN meets CAD – Extending the TIN Concept to GIS. In:
Spatial Data Infrastructure Research 2 (1), 90–111. Sloot, P.M., Tan, C.K., Dongarra, H.J., Hoekstra, A. (Eds.), Computational science-
Egenhofer, M.J., 1989. A formal definition of binary topological relationships. In: ICCS 2002. Proc. of the International Conference on Computational Science,
Litwin, W., Schek, H.-J. (Eds.), Foundations of Data Organization and Algorithms. Amsterdam, The Netherlands, April 21–24, 2002, Part III. Springer, Berlin, New
Proc. 3rd International Conference, FODO, Paris, France, June 21–23, 1989. York, pp. 135–143.
Springer, Berlin Heidelberg, pp. 457–473. Vretanos, P.A., 2004. Web Feature Service Implementation Specification, Version:
Egenhofer, M.J., Frank, A.U., Jackson J. P., 1990. A Topological Data Model for Spatial 1.1.0. Open Geospatial Consortium, OGC Doc.-No. 04–094.
Databases. In: Buchmann, A.P., Günther, O., Smith, S., Wang, Y.F. (Eds.), Design Widom, J., Ceri, S., 1996. Active Database Systems. Morgan Kaufmann.
and Implementation of Large Spatial Databases. First Symposium SSD ‘89 Santa Zlatanova, S., 2000. 3D GIS for urban development. PhD Thesis. ITC Dissertation 69,
Barbara, California, July 17/18, 1989 Proceedings. Springer, Berlin, Heidelberg, International Institute for Aerospace Survey and Earth Sciences, Enschede, The
pp. 271–286. Netherlands; Institute for Computer Graphics and Vision, Graz University of
Foley, J.D., van Dam, A., Feiner, S.K., Hughes, J.F., 1995. Computer Graphics: Technology, Graz, Austria.
Principles and Practice, 2. ed. Addison Wesley Publishing Company; Addison- Zlatanova, S., Holweg, D., 2004. 3D Geo-information in emergency response: a
Wesley, Boston. framework. In: Proceedings of the Fourth International Symposium on Mobile
Garey, M., Johnson, D., 1979. Computers and Intractability. A Guide to the Theory of Mapping Technology, March 29–31, Kunming, China.
NP-Completeness. WH Freeman.

You might also like