PROCEDURAL MODELLING OF MONUMENTAL BUILDINGS

FROM TEXTUAL DESCRIPTIONS

Roberto Rodrigues, António Coelho

DEI/FEUP/ INESC Porto, Rua Dr. Roberto Frias, s/n, 4200-465 Porto, Portugal

Luís Paulo Reis

DEI/FEUP/ LIACC, Rua Dr. Roberto Frias, s/n, 4200-465 Porto, Portugal

Keywords: Procedural Modelling, Natural Language Processing.

Abstract: The generation of three-dimensional models of urban environments using procedural modelling is presented

as being a solution which allows financial and temporal gains, maintaining an acceptable visual fidelity

level. Nevertheless, the modelling of anchor buildings (or monumental), identifying certain urban areas,

needs a more careful modelling due to the high level of detail necessary, using, generally, manual

modelling. We present an automation proposal of the building modelling process through the introduction of

additional knowledge from textual descriptions in a procedural modelling system. The results show that the

data model is flexible enough to build distinct models of churches. The data model can also provide an

initial structure for high level modelling, providing the global shape for the building and the location of

doors, windows or other structures. High detailed models can be built from this initial structure. The results

demonstrate also that it is possible to create a 3D model from a text and thus permitting that non-specialised

users may increase effectiveness using a procedural modelling system.

1 INTRODUCTION

Virtual Reality is being used in diverse areas, from

leisure applications, such as digital games until more

technical uses, like urban planning. The

technological advances have lead to the

vulgarization of its use, showing the way to an

improvement of the services offered and to cost

optimization.

Although there are several interactive tools for

3D building reconstruction, these demand

specialized technicians and great amount of

informatics resources. The use of procedural

modelling tools for generating urban buildings,

allows a great reduction of costs, either by the non-

use of specialized technicians, or by the reductions

of the time spent during its execution.

Among the most representative buildings of an

urban environment, the monuments are the ones that

need more detail, in order to be easily recognized,

increasing the modelling time, and being therefore

more expensive.

Although procedural modelling allows saving

time and money, the definition of the modelling

processes for monumental buildings can be quite

complex, because most of these tools are based on

formal grammars which also demands for highly

trained professionals. This way, we introduce a

proposal for procedural modelling of buildings from

a descriptive text that can be used on the modelling

process. The descriptive text is converted for an

intermediate format in XML, being then converted

to cityGML (Kolbe, Gröger et al. 2005).

The article is structured like this: in section 2 we

will present some related work in this area, showing

the architecture proposed in section 3. Section 4 will

display some of the results and in section 5 will

present the conclusions and future developments.

2 RELATED WORK

2.1 Three-dimensional Modelling

The creation of three-dimensional models of buil-

130

Rodrigues R., Coelho A. and Paulo Reis L. (2010).

PROCEDURAL MODELLING OF MONUMENTAL BUILDINGS FROM TEXTUAL DESCRIPTIONS.

In Proceedings of the International Conference on Computer Graphics Theory and Applications, pages 130-133

 SciTePress

dings for urban environments is a long lasting

process that can be used in archaeology (Pollefeys,

Proesmans et al. 2000), (Uotila and Sartes 2000),

(Bernardes and Martins 2003) and (Frischer 2008),

in environmental simulations and in architecture or

urban planning (Shiode 2001) (Döllner, Kolbe et al.

2006).

The proposal cityGML (Kolbe, Gröger et al.

2005), covers the geometric, topologic and semantic

aspects of three-dimensional urban models. It is a

free format and based in ISO standards and sustains

3D simple or complex geometry and topology. It

stores not only the geometric information, referring

to shape and topologic characteristics of buildings,

but also semantic information, with a

correspondence between the geometric object and

the semantic object.

2.2 Procedural Modelling

The costs and time used to model urban

environments have led to the appearance of some

research works for automating those tasks, aimed to

the creation of procedural modelling solutions.

There are two guidelines for this process,

automating the collecting process and data

interpretation or enlarging the original data through

a wider base of knowledge.

In the first case, it is necessary to collect a wide

set of information about the environment to be

modelled. Beside photos and video, the data can also

be gathered from three-dimensional point meshes or

from laser or x-ray sweeping.

In the second case, the widening of the original

data is obtained through mathematical tools, which

amplify the original information. The L systems and

shape grammars are among the tools that allow a

high level of data amplification.

With the proliferation of informatics systems,

information has become available in several sources

and often, not compatible. A proposal of semi-

automatic expeditious modelling (Coelho, Bessa et

al. 2007) proposes a system for modelling urban

environments, with access to different information

sources, and that uses L systems to carry out the

modelling.

The works of Wonka and Müller, from the

definition of split grammars (Wonka 2003) to the

development of CGA Shape grammars (Mueller,

Wonka et al. 2006) allows the creation of detailed

building facades (Mueller, Zeng et al. 2007).

In the range of graphic computing, a system (Liu,

Xu et al. 2006) was presented for the reconstruction

of old Chinese houses, using a semantic model. This

system converts the geometric components, like

points, lines or triangles into semantic components,

such as, streets, blocks and houses.

3 PROCEDURAL MODELLING

OF BUILDINGS FROM

TEXTUAL INFORMATION

In this section we describe a modelling process for

extracting all relevant information, from textual

descriptions, in order to create a 3D model. Since

this is a complex process it has been divided in

several development phases.

In the first phase, the system reads a text that

describes a simplified monumental building using a

simple grammar.

The proposed modelling process will generate a

3D model of a church, in cityGML format from a

natural language text, written in Portuguese.

Figure 1 presents the general architecture scheme

of the modelling process.

Figure 1: General architecture.

The modelling process is divided in two stages:

 Extraction of the model’s characteristics,

contained in the text file written in natural

language, for an interoperable format and stored

as a XML file. This file should obey the

specifications of the XML scheme from which

we can see an extract in figure 2;

 Conversion of the data model to a 3D model, in

cityGML format.

PROCEDURAL MODELLING OF MONUMENTAL BUILDINGS FROM TEXTUAL DESCRIPTIONS

131

Figure 2: Extract of XML scheme of the internal model.

3.1 Natural Language Processing

The information extraction module, was developed

using NooJ (Silberztein 2004). This grammar uses

the recognition resources implemented by the portal

Port4NooJ (Ribeiro 2008), having been developed

specific resources for this area.

The formats of churches and their descriptive

texts have been analyzed, and a typical text has been

created for this type of building that describes it

realistically.

Therefore, the text should indicate the type of

building being analyzed, describing its constitution

in terms of buildings or towers. The text should also

indicate the relative position of the towers.

A generic church can be described by a text as

the one presented in figure 3.

Figure 3: Textual representation of a church.

The information extraction module, retrieving

the related information for the data model, processes

this text.

3.2 Data Model

The data model can store one or more buildings, but

the first building is set as the origin of the

referential. All buildings can be composed by a main

body (or central wing), one or more towers and one

or more entrances.

All buildings have, necessarily, a main body.

Figure 4 shows an extraction of the XML internal

model created from the Figure 3 text. The

description of a building must always be carried out

with the observer facing the main entrance.

For each tower, besides its attributes (length,

width, height and span) one must also refer the

position in relation to the main building and axis X

and Y. Then, the posX may assume the value “left”,

“centre” or “right”, being also indicated, in

percentage, how much of the tower is (or not) inside

the main building. In the case of posY, the possible

values it can assume are “front”, “middle” or

“back”. In case it is not mentioned in the text what

the value of this percentage is, the natural language

processing module assumes that the tower is leaning

against the building. The shape of the roofs is also

automatically associated to the type of roof, having a

pyramid shape in the towers and a triangle in the

others.

Figure 4: Extract from XML file, representing a church

with two towers.

3.3 Creating the 3D Model

The data model contains semantic information that

identifies the type of building (main or tower) and

also information about its position. A geometric

shape corresponds to each type of building, as

defined for this prototype phase. This way, it is

possible to convert this data model to a 3D

representation. We choose cityGML because it is

possible to combine semantic and geometric infor-

</location>

</main>

<posX deviation="-1.0">right</posX>

<posY deviation="1.0">front</posY>

</position>

…

< bell / >

</features>

</tower>

…

<xs:element name="building">

<xs:complexType>

<xs:sequence>

<xs:element ref="type" />

<xs:element ref="name" />

<xs:element ref="main" />

<xs:element ref="towers" />

</xs:sequence>

</xs:complexType>

</xs:element>

“The church of XYZ has a main body and 2 towers.

The bell tower is on the right and front.

The tower is on the left and front.”

GRAPP 2010 - International Conference on Computer Graphics Theory and Applications

132

mation and has several levels of detail.

We created two types of buildings that

correspond to the two main structures presented in

the data model, the main building that represents the

church core building and the tower that in most of

the cases has a bell or a clock.

This module explores each XML node from the

data model, checks if the node represents a structure,

and creates the cityGML building represented in that

node. The type of structure and the level of detail

used define the final shape of the building. Highly

detailed templates can be used for each component

(door, window or facade) or basic solid geometries.

We have developed a module that converts the

data model into cityGML. This module was created

using a free library provided by cityGML and called

citygml4j.

4 RESULTS

To test the prototype a couple of churches from

Oporto were created. Figure 5 shows some

cityGML3D models.

Figure 5: CityGML models created by the prototype.

5 CONCLUSIONS AND FUTURE

WORK

The results demonstrate that the data model of the

prototype is flexible enough to build countless

church models, with an acceptable level of realism

in LOD3. The results also show that it is possible to

create a 3D model from a simple text.

This way, future work will concentrate in

increasing the level of detail of the model adding

textures and evaluating the use of the model for high

level modelling, since detailed doors and windows

would increase the level of visual fidelity.

REFERENCES

Anslow, C., S. Marshall, et al. (2006). Evaluating X3D for

use in software visualization. Proc. of the 2006 ACM

Symp. Software Visualization. Brighton, United

Kingdom, ACM: 161-162.

Bernardes, P. and M. Martins (2003). Computação Gráfica

e Arqueologia Urbana: O caso de Bracara Augusta.

Actas do 12º Encontro Português de Computação

Gráfica, ISEP, Porto.

Coelho, A., M. Bessa, et al. (2007). Expeditious

Modelling of Virtual Urban Environments with

Geospatial L-systems. Computer Graphics Forum

26(4): 769–782.

Döllner, J., T. H. Kolbe, et al. (2006). The Virtual 3D City

Model of Berlin - Managing, Integrating, and

Communicating Complex Urban Information.

Proceedings of the 25th Urban Data Management

Symposium UDMS, Aalborg, DK.

Frischer, B. (2008). The Rome Reborn Project. How

Technology is helping us to study history. OpEd,

November 10. University of Virginia.

Kolbe, T. H., G. Gröger, et al. (2005). CityGML -

Interoperable Access to 3D City Models. International

Symposium on Geoinformation for Disaster

Management.

Liu, Y., C. Xu, et al. (2006). Semantic modeling for

ancient architecture of digital heritage Computers &

Graphics 30(5): 800-814.

Mueller, P., P. Wonka, et al. (2006). Procedural Modeling

of Buildings. Proceedings of ACM SIGGRAPH.

Mueller, P., G. Zeng, et al. (2007). Image-based

Procedural Modeling of Facades. Proceedings of ACM

SIGGRAPH.

Pollefeys, M., M. Proesmans, et al., Eds. (2000).

Acquisition of Detailed Models for Virtual Reality.

Virtual Reality in Archaeology, British Archaeological

Reports, Int. Series #843.

Ribeiro, A. (2008). Port4NooJ: an open source, ontology-

driven Portuguese linguistic system with applications

in machine translation. Proc. of the 2008 Int. NooJ

Conference (NooJ'08), Budapeste, Hungary, 8-10

June.

Shiode, N. (2001). 3D urban models: Recent

developments in the digital modelling of urban

environments in three-dimensions. GeoJournal (52):

263-269.

Silberztein, M. (2004). NooJ: A Cooperative, Object-

Oriented Architecture for NLP. INTEX pour la

Linguistique et le traitement automatique des langues.

Cahiers de la MSH Ledoux, Presses Universitaires de

Franche-Comté.

Uotila, K. and M. Sartes (2000). Medieval Turku – The

Lost City. A Project trying to reconstruct a Medieval

Town in Finland. Virtual Reality in Archaeology. J. A.

Barceló et al. eds, British Archaeological Rep., Int.

Series #843: 219–223.

Wonka, P. W., M.; Sillion, F.; Ribarsky, W (2003). Instant

Architecture. ACM Transactions Graph 22(3): 669–

677.

PROCEDURAL MODELLING OF MONUMENTAL BUILDINGS FROM TEXTUAL DESCRIPTIONS

133