HAL Id: halshs-00264181https://halshs.archives-ouvertes.fr/halshs-00264181

Submitted on 1 Apr 2008

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.

Semantic-based modelling and representation ofpatrimony buildings

Livio De Luca, Philippe Véron, Michel Florenzano

To cite this version:Livio De Luca, Philippe Véron, Michel Florenzano. Semantic-based modelling and representationof patrimony buildings. SVE Worksop towards Semantic Virtual Environments, Mar 2005, Villars,Switzerland. pp.1-11, 2005. <halshs-00264181>

Semantic-based modelling and representation ofpatrimony buildings

Livio De LucaLaboratoire MAP

UMR CNRS/MCC 694ldl@map.archi.frwww.map.archi.fr

Philippe VéronLaboratoire LSISUMR CNRS 6168

Philippe.Veron@aix.ensam.frwww.lsis.org

Michel FlorenzanoLaboratoire MAP

UMR CNRS/MCC 694mfl@map.archi.frwww.map.archi.fr

AbstractThis article presents a methodologicalapproach for the semantic description ofpatrimony buildings based both on theoreticalreflections and on research experiences. Todevelop this approach, a first process ofextraction and formalisation of architecturalknowledge based on the analysis ofarchitectural treaties is proposed. Then, theidentified features are used to produce atemplate shape library dedicated to thebuildings surveying. Finally, the problem ofthe overall model structuration andorganization using semantic information isaddressed for user handling purposes.

Keywords: architectural heritage, architecturalknowledge, feature-based modelling, virtualenvironments, semantic shape.

1. IntroductionUnderstanding studies of architectural heritagecan be today widely improved by the use of 3Dreconstruction of the real buildings. This wayis also efficient to document historic buildingsand sites for their reconstruction or restoration,to create resources for researchers and toanalyze their historical evolutions.The spatial data acquisition techniques haveknown in these last years large improvementsdue to the introduction of laser scanningtechnologies [1]. Nevertheless, if the digitizingphase can profit of automatic and high-speedprocedures, the 3D point clouds postprocessing requires many manual actions whenthe objective is to produce an analyticalrepresentation of architectural objects. In fact,if this technical solution theoretically allowshaving an accurate geometric reproduction of abuilding with all its details, it remains still theproblem of producing a semantically enrichedmodel [2]. A triangulated-model can be widely

appropriate to represent artistic objects such assculptures or facing relief [3]. But, when thedefinition of geometrical relations and rulesbetween dimensions are required to describethe proportions of typical objects participatingto an architectural project, this type ofgeometrical representation is not efficient.With regard to the set of classical architecturalshapes, relevant information, which must beacquired, recognized and modelled, can befound in a set of relevant profiles describingthese shapes on the basis of the conventions ofarchitectural drawings.Whatever the digitizing device used, beforestarting the treatment of the 3D data sets, it isessential to take into account the representationproblems, because the reconstruction processcarried out makes a geometrical interpretationof the architectural elements. Indeed, if theactual reconstruction techniques provide goodsurface reconstructions, they do not necessarilyintegrate the semantic interpretation of thearchitectural shapes. In this sense, the efficientproposed way to define an architecturalsurveying is to consider it as a rebuildingproblem [4]. Architectural surveying is areverse process which starts from the realobject, rebuilds a digital model and interpretsthe idea which is upstream of its realization. Inan architectural building project two essentialpoints of views can be characterized. The firstone is related to the architect who has an idealand perfect design reference and the secondone relates the concrete and imperfect realrealization. In the same manner, thearchitectural surveying can be articulate aroundtwo points of views. The first one is related tothe crude processing of the 3D data sets whichproduce an imperfect digital model and thesecond one is related to the exploitation ofsemantic information producing an enhanceddesign model of the real building.

In this way, the extraction of relevantinformation from the 3D digitized data,according to the architectural drawingconventions (related to the different historicperiods), forms an important problem whichmust be addressed. To this end, semantic-basedrepresentations of patrimony buildings canthereby play an important role for thedevelopment of information systems in thearchitecture field. Beyond survey, in fact, thedevelopment of qualitative descriptions ofarchitectural objects is a major research area.Nevertheless, if the scientific analysis of crudedocumentary sources has benefited from thedevelopment of various data managementtechniques, it has most often been done withouta concern for graphical visualisation or for aclear reference to the architectural morphology.Architectural surveying and documentationseems to relate to totally independent researchareas whereas they have naturally commonproblematic around the integration ofinformation and knowledge associated to anedifice. Consequently, today it appears asessential to take into account and integratequantitative information supported by thesurveying process and qualitative informationor knowledge extracted from documentationsources. The approach proposed, sets a methodto formalize the theoretical knowledge ofpatrimony buildings which is then used tosupport the different investigation stages of anedifice, from 3D surveying to representationand handling in a 3D real-time environment.The paper is organized as follow. Section 2discusses the problem of the semanticdescription of patrimony buildings. Section 3describes the knowledge-based modelling toolsdeveloped, starting from a method for thefeature formalisation from architecturaltreaties. Section 4 presents a result example ofsemantic-based representation for informationinterfacing. Finally, section 5 concludes thepaper and suggests some future works.

2. Architectural knowledge system The study of the drawing convention in thehistory of the architectural representation has adouble finality: the first one leads to therepresentation and the second one leads to theobject surveying. These two analysis points ofviews of the architectural elements are strictlyinterdependent [5]. The knowledge extractionproblem consists in identifying the genesis of

the element shape to define, both, theappropriate way for its measurement and for itsrepresentation.

Figure 1: extract of the Palladio’s treaty ofArchitecture

To this end, architectural knowledge rules haveto be formalised. An architectural knowledgesystem can be described as a collection ofstructured objects, identified through a precisevocabulary. Several studies led to thedefinition of classification methods forclassical architectural elements.

Figure 2: synthesis of the elements participatingto a 3D building reconstruction process: (a)

documentary sources, (b) formalized knowledge,(c) real object, (d) semantic description

These are based on a structure of different levelof abstraction of architectural space [6]. Theseclassifications are based on the study of thearchitecture treaties, which organize the ‘art tobuild’ knowledge relatively to the differenthistorical periods. Many treaties develop anidentity coding of architectural elements. Thisidentity is normally expressed through ahierarchical description of all the elements,which make a build unit (Fig.1).In [7], by means of a representationconvention, each architectural element isexpressed by a geometrical description level(i.e. lines, curves), a topological relations level(i.e. parallelism, concentricity, etc) and aspatial relations level (i.e. proportions,harmonic reports/ratios). The problematic ofthe semantic description of patrimonybuildings return to the extraction of these threelevels starting from the 3D digitized data of areal building. It starts from the analysis ofvarious sources (Fig.2 a) of knowledge(including the study of particular cases) toextract drawing rules. These rules areformalised (Fig.2 b) to produce theirappropriate digital translation into a 3Dsemantic-based description (Fig.2 d). Thesemantic description of an architecturalelement gathers modelling functions,constraints links and hierarchical relationships.

2.1 Architectural vocabulary

Figure 3: a Roman temple of Corinthian order:the Capitol of Dougga inTunisia

To explain the notion of semantic descriptionof patrimony building, we take as example theanalysis of a teatrastyle, prostyle temple ofCorinthian order (Fig.3). The cella is precededby a pronaos composed by four columns in

frontage; this pronaos is closed on the sides ofthe cella from which a side column separates it.The temple rises on a stylobate. That finisheseach side, and in front of the pronaos by twolong pedestals including between them thestairs. The bases of the Corinthian order arecomposed of a surmounted plinth of a lowertorus and a higher torus connected by twoscoties separate one of the other by a rod andtwo l i n t e l s . The architraves are simplyprofiled. The cornice of the entablature isrichly decorated.Based on the previous description, the firstphase carried out a reasoned decomposition ofthe building using a vocabulary analysis [8].The goal is to identify the vocabulary and todescribe in a theoretical way an architecturalentity (i.e. definition of its model) (Fig.4).This analysis allows expressing themorphology of the sub-elements thatparticipate to the architectural unit. The goal ofthe next phase is to identify the compositionrules which govern the building arrangement.

Figure 4: semantic description of a Corinthian

order using the appropriated vocabulary (basis,capital, entablature)

2.2 Architectural grammar

By looking at the frontage of the temple, onefinds the distinction given by Vitruve [9],which says that the building must be ordered:

• with symmetry ([9] book I, course, II),i.e. its various parts respect an aspectreport/ratio (the module) which drivesa large number of the buildingdimensions in the same way that forthe human body with the report/ratiobetween the feet and the hands forexample;

• with the suitable proportion ([9] bookIII, course I), i.e. a relationshipbetween the sub-elements and thewhole building, the overall aspectreport/ratio.

The relations which connect the sub-elementsbetween them are expressed throughconstraints between surfaces (i.e. faces or axiscoincidences, distances, angles).

Figure 5: the eustyle rule for example

Four kinds of constraints are used: position,orientation, scale and contact. For example:the eustyle is the rule guiding the “between-columns”. These intervals must be of 2.25diameters (Fig. 5).Several rules of this type constitute thegeometry layouts which are used forscheduling and dimensional coordination. Theyconstitute the geometricae rationes, which arebased on the properties of the Pythagorastriangle about which Vitruvius [9] speakswithout exposing the details: "its proportionsare useful in many cases, both for the design ofthe buildings and for their surveying".For example, the lines resulting from the torusof the base of first column and parallel with thehypotenuse of the Pythagoras triangle cut theaxis of the third column in various pointswhich give the height of the architrave, theplank and of the cornice (Fig. 6).

Figure 6: the geometricae rationes based on thePythagoras triangle properties

As example, Figure 7 illustrates the completesemantic description of the analysed RomanTemple.

Figure 7: semantic description of the Capitol ofDougga

3. Architectural knowledge-basedmodelling toolMaking of a 3D building model requires ageometrical representation of the objects that

make it up, then the determination of the aspectof its surfaces. The developed tool presentedhere is an element of a complete approach for3D surface reconstruction. This approach isfounded on a hybrid registration of two typesof digitalized sources (point-clouds and digitalimages) in order to provide a complete supportfor the 3D modelling process (Fig.8). Itincludes three distinct concerns: the first one isthe surveying of the object under these metricand photometric points of views. The secondone is the construction of its geometrical modeland its enrichment by photometrical anddimensional information. The last one is thesemantic organization (i.e. structuration) of thebuilding.For each concern, the approach introduces thearchitectural knowledge and its exploitation asa support to interpret and reproduce shapes.Combining range-based and image-basedtechniques, the surface reconstruction processuses: the 3D point-clouds to extract relevantprofiles and to create triangulated-mesh; thedigital images to cover shadow areas in the 3Dpoint-clouds, to recover additional coordinatesand to extract textures for their future mapping.

Figure 8: hybrid registration of point-clouds anddigital images

In the modelling module, 3D point-clouds anddigital images constitute respectively themetric and the visual (i.e. graphic) support[10]. The knowledge-based modelling modulepresented here allows the surfacereconstruction focusing on relevant profiles

and using a library of architectural featureshapes in order to produce semanticrepresentations of architectural elements and toorganize them into hierarchical structures.

The next section describes the knowledgeformalisation process proposed. Section 3.2presents the architectural feature shape librarydeveloped and the section 3.3 illustrates thefeature instantiation process using dimensionalinformation extracted from the digitized data ofa real object.

3.1 Knowledge formalisation process

Based on an interpretative study ofdocumentary sources or on a comparativeanalysis of different real elements, theformalization of an architectural element startsby the definition of the relevant shape profiles.A network of B-Splines curves associated witha modelling procedure is defined andconstitutes the geometrical description level ofthe architectural element.The method proposed to formalize thearchitectural primitives starts from an analysisof historical sources such as, for example, asimple frontage element (Fig. 9).

Figure 9: extract of an architectural treaty of thefive classical orders

The (MEL) Maya Embedded Language’s coreuses a data flow paradigm [11]. This core isincorporated in the dependency graph (DG).The data and their operations are encapsulatedin the DG as nodes. In order to perform somecomplex modification to some data, a networkof simple nodes is created. Each node has oneor more properties associated with it. Theseproperties are commonly defined as attributes.An entire geometric mesh or Non-Uniform

Rational B-Splines (NURBS) surface can bestored as an attribute in a node.Thus, the first level of primitives is defined asset of simple nodes: the mouldings in oursimple frontage example (Fig.10). Each nodecontains the essential attributes according withthe entity position, the orientation of its centrepoint, plus its width and height.These nodes are connected to children nodesdefining the geometrical representation of theentities using B-Splines curves of differentdegrees. Five types of first level primitives arepresented below.

Figure 10: a set of mouldings

In the same manner, a set of extrusion pathswhich drives the surface generation is defined.By connecting simple nodes, a chain ofcomputation can be created to produce the finalresult [11].To set up the concept of hierarchy, a DirectAcyclic Graph (DAG) is used. The DAGdescribes a hierarchy in which a node can notbe both a parent and a child in the samelineage. The second level of primitivesrepresents profiles defined by progressivecombination of several first level primitives(Fig.11).

Figure 11: a combined profile: the second level ofprimitives

A constraint node links the last curve’s controlpoint to the last one of the precedent entity.Thus, in the DAG paths, the relationshipbetween a 3D object local reference (x, y, z)and a 2D (u, v) parametric space is known.This property is used to generate surfaces bycreating a connection between a profile node, amodelling function and a path node. Each firstlevel primitive of a profile node is extruded

along the parametric space of the associatedpath node (Fig.12).

Figure 12: example of an extrusion node

In the same manner, but in an upperhierarchical level, the third level of primitivesis created using Boolean operations (Fig.13).

Figure 13: Boolean operations applied to twoextruded surfaces

The DG is based on a push-pull model. So,when one pull information from a node, itpropagates that request through all the nodesthat provide input connections to the node(Fig.14).

Figure 14: hierarchical organization of theformalized entities

3.2 Architectural feature library

A MEL command makes it possible tointroduce a Web browser inside an interfacelayout. This browser allows receiving MELscript since a distant site. A simple script cancontain the procedure for the surfacegeneration, the parameter setting and the

structuration of a formalized entity. One of thefundamental properties of NURBS surfacesallows a deformation process to produce ashape which respect the parameters defined.This property is used to locate the primitiveinto the 3D space containing the 3D point-clouds and to adapt it precisely to the realshape of the digitized object. Such aninstantiation process opens an interesting wayaround the structuration of a library ofarchitectural entities taking again the treaties ofarchitecture of different periods, whosedescribed elements are translated intoparametric modelling terms. The proceduredeveloped to reconstruct architectural elementsusing an instantiation process based on afeature shape library is divided into three steps:

• Selection and generation of anarchitectural feature shape since adistant site,

• Fast positioning in the 3D point-cloud,

• Dimensional adaptation of the surfaceto the 3D point-cloud.

Imported once, the feature template isintroduced automatically into the 3D scene andassigned at the geometric centre of a group ofcameras with orthogonal projection. Then theuser can locate the feature on different distinctwindows (also including a calibrated image asshow in Fig.15). The camera planes are locatedat centre of each relevant profile of theprimitive. A device of “deformation tools”allows applying transformations (i.e.translation, orientation and scale factor) to thefeature parameters. Transformations can beapplied to an entire profile, to a single profile’smodule or to a simple parametric control point.

Figure 15: illustration of the feature positioningstep using a 3D handler

The constraint set encapsulated in the featuretemplate produces a propagation of thedeformations applied, to keep the coherence ofthe overall entity shape according with thearchitectural rules involved.

3.3 Extraction of dimensional information

One of the most interesting exploitation of theknowledge-based modelling tool consists in thepossibility of creating, at the same moment ofthe positioning of the feature shape, anorganized structure of drifting measurement.This is possible extracting the dimensionalvalues associated the parameters of theprimitives instanced. Modelling byarchitectural primitives becomes in thisdirection a way to structuring an organizedabacus of dimensional information (Fig.16).We work actually to the management of theseinformation in relational databases.

Figure 16: extraction of dimensional information

To allow the future handling of the 3D buildingdigital mock-up, the approach proposedensures the structuration of the model of thearchitectural object in parts and sub-parts, andthe identification of their reciprocal relations toguarantee a correspondence between thetopological description model of thearchitectural object and its 3D geometricalrepresentation. This aspect is very important inthe objective of a semantic structuration of theelements which compose the scene for 3D real-time visualization as presented below.

4. Semantic 3D representation forinformation interfacingIn the “3D Architectural Information System”(Sia3D) prototype [12] developed to illustratesthe approach proposed on the basis of theremains of the Roman Theatre in Arles (in thesouth of France), the issue of how to betterexploit, in terms of readability for architects,the results of a surveying process based on animage-based modelling techniques is

addressed. According with the principlesestablished in [13], we consider it is necessaryto identify and organise non-ambiguousmorphological elements to which we willattach various information, including rawresults of the surveying campaigns.

Figure 17: remains relocated in the space of theircorresponding theoretical elements

The results of the surveying and modellingprocess can be contextualized in the followingway:

• by locating in a same reference spaceof both the elements surveyed and thetheoretical elements (Fig.17),

• by positioning them with regards totheir associated knowledge.

For example, once surveyed, a cornice’sremain can be displayed in a 3D model of thewhole building with indications concerning:

• its belonging to an object typology (i.e.a stylistic references, a role in thebuilding),

• its hypothetical position, marked assuch, in the virtual buildingreconstruction,

To illustrate the previous contextualizationconcept, the following part of this sectionfocuses on the final output structure of a 3Dinterpretative model of a Roman Theatre.

Figure 18: the theoretical model of a RomanTheatre in the Augustinian period

The surveyed remains are localised in thereference space of a theoretical Roman Theatredefined through the relevant architecturalvocabulary previously analysed. The analysismade aims at identifying non-ambiguousconcepts that on one hand correspond tophysical objects which exist in the building(base, capital, …) and, on the other hand, havea significant role in the edifice’s composition.The vocabulary produced is used to rebuild,level after level, the theoretical model of aRoman Theatre in the Augustinian period(Fig.18). These levels can be read as rebuildingpoint of view starting by individual elementssuch as a capital to whole groups such ascolonnade. Thus the levels defined match theconcept of architectural scales.The building is thus described by a hierarchicalstructure that derives form the vocabularyanalysis (Fig.19). The hierarchical structureformalizes as composition links the relationsbetween the elements in a five (for the case ofa Roman Theatre) level hierarchy (example:capital / part-of / column / part-of / colonnade /part-of / …). Once the elements are classifiedas shown above, and once the compositionalrules are expressed, the understanding of theedifice is improved. The hierarchical structurealso allows establishing bilateral relationsbetween the information sources and the 3Dmodel. Indeed, to each hierarchical level

specific information sources are associated.Thus, each element of the hierarchy acts as afilter in the handling of the 3D model since it isrepresented by its own geometry or by sub-elements.

Figure 19: the five-level hierarchicalstructuration of the architectural elements of a

Roman Theatre

The hierarchical structure controls transitionsbetween levels, either inside the 3D sceneswhere geometrical representations of objectsare grouped according with the level displayed,and also inside the information sheets. Theseinformation sheets gather data pieces ontheoretical elements, but also descriptions ofthe surveyed remains (current localisation,conservation state, materials).The 3D model is finally a digital mock-up ofthe building with an interface that allows theuser:

• to select an element by its name in thehierarchical structure and to display itsposition in the scene,

• to get the information sheets related tothis element,

• to select an element into the 3D modelitself and to visualize its position in thehierarchical structure,

• to switch between levels accordingwith the user requirements,

• to show / hide elements or groups ofelements,

• to interactively control transparencylevel of selected elements or groups ofelements,

• to show / hide selected remains, and soone.

Figure 20: illustration of the web interfacedeveloped in the Sia3D prototype

The interface developed (Fig.20) is accessibleon a standard Web browser with the Virtoolsplug-in to read the 3D scene. Scene/windowsinteractions are written in JavaScript, enablingeasy updating of the various linksimplemented. Indeed, a wide variety ofinformation sheets links can be implementedsuch as surveying results or purelybibliographic sources.

5. ConclusionAn interesting development prospect relates tothe problem of the multi-representation.Indeed, the proposed approach allowsintroducing the concept of objective or point ofview, which governs the choices betweenvarious representations of the same object (i.e.multi-representation). These variousrepresentations will be able to correspond atvarious levels of consultation and necessarycomprehension according to the user profiles.Moreover, the geometrical model associatedwith the object must also be able to take intoaccount the temporal dimension of the building

(i.e. its evolution in the time). Within thisframework, the 3D structured representation ofbuildings becomes a privileged support forreal-time handling and to interact with theassociated documentary sources (i.e. searchand consultation of documents or information).These information and documents will be ofcourse associated with the geometrical modelof the object in a structured way and accordingto various user points of views (i.e. architects,historians, archaeologists, general public, andso one).

AcknowledgementsWe thank MENSI Company for helpfulcomments; Prof. Alberto Sdegno of IUAV forthe survey of “Convento della Carità” inVenice; the “Musée de l’Arles et de laProvence Antiques” for providing usdocumentary sources; Francesca De Domenicofor providing us images from herimplementation work (Fig.17,18,19,20).

References[1] D. Barber, J. Mills, P. Bryan. Laser

Scanning and Photogrammetry: 21stcentury metrology. CIPA Symposium,Potsdam, Germany 2001.

[2] C. Trevisan. Proporzioni e vera forma diparticolari architettonici rilevati conscanner 3D: caratteristiche di un softwarespecifico. in Disegnare idee immagini, 24(2002), pp. 44-49.

[3] B. Curless, M. Levoy. A volumetricmethod for building complex models fromrange images. Proceedings SIGGRAPH96.

[4] R. Migliari. Per una teoria del rilievoarchitettonico; primi appunti. Website:http://www.rappresentazione.net

[5] M. Docci, R. Migliari. Geometria eArchitettura. Gangemi Editore. Roma 2000

[6] A. Tzonis, L. Lefaivre. ClassicalArchitecture - The poetics of Order. MITPress. Cambridge 1986

[7] A. Palladio. The four books ofArchitecture. Dover publications, NewYork 1965 (original: Venezia 1570)

[8] J.M. Perouse de Monclos. Architecturevocabulaire. Principes d’une analysescientifique. Imprimerie Nationale, Paris1988

[9] Vitruvius Pollio. Ten books onarchitecture. Cambridge University Press,2001

[10] J-Y. Blaise, L. De Luca, M.Florenzano. Architectural surveying - froma point cloud to a 3D model. EVA2004 -Electronic Imaging and the Visual Arts.Florence 2004

[11] D. Gould . Complete MayaProgramming: An extensive guide to MELand C++API. Morgan Kaufmann,Paperback, 2002.

[12] J-Y. Blaise, F. De Domenico, L. De Luca,I. Dudek. Architectural Modelling andInformation Interfacing: Learning from ThreeCase Studies. ITI 2004 - 26th InternationalConference on Information TechnologyInterfaces. Cavtat / Dubrovnik, 2004

[13] I. Dudek, J-Y. Blaise. Newexperimentation of a generic framework forarchitectural heritage data visualisation.Proceeding WSCG 2003, pp109-117.