Reconstructing Conimbriga

Digital Cantaber

César Ferreira

, Nuno Rodrigues

1, 2

, Alexandrino Gonçalves

1, 2

and Virgílio Hipólito-Correia

School of Technology and Management, Polytechnic Institute of Leiria, Leiria, Portugal

Computer Science and Communication Research Centre, Polytechnic Institute of Leiria, Leiria, Portugal

Conimbriga Monographic Museum, Condeixa-a-Velha, Portugal

Keywords: Virtual Reconstructions, Cultural Heritage, Conimbriga, House of Cantaber, Kinect.

Abstract: Being the ancient cultural heritage structures and artefacts so full of detail and relevance to studies related to

our past, it would be desirable to have precise virtual replicas that could be freely explored without

endangering important pieces of history. However, the costs to produce three-dimensional environments are

sometimes discouraging, due to the significant cost of some 3D authoring tools, and to the time necessary to

manually produce the models. This paper presents a low-cost alternative to the classic manual modelling

process, towards the production of highly detailed virtual models, by using open source software and a low-

cost moving depth camera. The ambition is the dissemination of our cultural heritage legacy, making it

accessible, not only to experts, but also to the general public, without requiring any high-performance

hardware, authoring software or professional 3D skills. For the visualization, our virtual reconstructions will

be available through a three-dimensional live model viewer, based on recent technologies such as HTML5

and WebGL. Those may be triggered from a wide variety of device, contributing in this way to a true

democratization of history knowledge. The proposed approach was applied to create a virtual model of the

so-called “House of Cantaber”.

1 INTRODUCTION

Over the years several tools have emerged, that

intended to respond to the need to have exact

replicas of archaeological heritage that can be

exploited without endangering them. However, these

have proved discouraging since they are associated

with high costs, time consuming processes and

obstacles regarding the online availability to target

audiences. As a solution to this problem, the creation

of a tool based on open-source software and a

moving depth camera is presented. This results in a

website, which provides an interactive model with

exact replicas of the artefacts, without the need to

install additional plugins in the browser.

Nowadays it is possible to perform three-

dimensional acquisitions without the need for major

infrastructures, using low-cost equipment, which

enables mapping real scenarios with a high level of

detail. This possibility opens up new prospects for

the virtual and augmented reality scenarios. These

technological advances allow us to reach levels of

reliable reconstruction of cultural heritage spaces.

The case study presented in this paper is located

in the ruins of an important Lusitanian city of the

ancient Roman Empire: Conimbriga. Excavated

since 1899, opened to the public in 1930 and with

less than 20% of its full length brought to the

surface, this site has a prominent role in the

Portuguese cultural heritage.

Discovered during excavations that took place

between 1930 and 1934, completed in 1938, the

House of Cantaber is one of the largest and richest

in the western part of the empire (V. N. H. Correia,

2011).

It is precisely the House of Cantaber that was

chosen for a virtual reconstruction to be made

available on the Internet so that it can be accessed

interactively everywhere in the world.

2 VIRTUAL

RECONSTRUCTIONS

In recent years the so-called Virtual Archaeology led

to a significant growth of virtual reconstructions of

145

Ferreira C., Rodrigues N., Gonçalves A. and Hipólito-Correia V..

Reconstructing Conimbriga - Digital Cantaber.

DOI: 10.5220/0004732701450152

In Proceedings of the 5th International Conference on Information Visualization Theory and Applications (IVAPP-2014), pages 145-152

ISBN: 978-989-758-005-5

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

cultural heritage spaces, including Conimbriga. As

an example, Eduardo Barragán presents on his

personal page (http://italicaromana.blogspot.pt/)

excerpts from videos with a possible reconstruction

of the city of Conimbriga, namely the Insula do vaso

fálico. There are several apparent problems in the

architectural model chosen, namely those of

Pompeii/Herculaneum, breaking down, in our view,

the historical realism desirable that must take into

account the particularities of the local architectures.

Another Roman site propitious to being virtually

rebuilt is Bracara Augusta (Bernardes et al., 2000)

and (Bernardes and Martins, 2003). These, despite

representing fundamental resources for research and

being extremely important for raising cultural

heritage consciousness, lack reality quality for

today’s standards.

Regarding the acquisition of three-dimensional

archaeological structures there are several tools

available for the task. For instance, in wide exterior

environments, the DeltaSphere 3000 time-of-ﬂight

laser scanner (3rdTech, 2000), used by some police

departments by the forensics team to scan crime

scenes, which uses readings from infrared lasers

combined with digital photographs, in order to

acquire textures. However, despite its good results it

is quite a slow process (Zhu et al., 2007). Another

solution is the Metric Vision LR200 Laser Scanner

which was the first device that combines software

acquisition by radar, laser and 3D. It was used in the

famous case of the “Plastico di Roma antica” (Guidi

et al., 2005) acquisition. Despite being a large model

in size it was tiny in detail and high on price.

To acquire smaller objects the NextEngine Multi-

Laser 3D Scanner (NextEngine, 2001) can be

considered a good solution, since it as several

advantages in recreating textures and colours of

objects, thus capturing an accurate picture of the real

object. Though, it has some drawbacks, such as the

high price and the fact that it needs a specific

scenario with a swivel base. This base has to run at a

constant speed, making the procedure very time

consuming.

There is also another type of device that allows

three-dimensional acquisitions at a significantly

reduced cost when compared with the

aforementioned solutions: the Microsoft Kinect

This piece of hardware was originally aimed at the

game industry, designed for the Xbox 360

video

game console. It allows interaction in a human

natural way (PrimeSense, 2013), but due to its

capabilities when it comes to capturing geometric

data, proved to be an asset in the process of

acquiring three dimensional objects. A

demonstration of the potential of the Kinect in three-

dimensional modelling of an indoor environment

may be observed in the work of (Henry et al., 2010).

Indeed, the potential of this equipment and its

growing popularity in a number of areas beyond the

video games led to the emergence of a new version,

called Kinect for Windows, largely identical to the

console model but with significant improvements in

catchment details on objects of small size (Pheatt

and McMullen, 2012).

The solution presented in this paper uses the first

version of the Kinect sensor, available at a relatively

low cost. However, the launch of the second version

of this device, the Kinect 2, is already planned by

Microsoft's. Amongst other features it will have a

resolution of 1080p (instead of the 480p from the

previous version). It is expected that with a few

adjustments the ConimbrigaCG, a small tool which

was developed to capture Conimbriga’s artefacts,

and using the soon to be released Kinect SDK 1.8 (a

free development kit developed for this version of

the Kinect sensor (Microsoft, 2011)), acquisitions

will become much more accurate due to the

improved depth perception of the new sensor.

When it comes to be able to interactively create

and control three-dimensional scenes, these may be

produced using specific modelling software.

However, some have high costs, not only monetary

but also associated with the learning time necessary

to master the software.

Table 1

compares some of the most frequently

used modelling software available in the market.

For this work Blender was chosen, since it is a

free open-source software that was created as an

alternative to high priced programs currently on the

market. Blender is in constant development since

1995 and currently provides many features matching

some of the main market competitors.

Finally, the representation of three-dimensional

models of cultural heritage sites through the Internet,

particularly Roman structures, has also been

highlighted by several authors. For example,

(Gonçalves and Mendes, 2003) and (Silva and

Gonçalves, 2004) present an interesting approach to

provide relatively realistic environments, with

special emphasis on the small size of the models,

using one of the popular standards at that time,

VRML. However, it was necessary to install a

specific plugin.

The solution presented in this paper to display

the three-dimensional heritage models is based on

HTML5 and WebGL and, for this reason, it only

requires a HTML5 compatible browser.

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

146

Table 1: 3D modelling software.

Application Mainly used for Licence Price

3DS Max

Modelling,

Animation (Video

Games), Lighting,

Rendering

Proprietary $3,495

Cinema

Animation,

Lighting,

Modelling, Visual

3D Effects,

Rendering,

Simulation

Proprietary

$995 -

$3,695

Maya

Modelling,

Animation (Video),

Lighting,

Rendering, Visual

3D Effects

Proprietary $1495

LightWave

Modelling,

Animation,

Lighting,

Rendering, Film

and Television

Previz, Videogame

Asset Creation

Proprietary $3,495

Blender

Animation,

Lighting,

Modelling,

Rendering, Video

Game Creation,

Visual 3D Effects,

Sculpting, Basic

Post-Production

Video Editing

GPL 2+ Free

SketchUp /

SketchUp

Pro

Computer Aided

Design

Proprietary

Free /

$495

3 HOUSE OF CANTABER

The house of Cantaber is an extraordinarily large

house, actually the larger one excavated in the town

that covers more than 3200 sq. m.

The house is articulated around five different

perystiles, and about 1/3 of its area was originally a

large garden. This architectural setting responds to a

social need for large urban houses to give to their

owners the commodities expected from country

houses, the rus in urbe of the Roman poet Martial.

The artistic side of this social and cultural fashion

has already been studied (Alarcão, 2011), and can be

dated in the Flavian period (69-96 AD), which is an

epoch of outstanding urban development in

Conimbriga.

Modern research about the house (V. H. Correia,

2001) has, to some extent, contributed with data

about the chronology, the original plan and the

actual construction of the house, supplementing the

poor conservation of the remains (with the research

and conservation problems already mentioned). This

has allowed for a theoretical reconstruction proposal,

based on the fact that the house was built in an

isolated insula in the town, with few, if any,

constraints due to earlier construction on site. A

“Vitruvian” structure could thus be posited; the

different sectors of the plan of the house were

assumed to be made evident also in the overall

development of the building and some expression of

the relative importance of individual rooms was

made to be clear in the elevations (vg. the main

triclinium). With the evidence available concerning

actual architectonic elements, on which the town is

poor, due to the fact that they were mostly brick-

and-mortar built and have not survived for the most

part, the final result is thought to be credible,

although within a margin of error that has to be

acknowledged (on the theoretical implications of

this issue (Panhuysen, 2011)).

Figure 1: House of Cantaber plan.

ReconstructingConimbriga-DigitalCantaber

147

Figure 2: Real photo of the House of Cantaber.

4 VIRTUAL RECONSTRUCTION

The virtual reconstruction of the House of Cantaber

was carefully prepared based on information

collected on the site and in the knowledge of

experts, through a collaborative process between the

areas of Archaeology and Informatics. The goal is to

provide a low cost solution for similar cases with

Figure 3: Reconstruction workflow.

high accuracy both historical and archaeologically,

which will make the results available via the Internet

efficiently. Figure 3 illustrates the workflow

followed for the virtual reconstruction of the House

of Cantaber.

The next three sections address the workflow

emphasizing on the main processes presented (in

italic) on Figure 1, namely the modelling of the

house (section 4.1), the acquisition of the artefacts

(section 4.2) and the three-dimensional

representation of the model (section 4.3).

4.1 Modelling the House of Cantaber

As aforementioned for the intended purpose the

modelling software chosen was Blender which

allowed to recreate the house three-dimensionally.

First of all we had to consult an expert in

archaeology which provided us with a detailed plan

of the house (Figure 1) on which we extruded all the

walls from the house of Cantaber, then a detailed

study of the roof structure of the house was also

supplied, as well as a document with the heights of

the walls and columns of the entire complex.

Upon completion of the extruding walls phase,

we moved to the modelling of the different types of

columns (both Jonic and Tuscan). Lastly we

travelled to the storage rooms of the Monographic

Museum of Conimbriga, where there are pieces of

mosaics that once belonged to the House of

Cantaber floor, which were photographed in order to

apply them as textures in our newly created three-

dimensional models.

4.2 Acquisition of the Artefacts

Although at the time the model of the House of

Cantaber was presented to experts it received a

favourable appreciation, though the fact is that,

without any kind of ornaments it does not give the

feeling that at somewhere in time someone actually

lived there. In order to deliver a more realistic

sensation it is important to have some kind of human

evidence, more than just the building structure of the

house. For this reason, furniture, decorations and

other kinds of artefacts are indispensable to provide

a more human feeling, something which is

unrealizable in the real Conimbriga location.

Since Conimbriga’s artefacts are very ancient

relics, it was necessary to travel to the Monographic

Museum of Conimbriga. There, it was possible to

carry out the 3D acquisition of the artefacts, to use in

the virtual model of the House of Cantaber, without

compromising the real pieces. The process consisted

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

148

of placing the objects on a tripod (one by one) and to

move the Kinect manually around them.

Conceptually, there are two equivalent

approaches: make an acquisition of a static object

while moving the sensor, or having the object

moving in a rotating tripod while the sensor is static.

This may be achieved with Kinect Fusion, an

Application Programming Interface (API) which

provides the ability to scan and create a model of

any three-dimensional object from the real world

using the Kinect sensor. This API supports both

approaches, but moving the sensor in a static object

is a more robust solution with a significantly lower

data loss.

4.2.1 Kinect Fusion

While mobile cameras depth are not a new concept,

Microsoft has made such sensors become available

to all at an affordable price through its Microsoft

Kinect. The quality of depth sensors, given the low

price and the nature of the technology which

requires a real-time high throughput, made it

immediately very popular among researchers and

enthusiasts. The user can create a scene with the

camera and simultaneously view and interact with

the detailed three-dimensional version of the reality.

Since it is a demanding computational work the need

to use GPGPU (General-Purpose Graphics

Processing Unit) arose to attain a fluid interaction

with the real environment.

The Kinect Fusion allows dense surfaces to be

reconstructed in real time (Newcombe et al., 2011),

with a much higher level and robustness that any 3D

modelling professional could do in hours or days of

work. Within seconds after the acquisition is

accomplished a file is ready to be freely manipulated

without posing any threat to the integrity of the real

object.

4.2.2 ConimbrigaCG

In order to use the Kinect Fusion a small application

was built, which reconstructs a single dense surface

model with smooth surfaces, by integrating the

depth data from the Kinect sensor over time from

multiple viewpoints. The camera pose is tracked as

the sensor is moved (its location and orientation) and

because we now know each frame's pose and how it

relates to the others, these multiple viewpoints of the

objects or environment can be fused together into a

single reconstruction voxel volume.

You can see a screenshot of our tool in

Figure

as we scan an ancient clay plate.

Figure 4: ConimbrigaCG (Conimbriga Computer

Generated) screenshot.

The reconstruction is made in three steps. The first

step is the depth map conversion. This takes the raw

depth from Kinect and converts it into floating point

depth in meters; The second step consists in the

calculation of the global camera position (location

and orientation) and tracks this position as the sensor

moves in each frame, using an iterative alignment

algorithm, so that the system always knows the

current sensor posing, relative to the initial starting

frame; The third step is the integration of the depth

data from the known sensor pose into a single

volumetric representation of the space around the

camera. As a moving sensor sees a surface from

slightly different viewpoints, any gaps or holes

where depth data is not present in the original Kinect

image can also be filled in and surfaces are

continuously refined with newer data as the camera

approaches the surface more closely.

We learned that small, slow movement in both

translation and rotation are best for maintaining

stable tracking. Dropped frames can adversely affect

tracking, as a dropped frame can effectively lead to

twice the translational and rotational movement

between processed frames. If this occurs we need to

reset the reconstruction and start over.

4.2.3 Conimbriga’s Artefacts

Several artefacts collected in Conimbriga were

chosen by experts to be acquired to virtual 3D

objects to complete the house model. These include

amphorae, dishes, bowls, pots, clay lamps and jars

amongst others.

The acquisition procedure resulted in overly

complex models, with some flaws. For this reason,

in a first step, arose the need to conduct tests, to

attain a geometric simplification, without

ReconstructingConimbriga-DigitalCantaber

149

compromising the visual quality of the artefacts.

Since the Kinect produces models with a high

number of polygons (see Figure 5), they are unfit for

the internet availability, since the geometric

complexity of the models will lead to a significant

degradation in performance when using them in an

interactive mode.

Figure 5: 3D scan of an amphora with 669’000 faces.

To reduce the complexity of the geometric model

without compromising the realism, a feature of

Blender was utilized: the Decimate modifier add-on.

This allows reducing the total number of

vertices/faces without significantly change the shape

of the object.

This feature has proved its value with a

configuration of 95% ratio of collapse. The model

shrank to 33’000 faces (Figure 6), without any

noticeable loss of quality.

Figure 6: Amphora after applying the Decimation

modifier.

In a second step the failures that occurred during the

geometrical capture of some of the artefacts with the

Kinect were corrected. In this particular case, a

Blender modifier called Remesh Modifier was

utilized to alter the topology of the model filling any

holes and to smooth the result by applying a smooth

shading. The final result may be observed in Figure

Figure 7: Final amphora model with the necessary

corrections.

Finally, textures were applied to enhance the realism

of the models. Figure 8 shows a clay plate attained

by the same method with the texture applied to it.

Figure 8: Three-dimensional model from the acquisition

(right) and the real object (left).

To complete the process, both the model of the

house and the artefacts were exported to a format

which allows the embedding in a web page.

4.3 Online Availability with WebGL

In the recent past, in order to create and/or visualize

3D models, it would be necessary to utilize

specialized three-dimensional modelling programs

(Maya, 3D Studio Max, SketchUp, etc...). These use

languages to create and export models or even to

compile stand-alone applications with them. These

tools have access to the Graphics Processing Unit

(GPU) of a given computer, which would allow it to

render effects such as shadows, reflections and

atmospheric distortions at speed rates that otherwise

would not be possible.

In web browsers however, when the standard

used to represent three-dimensional models were

VRML or even X3D, not all of these features were

available and plugins were necessary to allow the

visualization (Gonçalves et al., 2005). Due to

advances in the current characteristics of the current

browsers, they enable the implementation of WebGL

(KhronosGroup, 2013), a low-level API in

Javascript that accesses directly the GPU

capabilities where the browser is running. Therefore

the process of three-dimensional visualization in real

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

150

time is easier. Since it is based on web standards it

does not require any specific plugin to present these

results. It also has the advantage of being fully

interactive, running on any personal computer, and it

is estimated that will shortly be available in

practically all prominent mobile devices. In this

sequence, the Three.js (MrDoob, 2013), developed

by MrDoob, is an extremely lightweight 3D engine

with a low level of complexity that allows to mix all

of the technologies aforementioned and create

interactive three-dimensional models which may be

freely navigated in any compatible browser.

To finalize the process, both the model of the

house and artefacts were exported using a Blender

plugin that is included in the Three.js library to a

format that allows the embedding of them in a web

page (Figures 9, 10, 11 and 12). The web page is

available at: http://cesarferreira.com/work/cantaber/.

Figure 9: Final result of the house exterior.

Figure 10: Final result of the house interior.

Figure 11 - House exterior overview.

Figure 12 – A look through the window of the dining

room.

5 CONCLUSIONS

In recent years there has been an increment on the

concern about the legacy left by our ancestors and,

with the development of powerful new technologies,

it is now possible to use the latest to explore new

ways to increase the knowledge about our past.

This work featured the creation of a three-

dimensional model of the House of Cantaber,

including virtual replicas of artefacts used at the

time, allowing interactive virtual tours online. This

process was implemented through manual modelling

using open source software and a low cost moving

depth camera.

Three-dimensional modelling presents itself as a

real reliable alternative making it an asset, not only

for historians and archaeologists but also for the

simply curious common user. Beside the low cost

advantages proposed by this alternative, the

combination of manual modelling and automatic

acquisitions gets the best of both worlds: the high

level of realism which can only be achieved with

manual modelling processes and the reduced time

necessary to have 3D models of objects.

Unlike other alternatives presented, this solution

is operating system agnostic, requires no additional

plugins for viewing in a browser and shows no

excessive costs.

ACKNOWLEDGEMENTS

This work was partially supported by the Portuguese

government, through the Foundation for Science and

Technology – FCT and the European Union

(COMPETE, FEDER) through the project

PTDC/EIA-EIA/114868/2009 (FCOMP-01-0124-

FEDER-015075) intitulado “ERAS - Reconstrução

Virtual Expedita de Sítios de Herança Cultural”.

ReconstructingConimbriga-DigitalCantaber

151

REFERENCES

3rdTech, I. (2000, July). Capturing Crime Scenes in 3D

and Color. Retrieved from

http://www.deltasphere.com/

Alarcão, V. N. H. (2011). A pedra e o risco: Arqueologia e

arquitectura na investigação do fórum de Conimbriga.

Actas Seminário Internacional de Arquitectura e

Arqueologia: Interpretar a Ruína. Contribuições entre

campos disciplinares--FAUP 2008, 157–165.

Bernardes, P., Ferreira, R., Lains, H., Martins, M., &

Marcos, A. (2000). Virtual Tour to Bracara Augusta.

In Computer Graphik Topics, Vol. 12, 5/2000,

Darmstadt, Germany.

Bernardes, P., & Martins, M. (2003). Computação gráfica

e Arqueologia Urbana: o caso de Bracara Augusta.

EPCG, Porto.

Correia, V. H. (2001). Conimbriga, casa atribuída a

Cantaber. Trabalhos arqueológicos 1995-1998.

Conimbriga: revista do Instituto de Arqueologia, (40),

83–140.

Correia, V. N. H. (2011). A arquitectura doméstica de

Conimbriga e as estruturas económicas e sociais da

cidade romana.

Gonçalves, A., & Mendes, A. (2003). The rebirth of a

Roman Forum: The case study of the Flavian Forum

of Conimbriga. In Proc. of CAA2003, Austria.

Gonçalves, A. J. M., Silva, F., & Mendes, A. J. (2005). An

approach to (virtually) recreate historical findings. In

Proc.of the 10th International Congress “Cultural

Heritage and New Technologies”, Austria.

Guidi, G., Frischer, B., De Simone, M., Cioci, A., Spinetti,

A., Carosso, L. & Grasso, T. (2005). Virtualizing

ancient rome: 3d acquisition and modeling of a large

plaster-of-paris model of imperial rome. In Electronic

Imaging 2005 (pp. 119–133).

Henry, P., Krainin, M., Herbst, E., Ren, X., & Fox, D.

(2010). RGB-D mapping: Using depth cameras for

dense 3D modeling of indoor environments. In the

12th ISER (Vol. 20, pp. 22–25).

KhronosGroup. (2013, July). WebGL, OpenGL ES 2.0 for

the Web. Retrieved from

http://www.microsoft.com/en-us/kinectforwindows/

Microsoft. (2011, July). Kinect for Windows. Retrieved

from http://www.microsoft.com/en-

us/kinectforwindows/

MrDoob. (2013). JavaScript 3D library using WebGL.

Retrieved from https://github.com/mrdoob/three.js/

Newcombe, R., Izadi, S., Hilliges, O., Molyneaux, D.,

Kim, D., Davison, A., Kohli, P., Shotton, J., Hodges,

S. & Fitzgibbon, A. (2011, October). KinectFusion:

Real-Time Dense Surface Mapping and Tracking. In

the IEEE ISMAR.

NextEngine. (2001, July). 3D Scanner HD. Retrieved from

http://www.nextengine.com/

Panhuysen, T. (2011). Metropolitan artistic models in

Lusitania: the examples of the domestic architecture of

Conimbriga, 681–692.

Pheatt, C., & McMullen, J. (2012). Programming for the

Xbox Kinect

sensor: tutorial presentation. Journal

of Computing Sciences in Colleges, 27(5).

PrimeSense. (2013, July). Prime Sense. Retrieved from

http://www.primesense.com/

Silva, F., & Gonçalves, A. (2004). House of the Skeletons-

A Virtual Way. In Proceedings of CAA2004. Italy.

Zhu, J., Humphreys, G., Koller, D., Steuart, S., & Wang,

R. (2007). Fast omnidirectional 3D scene acquisition

with an array of stereo cameras. In 3-D Digital

Imaging and Modeling, 2007. 3DIM’07. Sixth

International Conference on (pp. 217–224).

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

152