3D modelling of laser scanned and photogrammetric data for ... · Mosteiro da Batalha monastery. Finally, in Sect. 6 we outline desirable solutions that would help the modelling process

SURVEY PAPER

3D modelling of laser scanned and photogrammetric datafor digital documentation: the Mosteiro da Batalha case study

Adriano Oliveira • Joao F. Oliveira •

Joao M. Pereira • Bruno R. de Araujo •

Joao Boavida

Received: 28 June 2011 / Accepted: 27 January 2012 / Published online: 6 March 2012

� Springer-Verlag 2012

Abstract Advances in both terrestrial laser scanning

hardware and photogrammetric systems combined are

creating increasingly precise and rich 3D coloured data. In

this article we show how computer graphics and visuali-

zation techniques have played an important role in real-

time visualization, data management, modelling, and data

fusion in an increasing number of applications such as

surveying engineering, structure analysis, architecture,

archaeology and cultural heritage. Specifically, we describe

the typical modelling steps involved in the creation of a

range of digital documents provided by the 3D digitization

company Artescan to customers. We present how these

modelling steps were applied in the context of creating

digital documents used in the preservation of Mosteiro da

Batalha.

Keywords Laser scanning � Photogrammetry � 3D

modelling � CAD � Orthophoto � Virtual reality

1 Introduction

Since the development of the first solid-state ruby ‘‘laser’’

(light amplification by the stimulated emission of radia-

tion) in the early 1960s, progressive technological

advances have allowed the widespread introduction of

laser-based surveying for engineering, construction, and

environmental sciences [19]. While light detection and

ranging (LiDAR) allows for the rapid acquisition and

creation of digital models of large surface areas of the

earth in airborne laser scanning (ALS), terrestrial laser

scanning (TLS) enables the acquisition of local fine-level

details of structures such as faults in buildings. In this

article, we focus on the role of computer graphics and

visualization in modelling procedures of terrestrial laser

scanning data.

In Sect. 2 we highlight the benefits of the use of 3D

laser scanning technology in various applications and

present considerations on data processing and manage-

ment. In Sect. 3 we provide an overview of the typical

steps required in the modelling pipeline to create various

types of digital documents, for instance topographical

charts, 2D cross-section plans of buildings or 3D-textured

models. In Sect. 4 we illustrate some of the digital docu-

ments produced for customers for their respective appli-

cation areas. In Sect. 5 we present as a case study, how the

various modelling steps reviewed in this article were used

in creating digital documents for the preservation of

Mosteiro da Batalha monastery. Finally, in Sect. 6 we

outline desirable solutions that would help the modelling

process.

A. Oliveira � J. Boavida

Artescan, 3D Scanning, IPN Incubadora,

Rua Pedro Nunes, 3030-199 Coimbra, Portugal

e-mail: [email protected]

J. Boavida


J. F. Oliveira

Escola Superior de Tecnologia e Gestao,

Instituto Politecnico de Portalegre Lugar da Abadessa,

Apartado 148, 7301-901 Portalegre, Portugal


J. M. Pereira (&) � B. R. de Araujo

VIMMI Group, INESC-ID,

Department of Information Systems and Computer Science,

IST/Technical University of Lisbon,

Rua Alves Redol, 9, 1000-029 Lisbon, Portugal


B. R. de Araujo


123

J Real-Time Image Proc (2014) 9:673–688

DOI 10.1007/s11554-012-0242-0

2 Background

In this section we focus on the following aspects: mobility

and safety benefits of laser scanning; high data integrity;

scanner operation; real-time processing and rendering; new

applications and data processing and management.

2.1 Mobility and safety benefits of laser scanning

Increasingly portable Laser scanning equipment have

facilitated the task of model acquisition in, for example,

topographical surveying and archaeology (Fig. 1).

As non-contact technologies, laser scanning and photo-

grammetry provide significant safety benefits for the

operator when for example making measurements of haz-

ardous sites such as a hydropower substation (Fig. 2) or of

sites with difficult access such as dams (Figs. 3, 7). In

addition class I laser scanners are totally harmless to

humans, unlike higher level laser classes (International

Electrotechnical Commission [16]) which require some

precautionary logistics for a safe site acquisition.

As active sensors and unlike passive digital photo-

graphic cameras, that capture the existing light, laser

scanners analyse the reflected light from an emitted laser

beam. This characteristic enables the acquisition of data in

the absence of light, which is the case in underground

tunnels (Fig. 6).

Fig. 1 Point cloud rendering and photogrammetric image from one

of the several laser scanner positions (in red) used for the digitization

of the roman oven of the Quinta do Rouxinol, Seixal, Portugal

Fig. 2 Hydropower substation 3D surveying and modelling of the

Cahora Bassa dam in Mozambique. Point clouds have been coloured

by projecting the photogrammetrically processed acquired images

Fig. 3 Contour lines extracted from point cloud data and structure

analysis of the Cahora Bassa dam

Fig. 4 Cross-section cut of the Monserrate 3D model

Fig. 5 Longitudinal profile of the Monserrate 3D model

674 J Real-Time Image Proc (2014) 9:673–688

123

2.2 High data integrity

Another advantage of 3D laser scanning and modelling is

the high data integrity associated with acquiring the full 3D

model. This for example can benefit 2D section analysis

coherency as it is shown in Figs. 4 and 5 where illustrations

of the Palacio de Monserrate, Sintra, Portugal, are

provided.

By acquiring the full 3D model, future surveys need

only to analyse a portion of the captured data, avoiding

potential errors of individual surveys.

These individual errors are typically due to the sub-

jective nature of pinpointing the exact location of the

features one wishes to measure. Specifically, when using

the optical scope of a total station measurement device in

traditional surveying, there is a degree of freedom for the

operator to choose the location of the measurement for a

feature, whereas with laser scanning the decisions of

placing a measurement follows a known sampling error.

2.3 Scanner operation

In the context of our approach, the available Riegl Z360i

laser scanner (Fig. 7) operates based on the time-of-flight

operational principle. Such a system enables the operator to

set the vertical and horizontal sampling intervals defined in

angle increments of for example 1/100 of a degree within a

vertical range of ±45 and a horizontal range of ±180. The

scanner thus provides points expressed in either polar

coordinates, with a range depth and intensity, or in Carte-

sian coordinates. The depth is typically calculated with the

time of flight of the returned echo pulse. It is important to

note that with this technology for a given angle pair only

one point is captured. For example, natural features such as

a foliage can hinder the acquisition of partially covered

man-built structures.

New online waveform analysis laser scanners solve this

problem, such as the Riegl VZ400 with a specified manu-

facturer’s measurement accuracy in the range of 3–5 mm.

By sampling, digitizing and storing the full waveforms, it

provides a solid basis for a thorough insight into the

interaction of the laser pulse with the targets hit by the laser

beam [32]. On the one hand it enables multi-target reso-

lution, meaning that it is possible to record multiple depth

points for a given angle pair. On the other hand, additional

parameters from the full waveform analysis are especially

beneficial to the classification task. Data storage is

increased with the adoption of this technology. However,

as the visualization and processing software and hardware

are evolving in parallel, there is no significant increase in

data processing tasks.

A digital camera can be mounted on top of scanner,

enabling the automatic acquisition of images from each

scan position. As the transformation parameters between

the coordinate system of the digital camera and the coor-

dinate system of the laser scanner are well known and

registered in the library of available camera calibration

files, it is easy to map coordinates from image space to

object space (Fig. 6). It is also possible to use photos

acquired using a freehand camera, and to compute their

exterior orientation parameters for each position using

spatial resection based on given control points, present in

the scene. An image with interior and exterior orientation,

in the context of this article, can be called an oriented

photo.

In addition, the scanner can be coupled with a global

navigation satellite system and a digital compass thus

providing its position and its orientation. These issues will

be further discussed in Sect. 3. The scanner thus provides

local reference frames for the camera and laser scanner

Fig. 6 Site reference frames: scanner position local reference frames

on surface and underground, with transformations for both for the

project reference frame. Site survey of Caneiro do Rio Tinto, Portugal

Fig. 7 Data acquisition of the Alto Ceira dam, Portugal, with a Riegl

Z360i coupled with a Nikon D100

J Real-Time Image Proc (2014) 9:673–688 675

123

head for each scan position, a project reference frame and

global reference frame (Fig. 6).

2.4 Real-time processing

Laser-scanned data are acquired at very high measurement

rates and is normally processed in real-time to allow for a

instantaneous visualization of the scanned scene. For

instance, in the case of the Riegl VZ-400, 125,000 mea-

surements per second are acquired and recorded.

In the Riegl V-Line scanners the processing of each

echo signal is performed in real-time, within the instru-

ment, through echo digitization processing and waveform

analysis. The implemented online waveform processing

allows for multi-target capability, just as the well-known

full waveform analysis systems, but it is done almost

instantaneously, thus making the processed data available

to the user in real time. In this practical case the operating

software RiSCAN PRO allows for a real-time visualization

and interaction of a lighter version of the data on the lap-

top, the monitoring data stream, at the same time that the

equipment is acquiring the data. This functionality is

essential for a fast evaluation of the coverage of the scan

and of the potential obstacles in the scene.

2.5 New applications

With advances in technology more types of features can

be acquired, and modelled. A typical laser scan captures

features of interest and unintended features that can be

studied in the future. This richness of data on the one

hand opens up a range of new applications to study these

acquired features of interest; on the other hand, extrane-

ous features in the data foster research in algorithms to

automatically remove them. In photogrammetry and

remote sensing some of the challenges remain the same at

different range scales and data sources (satellite imagery,

aerial imagery and LiDAR, terrestrial laser scanning and

close range photogrammetry). For example, one applica-

tion with increasing interest is the automatic segmentation

and classification of features in the data such as roads,

buildings, cars, people and vegetation. In particular, where

laser scanning is concerned, both aerial and terrestrial,

segmentation is considered as a means to organize points

into homogeneous groups, whereas classification is con-

sidered as a means to provide a class attribute for each

segment in the context of a specific application [25].

Very often the shear volume of data makes it very

impractical to have a user select and classify manually

these data sets. On the other hand, real-world objects and

scenes, represented by dense 3D point cloud, are of dif-

ficult generalization for creating totally automated

segmentation.

2.6 Data processing

Several commercial and open source solutions exist for

geometric processing of large scanned data sets. It is

beyond the scope of this article to make a detailed review

of such systems. However, for a comprehensive list of

available resources we refer the reader to [17].

There is a large set of functions related to the processing

of laser scanner acquired data, and typically each solution

is optimized for a particular goal or functionality. In the

following items we describe processing functions focusing

on open source examples.

Importing sets of point clouds and meshes, visualizing

and manipulating them are basic functionalities, which are

implemented in most of the related software. However,

considerations regarding whether a complete or partial

model can be visualized during editing are made in the data

management and real-time rendering section.

Alignment tools normally use the well-known iterative

closest point (ICP) algorithm [6] or as in the case of

Scanalyze, one of its variants [28]. This system originated

from Digital Michelangelo Project [20] and was designed

for viewing, editing, aligning, and merging range data. An

additional tool for robust surface registration [2] is adopted

in MeshLab, an open source editing software for process-

ing and editing of 3D triangular meshes and point clouds

[10].

Merging is additionally implemented in Scanalyze and

Meshlab tools mentioned above as examples. In the former

automatic merging uses the VRIP [11] algorithm, whereas

the latter uses Poisson surface reconstruction.

Surface reconstruction algorithms from point clouds can

produce triangle soups that are noisy, with disjoint surfaces

if the input data is also noisy and not spatially uniform.

Hence, it is necessary to apply mesh cleaning, mesh deci-

mation and mesh smoothing to the result. Amenta et al. [4]

present an algorithm for triangulating point clouds based

on 3D Delaunay triangulation, whilst Isenburg et al. [18]

present a robust out-of-core 2D Delaunay triangulation

method for huge point sets. For instance, Paraview [1] is a

system that relies on the Visualization ToolKit VTK [29]

and offers a powerful set of filters to perform decimation,

Delaunay triangulations, subdivision, smoothing and mesh

fairing. While these algorithms take advantage of parallel

processing and rendering, interactive editing relies only on

discrete Level of Details (LoDs) of coarser volume repre-

sentations. Alternatively, Meshlab provides a wide set of

powerful surface reconstructions approaches from point

clouds. In the context of mesh decimation, Scanalyze

provides automatic model simplification using a quadric

error approach [13].

Mesh cleaning functionality enables the user to create

coherent meshes, i.e. removing duplicate vertices and


123

self-intersection or converting non-manifold representa-

tions as well as flipping inconsistent triangle orientations

[23]. In addition, if occluded areas (no point data) exist in

the surveyed object, hole-filling techniques can be applied

to repair the incomplete meshes. In Meshlab these func-

tionalities are available through the Visualization and

Computer Graphics library (VCGLib), which includes

several mesh editing techniques. The Scanalyze tool pro-

vides a set of mesh cleaning utilities, one of which enables

the completion of holes using a volumetric algorithm [12].

Mesh texturing functions are available in some solutions

(e.g. Paraview), whereas others only support the visuali-

zation of the previously textured meshes (e.g. Meshlab)

indicating that some approaches do not support textures at

all (e.g. Scanalyze).

2.7 Data management and real-time rendering

Software supplied with Laser scanner systems such as

Riegls RiSCAN PRO, typically manage the different data

files associated with a survey project, provide editing and

modelling functions and have a degree of visualization

support where changes of user view-points can be rendered

in real time. This software follows an external memory

management paradigm similar to that of [9] where pointers

to files residing in secondary memory are loaded and

unloaded to and from main memory in turn by the user and

operations such as triangulation and stitching together of

meshes is performed to create a larger combined model.

Another simple solution that is used to obtain real-time

rendering is to simply skip in regular intervals the data to

be rendered when mouse activity is detected.

Several point-based rendering tools exist for interactive

visualization of whole out-of-core point cloud data sets

(e.g. QSplat [27], XGRT [33]). An out-of-core tool for

Delaunay Triangulation of huge well-distributed point

clouds using streaming and spatial coherence is presented

by Isenburg et al. [18]. In addition [34] enables editing

operations on parts of the model whilst viewing at the same

time the whole model using a multi-resolution rendering of

out-of-core hierarchical point sets. In contrast, for huge

out-of-core triangulated meshes, existing visualization

solutions provide little or no editing capabilities whilst

viewing the full model. Namely, level-of-detail techniques

are applied to a finished model before the model can be

visualized. Borgeat et al. [8] divide a textured scanned

model into smaller surface area parts, and create discrete

LoDs for each part, during visualization discrete LoD are

switched with geomorphing to avoid popping artefacts.

Gobbetti et al. [14] enable the interactive visualization of

complex isosurfaces, CAD and scanned models by dividing

a scene into voxels and pre-compute shaders that enable

one to model the original appearance of the geometry in

each voxel, at run-time far away voxels are splat and tri-

angles in close by voxels are rendered normally.

3 Modelling pipeline

In this section we describe the modelling steps involved in

the creation of the examples highlighted in the previous

sections as well as the creation of the digital documents to

be presented in Sect. 4. Figure 8 provides an overview of

the digital document creation process starting with the

acquisition of acquisition of point clouds and oriented

photos and showing the different modelling pipelines. We

note that several digital documents use only a small subset

of modelling steps, and that Fig. 8 only depicts the main

modelling pipelines currently used in Artescan.

3.1 Acquisition of point clouds and oriented photos

The data acquisition process is of extreme importance, as it

will affect the overall quality and accuracy of the final

results.

An important aspect to be considered is the coverage of

the object under study. To obtain a whole coverage, the

data must be acquired from different scan positions

ensuring that there is sufficient overlap between successive

scans to facilitate registration and avoid occlusion in the

data.

With systems such as the Riegl Z360i shown in Fig. 7,

the data acquired from each scan position will consist of a

point cloud (normally with a data size of millions of

points). These data include a binary file with XYZ together

with intensity (I) information for each point, as well as

digital image files with recorded orientation parameters

(oriented photos). An example of such an output is illus-

trated in Fig. 9.

Subsequently, it is important to consider how the data is

acquired from different scan positions and hence local

reference frames are then transformed onto a unique

common coordinate system. This transformation is known

as registration or relative orientation when the used com-

mon coordinate system is an arbitrary one. When all the

data sets are transformed to a meaningful coordinate sys-

tem the above-referred transformation may be called as

global registration or absolute orientation. If the mean-

ingful coordinate system relates to earth the process can be

called georeferencing. Specifically for each scan position,

point cloud position coordinates and image orientation

parameters are expressed in the instrumental coordinate

system. As a result, it is essential that the data from these

different scan positions are registered onto a single com-

mon coordinate system. For this purpose different regis-

tration methodologies can be adopted [3, 5, 15].


123

For example, indirect registration can be performed

with surface-matching techniques provided that the partial

point clouds present good overlap with geometrically well-

identified features. High precision can be accomplished

using a different technique where correspondences between

data sets are established by 3D tie points. One approach of

this method is to distribute a set of tie points implemented as

retroreflectors that are variant in planimetry and altimetry,

around the area of interest. These retroreflectors are sur-

veyed within the laser scanning acquisition process and

subsequently, automatically detected by the acquisition

software. Given that, if the instrumental coordinates for each

scan position are well defined, scanning of the same set of

retroreflectors (in a minimum of 3) makes it possible to infer

the correct 3D transformation matrix that performs the

coordinate system registration. This is similar to photo-

grammetry where tie points are used to relatively orient two

data sets (two point clouds instead of two images). However,

it is indicated that georeferencing will demand knowledge

of the tie point coordinates in a meaningful coordinate

system. In addition, it is noted that this is relevant to the

point-transfer coordinate problem. In photogrammetry,

Fig. 8 Overview of digital

document creation process,

starting with acquisition (a);

main modelling pipelines

(b–d) and resulting digital

documents (e)

Fig. 9 Point cloud and oriented photo of Monserrate exterior


123

geodetically known control points present high significance

for datum definition. In the TLS scenario retroreflectors can

be measured with the use of conventional topographic

methods, using, for example, total station or global naviga-

tion satellite system techniques (GNSS) in a meaningful

coordinate system (e.g. WGS84 or a national datum and

subsequently convert this to a local reference system).

Direct georeferencing approaches, for example, calcu-

late the position and attitude of the device with additional

sensors such as the ones incorporated in Inertial Navigation

Systems (INS) and GNSS systems. These are usually

integrated and calibrated with the laser scanner (when, for

example, a kinematic survey scenario is considered).

A rapid ‘stop and go’ registration method would be to

approximate position and orientation starting values for

each scan position to those collected from a GPS mounted

on a TLS (for the position parameters) and digital compass

and tilt sensor (orientation parameters). These parameters

can subsequently enter a multi-station adjustment that will

iteratively adjust these position and orientation matrices to

already registered positions with the control points

approach described earlier. Bundle adjustment can as a

result improve the final measurement quality of the results.

A huge step was taken with the recent development of

several mobile laser scanning systems that allow for a great

amount of data to be acquired in a fast and efficient way.

One example of such a system is the Riegl VMX-250

composed of two Riegl VQ-250 scanners, an INS-GNSS

unit and a control unit [26]. This compact system can be

mounted on a roof-rack of a vehicle and acquire up to

600,000 measures per second at normal traffic speeds.

3.2 Modelling pipeline b

It has already been mentioned (see Sect. 2.4) that it is often

the case that applications using LiDAR and TLS face

similar challenges in the context of classification or elim-

ination of potential undesired features in the data. A digital

surface model (DSM) can be collected from aerial or ter-

restrial platforms when active systems are considered. Data

segmentation will generally be followed by a classification

scheme which can be application dependent. For instance,

such a case scenario example is the application of forest

inventory. In this case, single tree parameters, such as tree

height, diameter at breast height or the height of crown

base are the typical parameters to be measured [31]. It is

indicated that for topographic (mapping) purposes, a digital

terrain model (DTM) is generally required, that represents

only the earth surface without the canopy. As a result,

‘unwanted’ objects such as vegetation, cars or people need

to be typically removed by combinations of automatic and

manual techniques. X Y Z LiDAR data in 2.5D can be

stored in an image representation suitable for image

processing and feature extraction, whereas X Y Z TLS data

can be spatially binned into an octree or grid structure. TLS

produces real 3D data that allows access to areas that are

typically occluded in a 2.5 representation, thus the filtering

and segmentation algorithms used are inherently different.

It is unfortunate, that algorithms for removing or classi-

fying features, generally require considerable manual

supervision and editing in order to ensure good-quality

results. Here and in the context of pipeline b, terrestrial

laser scanning is considered for 2D and 3D topographic

map construction. In the case of natural environments, the

original data undergoes a semi-automatic filtering process

based on robust local surface fitting and outlier detection.

Figure 10a shows an example of unfiltered data with veg-

etation whereas Fig. 10b provides an illustration of the

same data after filtering.

Once the data has been filtered, the point clouds are

spatially organized into an octree (Fig. 11a). Oriented

photos are then used to manually generate vector features

in a CAD environment.

Fig. 10 Vegetation filtering of a valley in Gois, Portugal. a Acquired

raw point cloud before filtering; b filtered point cloud


123

Examples of such features can be seen in Fig. 12, where

terrain discontinuity lines, also referred to as break lines

(illustrated in red) are used to create a digital terrain model

(Fig. 11b), that in reference to pipeline b can be repre-

sented by a triangulated irregular network (TIN). Such a

DTM can be utilized to compute contour lines such as the

ones extracted in the subsequent Fig. 11c.

In surveys where the area of interest contains man-made

structures, such as buildings, 3D representation can be

additionally generated with manual editing procedures

(depicted as blue coloured lines in Fig. 12), and subse-

quently incorporated in a topographic map.

3.3 Modelling pipeline c

In this section we describe the main modelling steps

required to generate orthophotos of, for example, a laser-

scanned building, and the production of 3D-textured mesh

models.

The first step involves partial point cloud (Fig. 13a)

triangulation utilizing a Voronoi-based triangulation algo-

rithm. The resolution of the triangulated result will gen-

erally include noise and exceed the required modelling

resolution. These are treated with decimation and

smoothing filters (Fig. 13b). Examples of such smoothing

options are the available Laplacian smoothing and Taubin

et al.’s [30] robust filter technique.

As with most modelling tasks, the end quality is often

improved with how one selects and groups the input data.

We select points into groups in such a way that all the

points have a bi-univocal relation with a reference plane

and use a 2D Delaunay Triangulation approach. However,

sometimes the object is more free-form and so it is nec-

essary to use more powerful surface reconstructions, as the

ones described in Sect. 2.6.

Oriented photos can thus be utilized to colour the point

cloud, or to texturize the triangulated mesh (Fig. 13c).

Although orthoimages have existed for more than half of

a century as a cartographic document, in the aerial

domain, they are now gaining an increasingly significant

interest in the area of architectural photogrammetry as

well.

Fig. 11 Topographic map creation from Vale do Tamega, Portugal. aTerrain point cloud spatially indexed in an octree; b digital terrain

model; c contour lines

Fig. 12 Manually defined break lines in red, and structures in blue

superimposed on oriented photo of Olgas dam, Portugal


123

Unlike conventional photos orthoimages provide metric

quality, particularly useful in surveying and modelling

applications. Specifically when orthoimages are imported

into a CAD environment, they can be used to register and

model architectural structures that, for example, may need

restoration and recovery of pathologies or can be used to

guide the vectorization of intricate architectural structure

details and facade features. Orthoimages thus provide

many benefits over traditional surveying methods with

regard to their reliability as well as the semantic and

positional information integration in existing surveying

data.

Such an orthophoto produced by orthographically pro-

jecting the textured model on a defined plane is illustrated

in the subsequent Fig. 14.

For complete production of a 3D-textured mesh model

of a particular large survey area, significant mesh merging

and hole filling steps are incorporated within the modelling

pipeline c procedure.

3.4 Modelling pipeline d

For architectural or engineering structure surveys it is of

great importance that the acquired data are detailed, precise

and accurate as it has already been discussed. Laser scan-

ning technology is therefore desirable. In these application

areas the representation of the collected information is

most useful when hierarchy and structure is added incre-

mentally during the modelling. For instance, the use of

information layers allows the user to visualize different

features of interest more clearly, by not rendering other

layers. The approach described in pipeline c, produces

triangulated meshes from point clouds accomplishing a

faithful representation of intricate surfaces. This, together

with the ability of Computer Aided Design (CAD) for

hierarchization and structuring of information, results in an

integrated layer and relational information to the geometry

data avoiding potential ‘triangle soups’.

The surveyed object can be modelled by combining

point cloud data with photogrammetrically oriented image

data, in a CAD environment using its vectorization tools,

Fig. 13 3D-textured model of the Mosteiro da Batalha, Portugal,

architectural survey. a Acquired raw point cloud; b triangulated mesh;

c 3D-textured mesh

Fig. 14 Orthophoto creation from the architectural survey depicted

in Fig. 13


123

traditionally known as monoplotting [21] (Fig. 15). A layer

structure is adopted for the classification of features such as

window, door, column, etc.

This kind of modelling presents some subjectivity in the

choice of level of detail. In our case the clients require-

ments are analysed before setting the most adequate level

of detail of the representation. Figure 15b depicts the

vectorization of a higher level of detail than that of

Fig. 15a.

After the 3D model has been built, texture mapping

techniques are used to enhance its visual appearance, thus a

photorealistic model is generated (Fig. 16).

The 3D model in this stage can be used for the gener-

ation of virtual products that will be discussed in Sect. 4. It

can also be used for the extraction of coherent profiles and

cross-sections, thus creating high integrity 2D documents

(Figs. 4, 5). In Fig. 17, the process of producing a complete

2D architectural document is depicted [22]. The construc-

tive vector lines are generated from automatic extraction of

2D vector data from the 3D model (Fig. 17a). This drawing

is superimposed in a CAD environment onto an orthophoto

from the same view (Fig. 17b), thus assisting the manual

vectorization of additional decorative details (Fig. 17c).

4 Digital documents

In Sect. 3 we described the main modelling steps for digital

document production, whereas here we provide an over-

view of the different supported types of digital documents.

A common digital product is a 2D topographic map that

can be derived following the methodology proposed in

pipeline b (Fig. 11c). However, nowadays 3D topographic

maps that can be visualized in digital platforms are gaining

popularity. In particular, DTMs from laser-scan data are

Fig. 15 3D CAD model creation of the Palacio de Monserrate. aManual 3D modelling of vector primitives using both superimposed

photogrammetrically oriented images (right) and point cloud data

(bottom left); b finer detail modelling of intricate architectural

features

Fig. 16 Exterior of the created 3D-textured CAD model

Fig. 17 Production of 2D architectural documents (Palacio de

Monserrate). a 2D vector data extracted from 3D model; bcorresponding orthophoto; c added 2D vector detail of rock, tiles

and masonry features


123

considered to be of high value in surface deformation

monitoring applications. Figure 18 (right) illustrates a

colour ramp representation of a DTM of the same geo-

graphical area over different time intervals that can be

particularly useful for the purpose of spatially continuous

surface monitoring. An additional product, with great

interest for the user, is the mesh visualization. An example

is given in Fig. 18, left, depicting a mesh from an earth

dam. The coloured arrows represent sliding behaviour of

the upstream face (red arrow), paths made by rolling

machinery (yellow arrows) and a strip where rocks were

removed and replaced after dam inspection (blue arrow).

These phenomena would not be easily recognized with

conventional engineering documents, and in some instan-

ces not even whilst present on site [7]. Another product is

for instance an orthophoto that can be generated from the

modelling pipeline c. In particular, we use the case study of

orthophoto production for the visual inspection of pathol-

ogies in a concrete wall of a dam [7] located in Alto Ceira,

Portugal (Fig. 19). Figure 20 presents an example of

combined methodologies of modelling pipelines b and d

comprising the 2D digital document of a vectorized street

facade. The illustrated architectural structures were man-

ually digitized using oriented photos and point clouds.

3D models resulting from pipelines c and d, can be also

exported to 3D PDF format documents that can be easily

visualized in an adobe reader. This platform enables the

easy sharing of information that has the additional benefit

of being encrypted. Furthermore, 3D PDF documents

enable the visualization of layered information. Figure 21

shows the viewer in action with a 3D PDF model for an

archaeological application.

Furthermore, it is indicated that the 3D-textured models

produced with pipelines c and d can be used in interactive

virtual reality applications. Immersive experiences can be

generated in virtual environments where the user can

interact and analyse the modelled information. Animations

can be created to simulate events (e.g. the water filling of a

dam) or to create a predefined tour around the object of

interest, as shown in the still frame of Fig. 22.

Fig. 18 Surface deformation monitoring of an earth dam located in

Lapo, Portugal. Left 3D visualization, right colour ramp representa-

tion (differences between epochs are expressed in meters)

Fig. 19 Pathology monitoring of concrete wall dam (Alto Ceira dam,

Portugal) top orthophoto, bottom vectorized CAD features drawn over

orthophoto

Fig. 20 Architectural document of street facade survey of Bairro da

Se, Porto, Portugal

Fig. 21 3D PDF of the resulting textured model of the survey of a

roman oven (Quinta do Rouxinol, Seixal, PT)


123

5 Case study: Mosteiro da Batalha

Following the technological developments, conservation

and restoration professionals are now starting to adapt

new digital documents in their analysis workflows (e.g.

please refer to Sect. 4). However, the conventional

architectural drawings like floor plans, elevations and

cross-sections still comprise main charts in monument

preservation surveys. This reality was present in the sur-

vey of the Mosteiro de Santa Maria da Vitoria, known,

among the population, as Mosteiro da Batalha monastery.

This exhaustive survey was awarded to the company

Artescan by the, at that time, Portuguese Institute for

Architectural Heritage (IPPAR) in 2004, with the goal to

create a set of detailed architectural drawings with the

potential to comprise the basis for the establishment of the

architectural interventions specifications.

5.1 Survey role in cultural heritage preservation

Cultural heritage is an asset of great importance for todays

society, not only for cultural reasons, but also for other

reasons such as the economy, tourism or ecology. It is

significantly essential for the authorities to assure that their

historical monuments and sites are well preserved.

Professional activity in cultural heritage preservation

focuses on intervention measures that can be considered to

be classified in three groups as follows: conservation;

restoration and renovation.

Other concepts, related to specific types of intervention

when applied to built heritage or architectural heritage, can

also be used: maintenance; repair and stabilization; reha-

bilitation and modernization; reconstruction and relocation.

For a further description of these processes please refer to

[24].

Before all these intervention projects take place, it is

essential to acquire valuable data of the monument to be

intervened. Together with historical investigations, there are

necessary structural and architectural studies that require

reliable dimensional and often pictoric surveys as a basis. In

this context, different surveying methods can and have been

applied, namely: hand sketching, total station surveying,

photogrammetry and, more recently, 3D laser scanning.

In addition to the above utilities, a ‘correct’ and realistic

representation of a monument comprises a documented

record that can add a significant value in future preserva-

tion of the monument.

Before we present the applied methods and the produced

deliverables, we mention two additional examples of dig-

ital documents with regard to their utility to the contractor

in the context of the Mosteiro da Batalha project:

• The stereotomy of stone was represented in front, rear

and both sides elevation drawings, allowing the tech-

nicians to identify and locate open joints, lacunae to be

filled, or eroded areas (Fig. 23).

• Figures 24 and 25 provide the representation of

distinctive architectural elements and ornamentations

details in the produced digital documents. These allow,

for instance, the establishment of the precise dimen-

sions and form of the existing windows and stained

glasses to be restored, and can therefore be used as a

basis for the specification tender (Figs. 24, 25).

5.2 Surveying and processing the data

The survey specifications required the production of sev-

eral documents that would allow the representation of the

whole monument together with its main architectural

characteristics.

The first stage required the production of four floor plans

at different levels and their corresponding ceiling plans, as

well as four elevations (front, rear and both sides), whereas

the second phase required the production of 64 cross-sections

for the purpose of the whole monument mapping (Fig. 23).

To acquire this huge amount of data, a terrestrial

imaging system technology, combining 3D laser scanning

and photogrammetry, was chosen due to its fast and

accurate data acquisition capabilities. Specifically, a Riegl

Z360i laser scanner was used coupled with a DSLR Nikon

D100 camera.

More than 700 scan positions were necessary to cover

all exterior, interior and top areas of the monument. This

resulted in a data volume of more than 100 GB of raw data

when considering both point clouds and digital images. At

each scan position both point clouds and images were

acquired. The utilized registration method was the one

based on the retroreflectors tie-points as described in

Sect. 3.1. These reflectors were placed all around the

monument in every division, corridor, courtyard and roofs.

Measurement and georeferencing were implemented with

basic tacheometric methods. Hence data coherency with

Fig. 22 Still frame of an animation sequence (Palacio de Monserrate)


123

regard to the data’s expression in a meaningful coordinate

system was ensured.

The processing phase was mainly implemented using

pipeline c for the orthophoto production assisted by pro-

jection planes coincident with the cross-sections, elevations

and plans. Subsequently, all the drawings were produced in

the CAD environment by superimposition of these images.

Point cloud information was finally utilized for cross-

section contouring and 3D visualization for clarification of

details.

6 Future work

In this section we point out areas that would further

improve modelling tasks, in terms of reducing costs and

time. Modelling of large surveys can be a lengthy process,

Fig. 23 Stereotomy of stone of the front facade of Mosteiro da Batalha

Fig. 24 Representation of architectonic elements of a window of

Mosteiro da Batalha

Fig. 25 Representation of ornamentation details and boundaries of

stained glass windows of Mosteiro da Batalha


123

where several intermediate modelling files with different

versions are created to produce a final model. Similarly to

digital photographs, the significantly large number of files

created warrant different search and indexing techniques.

Specifically, it would be desirable to have tools that help

asset management, for example establishing possibly

through static renderings which files map to which survey

areas with alpha blending.

In addition, as mentioned in Sect. 2, most LoD algo-

rithms for interactive real-time visualization of whole tri-

angle meshes are based on the precomputation of the final

model. However, intermediate modelling files can also

have a considerable size and the time for LoD precompu-

tation of several inherently temporary versions of meshes

might not be tolerable by the user.

Future considerations involve having more robust seg-

mentation and classification algorithms where less manual

intervention would be required. In the context of the

growing field of mobile laser scanning, the huge amount of

data that is collected will need efficient and more auto-

mated processing. For both terrestrial and mobile laser

scanning, and in what concerns the data acquisition phase,

improvements on real-time processing of the data to allow

for instantaneously more detailed visualization, would be

desirable. Finally, it would be desirable to have better

integration of CAD primitives and triangulated meshes.

Acknowledgments This work was supported by FCT (INESC-ID

multiannual funding) through the PIDDAC Program funds. All laser

scanning and modelling presented in this article were carried out by

Artescan, 3D Scanning. The collaboration between Artescan and

INESC-ID Lisboa was partially funded by Artescan and by the Por-

tuguese Foundation for Science and Technology (FCT), VIZIR pro-

ject grant (PTDC/EIA/66655/2006). In addition, Bruno Araujo would

like to thank FCT for doctoral grant reference SFRH/BD/31020/2006.

Finally, we would like to thank Dr. Margarita Rova for invaluable

comments.

References

1. Ahrens, J., Geveci, B., Law, C.: ParaView: An End-user Tool for

Large Data Visualization. Visualization Handbook. Elsevier

(2005)

2. Aiger, D., Mitra, N.J., Cohen-Or, D.: 4-Points congruent sets for

robust pairwise surface registration. In: ACM SIGGRAPH 2008

papers, ACM, pp. 85:1–85:10 (2008)

3. Alba, M., Scaioni, M.: Comparison of techniques for terrestrial

laser scanning data georeferencing applied to 3-D modelling of

cultural heritage. In: The International Archives of the Photo-

grammetry, Remote Sensing and Spatial Information Sciences,

36(5/W47), (on CD-ROM, p. 8 (2007)

4. Amenta, N., Bern, M., Kamvysselis, M.: A new Voronoi-based

surface reconstruction algorithm. In: Proceedings of the 25th

Annual Conference on Computer Graphics and Interactive

Techniques, ACM, SIGGRAPH ’98, pp. 415–421 (1998)

5. Beinat, A., Crosilla, F. Generalized procrustes analysis for size

and shape 3-D object reconstruction. In: Proceedings of Optical

3D Measurement Techniques, pp. 345–353 (2001)

6. Besl, P.J., McKay, N.D.: A method for registration of 3-D shapes.

IEEE Trans. Pattern Anal. Mach. Intell. 14, 239–256 (1992)

7. Boavida, J., Oliveira, A., Berberan, A.: Dam monitoring using

combined terrestrial imaging systems. Civil Engineering Sur-

veyor, pp. 33–38 (2009)

8. Borgeat, L., Godin, G., Blais, F., Massicotte, P., Lahanier, C.:

Gold: interactive display of huge colored and textured models. In:

ACM SIGGRAPH 2005 Papers, ACM, pp. 869–877 (2005)

9. Cignoni, P., Montani, C., Rocchini, C., Scopigno, R.: External

memory management and simplification of huge meshes. IEEE

Trans. Vis. Comput. Graphics 9, 525–537 (2003)

10. Cignoni, P., Callieri, M., Corsini, M., Dellepiane, M., Ganovelli,

F., Ranzuglia, G.: Meshlab: an open-source mesh processing

tool. In: Sixth Eurographics Italian Chapter Conference,

pp. 129–136, (2008) http://vcg.isti.cnr.it/Publications/2008/CCC

DGR08

11. Curless, B., Levoy, M.: A volumetric method for building com-

plex models from range images. In: SIGGRAPH ’96, ACM,

pp. 303–312 (1996)

12. Davis, J., Marschner, S.R., Garr, M., Levoy, M. (2002) Filling

holes in complex surfaces using volumetric diffusion. In: 3DPVT,

IEEE Computer Society, pp. 428–438

13. Garland, M., Heckbert, PS.: Surface simplification using quadric

error metrics. In: SIGGRAPH ’97, ACM Press/Addison-Wesley,

pp. 209–216 (1997)

14. Gobbetti, E., Marton, F.: Far voxels: a multiresolution framework

for interactive rendering of huge complex 3D models on com-

modity graphics platforms. In: ACM SIGGRAPH 2005 Papers,

ACM, pp. 878–885 (2005)

15. Gruen, A., Akca, D.: Least squares 3D surface matching. In: The

International Archives of the Photogrammetry, Remote Sensing

and Spatial Information Sciences, 34(5/W16), (on CD-ROM

(2004)

16. International Electrotechnical Commission: International stan-

dard iec 60825-1 corr1, safety of laser products, 1st August 2008

(2008). http://webstore.iec.ch/p-preview/info_iec60825-1

17. International Society for Photogrammetry and Remote Sensing,

Commission V: Close-range sensing: Analysis and applications,

working group v/3: Terrestrial laser scanning and 3D imaging.

(2009) http://www.commission5.isprs.org/wg3/

18. Isenburg, M., Liu, Y., Shewchuk, J., Snoeyink, J.: Streaming com-

putation of Delaunay triangulations. ACM Trans. Graph.

25:1049–1056 (2006). http://doi.acm.org/10.1145/1141911.1141992

19. Large, A.R.G., Heritage, G.L., Charlton, M.E.: Laser Scanning:

The Future, Wiley-Blackwell, pp. 262–271 (2009)

20. Levoy, M., Pulli, K., Curless, B., Rusinkiewicz, S., Koller, D.,

Pereira, L., Ginzton, M., Anderson, S., Davis, J., Ginsberg, J.,

Shade, J., Fulk, D.: The digital Michelangelo project: 3D scan-

ning of large statues. In: SIGGRAPH ’00, ACM Press/Addison-

Wesley, pp. 131–144 (2000)

21. Luhmann, T., Robson, S., Kyle, S., Harley, I.: Close Range

Photogrammetry Principles, Methods and Applications. Whittles,

Scotland (2006)

22. Oliveira, A., Boavida, J., Cortesao, L., Gomes, S.: Laser scanning

aplicado ao levantamento do patrimonio edificado—o caso de

estudo do palacio de monserrate. Revista Arte & Construcao,

pp. 44–48 (2009)

23. Oliveira, J.F., Steed, A.: Determining orientation of laser scanned

surfaces. In: SIACG, pp. 281–288 (2002)

24. Petzet, M.: Principles of preservation: an introduction to the

international charters for conservation and restoration 40 years


123

http://vcg.isti.cnr.it/Publications/2008/CCCDGR08

http://vcg.isti.cnr.it/Publications/2008/CCCDGR08

http://webstore.iec.ch/p-preview/info_iec60825-1

http://www.commission5.isprs.org/wg3/

http://doi.acm.org/10.1145/1141911.1141992

after the venice charter. In: International Charters for Conservation

and Restoration, Monuments & Sites, vol. I, ICOMOS, Munchen,

pp. 7–29, (2004). http://openarchive.icomos.org/432/

25. Pfeifer, N., Briese, C.: Geometrical aspects of airborne and ter-

restrial laser scanning. In: International Archives of Photogram-

metry, Remote Sensing and Spatial Information Sciences, vol. 36,

part 3/W52, p 311 (2007)

26. Rieger, P., Studnicka, N., Pfenning-bauer, M., Ullrich, A. Advan-

ces in mobile laser scanning data acquisition. In: FIG Congress

2010. Facing the Challenges Building the Capacity (2010)

27. Rusinkiewicz, S., Levoy, M.: Qsplat: A multiresolution point

rendering system for large meshes. In: SIGGRAPH ’00, ACM

Press/Addison-Wesley, pp. 343–352 (2000)

28. Rusinkiewicz, S., Levoy, M.: Efficient variants of the ICP algo-

rithm. In: International Conference on 3-D Digital Imaging and

Modeling (2001)

29. Schroeder, W., Martin, K., Lorensen, B.: The Visualization

Toolkit: An Object-Oriented Approach to 3D Graphics, 4th edn.

Kitware Inc. (2006)

30. Taubin, G., Zhang, T., Golub, G.H.: Optimal surface smoothing

as filter design. In: Proceedings of the 4th European Conference

on Computer Vision, vol. I, Springer, London, UK, ECCV ’96,

pp. 283–292 (1996)

31. Thies, M., Spiecker, H.: Evaluation and future prospects of ter-

restrial laser-scanning for standardized forest inventories. In:

Proceedings of the ISPRS Working Group VIII/2 Laser-Scanners

for Forest and Landscape Assessment (2004)

32. Ullrich, A., Pfennigbauer, M.: Echo digitization and waveform

analysis in airborne and terrestrial laser scanning. In: Fritsch, D.

(ed.) Photogrammetric Week ’11, pp. 217–228 (2011)

33. Wand, M., Berner, A., Bokeloh, M., Fleck, A., Hoffmann, M.,

Jenke, P.: The xgrt system. (2007). http://www.gris.uni-tuebingen.

de/xgrt/

34. Wand, M., Berner, A., Bokeloh, M., Fleck, A., Hoffmann, M.,

Jenke, P., Maier, B., Staneker, D., Schilling, A.: Interactive

editing of large point clouds. In: Chen B, Zwicker M, Botsch M,

Pajarola R (eds.) Symposium on Point-Based Graphics 2007 :

Eurographics/IEEE VGTC Symposium Proceedings, Prague,

Czech Republic, pp. 37–46 (2007)

Author Biographies

Adriano Oliveira is the

Technical Director at Arte-

scan—3D Scanning. He gradu-

ated in Geographic Engineering

at the University of Coimbra

and developed a research pro-

ject in Classification and The-

matic Mapping using GIS at

the Cartographic Laboratory of

the Excellence Centre in Tele-

geomatics (University of Trie-

ste, Italy). He has a post

graduate degree in Remote

Sensing from the University of

Porto. In Artescan, since 2004

he has been actively working with the planning, acquisition and

processing of terrestrial and mobile laser scanning as well as high-

resolution imagery, developing solutions for applications in the

areas of heritage, surveying, reverse engineering, geology and vir-

tual 3D modelling. He has been involved with several national and

international research and development projects in the above-refer-

red areas.

Joao F. Oliveira holds a PhD in

Computer Science (Geometry

reduction) from University Col-

lege London. He recently finished

his post-doc in out-of-core visu-

alization techniques at INESC-ID

(Computer Systems Engineering

Institute). He is currently adjunct

Professor at Instituto Politecnico

de Portalegre—Escola Superior

de Tecnologia e Gestao, where he

teaches 3D programming. His

main interests are point cloud

processing and 3D visualization

systems.

Joao M. Pereira is Associate

Professor at the Computer Sci-

ence Department of the Technical

University of Lisbon (Instituto

Superior Tecnico—IST/UTL)

where he teaches Computer

Graphics. Joao Pereira holds a

PhD in Electrical and Computers

Engineering (Computer Graph-

ics) from IST/UTL, Technical

University, December 1996. He

received also a MSc and a BsEE

degrees in Electrical and Com-

puters Engineering from IST/

UTL, respectively, in 1989 and

1984. He coordinates the Visualization and Simulation action line of the

VIMMI group at INESC-ID (Computer Systems Engineering Institute).

His main research fields are Real-Time Rendering, 3D Game Program-

ming, Serious Games, Networked Virtual Environments, Augmented

Reality and Parallel Computer Graphics. He won several prizes like the

international 2nd Prize at ‘‘1994 MasParChallenge Contest’’ and the

national 1st prizes ‘‘Best Graphic Interactive Demonstrator’’ in 2004 and

2002, respectively, with the games ‘‘Lost Ages—A Massive Role Playing

Game’’ and ‘‘Peace and War Games: Large Scale Simulation Over the

Internet’’. He has been involved with several European (RESOLV,

IMPROVE, RTP11.13, MAXIMUS, SATIN, INTUTION, TARGET,

GALA) and National projects. He was also proposal evaluator of the FET

during 2009. He is author or co-author of more than 100 peer-reviewed

scientific papers presented at national and international events and

journals. Professor Pereira is member of the Eurographics Association

and EuroVR Association.

Bruno R. de Araujo is a PhD

Student at the Instituto Superior

Tecnico from the Technical

University of Lisbon. He is a

researcher at INESC-ID and the

Intelligent MultiModal Inter-

faces Group. He participated on

European Projects such as

SMARTSKETCHES research-

ing advanced interaction tech-

niques for 3D surfacing using

Calligraphic interfaces and the

IMPROVE project proposing

innovative interfaces for im-

mersive and mixed reality. He is

interested in large-scale display-based visualization using PC cluster

and multi-projector systems, and participated in the LEME (Laboraty

in Mobility and Excellence) at IST related with intelligent ambient


123

http://openarchive.icomos.org/432/

http://www.gris.uni-tuebingen.de/xgrt/

http://www.gris.uni-tuebingen.de/xgrt/

and tiled display visualization technology. http://immi.inesc-id.pt/*brar/ or you can email at [email protected].

Joao Boavida is the Managing

Director at Artescan—3D

Scanning. He graduated in

Geographic Engineering. He

worked in GPS double fre-

quency observables and their

computations at the Geodesy

group of the Technische Uni-

versitat Berlin. At the Geofors-

chung Zentrum Potsdam

(Germany), he processed gravi-

tational data in the AGMASCO

EU project. He worked for the

Potsdamer Platz and Reichtag

renovation, as well as for the

cadastre of the renewed underwater system in eastern Berlin. From

1999 to 2003, J. Boavida was the technical director of UNAVE, a

Geoinformation and Remote Sensing laboratory at Aveiro University,

Portugal. From 2000 to 2001 he completed his post-degree in Remote

sensing by the University of Porto, Portugal. In 2003, he founded

Artescan, which uses Combined Terrestrial Imaging Systems and

produces detailed 3D models in the engineering, mining and heritage

domains.


123

http://immi.inesc-id.pt/~brar/

http://immi.inesc-id.pt/~brar/

Documents

3D modelling of laser scanned and photogrammetric data for ... · Mosteiro da Batalha monastery. Finally, in Sect. 6 we outline desirable solutions that would help the modelling process