2013 Digital Heritage International Congress
(DigitalHeritage) federating the 19 Int’l VSMM, 10 Eurographics GCH, & 2nd UNESCO Memory of the World Conferences, plus special sessions from CAA, Arqueológica 2.0, Space2Place, ICOMOS ICIP & CIPA, EU projects, et al. th
th
Volume 1
28 Oct – 1 Nov 2013 Marseille, France
Copyright ©2013 by the Institute of Electrical and Electronics Engineers, Inc. All rights reserved Copyright and Reprint Permission: Abstracting is permitted with credit to the source. Libraries are permitted to photocopy beyond the limit of U.S. copyright law for private use of patrons those articles in this volume that carry a code at the bottom of the first page, provided the per-copy fee indicated in the code is paid through Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA01923. For other copying, reprint or republication permission, write to IEEE Copyrights Manager, IEEE Operations Center, 445 Hoes Lane, Piscataway, NJ 08854. All rights reserved.
IEEE Catalog Number: CFP1308W-USB ISBN: 978-1-4799-3169-9
I
DigitalHeritage 2013 Congress Committee Honorary Chairs Francesco Bandarin, Assistant Director General for Culture, UNESCO Janis Karklins, Assistant Director General for Communications & Information, UNESCO Vincent Berjot, Director of Heritage, French Ministry of Culture and Communications Patrice Bourdelais, Director, Institute of Humanities and Social Sciences, CNRS Michel Vauzelle, President, Regional Council of Provence-Alpes-Côte d’Azur & President, Villa Méditerranée Bruno Suzzarelli, President, MuCEM Federated Event Chairs ArchaeoVirtual2013: Sofia Pescarin, CNR – Italian National Research Council (Italy) Arqueologica2.0: Victor Lopez Menchero, SEAV (Spanish Society of Virtual Arch.) CAA2013 Fall Workshop, Jeffrey Clark, North Dakota State University (USA) Digital Art Week: Art Clay, ETH-Zurich (Switzerland) GCH2013: Dieter Fellner, Franuhofer Inst. (Germany) ICOMOS/ISPRS CIPA: Mario Santana, Carleton Univ. (Canada) ICOMOS ICIP: Claudia Liuzza, Stanford Univ. (USA) Space2Place: Maurizio Forte, Duke Univ (USA) & Stefano Campana, Univ. of Siena (Italy) VSMM2013: Lon Addison & Livio De Luca, VSMM Society Congress Co-Chairs Alonzo C. Addison, VSMM Society Livio De Luca, French National Center for Scientific Research (CNRS-MAP) Sofia Pescarin, Italian National Research Council (CNR ITABC) Program Committee Program & Papers Chair: Gabriele Guidi, Polytechnic of Milan (Italy) Scientific Chairs: Roberto Scopigno, CNR – Italian Nat’l Research Council Scientific Commissioners: Digitization track: J. Angelo Beraldin, Canadian Nat’l Research Council (Canada) Analysis track: Robert Sablatnig, Technical University of Vienna (Austria) Visualization track: Holly Rushmeier, Yale University (USA) Policy track: Hal Thwaites, Universiti Malaya (Malaysia) Preservation track: Julian Richards, Archaeology Data Service (UK) Projects track: Maurizio Forte, Duke University (USA) Built Heritage theme: Lisa Fischer, Colonial Williamsburg Foundation (USA) Museums & Collections theme: Sarah Kenderdine, City Univ. of Hong Kong (China) Documentary Heritage theme: Joie Springer, UNESCO Panels Chair: Costis Dallas, University of Toronto (Canada) Workshops Chair: Denis Pitzalis, UNESCO CFP & Posters Chair: Holger Graf, Fraunhofer IGD (Germany) Reviews Chair: Sorin Hermon, Cyprus Institute (Cyprus) Exhibition Chair: Art Clay, ETH Zurich (Switzerland) EU Meetings Chair: Erik Champion, Aarhus University (Denmark) VII
Organizing Committee Local Chair: Livio De Luca, CNRS, Marseille Local Board: Pierre Alliez, INRIA Sophia Antipolis – Mediterranée, Denis Chevallier, MuCEM, Marseille Marc Daniel, LSIS, Aix-Marseille University, Marseille Philippe Jockey, CCJ, Aix-Marseille University, Aix-en-Provence Jean-Marc Vallet, CICRP, Marseille Philippe Véron, LSIS, Arts et Métiers ParisTech, Aix-en-Provence
VIII
DigitalHeritage 2013 International Program Committee Alonzo C. Addison, USA Pierre Alliez, FR Carlos Andujar, ES David Arnold, UK Alessandro Artusi, ES Adriana Bandiera, IT Francesco Bellotti, IT Jean Angelo Beraldin , CA Jean -Yves Blaise, FR Ulrich Bockholt, DE Jean -Luc Bodnar, FR David Bommes, FR Monica Bordegoni, IT Davide Borra, IT Michael Brown, SG Stefan Bruckner, NO Paul Bryan, UK Patrick Callet, FR Marco Callieri, IT Stefano Campana, IT Vito Cappellini, IT Vittore Casarosa, IT Maria Luisa Catoni, IT Paolo Cignoni, IT Antonio Cisternino, IT Jeffrey T. Clark, USA Arthur Clay, CH Sabine Coquillart, FR Andrea D’Andrea, IT Marc Daniel, FR Livio De Luca, FR Matteo Dellepiane, IT Vincent Detalle, FR Julie Digne, FR Michael Doneus, AT Pierre Drap, FR Luciana Duranti, CA Rand Eppich, ES Stephen Fa , CA Bianca Falcidieno, IT Mercedes Farjas, ES Mohamed Farouk, EG Isabelle Fasse -Calvet, FR Francisco Feito, ES Dieter Filner, DE Francesco Ferrise, IT Lisa Fischer, USA Julian Flores, ES Mariano Flores, ES Maurizio Forte, USA Christoph Franzen, DE Bernard Frischer, USA Francesco Gabellone, IT
Fabio Ganovelli, IT Andreas Georgopoulos, GR Gesquiere Gilles, FR Jeffrey Glover, USA Enrico Gobbetti, IT Guy Godin, CA Sanjay Goel, IN Holger Graf, DE Alfredo Grande, ES Pierre Grussenmeyer, FR Francois Guena, FR Antonella Guidazzoli, IT Gabriele Guidi, IT Gilles Halin, FR William Hanson, UK Sven Havemann, AT Sorin Hermon, CY Luis Hernandez, ES Mona Hess, UK Katsushi Ikeuchi, JP Doug Jarvis, CA Erszebet Jerem, HU Philippe Jockey, FR Yvonne Jung, DE Hans Kamermans, NL Jaime Kaminski, UK Martin Kampel, AT Eric Kansa, USA Sarah Kenderdine, HK Min H. Kim, KR Florent Lafarge, FR Guus Lange, NL Urich Lau, SG Hasup Lee, JP José Luis Lerma, ES Fotis Liarokapis, UK Maria Liouliou, FR Ioannis Liritzis, GR Claudia Liuzza, USA Gary Lock, UK Victor M. Lopez -Menchero, ES Scott Madry, USA Nadia Magnenat -Thalmann, SG Ricardo Marroquim, BR Asla Medeiros E Sa, BR Javier Melero, ES Despina Michael, CY Mark Mudge, USA Michela Ott, IT Gianpaolo Palma, IT Zhigeng Pan, CN George Papagiannakis, GR Sofia Pescarin, IT
IX
Marc Pierrot Deseilligny, FR Denis Pitzalis, FR Daniel Pletinckx, BE Axel Posluschny, DE Dominic Powlesland, UK William Puech, FR Adam Rabinowitz, USA Romain Raffin, FR Fabio Remondino, IT Alejandro Ribes, FR Julian Richards, UK Karina Rodriguez, UK Pablo Rodriguez -Navarro, ES Maria Roussou, GR Holly Rushmeier, USA Michele Russo, IT Robert Sablatnig, AT Fathi Saleh, EG Donald Sanders, USA Pedro Santos, DE Martin Sauerbier, CH Pasquale Savino, IT Christopher Schwartz, DE Roberto Scopigno, IT Michela Spagnuolo, IT Joie Springer, FR Stephen Stead, UK Andre Stork, DE Didier Stricker, DE Daniel Thalmann , SG Maria Theodoridou, GR Hal Thwaites, MY Corey Toler -Franklin, USA Juan Carlos Torres, ES Jean -Marc Vallet, FR Giorgio Verdiani, IT Philip Verhagen, NL Philippe Véron, FR Krzysztof Walczak, PL Martin White, UK Alexander Wilkie, CZ Ryan Williams, USA Theodor Wyeld, AU Hyun Seung Yang, KR Jiang Yu Zheng, USA
Multi-scalar 3D digitization of Cultural Heritage using a low-cost integrated approach Anna Maria Manferdini
Michele Russo
Dept. of Architecture University of Bologna Bologna, Italy
[email protected]
Dept. Design Politecnico di Milano Milan, Italy
[email protected]
During the last decade, the developments in the highresolution digital photogrammetry field [6-8] has led to the use of 2D images as tools able to derive detailed and accurate 3D reality-based information. The achievements of this approach and the contemporaneous lowering of costs of range sensors are actually encouraging the widespread of 3D digitization and are stimulating investigations on the possibilities of their integration to optimize final results and the whole pipeline [914]. This last aspect, in particular, represents a fundamental task in the field of Cultural Heritage survey, since artifacts usually present multi-scalar geometrical complexities and variations which don’t allow to define standard methodologies and procedures for every situation and context. In the last years, many interesting applications have been developed [1516], that confirm, on one hand, that the optimization of results is the best practice within multi-scalar contexts; on the other hand, they show that the pipeline is often complex and requires expertise. In order to overcome these critical aspects, recently in the computer vision field, researches have been held based on the Structure from Motion approach. This latter is based on the principle that the structure of four non-coplanar points is recoverable from three orthographic projections [17]. Following this statement, scientific research developed technologies that allow the possibility of reconstructing 3D scenes and camera motion through sets of 2D images. The process mainly consists in image orientation and camera selfcalibration. In computer vision, these processes are based on projective geometry and consist in the derivation of parameters of interior and exterior orientation of cameras. Since in this field automation is more important than accuracy and, in addition, in common situations it is not always possible to use targets, many investigations have been held in order to perform automatic markerless orientation of 2D images [18-22]. In the last years, different solutions and algorithms have been developed [23-25]; some of these have been integrated inside web-servers (Arc3D, Hypr3D, Autodesk 123D Catch, My3DScanner, etc.) [26], while others have been integrated in free open-source packages (VisualSFM, Apero, Insight3D, Bundler, etc.). These solutions, along with the possibility to use free
Abstract—In the architectural survey field, one of the main aspects to consider during a 3D digitization is the multi-scalar geometrical complexity of the artifact to acquire, besides other fundamental factors connected with the different aims of communication. Since the widespread of range-sensors has provided extremely versatile instruments able to easily acquire huge amount of data that can be processed for different uses and users and changing communication aims, the possibility to survey and restore high-quality 3D multi-resolution models has become an urgent need. Despite these developments, these technologies are still very expensive, need expertise and present persistent bottlenecks both in the reverse modeling process and in time consuming. In order to overcome these critical aspects and taking advantage of recent improvements of automated image-based technologies based on the Structure from Motion approach, this contribution presents some first results of investigations on the reliability of these low-cost technologies for the 3D digitization of Cultural Heritage. One of the main aims of these investigations rely on developing a procedure that could ease the work of surveyors called to represent artifacts at an architectural scale using fast and low-cost technologies. 3D models derived using the selected low-cost image-based technologies were compared among each other and with a 3D laser scanner gold standard acquisition. These investigations led to qualitative and quantitative evaluations and to considerations on times and skills required by all tested technologies. Strengths and weaknesses are highlighted, suggesting the best solution with respect to the optimization of all considered aspects. Finally, integration of different technologies are presented, as it represents the best solution in multi-scalar contexts. .H\ZRUGV—Multi-scalar, SfM, 3D Scanning, Low cost, Data comparison, Process optimization
I. INTRODUCTION Since the widespread of range sensors in the Cultural Heritage field [1-5], the possibility to acquire in easily and quickly way huge amount of 3D data for different aims and in different times has changed from one side the expectations of users, called to collect information about sites and artifacts, from the other has widened the need to acquire complex and versatile data to be used in different contexts. Despite these achievements, range sensors are still very expensive, need expertise and present persistent bottlenecks both in the reverse modeling process and in time consuming.
978-1-4799-3169-9/13/$31.00 ©2013 IEEE
153
production of these well preserved terracotta elements that have been fabricated using moulds, assures the repetition of constant geometries, as if they belonged to an abacus that composes the whole facade. The check on the possibility for low-cost technologies to survey their small-scale details and their repetitive shapes were some of the main aims of our investigations, since low-cost technologies usually present some problems in the modeling pipeline using high-definition images and in acquiring recurrent geometries.
software for the post-processing of data (Meshlab, MicMac, etc.) and with the sharing of 3D data through the web (X3DOM, OSG4WEB, Google O3D, Canvas3D JS Library, CubicVR, Three, SceneJS, SpiderGL, etc.) are actually enhancing their intense use for different aims ranging from simple visualizations for information purposes, to the promotion of Cultural Heritage sites [27]. Some aspects that do not enhance the use of these survey technologies for scientific purposes, are mainly due to their extreme automation that limits the possibility to control the processing of data. In addition, they often produce unpredictable results that are related to shooting problems (lighting changes, bad sequence of images, poor overlap, repetitive elements, homogeneous textures, etc.). The potentialities offered by these tools in contexts where accuracy and precision are indispensable tasks are actually being investigated [28-30] and one promising field of application seem to be the architectural one [31], where accuracy and definition is generally under pre-defined thresholds. Starting from these considerations, the aim of this paper is to evaluate if SfM technologies could be used as reliable instruments in the architectural field, where one of the main aspects to consider during a 3D digitization is the multi-scalar geometrical complexity of the artifact to acquire, besides other fundamental factors connected with the different aims of communication. Within the developed methodology, 3D models derived by 2D images were compared among each other and with a 3D laser scanner gold standard acquisition. Strengths and weaknesses of all processes are highlighted, as well as evaluations on times and skills required by all tested technologies, suggesting the best solution with respect to the optimization of all considered aspects. Finally, integration of different technologies are presented, as it actually represents the best solution for multi-scalar purposes.
III. METHODOLOGY The methodology described in this paper deals with the use of different 3D acquisition and modeling techniques, based both on 2D and 3D information and developed in order to suggest an optimized and integrated multi-scalar approach to 3dimentional digital surveys. As already mentioned, the aim of the paper is to suggest a reality-based digitization pipeline that might be easily adopted to provide information for different uses, with different levels of details. In order to reach this purpose, the first crucial aspect consisted in the selection of the most appropriate scale of detail. Normally this definition depends on several aspects that mainly reside in communication aims, but also deeply depend on the spatial characteristics of the artifact (overall dimensions, minimum dimension of small-scale details, their definition and density within the whole architectural framework, etc.).
II. CASE STUDY Our investigations were conducted on the small St. Spirito Church that is located in the historical center of Bologna, Italy and that was commissioned by the Celestine Fathers and built in 1481. At the end of the XVIII century, after the abolition of the Celestines monastic order, the church was abandoned and sacked, until it was restored at the end of the XIX century by Alfonso Rubbiani. Although the church is actually private, off the beaten tracks and almost inaccessible, it is an extraordinary example of the use of terracotta ornaments on architectural façades. As a matter of fact it is completely decorated using XV century elements that were attributed to the Bolognese sculptor Vincenzo Onofri or to Sperandio, the medallist of Mantua [32]. Besides its artistic relevance, within our research, we selected this architecture as interesting case study because of its multi-scalar complexity. The use of very detailed ornaments and their recurrences on the façade contribute to the definition of the overall architecture, where each single element is essential inside the whole composition. In addition, the
Fig. 1. View of the façade of the St. Spirito church.
154
fulfill the low-cost approach, we compared 3D models built using different SfM approaches with a gold standard survey performed using a Time of Flight laser scanner. We analyzed different aspects, ranging from work distances to time consuming for acquisition and data processing, evaluating the quality of restored data in terms of resolution, accuracy and cleanness of geometrical information. These analysis led to the selection of the best solution for each scale of complexity and to considerations on their deep integration.
In the presented case study, before survey planning, we singled out two different levels of detail. The first one refers to the 2D technical drawings creation, in order to document architectural and decorative characteristics of the whole building to be used for examples by Superintendence employers. The small overall dimensions of the church led to the selection of 1:10 scale as the most appropriate one to document the main features of its structure, such, as, for example, its main volumes, its openings and the geometry of planar surfaces on which decorated elements are inserted. A second level of detail was selected in order to provide representations of single terracotta elements that could be used, for example, by restores. In order to fulfill this aim, we selected a scale ranging from 1:1 to 1:2, depending on their distance from acquisition location. Within this framework, we tested different 3D survey technique based on 3D laser scanner and 2D images in order to develop a methodology and single out a procedure that could be adopted within other multi-resolution contexts. In order to
IV. PROCESS A. 3D acquisition The range-based survey was performed using a Leica C5 laser scanner in order to evaluate the reliability of SfM technologies for architectural surveys. The survey campaign was planned to acquire 3D shapes at 1 cm of resolution. In order to reduce costs and time for survey campaign and
Fig. 2. Scheme of survey planning using a Leica C5 ToF laser scanner (scan stations A and B) and using 2D images at two different scales of complexity. On the right, overlapping areas of images captured to survey large scale (1:10) elements. On the left, at the small-scale, relationship between the focal length and distances from the element to measure.
155
data process, the scanner was not positioned at different heights; we planned two station positions (see points A and B in Fig. 2) to cover horizontal undercuts. Given the main characteristics of this kind of laser scanner (accuracy of single measurement: position: 6 mm, distance: 4 mm, at 1÷50 m range, one sigma; 2 mm std. deviation for target acquisition), we fixed the resolution of this survey campaign at 1x1 cm at 10 m distance, while scan positions were located 3 m far from the building. The image survey campaign was carried out with a standard digital camera (NikonD90), equipped with a AF-S DX Nikkor 18-105 mm, in order to provide images with different resolutions that could be processed using some of the main widespread and low-cost SfM tools. In particular, we selected a commercial tool, Agisoft PhotoScan and two free ones, Visual SFM and Autodesk 123D Catch, that do not require particular skills for the post-processing of data. During the planning of survey campaign, we took into consideration the main shooting rules required by each technology (minimum overlap, convergence, etc.), in order to evaluate the number and distribution of images required to cover the whole surface of the artifact in advance. Figure 2 reports the layout of images at two scales of complexity. The 2848 x 4288 pixel at 300 dpi images determined the resolution of acquisition with which we had to define the most appropriate focal length with respect to the two scales of
representation and to the distance of the artifact. In particular, as far as the 1:10 scale is concerned, image resolution determined single pixel dimension (0.085 mm) that was compared with the dimension of the object to be represented at that scale (for example, the 5 m large façade should measure 50 cm). This relationship determined the number of images and the related focal length necessary to capture the object to survey at a pre-defined distance (3 photos captured at 3 m distance, using the 18 mm focal length, to which the requested minimum overlap and images mandatory to survey undercuts have to be added). As we wanted to pursuit the low-cost aim and testing our methodology in standard and recurrent survey conditions, we did not use any additional devices (i.e. drones, scaffolding) to take pictures from different heights. Moreover, similarly as recurrent situations, the presence of a fixed obstacle (barrier) that surrounds the building, further constrained camera locations. B. Data process 3D data acquired using laser scanner were processed following a well-know pipeline: each scan was cleaned and aligned using the ICP algorithm, defining a first global point cloud of about 9 millions points. Afterwards these data were merged and meshed, creating a first raw polygonal model, that was edited and optimized generating a final model of 2.3 million polygons. This first model was used as reference in the comparison step.
Fig. 3. Rendering of the texturized models obtained from the whole façade and exemplification of different single ornaments.
156
deviation that is clearly below the standard deviation parameter of the 3D scanner. The other two ornaments (cornice and medallion) that are more distant from camera location, present some problems that are probably due to errors in image alignment. This is evident in the asymmetrical distribution of deviations (123D Catch). The same analysis was reiterated for the whole façade (Fig. 5). In this case, the Photoscan model is more reliable than the others, as it is closer to the reference one. Table I shows the evident differences of polygons number of 3D models that affects, on one side, memory consumption and computational costs. Table I also shows the progressive increase of Standard Deviation in relation with the distance of each architectural element from camera locations. The second kind of geometrical evaluation can be performed using sections of the façade that clearly highlight
This reverse modeling approach requires expertise in order to reach good quality results and face the different bottlenecks that are still present in the editing and optimizing process. In addition, this is a very time consuming process, as it requires to clean the polygonal surface, to correct topological errors and to properly fill holes. After this process, as this polygonal model does not restore a good quality chromatic information, it needs to be manually textured. This further step is generally rather complex and time consuming, as it consists in the projection of different 2D images acquired from different points of view on the surface. In case of complex shapes or non visible areas, this process does not allow to cover the whole model with images, as it is generally difficult to capture images from all possible points of view that it is possible to assume during the exploration of the model in a digital environment. In addition, as real objects are always inserted in a specific illumination context that is generally characterized by shadows and lighting changes, while 3D models should be used in different lighting contexts, the blending process aimed at attenuating sudden lighting changes, is generally a very long and delicate step. The sequences of images related to the whole façade and to the selected ornamental portions were processed using Visual SFM, Autodesk 123D Catch and Agisoft Photoscan. These systems based on the SfM approach allow to automatically recognize all tie points correspondences between images, calculate internal and external parameters and finally obtain sparse blocks. Afterwards, colored point clouds (VisualSfM) and texturized polygonal models (123D Catch, Photoscan) are automatically generated. Within this research, four different sets of images were processed using the aforementioned packages and 12 texturized models were reconstructed. Figure 3 shows four of them, as exemplification of the final models obtained from the image-based modeling process. C. Data comparison The analysis of 3D data derived by laser scanner and using the SfM approach regarded different aspects. First of all, metrological aspects were inquired by measuring deviations among different polygonal models, as well as by comparing different 2D sections. This step was performed in order to highlight the reliability of the adopted SfM techniques with respect to the range-based one and also to evaluate their ability to acquire the principal architectural outlines that can be used, for example, in technical drawings. A crucial aspect that is mandatory before the metric comparison among the rebuilt 3D models is the definition of the absolute dimensions of digitized artifacts. Although in some accessible areas, measures were directly surveyed using traditional systems (i.e. measuring tape and laser distance meter), reference measures of each architectural element were extracted from the gold standard 3D model in order to reduce errors due to scale factors and perform a more reliable comparison. The analysis of mesh deviations reported in Fig. 4 highlights, for example, that the pilaster that was the closest element to camera locations, presents a well-distributed
Fig. 4. Deviation maps between the 3D model of each ornament derived different SfM packages and 3D gold standard. Left, height location of each ornament within the façade.
Fig. 5. Deviation maps of the façade in the different software
157
TABLE I. TABLE OF ALL CONSISTENCIES. 3D survey 3D postprocessing 2D survey 123D Catch processing VisualSfM processing Photoscan processing
Façade 2
Pilaster /
Cornice /
Medallion /
# Poly (x 103)
2 291
11
49
34
# Images Focal lenght # Poly (x 103) Std. Dev. # Poly (x 103) Std. Dev. # Poly (x 103) Std. Dev.
55 18 211 0,0213 6 078 0,0187 2 130 0,0188
17 50 56 0,0025 1 800 0,0021 1 800 0,0024
7 108 56 0,0026 735 0,0028 2 265 0,0034
8 108 27 0,0038 273 0,0043 791 0,0047
# Scan
the reliability of each package in the derivation of the main architectonical outlines. The comparison exemplified in Fig. 6, focuses the attention on one important aspect related to the geometrical characteristics of this architecture. If one purpose of this kind of survey is the possibility to create orthogonal projections to be used within technical drawings, the acquisition of all building edges becomes an essential task. From this point of view, the closest section to the gold standard one was obtained using VisualSfM; nevertheless this latter is defined by a sequence of portions of curves. The Photoscan model is instead very smoothed, in spite the set out sharp settings, however it represents the best solution in terms of deviation (Fig. 4). A third kind of analysis was carried out comparing the ability of each technique to restore the shape of sculpted terracotta ornaments. Figure 7 allows visual evaluations on the quality of restored data. Autodesk 123D Catch generated very clean geometries that, on the other hand, appear quite smoothed in their sculpted elements. Photoscan provided less smoothed 3D models but also generated several mesh incoherencies, while VisualSfM represents the best solution in terms of edge conservation. Unfortunately, geometries built using this last tool present several lacks. The overlapping of chromatic information flattens these differences and increases sculptural effects. The same analysis were repeated for the other details of the façade. These further comparisons confirm the aforementioned evaluations and highlight the rising of problems due to the increasing distance from camera locations. This quality check demonstrates on one side that the VisualSfM approach represents the best solution when applied at short distances; on the other hand, it requires more complex and longer post-processing of data (Table II). Autodesk 123D Catch is quite invariant with respect to the distance, but it extensively fills lacks by automatically interpolating the acquired data. Photoscan models can therefore represent the best solution, as it mediates all these different aspects. Besides these quality and metric evaluations, it is important to take other factors into consideration in order to evaluate the best solution to adopt. Some of these are related to peculiarities of single packages, such as, for example, automation, image processing time, level of complexity of restored data.
Fig. 6. Orthogonal projection and horizontal section (A) of the façade. Right, close up of A-A’ section.
TABLE II. TABLE OF TIME/MODEL RATIO LASER SCANNING ARTIFACT
Façade
Survey 1.5
Processing 14
(man-hours) Texture mapping 3.5
TOTAL 19
STRUCTURE FROM MOTION Survey
Processing
ARTIFACT
Visual SFM Façade Pilaster Cornice Medallion
Fig. 7. Comparison between models obtained using different reverse modeling and dense matching techniques, with relative chromatic information.
158
1 0.5 0.5 0.5
2.5 1.5 1.5 1.5
(man-hours) Processin g + Texture mapping Autodesk 123D Catch 1.5 0.5 0.5 0.5
Processing + Texture mapping TOTAL Agisoft PhotoScan
average)
2.5 1.5 1.5 1.5
3.17 1.67 1.67 1.67
DATA PROCESS
SURVEY
TABLE III. TABLE OF MAIN ASPECTS AFFECTING 3D SCANNING AND SFM Factors of influence
3D Scanning
SFM
Instruments portability
low
high
Instruments price (hardware)
high
low
Dependence on environmental factors
low
high
Survey timing (man/hours)
low
low
Operators’ skills
medium
low
Instruments price (software)
high
low
Data process timing (man/hours)
high
low
Operators’ skills
high
low
VI. CONCLUSIONS AND FUTURE WORKS The presented research was undertaken in order to investigate if low-cost technologies can be adopted as reliable survey tools in architectural multi-scalar contexts, where metric resolution and accuracy are under pre-defined thresholds. In order to fulfill this aim, some widespread and low-cost technologies were selected and compared with range-based acquisitions. Besides metric assessments, different aspects were analyzed in order to optimize both the results and every single process of the pipeline. This contribution suggest the combination of different technologies and 3D models acquired at different resolutions, that represents an interesting solution to ease post-processing of data and limit the need to acquire and manage redundant data. In case of artifacts with complex sculptural characteristics that have to be documented at an architectural scale, the SfM
Other aspects are instead related with 3D modeling process, such as, for example, mesh optimization, remote processing, presence of lacks, incoherencies, times and costs (Tables II, III). Taking all these aspects into consideration, 123D Catch represents the best solution for an extensive application on the whole façade V. INTEGRATION RESULTS Two different integration procedures were performed, involving the use of both active and passive techniques. The integration between 3D laser scanning, that was used to acquire the general architectonical framework of our case study and the selected SfM approach (Autodesk 123D Catch) aimed at surveying small-scale elements represents an interesting solution for several reasons. Generally 3D acquisitions of complex architectural façades with a ToF technology can be planned following two different approaches. On one side, it is possible to perform very detailed information and acquire huge point clouds, from which detailed ornaments and sculptures can be extracted. Following another approach, a global and low-resolution acquisition can be performed and can be afterwards integrated with close-ups surveys of each decorative element. In both cases, long times and expertise are required, in particular as far as survey campaign is concerned and in the post-processing of data. Within our research, we evaluated that an integration of 3D laser scanning and SfM can reduce both the acquisition and the post processing times and can, at the same time, allow to restore an optimized mesh that preserves the pre-defined quality of the acquired data. The integrated model presented in Fig. 8a shows that multi-resolution modeling is also strictly related to the visualization distance, since different levels of detail can be appreciated only from close distances. Integration of 3D models acquired using the SfM approach provides an interesting solution, as it can avoid the use of laser scanner and therefore sensibly reduce costs and times (Fig. 8b). In this case, particular attention has to be paid to transition areas between low and high-resolution portions, since sharp edges tent to be rather smoothed in the less detailed one.
Fig. 8. a) Points and shaded visualization of the model of façade portion obtained with 3D laser scanner data and SfM integration; b) integration close-up using 123D Catch at different resolutions.
159
approach can be considered a useful and reliable solution when the acquisition process is carefully planned. In particular, much attention should be paid to illumination conditions and to the selection of the most appropriated resolution with respect to distance and to the level of detail of representation. Future advancements can be singled out in the further optimization of 3D models aimed at their simplification in regular and almost flat areas. Other investigations should also be held in order to develop a methodology to easily texture integrated models using low-cost technologies.
[15] G. Guidi, F. Remondino, M. Russo, F. Menna, A. Rizzi, and S. Ercoli, “A multi-resolution methodology for the 3D modeling of large and complex archeological areas”, in Int. Journal of Architectural Computing (IJAC), special issue, 2009, pp 39-55. [16] M. Callieri, A. Chica, M. Dellepiane, I. Besora, M. Corsini, J. Moyés, G. Ranzuglia, R. Scopigno, and P. Brunet, “Multiscale acquisition and presentation of very large artifacts: The case of Portalada”, in Journal on Computing and Cultural Heritage (JOCCH), 3(4), 2011, pp.1-14. [17] S. Ullman, S., “The interpretation of Structure from Motion”, Proc. R. Soc. Lond. B, vol. 203 (1153), 1979, pp. 405-426. [18] A. Gruen and T.S. Huang, Calibration and Orientation of Cameras in Computer Vision, Springer, Berlin/Heidelberg, 2001. [19] F. Remondino and C. Ressl, “Overview and experiences in automated markerless image orientation”, in Int. Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, 36(3), 2006, pp. 248-254. [20] L. Barazzetti, M. Scaioni, and F. Remondino, “Orientation and 3D modeling from markerless terrestrial images: combining accuracy with automation”, in The Photogrammetric Record, 25(132), 2010, pp. 356-381. [21] S. Del Pizzo and S. Troisi, “Automatic orientation of image sequences in cultural heritage”, in Int. Archives of Photogrammetry, Remote Sensing and Spatial Information Science, 38(5/W16), 2011, pp.293-300. [22] C. Wu, S. Agarwal, B. Curless, and S.M. Seitz, “Multicore Bundle Adjustment”, Proc. CVPR (2011), 2011. [23] M. Pollefeys, R. Koch, M. Vergauwen, and L. Van Gool, “Metric 3D Surface Reconstruction from Uncalibrated Image Sequences”, 3D Structure from Multiple Images of Large-Scale Environments ,vol. 1506, 1998, pp. 139-154. [24] N. Snavely, S.M. Seitz, and R. Szeliski, “Photo tourism: exploring photo collections in 3D”, SIGGRAPH’06, 2006, pp. 835-846. [25] S. Agarwal, N. Snavely, I. Simon, S.M. Seitz, and R. Szeliski, “Building Rome in a day”, Proc. of Int. Conference on Computer Vision, Kyoto, Japan, 2009, 8 pages. [26] M. Vergauwen and L.V. Gool, “ Web-based 3D reconstruction service”, in Mach. Vis. Appl., 17(6), 2006, pp. 411-426. [27] A.M. Manferdini, “The Use of Low-Cost Technologies for the Promotion of Cultural Heritage Sites. The Case Study of Veleia”, Proc. of the VSMM 2012, vol. CFP12137-USB, 2012, pp. 483-490. [28] S. Hermon, D. Pilides, N. Amico, A. Dšandrea, G. Iannone, and M. Chamberlain, “Arc3D and 3D laser scanning a comparison of two alternate technologies for 3D data acquisition”, Proc. of CAA 2010, 2010, pp.55-58. [29] D. Abate, G. Furini, S. Migliori, and S. Pierattini, “Project Photofly: New 3D Modeling Online Web Service (Case Studies and Assessments)”, 3D-ARCH 2011. Proc. of the 4th ISPRS Int. Workshop, 2011, pp. 391-396. [30] M. Dellepiane, N. Dell’Unto, M. Callieri, S. Lindgren, and R. Scopigno, “Archeological excavation monitoring using dense stereo matching techniques”, in Journal of Cultural Heritage, in press, pp. 1-10. [31] A.M. Manferdini and M. Galassi, “Assessments for 3d reconstructions of cultural heritage using digital technologies” Int. Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences», XL-5/W1, 2013, pp. 167-174. [32] L. Bortolotti, Bologna dentro le mura. Nella storia e nell'arte, Bologna, La grafica emiliana, 1977, pp. 51.
REFERENCES [1] F. Blais, “Review of 20 years of range sensor development”, in Journal of Electronic Imaging, 13(1), 2004, pp. 231-243. [2] G. Guidi, B. Frischer, M. Russo, A. Spinetti, L. Carosso, and L.L. Micoli, “Three-dimensional acquisition of large and detailed cultural heritage objects”, in Machine Vision and Applications, vol. 17-6, 2006, pp. 349-360. [3] G. Guidi, M. Russo, and J-A. Beraldin, Acquisizione e modellazione poligonale, Ed. McGraw Hill, 2010. [4] G. Vosselman and H.-G. Maas, Airborne and Terrestrial Laser Scanning, Whittles Publishing, 2010. [5] M. Russo, F. Remondino, and G. Guidi, “Principali tecniche e strumenti per il rilievo tridimensionale in ambito archeologico”, in Archeologia e Calcolatori, 22, 2011, pp. 169-198. [6] A. Grün, F. Remondino, and L. Zhang, “Photogrammetric Reconstruction of the Great Buddha of Bamiyan, Afghanistan”, in The Photogrammetric Record, 19(107), 2004, pp. 177-199. [7] F. Remondino and S. El-Hakim, “Image-based 3D modelling: a review”, in The Photogrammetric Record, 21(115), 2006, pp. 269-291. [8] A.M. Manferdini and F. Remondino, “Modellazione 3d da immagini. Pipeline fotogrammetrica”, Modelli digitali 3D in archeologia: il caso di Pompei, Edizioni della Normale, Pisa, 2010, p. 162-199. [9] A. Velios and J.P. Harrison, “Laser Scanning and digital close range photogrammetry for capturing 3D archeological objects: a comparison of quality and practicality”, G. Burenhult (ed.) Archeological Informatics: Pushing the envelope, CAA 2001, Oxford, 2002, 205-211. [10] G. Guidi, G. Tucci, J-A. Beraldin, S. Ciofi, V. Damato, D. Ostuni, F. Costantino, and S.F. El Hakim, “Multiscale archaeological survey based on the integration of 3D scanning and photogrammetry”, CIPA WG 6 Int. Workshop on Scanning for Cultural Heritage Recording, Korfu, Greece, 2002, pp.58-64. [11] W. Böhler and A. Marbs, “3D Scanning and Photogrammetry for Heritage Recording: A Comparison”, Proc. of the 12th International Conference on Geoinformatics, Gävle, Sweden, 2004, pp. 291-298. [12] S.F. El-Hakim, J.A. Beraldin, M. Picard, and G. Godin, “ Detailed 3D reconstruction of large-scale heritage sites with integrated Techniques”, IEEE Computer Graphics and Application, 24(3), 2004, pp. 21-29. [13] F. Remondino, “Heritage Recording and 3D Modeling with Photogrammetry and 3D Scanning”, in Remote Sensing, 3, 2011, pp. 1104-1138. [14] A.M. Manferdini and F. Remondino, “A Review of RealityBased 3D Model Generation, Segmentation and Web-Based Visualization Methods”, in Int. Journal of Heritage in the Digital Era, vol. 1, 2012, p. 103-124.
160
