VALIDATION OF LOW-COST SURVEY TECHNIQUES

The Monumento a la Tolerancia is an emblematic sculpture realized by Eduardo Chillida, a perfect example for the documentation of sculptures with uniform textures, non-reflective colours and with poorly elaborated shapes through the implementation of various photogrammetric tools, as well as using different applications for processing phase. The basic data are photos taken quickly and without an accurate previous study, which is why the implementation of any target was not foreseen. In order to prove the results different kind of analysis were conducted. The first type was carried out analysing the results obtained from different software, with the use of the same instrument. The second one was a comparison of the models obtained with different tools, elaborated with the same software and the last one a validation of photogrammetric model with the laser scanner one. The complete results will be presented and critically discussed to validate these tools and applications.


INTRODUCTION
Architectural survey with the implementation of digital technologies represents today one of the main methods for as-built graphic reproduction, both in the architectural (Molero et al. 2013: 673), archaeological (Fioretti et al. 2020: 2) and sculptural field (Loaiza Carvajal et al. 2020: 234). The possibility of reaching rigorous measurements and the massive nature of the technology, in fact, allow interrogating repeatedly the documentation and to obtain technical elaborations with different degrees of resolution. Based on previous studies (Morena et al. 2019: 135) and SfM comparative analysis (Rodríguez-Navarro 2012: 100;Gil-Piqueras et al. 2019: 645), this contribution aims to analyse not only the opportunities that the new low-cost technologies guarantee in the field of sculptural heritage (with homogeneous textures, simple geometric shapes and non-reflective materials), but also comparative evaluations of surveys of the same object when the variables are not only the instruments but also the times and above all the operators. The progress achieved, both hardware and software, in the field of surveying makes it increasingly easy to generate models even from acquisitions that are not perfectly pondered.
The possibility of implementing photos acquired through technologies that are now widely used, compact cameras and smartphones (Shults 2017: 481), as well as the availability of lowcost and user-friendly software, in fact, allows pursuing sufficiently accurate results. This ensures greater digitization of the heritage (Tabone 2019: 48), even if, only for divulgation purposes: web visualization, virtual tours, as well as innovative ways to catalogue and store documents in eminently graphic, technical and architectural history databases. The aim of this paper is the analysis and comparison of threedimensional models of the Monumento a la Tolerancia ( Figure 1). The attention is focused on this monument following the collection of many survey experiences conducted by different people (even non-experts) obtained during the years with various instruments (FARO focus 3D X330, Nikon COOLPIX L11 and Apple iPhone 11) and processed with some software (SCENE,PhotoModeller,Metashape,eyescloud3D,3DF Zephyr and ReCap Photo). To prove the accuracy of these data, several analyses have been carried out, the first one to evaluate the difference pursued by the various software in the processing of data coming from the same instrumentation.
A second analysis was carried out, instead, comparing the best results obtained with the two different instruments processed with the same software. Finally, the need to practically prove commercial and usual technologies for the restitution of models inevitably involved the comparison of the data acquired with a laser scanner characterized by high precision and accuracy ( Figure 2). Although the results of this investigation are achieved with scientific rigor, the data of this research are presented here with a predominantly informative intent.

HISTORICAL INTRODUCTION
The Monumento a la Tolerancia is an emblematic sculpture located in Seville and signed by the author Eduardo Chillida (San Sebastián, 1924-2002. It was inaugurated in 1992 at the Universal Exhibition in Seville in memory of the Edict of Granada, with which the Jews were expelled from the Catholic monarchs. The work, financed by the Fundación Amigos de Sefarad, is located on the "Muelle de la Sal", Paseo del Alcalde Marqués del Contadero, next to the Puente de Isabel II, popularly known as Puente de Triana (Rabe 2014: 137). Made entirely of reinforced concrete, it rises to a total height of about 4 m and extends 12 m wide. Opposing the violent and suffering past, it presents itself as a great embrace between different cultures, to encourage understanding and tolerance (Matos 2015: 99). Despite its apparent simplicity, seen from different perspectives, it always assumes different configurations, once the left arm protrudes, another one prevails the right one, seemingly every portion of the work is "always different but never the same" (Chillida 2016: 103). The absence of visual uniformity of the work is the very emblem of tolerance, there is no single point of view, only respect for diversity that can coexist together. Despite its artistic and symbolic importance, following several acts of vandalism, the sculpture had an extremely degraded appearance, which is why it was necessary to clean and consolidate it. This, however, was not enough to limit further acts of vandalism that replicated themselves on the structure, underlining even more the need to document, detect and digitize Fig. 2. Correspondence view based on point clouds laser acquired with FARO Focus 3D X330. Source: own elaboration 2020. our heritage to preserve, divulge and catalogue it by generating an archive of data available for any eventuality.

DATA ACQUISITION AND PROCESSING
The data acquired by different operators and in different time were reached by three different instruments and processed with various software, described below (the information contained is mainly taken from the respective manuals). It is also essential to underline that some acquisitions were carried out by inexperienced personnel and only for documentary purposes. The choice of photogrammetric programs for data processing fell on both professional and user-friendly applications where the interaction with the user during the process is strongly limited. In particular, the software implemented are Metashape by Agisoft and 3DF Zephyr by the Italian 3DFLOW. They are applications that operate in Structure for Motion (SfM) for the pseudo-automatic processing of three-dimensional photo-realistic models. In the second case, the choice fell on eyescloud3D and ReCap Photo. The first consists of an app connected to the cloud, accessible via both mobile phone and computer, of the Spanish eCapture3D. It automatically proceeds at the generation of point cloud and mesh, without the need of high-performance computers, and, at the end, it provides to display the model on a networked platform. With a very simple interface, it also allows different iterations of measurement and editing, as well as the possibility to geo-reference and download the model. ReCap Photo is a service included with Autodesk's ReCap Pro, which allows processing close-range or drone photos for the generation of point cloud, mesh and orthophoto. With a simple graphical interface, also in this case the process of the data is performed on the internet, allowing data editing once the model has been processed. However, given the characteristics of the monument and the close-up shots, the best results were obtained by operating in object mode, which automatically generates a textured mesh. The workflow with the use of Metashape and 3DF Zephyr software is mainly based on five consecutive steps: cameras orientation and images alignment with the generation of a sparse cloud, detection of control points (extracted from the laser scanner model) to locate and scale the model, dense cloud construction, mesh generation and model texturization. Differently to the previous cases, with the eyescloud3D and ReCap Photo applications, we proceeded to select the photos and start the process and, only after the generation of the model, we could scale and export it. At general level, the main difficulties encountered were: the alignment of the cameras, despite a sufficient overlap of about 70% with both COOLPIX L11 and iPhone 11, and the absence of targets related to a monument characterized by homogeneous texture and simple geometric shape. The selected Control Points were extracted from the laser scanner model. The coordinates of 7 points were identified to locate and scale the photogrammetric models ( Figure 3). Finally, the absence of a photographic pole or a UAV flight made impossible to return the upper part of the model, which is why all the processed data have gaps in the upper part. Tables visible in the figures 4 and 5, schematize the main data processing obtained during the alignment phase, dense cloud and mesh generation respectively.

FARO FOCUS 3D X330
Phase difference laser produced by the CAM2 company, is characterized by its weight (only 5 kg) and compactness, properties that guarantee an easy management of the device facilitating its movement between the various positions. The field of view covers an opening of 360° horizontally and 300° vertically, with an acquisition depth from 0.6 m to 330 m (Gollob et al. 2019: 4). The photographic sensor, which allows colour acquisition during scanning, is integrated in the system and allows further limitation of errors during data processing. To facilitate and speed up the acquisition of scans, realized in 2015, a first 360° pre-acquisition has been carried out for each station, with a distance between points of 15.35 mm at 10 m (1/10 resolution and 4× quality). Only afterwards, a higher resolution acquisition (1/4 with the same 4× quality), with a points distance of 6.14 mm at 10 m and with the addition of colorimetric information, was carried out. Data management and processing were conduct in 2020 with the proprietary software of the scanner, SCENE (version 2019.0.0.1457). The information collected directly from the instrument during acquisition (altimeter, compass, inclinometer, and GPS) guaranteed easier management of the 7 scans used to generate the reference point cloud. At a first manual recording, through the identification of homologous elements in the scans, an automatic cloud-to-cloud recording followed with the aim of making improvements in the final global spatial registration of the model. The generated cloud has about 58,000,000 points, with an average point error  of about 8 mm and a 20% overlap of scans. Finally, through a Clippinbox, the area of interest has been selected to export it for comparison on third-party software.

COOLPIX L11
Compact Nikon camera featuring 1/2.5" CCD sensor of 6.2 MP, a variable focal length between 6.2-18.6 mm and a focal aperture of f/2.8-5.2. Its simplicity of operation and its particular sensitivity to light with an ISO 800 make it easy to use even for non-professional people. The photographic acquisitions made on site in the 2008 were made only for documentation purposes and, therefore, without the use of targets. The inspection concluded with the acquisition of 77 photographs of the sculpture from different angles. The Metashape processing, carried out by setting the Highest accuracy (maximum resolution of the photos) and a key point limit of 40,000, ended with the alignment of 74/77 photos. A dense cloud, consisting of about 42,000,000 points and a mesh model of 8,000,000 polygons, was then created. In the case of the 3DF Zephyr application the best results were achieved by setting, during the orientation phase of the cameras, key point limits of 7,500 and a 100% resolution of the images guaranteeing an alignment of 70/77. The results obtained are a dense cloud of about 2,000,000 points and a polygonal mesh of as many faces, characterized by several gaps especially near the left arm.
The data processing through the eyescloud3D application gave better results with the selection of only 50/77 photos, generating a cloud of about 1,500,000 and a mesh of 170,000 polygons, with the presence of various gaps scattered in the model. Finally, the data was also processed with ReCap Photo, which allowed the generation of a mesh model of about 870,000 faces but with various deformations especially near the arms.
A further software, implemented in 2008, to manage this data was PhotoModeler v.6. In this case, the calibration of the camera was carried out successively to the survey: it consists in a process where it is necessary to take photos (maintaining fixed the parameters) of a grid with a series of points with theoretical coordinates provided by the software. The comparison of these points with the reality captured by our camera allows us to define the degree to which the field images are deformed by the lens, for subsequent correction. Successively, we proceed in the generation of the three-dimensional scene with the detection of homologous points between the images, by which the positions of the cameras in the space are calculated now of making the shot. This is an arduous process which Structure from Motion systems (also available in newer Photomodeler releases) have improved. Once all the necessary images have been oriented, the model is constructed on which, subsequently, the surfaces representing the monument are defined.
Finally, the textures are projected on the surfaces, ensuring that they come from the images in a more orthogonal situation, and without obstacles, for a better photorealistic quality.

IPHONE 11
Smartphone designed and manufactured by Apple characterized by a 1/2.55" CMOS sensor and a double camera of 12 MP, one wide angle and one ultra-wide-angle. Very popular in recent years, it guarantees the production of good quality photos even by inexperienced users. The inspection, once again carried out without a properly significant purpose and in the absence of targets, it was conducted in the 2020 and ended with the realization of 83 photos.
The parameters used in Metashape, very similar to the previous ones, with maximum accuracy and resolution of the photos during the orientation phase, allowed to obtain an alignment of 82/83 photos. Despite the higher number of acquisitions, the generated cloud has a number of points equal to 38,000,000 and a mesh of about 8,300,000 faces similar to the model with the Nikon. In the same way of 3DF Zephyr, the best results were obtained by setting key point limit values at 7,500 and maximum photo resolution, ensuring an alignment of 82/83 photos, a cloud model with 1,050,000 and a mesh of about 12,000,000 faces. Using 83/83 photos it was possible to proceed with the creation of the model also in eyescloud3D, a cloud of about 2,000,000 and a mesh of 405,000 polygons was generated. Finally, we use ReCap Photo with the generation of a mesh consisting of about 1,000,000 polygons and deformations evident in various parts of the model.

COMPARISON
The processed data were compared to validate what was attained. A first analysis was carried out by verifying the best results achieved using the same tool and processing the data with different software. Subsequently, based on what had been obtained, the main differences between two of the best models realized with COOLPIX L11 and iPhone 11 were evaluated and finally everything was validated by comparing the preferable model with the one detected by laser scanner. However, while in the case of professional applications such as Metashape and 3DF Zephyr the comparison can be based on statistical data extracted and processed directly by the software during the elaboration time, the same cannot be conducted in the case of the remaining two applications. To make further comparisons, the models were exported in universal format and compared in a third-party software, CloudCompare V.2.10.2. For a comparison between dense clouds, "Cloud-to-Cloud Distance computation" (C2C) was used. It is the default way to compute distance between dense clouds in the software: for each point of the compared cloud, it searches the nearest point in the reference cloud and computes their (Euclidean) distance. In the case of point cloud and mesh comparison, necessarily we should have proceeded with "Cloudto-Mesh Distance computation" (C2M) in which for each point of the compared cloud, the software should identify the closest triangle in the reference mesh. However, operating in this way, the software will automatically identify the mesh as the reference element. To invert the models and consider the cloud generated by Metashape as a reference, it was necessary to first sample the mesh and, only then, compare it.

SOFTWARE COMPARISON
Tables ( Figure 6 and 7) show, for both COOLPIX L11 and iPhone 11, the re-projection errors of the x, y, z coordinates for each Control Point used in   figure 3). As can be seen, in the case of Nikon tool, the 3DF Zephyr software has slightly higher deviation values than Metashape, reaching a medium value of error of about 1.50 cm respect to 1.38 cm of Metashape, with maximum value of 3.0 cm for the point 3.
More difference, instead, are visible in the case of iPhone 11 instrument with the point 2 that reach an error of about 5.00 cm in the case of 3DF Zephyr software.
A further comparison, as previously specified, has been carried out within the CloudCompare software to evaluate the existing deviations between the model surfaces (Barba et al. 2019: 149). The Metashape point cloud achieved the best results and was taken as a reference for comparison with the remaining models extracted from the software. In the figures 8, 9 and 10 it is possible to visualize the result of the above analysis. The column on the right of the image represents the range of the existing deviation between points belonging to two different models which are, in fact, represented in false colours. The maximum deviations (in red) reach values equal to 10 cm and are visible in the first comparison only in a portion of the left arm.
Satisfactory, are also those obtained with eyescloud3D software, instead, obvious    deviations, are mainly shown in the model returned with ReCap Photo. The results of the point-topoint comparison of the models generated by the different software used to process the COOLPIX L11 data have been schematically shown in the graphs where on the x-axis are the deviation values in m and on the y-axis the number of points to which this value corresponds ( Figure  11 and 12).
From what we can see, the best results were achieved in comparison with 3DF Zephyr which has the lowest standard deviation value (1.3 cm), however, good results were also achieved in the case of eyescloud3D with standard deviation values close to 1.8 cm as well as the return of a complete model without significant gaps. Less satisfactory are those generated by ReCap Photo with a standard deviation reaching a value of about 3 cm.
The peak of the curve is close to 0.15 cm for comparison with 3DF Zepyr and the analysis of the graphic show as the 90% of the point have a deviation less than 2 cm. In the case of eyescloud3D, most points have values of deviation of 0.80 cm with around 70% of with deviation less than 2 cm. Further comparisons have been made with the iPhone 11. From the analysis of the figure 13, we can be seen that the comparison of the Metashape and 3DF Zephyr models has more deviations than the previous comparison with the Nikon. Values of about 10 cm occur mainly near the left arm, they are confirmed by an incorrect alignment in 3DF Zephyr model.
Better results, in this second case, came from the eyescloud3D application, which allowed to generate a dense cloud with slight deviations from Metashape, reaching a standard deviation value of about 1.4 cm ( Figure 14). Once again, the distribution is schematically shown in the graphic visible in the figure 15, with the highest frequencies near the value of 0.15 cm and around 80% of points with deviation less than 2 cm. Otherwise, ReCap Photo has returned a model with a strong deviation from the reference one, not only on the arm, but in the general model, reaching in some points significant differences more than 10 cm ( Figure 16).

INSTRUMENTS COMPARISON
As previously mentioned, a further comparison was made between the results obtained  with Metashape from the data acquired with two different COOLPIX L11 and iPhone 11 instruments. The analysis of the re-projection error for the coordinates of the control points, visible in the previous tables in the figure 6 and 7 shows a higher value for the data processed with the smartphone, as well as a cloud of points and a mesh model with some gaps. The study of the deviation between the two surfaces was conducted within the CloudCompere software, operating with the C2C algorithm. From what we can see in the figure 17, the two models have a quite similar distribution. The standard deviation that can be extracted from the comparison is equal to a value of 1.7 cm and the distribution graph reaching the highest point near the value of 0.40 cm. These results, therefore, allow to confirm the similarity existing between the two models, in fact more than 90% of points have a distance less than 3 cm except for some present mainly on the arms and on the edges of the monument.

MODEL VALIDATION
In order to prove the results obtained, a further comparison was made by comparing the cloud extracted from Metashape made from the data acquired with the COOLPIX L11 and those obtained from the FARO Focus 3D X330 laser. The choice fell on the Nikon model for the best results, both metric and geometrically, presenting a complete cloud without significant gaps. The analysis has been carried out again with the C2C algorithm of the third-party software CloudCompare. The results pursued can be seen in figure 18. Surface deviations with values close to 10 cm are almost absent if not in some points near the edges, mainly due to cloud noise. The standard deviation pursued in this comparison, in fact, reaches values close to 1.3 cm with 90% of the points having a value of less than 3 cm. The graphic present the highest intensity of points with a deviation of about 0.35 cm.

CONCLUSION
Before proceeding with the conclusions, it is necessary to highlight and specify more important differences between the technologies used. First, the different tools in the data acquisition phase. In the case of photogrammetric surveys, we opted for two economic instruments that are easy to use and widely available on the market. These characteristics also allow nonspecialist operators to access equipment that offers satisfactory results for photogrammetric activities, without great expense; obviously, considering the objectives of each project.
The laser scanner, on the other hand, is a technological and specialized tool with a much higher price than the previous ones. It can acquire dense point clouds (976,000 points per second, according to FARO specifications) that makes its MORENA, S.; MOLERO ALONSO, B.; BARRERA-VERA, J.A.; BARBA, S. manipulation, interpretation, and management difficult. Another clarification to be considered is the characteristic of the sculpture that presents a homogeneous texture and simple geometric shape. Such peculiarities, in fact, make difficult for photogrammetric software to align and orient the images, even more if the survey was conducted by an inexperienced user without the presence of targets. Finally, in the analysis carried out we decided to implement two software that operate in cloud where the management of parameters is minimal if not nothing at all, consequently manageable by an increasingly wide audience.
By critically analysing the data pursued it is possible to see how, despite the implementation of low-cost and commercial technologies, the comparison for some models (Metashape obtained with Nikon photos vs Scene FARO) has generated satisfactory differences in the point cloud. The same can be said with the point cloud generated by eyescloud3D for the data from iPhone 11. The two models not only look complete and without gaps, but also have acceptable deviations from the reference point clouds. In the case of the remaining models, it is possible to underline the presence of greater errors but still acceptable as data to use for the generation of models for viewing via web or for 3D prints (Figure 19), above all given their greater lightness and manageability. Following the great progress achieved in the photogrammetric field, in fact, it is possible to achieve increasingly satisfactory results, often compensating for instrumental deficits or the inexperience of operators. On a general level, it is possible to appreciate a high level of phaseshift between the models in the areas of the monument's arms due certainly to the lack of total station control points in the photogrammetric survey. It is therefore an aspect of easy solution, but this decision is the result of the objective of comparing cheap, wide diffusion tools and easy to use by non-expert people. The three-dimensional model of the laser scanner, as expected, provides accurate metric information and a high degree of surface detail. This model can be preserved as a true digital copy of the real monument. In short, it will be a valuable database from which to analyse or carry out further studies in the office. It can be inferred, from what was previously exhibited, how the potential of these technologies, even if low-cost, shows a promising present and future for culture and heritage conservation. In recent years, the potential of photogrammetry together with advances in SfM computational visualization and technology affordable for many, provide us to generate satisfactory results in this area, always related to the objectives to be pursued, taking into account that without using control points and professional equipment and procedures the results are not suitable for metric or scientific purposes.
The most complex phase of SfM photogrammetry is to take photos: anyway, in the work we wanted to evaluate photographs taken by nonprofessionals and thus to analyze the presence of gaps, aberrations, etc. Likewise, to better take into account the economy and simplicity of SfM photogrammetry, we worked with consumer software (a possible development would be the use of other free software) and cameras with different quality lenses. The deviations shown in the article -many inadmissible -are referred to a small architectural element and do not allow generalization of the results in the case of a large construction: where, on the other hand, the survey would be carried out only by professionals. In conclusion, it is possible to observe that the use by non professionals can bring low quality results that, although, they can be used for consumption, nothing to do with professional, scientific surveys, where the operator handles many more complex factors that will guarantee the quality, rigor and fidelity of the models obtained. All of this forces us to emphasize the contribution of engineers and/or architects, also in the use of automated resources. Finally, we can argue that we do not do survey with SfM photogrammetry, which is an instrument, but with the head, with the brain, with the mind. , esta contribución tiene como objetivo analizar no solo las oportunidades que garantizan las nuevas tecnologías de bajo costo en el campo del patrimonio escultórico (con texturas homogéneas, formas geométricas simples y materiales no reflectantes), sino también evaluaciones comparativas de levantamientos de un mismo objeto en función de distintas variables. Tanto tomando como variables los instrumentos, como los tiempos y, sobre todo, los operadores. Los avances logrados, tanto en hardware como en software, en el campo de la topografía hacen que sea cada vez más fácil generar modelos incluso a partir de trabajos de campo que no están debidamente planificados con un propósito fotogramétrico profesional.
Satisfactorios son también los resultados obtenidos con el software eyescloud3D. En cambio, desviaciones obvias se muestran principalmente en el modelo conseguido con ReCap Photo. Los resultados de la comparación punto a punto de los modelos generados por los diferentes programas utilizados para procesar los datos de la COOLPIX L11 se han mostrado esquemáticamente en los gráficos donde en el eje X están los valores de desviación en m, y en el eje Y, el número de puntos a los que corresponde este valor (Figura 11 y 12).