Integrated Information System for 3D Interactive Reconstruction of an Archaeological Site

Documenting is one of the most important tasks both during archaeological excavation and once the finds have been collected and are under study in the lab. In any case, these meticulous processes have to be paired with an accurate methodology to record all the relevant data about the finds, gathering information such as the material they are made of or their dating. However, there are additional aspects, such as their position on the site or even the arrangement of the pieces on a given location, that can be also crucial for further analysis. Starting from the premise that excavation is destruction, the ever-present objective of recording is preventing any loss of information that might be missed in the future.

Traditional recording methods, mainly manual, have been gradually enriched with digital tools that allow feeding an archaeological database. The nature of the data provided by these tools is often dissimilar. The textual information of the records is frequently complemented with digital pictures, handmade sketches, and so on. Although these two-dimensional (2D) inputs are very enriching, they imply a level of abstraction by representing a three-dimensional reality in a two-dimensional way.

This drawback has been partly solved nowadays by means of techniques based on three-dimensional (3D) capturing. Some of them are based on photogrammetry, laser-scanning, or light detection and ranging (LiDAR). Three-dimensional modeling offers numerous advantages, mainly in terms of visualization but also for collaborative analysis and data exchange. These technologies can also solve the lack of geometric accuracy that planes and sections might have, or the morphology correctness of the pieces. Definitely, 3D documentation is able to reflect the features of archaeological vestiges in a more natural way.

However, fully integrating 3D information with the records of a traditional database is still challenging. 3D recording must go beyond mere three-dimensional visualization. The shape, appearance, and position of the virtual objects in the resulting scene should reproduce as closely as possible the reality they are representing. This entails a new approach that generates advanced knowledge compared to classic recording methods.

We propose in this article a methodology for 3D documenting archaeological sites that additionally allows generating a 3D model of the site before completing the extraction of the pieces. In short, we aim at (1) re-creating the moment of the discovery for experts who were not present at the time and place of the discovery, (2) preserve some informational elements that are lost after the excavation, namely, the spatial layout of the finds, and (3) allowing experts to observe and access the virtual re-creation of the site from anywhere.

Our proposal provides the necessary mechanisms to establish an accurate correlation between real finds or structural elements with their virtual representation. The result is a synthetic scene with 3D models of the vestiges placed at their original position and orientation in which the soil has been removed. For that purpose, we first acquire a 3D model of the archaeological site during the excavation and before the finds are removed. Different techniques for this 3D capture have been tested. Then, the pieces are cleaned and processed with a 3D scanner. Finally, the resulting models are placed at their original locations using the model of the archaeological site as a reference. For that purpose, the user is provided with an interface to manipulate the virtual elements, allowing them to translate and rotate each model. The interactive tool is supported by a database that allows to store and retrieve information of different nature. In addition, the information in the database is updated according to the changes made to the 3D scene by the users with the interactive tool. The system architecture allows final users to visit the virtual reconstruction of the archaeological site from a computer connected to the Internet by means of an application implemented in the Unity game engine. The result is an intuitive and user-friendly interface not only for archaeological recording purposes but also for consultation and analysis.

The rest of the article is structured as follows. Section 2 reviews the state of the art. Section 3 describes the main features of our proposal, including the system architecture. Section 4 details how the data is acquired and processed before being stored in the spatial database. Section 5 describes the process to virtually reconstruct the archaeological site as well as the spatial analysis that can be performed on the virtual site. Section 6 discusses some results, and Section 7 presents our conclusions and plans for future work.

Archaeological documentation techniques are currently evolving toward a multi-faceted approach. The primary objective is to save for posterity all the knowledge that disappears when the finds are extracted during excavation [25]. For that purpose, traditional archaeological recording methods are being improved by adding some of the following features:

In order not to destroy the context, all the features described above are desirable to re-create the original archaeological site. This way, the assertion “excavation is destruction” is replaced by “excavation is digital reconstruction.” Thus, an expert who did not participate in the excavation could reliably know the details of the archaeological site as if they were there.

Nowadays many possibilities to capture the details of sites and finds in 3D are at our disposal. Laser scanners [26], LiDAR [32], or structure from motion (SfM) photogrammetry [17] are widely used techniques in Archaeology [2, 18]. Their capabilities can be used to observe the features of a piece to determine its material, its dating, the presence of possible inscriptions or for examining in detail its deterioration. However, one additional aim is also capturing the underlying spatial information in an archaeological site. The original position of the pieces and their disposition regarding the surrounding environment, including other vestiges, is necessary for virtually reproducing the site [16]. For example, obtaining a collection of topologically connected fragments from a site provides a good starting point for later reconstruction of pottery [1, 15]. As another example, the way in which the remains were deposited in human burial sites can provide relevant clues to understand religious customs or social relationships in a human settlement.

Compared to a 2D pictorial representation, 3D virtualization enriches the experience of archaeologists [23]. Most 3D models are thought for a better user experience in visualization, for example using Augmented Reality [5, 6, 33]. In addition, greater geometric precision is achieved when compared to the drawings [21]. Photographs shot during excavations have limitations, because they are taken from a specific angle, allowing us to visualize only one side of the objects, and those parts of them that are still buried are not displayed. In our framework, the inclusion of 3D models, both of the environment and of the pieces, allows us to overcome this drawback. In a 3D scene of an archaeological site, given the appropriate Graphical User Interface (GUI), the user can move freely, i.e., the camera can be located at any position around the scene to allow observing it from any point of view. In addition, the soil can also be removed if necessary (for example, if the user wants to view the scene from below).

However, 3D virtualization of an archaeological site still has significant shortcomings to address. On the one hand, a 3D scan of the site can produce very heavy models, with millions of points, that are difficult to handle by standard computer hardware. Frequently, the software capable of visualizing these models is decoupled from the database manager. Therefore, the 3D model is usually complementary information that lies outside the information system. Something similar happens with models that are scanned individually once they are removed from the site. They are stored outside the database, or the database just stores a link to their file location. In the few cases in which they are integrated in the system, 3D models are used only for visualization and analysis, but detached from the place of discovery [34], or conceived just for later dissemination [31]. Thus, they do not include its original position and the spatial relationships with the surrounding finds. In short, in the literature we find a dissociation between these 3D models and the rest of the information system [20]. This way, it is difficult to establish the actual spatial relationships between the pieces and their environment. This can be solved by adding topological relationships to enable this spatial analysis. Adding topological operators allows archaeologists to recover crucial information that disappears after excavation [3, 39].

One solution to add spatial information to the archaeological record comes from Geographical Information Systems (GIS), which integrate data management and visualization [8, 37]. 3D GIS outperform traditional 2D systems by incorporating spatial data with three-dimensional models under the same framework [22, 24]. There are, however, some problems underlying these tools. They provide generic solutions with a high level of abstraction, that is away from archaeological methodology [32]. 3D GIS are still in early stages of development, with limited capabilities in terms of 3D model integration, friendly interfaces or ubiquity. This is usually the reason why in many research projects, like GRAVITATE, the choice is to develop their own software tools [9].

In this article, we introduce a methodology that makes possible the virtual reproduction of an archaeological site with a wide functionality. The prototype system has been designed not only for visualization [19]. This allows improving the integration between the spatial database and the graphical interface representing the site. Our system allows reproducing the original spatial distribution of the pieces as they were found [36]. To achieve this objective, we have designed a 3D GUI that facilitates the process of placing the individual 3D scanned models at their correct positions. This process is conducted by the archaeologist using well known 3D manipulators for translating and rotating the models (scale is set in advance). For that purpose, the site ground surface is digitized during the excavation process, that is, while pieces are still embedded in the ground but partially out and their shapes are distinguishable. Given a find model, once its scale is adjusted to match the one of the site ground surface model, the user translates and rotates it to match the part of its geometry that is visible from the site ground surface model. The process is easy to perform and no experience is needed to do it, since the visual reference of the site ground surface model helps to achieve the desired precision.

Some other techniques could also be taken into consideration to perform this adjustment automatically. In particular, the Iterative Closest Point (ICP) algorithm [7] has been already applied in Archaeology [38]. This method iterates trying to approximate positions between a floating model A (the find model) and a fixed model B (the site ground surface model). In each ICP iteration, the best geometric transformation that aligns A and B is calculated. This is done by minimizing the squared error (MSE), computed as the sum of the squared distances between the points in A and the closest points to them in B. In each iteration, a combination of a translation and a rotation of the floating model is computed. However, proper results are only obtained when one model is a subset of the other [35]. Regarding our proposal, the differences between the models can be too significant for applying this algorithm, as the site ground surface model covers much more space; moreover, it only contains the parts of the finds that have been already uncovered, and these can still have remains of materials stuck; on the other hand, the find models are acquired in the lab, once the pieces have been cleaned, and the resolution of the meshes will be typically higher. Many different modifications to the algorithm have been proposed in the literature for better accuracy [27], including genetic algorithms [35]. However, all of them work with the models when they have already been extracted, not facing the problem of alignment with ground semi-embedded objects.

On the contrary, our method can be considered agile, as it gives the expert the opportunity to manipulate each object individually. Additionally, the use of mobile devices or smartphones, which are able to create 3D models of the site in a few minutes [13, 40], makes this procedure very useful in Archaeology. To achieve such an objective, we have implemented a client-server architecture prototype. On the client-side, an expert can manage and interact with the virtual finds using the user-friendly interface provided by Unity, a video-game engine. In contrast with other solutions developed for dissemination or the general public [10], not only the user interacts and manipulates the 3D models through the GUI and updates their spatial disposition on the screen but also updates the database to save the new spatial configuration. Therefore, the resulting three-dimensional interface is not a separate module of the system. Instead, it is connected to the database, allowing full interaction and feedback between both. These and similar features are considered the challenges to conform the new Digital Cultural Heritage [23].

Our system aims at providing archaeologists with a simple and intuitive interface for creating virtual reconstructions of the sites under study that can help improve the results of the archaeological analysis. To achieve this goal, the architecture of the system is as follows (Figure 1):

Even though Unity is a general purpose video game engine, its development environment (using C# as scripting language) covers all the requirements for developing a tool for archaeological site re-creation. Unity core provides standard functionality for creating virtual scenarios with 3D models that can be created, edited, or imported. It also includes lighting, animation, physics, artificial intelligence or sound. We have taken advantage of this tool to create a virtual 3D environment that (1) enables free navigation and interaction using the mouse and keyboard, (2) is also capable of importing 3D models, and (3) uses standard protocols to communicate with an external database. Additional functionality that is not directly included in Unity core was added to the system through third-party asset packages.

The diagram in Figure 2 represents how all the aforementioned elements are combined in the graphical interface of our system. Both standard and external functionality are shown. We use four asset packages to provide advanced capabilities. In particular, the package (1) Modern UI Pack is used to improve the appearance and behavior of the graphical interface, (2) the OBJ Import pack is used for loading OBJ files in runtime, (3) the Runtime Transform Guizmos (RTTG App) pack provides runtime gizmos for modeling operations (translation, rotation and scaling), as well as camera behaviors, and finally (4) the scripts from the Optimized ScrollView Adapter (OSA) pack allows optimizing the use of lists, making them more efficient.

To get the most from this system proposal, we propose the following three-stage workflow:

The virtual site model obtained allows accessing the database information through 3D interaction instead of standard SQL queries or textual or numerical data filters. This results in a much more intuitive tool that also allows browsing through the timeline of the excavation, providing a much more complete perspective of the recorded data.

As stated in Sections 1 and 2, technology for obtaining 3D models of both the archaeological site and the pieces found on it is becoming increasingly cheaper and more agile. This makes it possible to speed up the recording methodology, since it may take only a few seconds to obtain a detailed representation of the work field and the finds. However, this acquisition process alone is not so useful if its results cannot be integrated into an information system. In fact, repeating the work field scanning at regular times during the excavation process would be very desirable to feed a four-dimensional data model. This 3D representation is used not only for further visual representation but as a structural frame to virtually rebuild the site with all its finds at their original positions.

In our tests, we have carried out the acquisition process on the testing site used by Archaeology students of our university. Despite not being a real archaeological site, the students follow the extraction process in a traditional way and the methods they use can be compared with those of the methodology proposed in this article.

This section describes the interactive process and mechanisms to virtually reconstruct an archaeological site through a graphical interface in the client device. This way, the interaction with the virtual 3D objects and the environment is the natural way to access the information stored in the database.

We have tested our system prototype with real data from the archaeological site of Castulo. Castulo was an Iberian-Roman city, one of the ancient capital centers in Southern Spain. From this site, we have obtained the information of areas, volumes, and layers. However, due to a series of setbacks in the excavation processes (mainly due to the COVID-19 pandemic), we have not been able to complete our data capture as required. Therefore, we have had to simulate the work on a layer in the following way:

Tests have been carried out on a desktop computer with an Intel Core i7-4790 3.80 GHz processor with 16 GB RAM memory DDR3-2400 and a NVIDIA GeForce GTX 970 graphics card with 4 GB VRAM. According to Figure 8, each scanned model can be directly displayed in the client-side user application. However, the database only stores simplified versions of these models. Figures 15 and 16 show some of these pieces with different resolutions. The models were acquired with the lowest resolution of the EinScan Pro 2X scanner, and for database storing, the meshes were decimated with Blender to a value between 5% and 30%. However, the visual results are very similar, as can be seen in the figures.

We have measured the time taken for loading the 3D models in a layer (Table 1) considering the number of models (column 1) and the time in seconds to load the full-resolution (columns 2 and 3) and simplified models (columns 4 and 5). In conclusion, we can determine that loading more than four millions of triangles corresponding to 100 virtual models is possible but it requires more than 5 min on our test computer. In contrast, if the system loads the simplified models, then the number of triangles to be loaded is about 200,000, and the time to have access to the layer is about 1 min and 30 s.

A standard layer with between 20 and 50 full-resolution models, and over one million triangles needs about 1 min to be loaded. In contrast, the same layer only needs between 12 and 50 s when the low-resolution models are used. From the point of view of an archaeologist, a full-resolution model is not always necessary, especially when placing the models at their original locations or performing spatial analysis on the virtual site. Thus, the visualization of an entire layer can be done with the simplified models. In contrast, when a specific piece is under thorough scrutiny, the full-resolution model should be used.

In this article, we have described an agile methodology to virtually reconstruct an archaeological site preserving the spatial arrangement of the finds for posterity. This process is implemented as an information system with a client-server architecture. The client runs a Unity application on any device connected to the Internet. It takes advantage of the 3D object manipulation capabilities of this game engine, which makes it possible to easily edit the position and orientation of the find models. We use the 3D site ground surface models acquired during the excavation and before the finds are removed as reference for fine adjusting the position and orientation of the find models. This scanning process can be performed easily in a few seconds by using the ToF technology of current smartphones. With this tool, it is possible to get a visually good result in just a few seconds, and therefore create a realistic replica of the site, as shown in Figure 17.

On the server side, a spatial database stores the classical tables on an archaeological site. In particular, we designed this database based on the entities of the database currently used in the archaeological site of Castulo (Jaén, Spain). Additionally, we include as spatial information for each find its location, its topological relationships with other finds, and a low-resolution 3D scanned model. The graphical interface in the client allows the user to fine tune the location and orientation of these models, therefore updating the transformation matrix associated with the corresponding element in the database.

Given the 3D models of the finds, the system is able to infer and propose to the user a series of topological relationships, based on the spatial distribution of the models. This process is based on 2D projections of the 3D models into the orthogonal planes XY, YZ, and XZ, using the DE-9IM model. Confirming these relationships as really meaningful is up to the expert user.

The procedure can be globally considered as agile, since depending on the number of finds for a given layer, it takes only a few minutes to create the virtual reconstruction once all the pieces have been scanned. In fact, the slowest part of the process is the 3D scanning of each individual piece in the laboratory. An additional result of this process is the achievement of an integrated information system. The graphical interface is linked to the data and makes it possible to navigate and interact with the 3D scene while the information is retrieved from the database transparently to the user. To optimize the performance of the interface, the full-resolution find models are used only when the expert user requires them for a detailed study. Experts in any part of the world can execute the client application and visit the site virtually, as if they were there. The user interface also allows visualizing the finds without the ground surrounding them. The navigation process provides a perspective from any point of view, and the database is queried by simply clicking the 3D objects.

As future work, we are interested in working with real finds of a layer from the archaeological site of Castulo, since due to the special circumstances caused by the COVID-19 pandemic, we have not been able to work with a real excavation. Thus, even though the information related to areas, volumes and layers is from this site, the field work has been simulated by using the facilities used by Archaeology students for their practicum on our university campus.

In addition to working with real data, we want to conduct a thorough study with potential users of the applications (namely, researchers and undergraduate students), to collect their feedback on the use of the application both with mobile devices and desktop computers, and include their suggestions and comments in future versions of the software. This study will also include aspects related to the time necessary not only for creating the virtual model of the excavation but also for querying the system. These results will be compared to the procedures currently in use for archaeological recording.

Finally, although initially we ruled out the ICP algorithm, because it cannot be directly used to compare the 3D find models to the 3D site ground surface model, we are considering the study of different adaptations or modifications of this algorithm to improve the results. Finally, we expect that the SFCGAL extension of PostGIS, which offers advanced 3D functions and volumetric support based on CGAL (Computational Geometry Algorithms Library), provides better spatial analysis features.