May 2014 (pdf.)

CONPRA project report Secondee: Seta Štuhec (PhD student at the University of Ljubljana, Slovenia) Host Institution: Via Magna s.r.o., Slovakia Period of secondment: April ‐ May 2014 Introduction Via Magna s.r.o. (Slovakia), one of the CONPRA project partners hosted a second early stage researcher (ESR) secondment in April and May 2014 coming from the University of Ljubljana, Slovenia. The main objective of the secondment was to expand the knowledge acquired at the previous secondment (October, November 2013) on the image‐based 3D modelling used for archaeological documentation. The first part of the secondment was focused on post‐processing of previously acquired 3D data and integration of new methods into the standard workflow of archaeological excavations on Bratislava and Čachtice castle. The result of this works was presented at the conference on Computer Applications in Archaeology (CAA) 2014 that took place in April 22nd – 25th in Paris, France. In the second part of the secondment it was decided to 3D digitize an ancient tomb in Brazda near Skopje, Macedonia. Field work was carried out in May 7th – 11th. The following calculations, post‐
processing and result visualizations were carried out by the end of the secondment. Integrating 3D image‐based digitization procedures into the standard workflow of archaeological rescue excavations Currently, rescue excavations make up a very big part of all archaeological activity, especially in a private sector. Archaeologists working at such excavations face financial limitations on the one hand and the demand for the most complete and efficient execution of dig on the other. Because of its destructive nature, high‐quality archaeological documentation is of utmost importance in archaeological excavations. Given the characteristic time pressure and resulting lack of on‐site interpretations during rescue excavations, accurate and exhaustive documentation becomes even more important as post‐excavation interpretations will be entirely based upon them. Although technologies such as terrestrial laser scanning enable a fast, accurate and rather complete acquisition of archaeological data for subsequent analyses, the lack of financial means and technical knowledge of rescue archaeologists generally hamper their incorporation into the documentation workflows established over the years (which are generally based on total station measurements combined with photographs and pencil drawings). However, the recent boom of computer vision programs that combine Structure from Motion (SfM) approach with dense Multi‐View Stereo (MVS) algorithms changed this situation. These packages enable the generation of accurate three‐dimensional (3D) models from a collection of photographs in a straightforward and cost‐effective manner without the need for extensive photogrammetric and computer vision knowledge of the user not information on the geometrical properties of the scene. As such, they became a common part of the documentation practice in many research and commercial fields, including cultural heritage monitoring and preservation. At the on‐going rescue excavations on the courtyard of the Bratislava castle (Slovakia), such an image‐based modelling approach was introduced into the existing documentation workflow. Within a reconstruction project at Čachtice castle, the complete castle was documented using this method. Traditional documentation Documentation of especially rescue excavations has to be quick and efficient. Ideally, archaeologist should be able to inspect the results of measurements, photographs and drawings on‐site, which would help him/her to decide on further course of the excavations. The nature of archaeological excavation methodologies that are based on recognizing separate stratigraphic units namely 1 demands a sufficient overview over the archaeological situation of the site. Therefore, spatial data and especially from them derived plans and maps are of the most importance already during the excavation process. Furthermore, the course of rescue excavations must usually be adjusted to the new building specifications, as the new constructions are many times built simultaneously on the already excavated part of the site. This means that the archaeological units are often documented part by part not only because of the nature of gradual excavation methods, but also and because of the separation of the site into sectors, which are subordinate to the new construction plans. The traditional documentation of archaeological sites is based on drawings, photographs, measurements, description forms, lists and diaries. In order to create previously mentioned plans of archaeological situation on the site, photographs are taken from above as vertically as possible using latter or pole (if possible also unmanned aviation vehicle (UAV) can be employed). The photographs are then rectified and georeferenced using ground control points measured by a total station or a GPS device. The method normally used for this process is a 1st grade polynomial transformation (that is independent scaling of x and y axis without local rectification). The resulting rectified photographs are sufficient when the documented archaeological situation is relatively flat. However, higher height differences (for example masonries) are not sufficiently represented on such orthophotoplans. A further problem for creating an accurate orthophotoplan is also an insufficient tangential and radial lens distortion correction. Rectified photographs are afterwards vectorized. This means that individual archaeological units and structures are outlined according to the on‐site interpretation. 3D image‐based documentation Computer vision based software packages that combine Structure from Motion (SfM) and Multi View Stereo (MVS) algorithms allow a straightforward generation of accurate three‐dimensional (3D) models from a set of photographs. To generate such 3D models, specific computer vision knowledge 2 is not needed, as the process is semi‐automatic. Furthermore, data acquisition is relatively quick, as no preliminary camera calibration is needed. Using SfM algorithms, the software calculates internal (focal length, lens distortion coefficients, principal point) and external (camera position and orientation) automatically and during this process generates a sparse point cloud. Afterwards MVS algorithm is used to generate a dense point cloud. Finally the point cloud is meshed into a solid 3D model onto which a texture can be applied. In terms of archaeological excavations this documentation technique means a relatively easy‐to‐use, cost‐effective alternative to 3D laser scanning that literally brings a new dimension to the documentation archive. Integration of this approach into the standard documentation workflow is not too demanding from the acquisition point of view; as digital photography and ground control point measurements are already employed, the only thing left to do is to accordingly acquire a higher amount of photographs. However, documentation used for further archaeological interpretation and preservation is (still) focused on 2D documentation. Rather to employ 3D models as such, georeferenced 2D digital elevation models (DEM), orthophotoplans and cross‐section plans are used. Georeferenced plans and maps derived from the 3D model are obviously more accurate than plans acquired by traditional techniques; rectification normally results in distorted photographs, whereas 3D models can be manipulated in a way that derives bird´s eye view without any further distortions and inaccuracies. Furthermore, georeferenced 3D models allow generation of different types of 3 views, cross‐sections, analytical calculations, etc. in the post‐excavation period that might help to answer the newly arisen question. To successfully integrate the new technique into the existing documentation workflow, certain storage standards for created 3D models have to be met. As all digital media, also 3D models run into the same problems on storage, interoperability and long‐term preservation. Questions such as: “Which file formats to use?”, “How to store and share data?”, “Which metadata should be included?” and “How to create a semantic 3D model?” arise. This is, however, a wider problem and will not be further discussed here. For the purpose of archaeological excavation documentation it is important, that the 3D models are stored in a structured, organized way. This means an efficient 3D database and registry should be created that would allow an overview over the 3D models and their metadata (size, location, key words, references...). At the moment there is no such database that would be tailored to the needs of archaeologists. However, some 3D software packages (such as aSPECT 3D from Arc Tron 3D) offer a construction of a 3D database. Another option is a construction of a Geodatabase using one of the GIS software packages. In any case there is a lot of space for improvements, which should ensure an efficient and flexible structure of a database as well as a simple manipulation of 3D models and their metadata. It is important that this kind of a database is user‐friendly and does not require any special preliminary knowledge. Namely, archaeologists usually lack of such technical expertise or under the constant time pressure of rescue excavations cannot afford to spend more time on it. Case study Before and during the secondment period the host institution Via Magna was involved in several projects. The project at Čachtice castle and excavations at Bratislava castle were selected as a suitable case study to complete the secondment program. Čachtice castle The Čachtice castle is today a ruin situated on the top of one of the limestone peaks (375 m) of the karstic Čachtice hills in Male Karpaty mountains between the villages Čachtice and Višňové. Former frontier castle served as a control point on the cross way of the Myjava valley to Váh valley. The castle was presumably built after the year 1260 under the patronage of Kazimir of the Hunt‐Poznan house. The oldest part of the castle is the residential tower situated at the front of the terrain disposition and it is dated to the 13th century. During the 14th and 15th century the building activity was intense and it resulted in a new defence area north of the main tower which consisted of new fortification features (cannon bastions), fore‐castle parts and masonries. Last major building activities date back to the 16th century when a complete reconstruction of the upper part of the castle was constructed and new architectural features were added to the fore‐castle. In 2012 the project of the castle restoration and static treatment was launched. Restoration works made archaeological excavations of the concerned castle parts inevitable. Excavations were limited to specified areas subjected to the masonry static treatment. During excavations bottom pavement levels of existing architectures as well as architectural fragments of earlier castle building phases were discovered. Remediation project together with archaeological and architectural research was finished January 2014. In order to document the excavation process as well as the castle itself, different were employed. Firstly, to create 3D models also older photographs that were not initially meant for 3D model generation were used. Afterwards, besides handheld photographing, photos were also acquired using a pole (pole aerial photography, PAP) and unmanned aerial vehicle (UAV). Measurements of 4 ground control points (GCPs) were measured by the total station (Nikon total station Nivo 5.M) in the national coordinate system (S‐JTSK). All 3D models were generated using Photoscan Professional. For post‐processing MeshLab and a trial version of Geomagic were used. For visualization purposes and cross‐section presentations a trial version of Global Mapper was employed. DEMs were generated by Photoscan Pro and ArcGIS. Plans were drawn in Civil 3D software package from Autodesk. Example 1 The 3D model of the interior of the main tower was created of photographs that were initially not meant for 3D model generation. However, 3D model with 2M faces was created from 25 photographs. 3D model Ortophotoplan – composite photogram of the main tower interior walls Example 2 5 The eastern palace was photographed using a UAV. 85 photographs were taken, generating a 3D model with 1.7M faces. 6 Bratislava castle Castle hilltop had been intensively settled since Aeneolithicum. It was used as a quarry, recent archaeological excavations revealed the remains of Roman architecture (which is a unique find north of the Danube), Maria Theresa strolled in the baroque gardens there and much more. The rich history has been with some interruptions excavated since 2009. Documentation of the excavation was carried out in small segments with the prime intention to generate the georeferenced ortho‐plans as soon as possible. Due to the processing time, particular masonries were divided into segments (e.g. facades) that were processed separately. Afterwards this segments were merged together in order to get the full overview over the complex structures. Combining the classical documentation approach with image‐based 3D modelling resulted in establishment of the so called photo‐sketch forms (see the image below). The documented part was sketched together with the order of the GSPs. Notes such as unit coding, description and number of images taken, were added. Photo‐sketch and TS/Rtk measurements code. Numbering the archaeological units. The sketch of the archaeological unit and GCPs markers. The markers are measured by means of a total station or GNSS Rtk Rover. A basic description of the archaeological structure and numbers of photos taken for the 3D model generation. 7 Example 1 As the excavations advanced 8 separate 3D models were lastly combined into one. Photographs were taken from the ground, using a ladder and a pole. 8 Do we really need 3D data? Because 3D digitization is a relatively new documentation approach at archaeological excavations, the question of necessity of such documenting arises, especially as there are still many limitations and problems to be solved. For example, as already discussed above, the prime output of excavations is a printed 2D report. Sharing and dissemination of 3D data are still very limited and have not yet become commonly used. 3D models have not yet fulfilled their full potential, as most of them above all have mainly a presentational role. They are, however, successfully used to extract 2D orthophotoplans and cross‐sections. The main advantage of 3D documentation on the site is speeding up the documentation process and enhancing its accuracy. Documenting in 3D namely means that every individual unit does not have to be measured separately, since the necessary measurements can be later then extracted from a georeferenced 3D model. This implies that we get more data in a shorter time period. On the other hand, a larger amount of data requires efficient storing and structuring of the data. Furthermore, it is almost impossible to post‐process everything on‐site, since the process is very time‐consuming. That is why a large part of the post‐processing is postponed to the “cabinet” stage of the archaeological works. This speeds up the documentation time on‐site, but prevents the possibility of later reassessment of the documentation on‐site. Moreover, the results are unforeseeable, which together with the lack of time prevents to pursue the classical vectorisation procedure (the new procedure goes as follows: 3D model – orthoplan – printed – vecorisation on‐site – vectorisation on the computer). 3D models, however, significantly assist by the interpretation and give the new notion about the explored site. Furthermore, using the Brown model for qualitative lens distortion correction and the 3D model itself bring the most accurate and comprehensive documentation approach that can be used at the rescue excavations today. Its advantage is especially noted when documenting complicated “volume” structures, such as masonries. 3D digitization of Gradište An ancient tomb The archaeological site Gradište is located on the northern periphery of the Skopje valley, in the village Brazda at the foot of the mountain Skopska Crna Gora. In 1985 archaeological excavations unearthed an ancient settlement dated in the 5th and 4th century BC. In that period the area was populated by Paionian tribe Arigans, but the pottery that was found during the excavations implies that Gradište was inhabited by Athenians who explored the rich natural resources of the environment. In 1986 archaeologist found a unique monumental structure which, according to the researchers, represents a tomb of a local dignitary. The tomb is dug deep in the ground, it consist of a rectangular room or chamber, and an entrance with six steps on the south wall. A long access corridor (dromos) steeply descends from the slope of the hill and leads to the entrance. The walls of the chamber are made of carefully shaped travertine stone blocks of different sizes, which were carried from the quarry in the village Gorno Svilare, 23 km away. The blocks are placed on top of each other in horizontal arrays. The roof of the chamber was constructed from wooden blocks. Because there were no finds discovered, archaeologist assume that the tomb was raided already in ancient times, when they dug even the floor tiles. 9 Data acquisition Even though an archaeological park Brazda was established in May 2013, the site was overgrown with enduring vegetation. Because of the occasional storms it took us an afternoon and the forenoon of the next day to clean up the monument. After approximately 30 GCPs makers were placed on and around the monument, it took us about three hours to take in total 497 photographs. At the time of data acquisition the sun was already high on the sky and the monument was throwing shadows on itself. More problems with the changing shadows were caused by the moving clouds. The wind did not seem strong, but it did cause some movement of the high grass, trees and other vegetation around the monument. The photographs were taken with the Nikon D5200 camera in the highest resolution. All of the photographs were taken with camera setting of aperture priority F5.6 in a JPG file format. The GCPs were measured by a local private company using a total station. 10 Data processing The 3D model generation was carried out in a commercial software package PhotoScan Professional (version 1.0.3 1832 build 64bit) from a Russian provider Agisoft on a PC workstation with the following specifications: Operating System CPU Main memory (RAM) GPU Alignment Microsoft® Windows™ 7 Professional 64‐bit Intel® Xeon® CPU E5‐2620 v2 @ 2.10GHz (24 cores) 128 GB NVIDIA GeForce GTX 780 For the process, 457 photographs were selected based on the image quality and the photographic perspective. Because of the conditions under which the photographs were acquired, the alignment was not a straightforward task. Wind caused movements of clouds and vegetation and consequently also of the shadows, therefore most of the background had to be masked out before an accurate camera alignment was accomplished. In addition to masking, several realignment processes had to be carried out, as well as alignment optimization using GCPs. At the latter stage, the point cloud was simultaneously also georeferenced. After removing background points, the final sparse point cloud consisted of 740 000 points. 11 Dense point cloud The dense point cloud was calculated on the medium quality to perform the process in a reasonable time period (5 hours, 15 minutes). The final point cloud consisted of ca. 17M points. Meshing The dense point cloud was afterwards meshed into a solid 3D model consisting of ca. 3M faces and 1.7M vertices. 12 Texturing In the final step, texture was created in the resolution of 8000 x 2 pixels. 3D model visualization After the 3D model was created we first created an enhanced 2D representation. For this purpose we used Global Mapper – analytical hill‐shading, contour visualisation and hypsometry. Afterwards we created two perpendicular cross‐sections. As a presentation of the site we created a fly‐through animated movie using Cinema4D software package. 13 Analytical hillshading: 14 Contour visualisation: 15 Hypsometry: 16 Cross‐sections: Final thoughts During the second secondment we successfully accomplished the main goals, which were expanding the knowledge on image‐based 3D digitisation gained during the first secondment. The program of the secondment was focused mainly on the application of 3D digitisation to the archaeological works, especially to the rescue excavations. A practical attempt on how to integrate the new methods into the standard workflow was carried out. Afterwards the results were evaluated, discussed and presented at the CAA 2014 conference. A more demanding example of image‐based 3D digitisation was the 3D documentation of an ancient tomb in Brazda, Macedonia. In this case we tackled the problems of accuracy of image‐based 3D digitisation and presentation of data for the archaeological purposes. The experience of the secondment was a valuable lesson from the practical and also commercial point of view on archaeological 3D documentation and the problems that accompany such procedures. At this point I would also like to thank Dr. Milan Horňak and all my colleagues from Via Magna s.r.o. that not only professionally fulfilled their part of the project, but also made my stay a very pleasant experience. 17