The UNESCO World Heritage Site of Lorsch Abbey is located in Hesse roughly between Heidelberg and Frankfurt/Main, Germany. Laser scanners generating 3D point clouds of the structure and photogrammetric Structure-from-Motion (SfM) approaches were used to study and create 3D models. Unmanned aerial vehicle (UAVs) were also employed to support restoration work. This project is continuing with current surveying, measurement and imaging this summer.
For several years the Interdisciplinary Center for Scientific Computing (IWR, Heidelberg University) and the administration of the UNESCO World Heritage Site Lorsch Abbey are cooperating in different topics. The IWR’s part was to support the three-dimensional documentation of several parts of the King’s Hall, especially the capitals.
The UNESCO World Heritage Site of Lorsch Abbey (Fig. 1) is situated in Hesse roughly between Heidelberg and Frankfurt/Main. Three parts of the abbey are preserved today: the fragment of the church, the abbey’s wall and the most famous building, the so-called “King’s Hall” (in German “Torhalle” or “Königshalle”) (Fig. 2). The latter is well preserved despite some rearrangements over the course of time.
This building, dating to the Carolingian epoch, is unique in its architectural structure. The ornaments are on the one hand prestigious and outstanding because of the shiny red stone hexagons contrasted by the white inlays in between, but on a closer look also detailed for example in the capitals or the frieze (Fig. 3).
In the course of the restoration, started in 2012, the King’s Hall has been investigated with methods of Building Archaeology. The main focus of attention is on building technology, classification and later rearrangements as well as general research in Art History. For example, a complete documentation with site measurement and detailed drawings (Fig. 4) is one of the tasks carried out by Dr. Katarina Papajanni, supervised by Prof. Manfred Schuller (Chair of Building History, Building Archaeology and Heritage Conservation, TU Munich), funded by the Deutsche Forschungsgemeinschaft (German Research Foundation).
3-D Data Acquisition
The work at the King’s Hall revealed the difficulties of scanning with a structured-light-scanner in somehow shaded daylight on a scaffold (Fig. 5). The results were not satisfying because unavoidable vibrations and minor movements of the scaffold caused large measurement and computing errors resulting in artifacts (false surfaces) in the 3-D model.
Simultaneously to the scanning we used the photogrammetric Structure-from-Motion (SfM) approach, in this case by the open-source available software Bundler by Noah Snavely (Cornell University) for the bundling and the PMVS/CMVS by Yasutaka Furukawa (Washington University). Lately we used VisualSFM by Changchang Wu (Washington University), whose models turned out very well, showing every detail of the capitals in a high resolution.
Two of the capitals were removed from their original places in October 2013 (as seen in Fig. 2 and 3), offering the chance to use the structured-light-scanner in a better setting indoors at the administration office of Lorsch Abbey – and of course having the possibility to record a complete 3-D model including the backsides. The models are very detailed and quite complete, although the narrow cutout of the upper sides of the covings made some problems as they were out of reach for the structured-light-scanner. Here the photogrammetry obviously had an advantage because of the free camera positioning.
We also did SfM models to compare them to the scanner’s models, which showed the very high precision of the SfM approach provided that the photos are taken thoroughly. These results will be published in a separate article.
In this case we used the octocopter of the author’s PhD-project called “ArchEye” (Fig. 6 and 7). This UAV has eight rotors and is roughly Y-shaped, developed by Frank Thiemig, (www.powerframe.de). The camera is, in contradiction to the most UAVs, not mounted below the frame, but in the opening of the Y on a level with the frame. Therefore we do not have to use a large landing gear and thus are reducing the risk for the UAV to cant over. At the King’s Hall we used the Sony NEX-7 camera with a lens of 30mm focal length and a 24MPx sensor. Since the eastern side of the King’s Hall was not open for the public, there was no need to block the area for our flights for security reasons.
To document the façade properly it was necessary to fly a previously calculated route parallel to it. Basically the route is from left to right and vice versa in different altitudes, so that there is a large overlap between the single photos in all four directions. The software we used for this task was originally developed for project ArchEye to document archaeological excavations. It provided the photo positions needed to cover the whole area with a photo-mosaic to calculate a high-resolution orthographic image of excavation-trenches.
This approach will not work for the roof because the software can not handle the slope and is therefore unable to calculate a route. To acquire this part, the photos were taken in manual flight from different directions all over the eastern side of roof.
The colleagues from the GIScience / Geoinformatics Research Group provided a Riegl VZ-400 Terrestrial Laser Scanner (TLS) (Fig. 7), operated by Martin Hämmerle, to do a complete 3-D scan of the building. The comparison of SfM results to the reference TLS data allows to assess the accuracy of the SfM models and is work in progress.
Since the work at the site is not fully completed, the presented results are of somehow preliminary character. As the western part of the King’s Hall has not been acquired yet, we will only show the eastern side and some detailed objects. The aerial documentation of the western side is scheduled for August 2014.
This model is the most recent of the presented ones, showing the whole façade in quite some detail (Fig. 8). The model consists of 818 single photos, 494 of them were taken by the UAV. The sparse point cloud for the reconstruction of the camera positions already shows a high grade of details consisting of over 300,000 points. The detailed model, computed in 20 hours on a high-end PC, shows all main elements in a sufficient way but clearly not with every detail. For this purpose we already had the models of the capitals. The red surfaces were holes filled by Hubert Mara's GigaMesh software framework we use at the IWR for the post processing and analysis of the models.
For comparison, Fig. 9 shows a detail of the drawing made by Katarina Papajanni and the same part of the 3-D model. Even though the surface is not very pretty at the moment, the match between the model and the drawing is clearly visible, whereas the model is sufficient as a template for basic (digital) drawing of a historic monument. This drawing can then be filled in on-site with all the additional informations the model can not provide.
Only the frieze of the eastern side was documented, showing the whole width of 18 meters along the façade in fairly good details, although the photos were partially taken freehand and under bad weather conditions (Fig. 10). We took three sets of photos from different monopod heights with a lens of 30mm focal length to provide a high vertical overlap in each picture. Some more photos from a first test run are also used.
All the eight capitals, four on each side, were documented in several sessions. The results especially on the western side show the heavy weathering, particularly in direct comparison with the capitals of the eastern side. For each of the capitals there were taken 70 to 180 photos with different cameras. The far best model is the one on the far left side on the eastern side, which is also shown here (Fig. 11 and 12). The model has serverd as a sample as Katarina Papajanni used screenshots taken from an orthographic view to improve her drawings of different parts of the capitals. Since a drawing is always the first step of the interpretation, it shows or ephasizes - or intentionally does not - different parts of the object. The model in contrast is far more objective and is therefore an additional result worth to keep.
The project for the authors PhD is slightly different, but yet tightly connected to the Archaeological Sciences. It is basically the next logical step in using a UAV for the documentation of buildings: to document them in a far more automated way. The basic idea is not to use just one camera, but two of them. Through the determined distance and angle between the cameras it is much easier to calculate a 3-D model. Using this Stereovision technique and a PC on board the UAV we can generate a model of a building and its surrounding area. With this information the UAV will be able to navigate automically and far more precise. It is also possible to decide where more 3-D information is needed. For example it is not necessary to document a flat wall with a high resolution, but at some architectural decoration like the capitals or the frieze from the King’s Hall it is. Considering all this information, we will calculate optimized trajectories for the flight to improve the duration as well as the overall quality and size of the 3-D model.
But besides the PhD work, the author will continue the work at Lorsch, since the western part of the King's Hall is waiting to be documented in 3D. Furthermore, there is a new software called "SURE" to do the calculation of the dense SfM -pointclouds, showing very nice results with less photos. It is developed by Mathias Rothermel and Konrad Wenzel (Institute for Photogrammetry, University of Stuttgart) especially for the use with UAV imagery, just as we plan to use the ArchEye UAV more often for future tasks.
The author would like to thank Dr. Susanne Krömker and Dr. Hubert Mara (both IWR), Dr. Hermann Schefers (Director of the Cultural Heritage Site Lorsch Abbey), Martin Hämmerle M.Sc. and Jun. Prof. Bernhard Höfle (both GIScience, Institute of Geography) and especially Dr. Katarina Papajanni for the outstanding teamwork and the trust in the methods used by the author.
For more information
About the Author