Community:Science/Architecture/DigitalSurveying/HotelAubecq

Acquisition and virtual rebuild of Victor Horta's “Hotel Aubecq”

Context

In 1949, despite numerous protests, one of the Art Nouveau masterpieces from Belgian architect Victor Horta, the “hotel Aubecq”, was demolished. It was one of the early victim of what is currently known as “bruxelisation”.

Tough, during the controversial destruction, thanks to the architect's wife and a former apprentice, elements of the front facade were saved and stored.

In 2010, following the unexpected rediscovery of the hotel Aubecq ironwork, a renewed interest arose to rebuild the facade as a memorial to the Art Nouveau movement.

During the last two years, on the demand of the DMS (monument and sites directorate) of Brussels, O'point associates has worked together with other experts to produce an accurate review of the state of the building remaining.

Original paper and complementary material Images and original paper in French can be found here

3D Acquisition

Each stone of the facade was then inventoried and scanned while the facade was rebuild horizontally in a warehouse and vertically in virtual space using Blender. To deal with the impressive amount of data from the 3d scan, we used Blender script API to produce simple but effective tools to set up a fast pipeline from the original 3d scans to the full reconstructed 3d model of the facade.

With these 600 scans, our objective was to recompose the entire façade as close as possible to the original. If we had worked on an existing building, the methodology would have been to measure first the main dimensions to create a basic framework, then to subdivide these to finally take care of the details. In this case, however, the usual approach, from the general to the specific, had to be inversed as we had to rebuild the façade from its constitutive elements with the risk of cumulating inaccuracies rather than reducing them.

The virtual reassembly was supported by rough sketches of the original architectural project, by pictures of the time and residues of mortar discovered on some of the stones. The advantage of open-source software – besides the fact that it is free, is the possibility to easily adapt the existing modules. For this specific project, most of the tools had to be created.

After the acquisition, all stone blocks were oriented randomly. Hence, we had to develop a number of semi-automatic tools to give each block its proper orientation. As block faces were often curved or decorated, it was more relevant to determine the blocks orientation based on the joint plans. So, our calculations were based on the upper courses and on the lateral faces so that the stones would look as close as possible to their rebuilt state.

We then automated the scan decimation at different resolution levels. We used the existing proxies system (external listing of stones) complemented by a redirection system that allowed us to select the resolution according to the level of visualization and precision required. As stones were integrated as external references, rather than as objects, selection was more precise. The system supports the visualization of the entire façade in low resolution and/or of certain areas of interest in high definition. Further, the working file only includes reference links to the geometries of the stones based on their positions, which only represents a few K octets. Hence, the risks of losing data were limited and several rebuilding configurations could easily be stored.

As Blender is not designed for architecture, it was necessary to program several tools to be able to calculate more precisely, movements, measures, positioning and alignments. Blender is normally used to create and process modeled information. Here, we are directly linked to the real material through an extraordinary surface definition (400 Gb of source data). This raises, amongst other things, the question on how to link the objects to the real 3D surface texture, taking into account the subsequent risk of imprecision. Ideally, primary surfaces should be reconstructed on the basis of the real surfaces so that each stone can be modeled. However, their number and complexity quickly convinced us to find a work around and leave space for human appreciation in the reassembly process. The assembly of the stones was globally satisfactorily, considering the horizontal courses and the joint plans that were better treated near the edges. Through this approach we were close to a true physical reassembly.

Finally, we created a virtual anchoring system to automatically align layers of stones while taking into account the joint spacing. This method worked well for the relatively straight parts of the façade. However it proved quite inefficient on the more complex sections (such as double curvature arches).

The reassembly was based on more or less precise pictures and drawings of the time. A preliminary layout of the stones also provided useful information on their positioning. We first used the available scaled drawings to locate the main entrance door and part of the ground floor windows (distance and width) to setup a global wireframe. From there, we derived the average value of the vertical joints from the first stone bed and from the dimensions provided by the drawing. As we did not have any vertical dimensions, we extrapolated this value from the horizontal joints. We first inserted each stone, individually, in a complete layer before adjusting their positioning in terms of rotation and distance. The stones were then reassembled, layer after layer, as a stonemason would do, in order to respect the alignment of each joint face and decoration as best as possible.

For a part of the façade (the upper right quarter), we did not have any scaled drawings. We then based our calculations on symmetry elements from the corner of the façade. Some manual touch up was necessary to obtain a correct assembly in the more complex sections. Other assemblies (fairly similar) of this part of the façade must exist; this is one alternative. We have resolved some inaccuracies thanks to the photos, particularly to back up some excessive joint distances (also visible on old pictures).

Note We did circumvent a few difficulties: We had to make some adjustments when closing the vaults. There were mortar residues on joint surfaces and we constantly had to go back to the photos taken on site to get a correct joint distance between two rough stones. However, some residues gave us precious hints and the quality of the scans allowed us to detect contact surfaces between some stones and deduct very precisely their respective position. The façade is very complex and shows, at places, curves along vertical lines. Aligning stones in these areas was more difficult. Numerous stones had been swapped compared to the first layout. This had to be corrected while reassembling. Thanks to the precision of the 3D assembly these positioning mistakes – difficult to see on site– could be detected. 15 stones were missing. These "holes" in the façade were variables that led to a certain lack of precision in the alignments. Some missing stones had been reproduced. Unfortunately their dimensions were sometimes wrong, which resulted in further difficulties in the reassembly.

Compared to the stonemason, we were lucky to be able to easily go back and “undo” what was wrong to solve all these issues. Slowly but surely, the façade was taking shape and after two months of assembly the result was beyond our expectations.

Ironworks

For interior work elements, we have used the Elcovision photogrammetry software to generate orthophotographs, from which we could redesign the ironworks in elevation view. These drawings were then rendered in 3D by controlling the thickness of the profiles. Based on the old photographs, on hollow traces and elements still in place, we managed to reposition the grids and frames.

Post-processing

3D rendering technologies, based on point clouds, mark a break with traditional architectural design, grounded on line work. Our objective was to generate a line drawing, automatically, from a 3D model view, whereby the evolution of the thicknesses would depend as well on the points of contact (outlines) as on the local curves (line form and texture) while taking into account the occlusions of the successive layers (ironwork, stones, woodwork).

The main difficulty lies in the fact that these edges and plans do not exist as such amongst 3D meshes. However, 3D model visualization and processing are key steps towards the final rendition and its future usage.

In the industry, reverse engineering is applied to extract and extend the primitive surfaces up to the creation of intersections. Such a methodology requires high skill levels and long manual processing times. Moreover, the geometry of the stones is highly heterogeneous. We found front faces with various textures, almost rough rear faces and some sculpted elements evolving into mouldings that ended up into the curved plan of the façade. It is clear that the formal Art Nouveau language, at the outskirts of arts and technique, prevented us from using Reverse Engineering within a reasonable cost and timing (the cost of this software is in the 20K Euros range).

Luckily, Blender already integrated functionalities that could be adapted to our needs. For instance, we have used the existing rendering engine to visualize the final result at different levels of details, each time with a specific line thickness. The different detail levels were then mixed together to provide a final view where the lines hierarchy presented both the overall geometry and the surface texture of the materials. The drawing then became multi-scale, including and structuring a high volume of graphic information. We could take advantage of this combination of line thicknesses in certain renditions. In a few simple operations (gaussian blur, substraction, sharpening), texture lines could be filtered from outlines, resulting in a fine control of the balance between the two types of lines.

The architect had also requested that we provide visualizations of the different stages of the façade in order to illustrate and support his study. The different parts of the building were rendered separately, so they could be integrated as separate layers in a classical photo touchup software (Gimp). This way, we could control at the same time, and for each sub-assembly, the thickness of the lines, the colors and the transparency This way, we managed to generated a 3D model that could be viewed from any angle, at various resolution level and in different phases. These line visualizations are a continuation of the traditional architectural language and provide a solid basis for comparison with the original documents of the project.