Jonathan Westin reports on scanning the Winter Station
Towards the end of Samuel Duse’s book about the expedition 1901-1903, he writes that ”if a Swedish polar expedition with solely scientific purposes would be carried out sometime in the future, I hope it will be conducted under more favorable conditions, with better funding, and first-class equipment, so it may result in even greater successes” (Duse 1905, p. 266, my translation).
I will not try to compare funding (or successes) but we did in many ways have more favorable conditions and we brought some equipment that surely would appear magical to Duse and his companions, and would have been great additions to the sort of work they performed. In addition to a host of cameras (including two GoPros and a 360-camera), computers and GPS devices, two drones (a DJI Phantom 4, and a DJI Mavic 2 Pro with a Hasselblad camera), sun panels, and automatic sensors and loggers for wind, radiation, and earth temperature, we also brought an EinScan Pro 2X Plus structured-light scanner from the Heritage Visualization Lab (GU), and a Faro Focus laser scanner from Centre for Digital Humanities (GU).
Early on, to be reasonably confident of leaving Antarctica with the data we needed, we decided to capture both the interior and the exterior of the Snow Hill Winter Station in 3D using two different techniques and technologies; Structure-from-Motion (also called photo scanning), and laser scanning (using the Faro Focus). For Structure-from-Motion, all you need is a camera. As the ground-based photo scanning could be performed with any of the 24 camera equipped devices we were bringing on the expedition, it is an extremely reliable technique that does not make us dependent on a single machine. It is also simple: one takes photos of the object or structure, and based on identified key points in the photographs and how these move from one image to the next, the software tries to construct a 3d structure by recreating how the photographer has moved around the object. Depending on the object, different procedures are followed when taking the photos. However, most objects, and even buildings, can be successfully scanned by moving in a circle around them while taking the photos, centering the object in each photo.
The Faro Focus scanning the interiors of the Winter Station. Photo: Jonathan Westin
The Faro Focus is primarily used for scanning whole environments and this particular model has been used in high profile projects such as the scanning of Rome and the city catacombs. It is mounted on a tripod and rotates slowly while measuring up towards a million points a second. A single scan can capture more than 700 million points, creating enormous point clouds when combining several scanning positions. However, it can only measure what is possible to beam laser points to from its fixed position in space. This means that the scanning of a furnished room often requires four or more separate scans to cover most of the interiors, and you might need even more at various heights to capture the underside of chairs and tables and the top side of high shelves.
A colour 360 panorama produced by the Faro Focus used to colour the point cloud. Image: Jonathan Westin
Both of these procedures require some planning to get a good result, and for the exterior scanning the weather conditions also affect the choices. For instance, a hard wind or rain would rule out the use of the Faro Focus while photo scanning could still be performed. Depending on the rain, a small lens, like that on a cell phone camera, is preferred since it is a smaller target for water drops that might warp the result. If snowing, none of these techniques are advisable, as there is a risk that the snowflakes will appear in the texture information.
To plan for the scanning using the Faro Focus, already in Buenos Aires I created a 3D model of the interiors and from that 3D model calculated where I would have to place the scanner to capture as much information as possible in as few scans as possible. The efficiency of the scanning was important, since each scan drains up towards 8 per cent of the battery on certain settings and we would have limited opportunities to recharge when camped on Snow Hill. There is also the time aspect: a scan averages 17 minutes at my preferred settings, meaning that a large number of scans would make the Winter Station inaccessible for the other researchers for days. Through this method and with these considerations in mind I came up with 28 scanning positions for the interiors of the station, which would amount to twenty one billion measuring points and could be finished in a day.
A detail from one of the scans. This is not a textured model or a photograph, but hundreds of millions coloured points making up a very dense point cloud that could be explored in VR and from which various measurements could be obtained. Image: Jonathan Westin
To latch on to Gunnar’s post on the subject, it is interesting to contrast using a laser scanner with traditional measuring methods as there occurs a shift from a physical place-based immediate analysis, to a digital remote-based delayed analysis. To save time, when planning the scanning I made use of both historical and modern photographs of the Winter Station, as well as plans by Goldberg, Wiklander and Capdevila (2001), to construct the interiors in 3D rather than wait until I myself had access to the physical place. And during the actual scanning, in Antarctica, I am shut out from the station and forced to wait inactive for the scanner to capture the data, postponing the analysis for the future. The scanning thus somewhat divorces me from the act of processing the Winter Station first hand, while a traditional measuring method would have forced me to closely work out every nook and cranny of the structure there and then. That said, a major advantage with the digital documentation is that it captures far more detailed information about the visual aspects of the Winter Station. The data from the scans will allow me and others to digitally investigate the station for years to come and do exact measurements of various details and analyse its visual qualities. Not only does this allow for countless revisits in less stressful surroundings (which in turn will allow for many more observations), but due to the sheer size of the dataset it also lets researchers answer questions we have not yet thought of: if a future researcher needs detailed information about the knots of the wallboards, the depth of the hammer marks surrounding individual nails or the roughness of the iron surface of the stove, they can obtain that information from the digital documentation. Manual documentation meanwhile only captures those aspects of the interiors that the researcher there and then deems important to include. Hence, the digital documentation is less an analytical process than an enabler for analytical processes, while the manual documentation is the artefact of an analytical process already begun.
Detail from model of the hill and winter station created from drone photography. Image: Jonathan Westin and Dag Avango
For the Structure-from-Motion photography of the interiors I took about 2400 photos. The advantage of the Faro in these types of complicated environments is that each measuring point provides reliable data, while a photo scanning process could result in erroneous data due to an inadequate number of images, dark or noisy photographs, repeating patterns (or no patterns at all!) that confuses the identification of necessary key points, or an unstructured way of taking the photos. While a square and well lit room could be easily captured in a hundred photos (or even half that number), the number of photos required quickly grows for each additional object in the room as these also has to be photographed from all possible angles. Furthermore, the technique you use when moving through each room, and between rooms, greatly affect the end result and each situation comes with its unique set of problems. Due to not having had access to a reliable power source until these last few days in Esperanza, I have yet to process these images into a model. A source of concern is the bright light that shone through the windows, causing reflections in the lens that will be hard for the software to make sense of. However, we anticipated that no technique would work in all conditions, hence the variation in technologies used. Even if the photo scan model will be incomplete, the data obtained through the Faro Focus scanning will be more than enough for all our research needs, providing us with a photo-realistic and measurable representation of the interiors.
Since it would be too resource intensive to process and check the results of each scan when camped (computer batteries drain fast in the cold, and the processing of point clouds is power hungry), preparations had started several months earlier: from the beginning of November I had done daily tests to learn how the Faro reacts in certain conditions and blindly be able to make an educated guess about the end result without processing the data. However, Antarctica offered some unique conditions I had not foreseen or been able to test: while the weather conditions seemed favorable for scanning also the exterior of the Winter Station with the Faro Focus, I was unable to get good colour information: the result was blurry in the samples I processed an oddly over saturated. Though I had done outdoors scanning in Gothenburg (a quite windy city) my first theory was that a combination of the wind and the ground was at fault, resulting in slight movements causing the image to miss-align. However, after some tests in nearly wind still conditions (a rarity in Antarctica), there seemed to also be other factors that affected the result. While I have yet to find a definitive answer, in all probability it is either connected with the cold temperature or the ground in combination with the natural movements of the scanner (or all three): the very soft and eroding moraine on top of the hill’s ice lens might have reacted to the rotational movements of the scanner making it sink slightly and unpredictably between each photo used to create the colour information. Furthermore, the repeating visual chaos of the moraine might have made it difficult for the Faro to stitch together the photos properly. This theory is supported by the fact that the quick laser data is sharp, and when processing the point cloud without using the colour information (just using the grayscale reflectance map) the result is very good (albeit colourless). Overall, I scanned the exteriors at four different occasions at different settings, resulting in 68 scans that also mapped the surrounding landscape within a 70 meter radius at a resolution of one measuring point per 1.5 millimeter.
The level of detail of the model will let us plan for future interventions to help stave off the erosion of the hill. Image: Jonathan Westin and Dag Avango
Once again, using two different technologies proved a good idea in this uncertain and untried environment: preliminary results from Gunnar’s ground-based photo scanning of the exteriors are promising and will together with the models of the hill and the surrounding landscape processed from the drone-based photo scanning complement the grayscale laser data.
References
Duse, Samuel 1905. Bland Pingviner och Sälar. Beijers bokförlagsaktiebolag.
Goldberg, F., Wiklander, L. & Capdevila, R. 2001. The Swedish Hut in Antarctica. The Stockholm Building Society.
