Friday, July 26, 2013

Justin J.L. Kimball, Olof Siljedahl & Jack Hope, students of our courses in Digital Archaeology (ARKN10-ARKN09), Lund University

Justin J.L. Kimball
Olof Siljedahl
Jack Hope

Subject: Winged Man of Uppåkra

Project Development and Data Acquisition Campaign

For our project, we have worked with the Winged Man of Uppåkra figure that was found in Uppåkra during the summer of 2011. This was an exciting venture for our group, particularly because there has been little published on this artefact. Nicoló had suggested this project to our group because he was aware that the Historical Museum was preparing a new exhibit to display this artefact and others. After getting in contact with the museum, the curator sounded enthusiastic about our project and we were given a green light to perform a data acquisition campaign of the artefact. Our basic plan was to digitally acquire the artefact, produce a 3D model of it, and perform a camera animation around the model to give the viewer a sense of the geometry. The museum was interested in having the artefact visualised on top of a small box; the currently favoured interpretation of the role of the Winged Man of Uppåkra figure.

We were permitted access to the artefact for three hours on the afternoon of 31 May 2013 (Friday) at storage facility in Gastelyckan. Owing to the type of material (which is a gilded bronze), complexity in shape, and size of the artefact [Fig. 1], it was decided that the best data acquisition method to use would be image-based modeling. Our group settled on using three cameras: Olof’s Canon EOS 1000D with a 55mm lens, Justin’s Nikon D7000 with a 55mm fixed focal length macro lens, and Jack’s point and shoot camera (primarily for video). Our other equipment included a sturdy tripod, two photography lamps with diffuser screens, and a photography table.
Fig. 1 - Photo of the Winged Man of Uppåkra
The data acquisition campaign was accomplished with little difficulty. The first technique to be attempted was framing the artefact in a way that there was no visible geometry in the camera frame other than the artefact itself: thereby allowing for the camera to be kept stationary while the position of the artefact itself was manipulated [Fig. 2 & Fig. 3].
Fig. 2 - The stationary-camera setup; the little box was used to frame the shot before the artefact arrived.

Fig. 3 - Justin and Jack preparing the stationary-camera setup
Fig. 4 - Mobile-camera setup
Although a great method in theory, we admittedly forgot to turn off the actual lights: a short-sight that effectively made useless the diffused photography lamp setup. Realising this, we then developed a new setup where the artefact was placed in a central point and the camera traveled around the object [Fig. 4]. Justin used a remote trigger in order to ensure that his shadow was not cast over the object resulting from having to stand in front of one of the lamps. In both setups, Jack acquired video of the artefact, using his camera to simulate how we wanted to try to visualise the camera movement for the final movie.

Modeling and Cleaning

After the acquisition campaign, we then set about creating the models in Photoscan. Again, because of the size of the object, we felt that our models would be best rendered using either the “High” or “Very High” settings for the geometry construction phase. Initially, the group was optimistic that the macro photos would render the most detailed model, and indeed the first model developed from the stationary-camera macro photos was quite impressive. However, this initial model was produced from photos that had not utilised the magnification power of the macro lens—as such the depth of field was quite deep, allowing for the generation of photos where the details were extremely crisp. Unfortunately however, Photoscan seemed to have severe difficulties generating geometry once photos where the magnification power of the macro lens was utilised were added as a chunk [Fig. 5].
Fig. 5 - Photoscan was confused
The result from this error was extremely unfavourable and was likely resulting from two factors: 1) the non-diffused room lights, and 2) the macro lens’ depth of field when used in high magnification. The first factor was probably negligible, but since there was no acquisition of the model with the room lights off and only the diffused lights operating, we cannot discount this entirely. The second factor however, was without any doubt a contribution to the failure in generation of geometry. The issue with macro lenses, is that they have an extremely shallow depth of field—there is a very small plane of focus. In other words, although the macro lens allows the photographer to magnify the details of an object, but only on a very thin region: anything that is too far into the foreground or background is subsequently out of focus and therefore useless for the development of a 3D model. Although this first set of photos produced a very impressive model, the detail was not captured as was hoped.

The second model produced was from the photos that Olof captured while moving the artefact and keeping the camera stationary. The settings were Very High Quality, Sharp, and with 250k faces. Using these settings and Olof’s photos, Photoscan produced a model with incredibly impressive geometry. Unfortunately, there were some major problems with this model that affected the figure’s face (arguably one of the most important features of this artefact). For some reason there was a severe loss of detail around the cheeks/sides of the face as well as under the nose and above the mouth. When the texture was developed, Photoscan left the model looking (quite disconcertingly we must add) eerily similar to one of the world’s most notorious dictators [Fig. 6]. Several attempts at re-processing this model only resulted in the same issues or worse, while cleaning the “best” model only ended in a large loss of detail from the face [Fig. 7].
Fig. 6 - Hmm ...
Fig. 7 - Removing the noise also removed much of the detail from the face. We would decide this was unacceptable for our final model
Justin first attempted to fix this loss of detail by reconstructing the face in 3DS Max. However, becoming displeased with the results, he eventually went back to his photos and attempted to produce a detailed model from the mobile-camera acquisition set. Although some photos were again plagued by the limitations caused by the depth of field, the non-macro magnified photos produced an impressive model with an excellent display of geometry (albeit less impressive than some of the geometry visible from the model composed of Olof’s photos). Fortunately, the model produced contained very few areas that needed to be cleaned; allowing for a model that was moved quickly from Photoscan to Meshlab (for cleaning) to 3DS Max for texturing work and scene setup. The favourable qualities of this last model were key behind our decision to use this as our display model for the artefact [Fig. 9].
Fig. 8 - Uncleaned final model

Scene-Setup and Animation
All three group members attempted to develop a box for the project. Running off of a group-developed theory that the Winged Man of Uppåkra figure may have been used also as a brooch at one time, Jack discovered through research that there was a specific ornate style of women’s jewellery that combined a brooch and a box. This combination was intended to both have a decorative, storage, and utilitarian (it was used to keep a cloak pinned to the shoulders) function. Although tempting to reproduce, these boxes were highly ornate and would have presented a challenge for our group to achieve with the limited amount of time left. Such a creation would have been entirely subjective and founded on nothing concrete from the associated excavation. Furthermore, we were worried that such an ornate box may mislead the audience and may have even subtracted attention from the true purpose of the scene—to highlight the Winged Man of Uppåkra figure and the impressive details visible on its surface. As such, it was decided that a very simple wooden box would be used instead.

Our basic scene is comprised of one camera and four lights. One of these lights was a spot light, positioned to contrast the geometry of the figure through highlights and shadows. The remaining three lights are omni-lights positioned to soften the shadows around the box and underneath the chin of the figure. For the animation, we had originally decided that it was important to have a camera that moved in a circular motion around the model. This was accomplished using a target camera set on a fixed point and bound to a path around the object. The path causes the camera to approach the model from the front, move around the model in an anti-clockwise direction, with the end of the path bringing the camera close alongside of the model to highlight the geometry on its back and face [Fig. 9]. From this animation, we then produced a 10 second movie (300 frames) which we rendered in .AVI format. The movie was produced in high-quality format with a resolution of 1536x1167. However, due to the quality and length of the film, it was much too large to upload to blogger. Instead we have included the original sample video (full length but very low quality) and a short 3 second sample of the high quality video. You will find the links to these movies after the following images.
Fig. 9 - Screenshot showing the path of the camera, the position of the lamps and a preview of the rendering
Fig. 10 - A rendered sample of the box and artefact placed on top
Fig. 11 - Olof hard at work rendering the high quality movie 

Future Directions

Although we accomplished our goals for this project, we recognise that there is still much room for improvement. This project has functioned to truly test our skills not only in using 3D modeling software, but in the actual data acquisition process as well. In fact, this is something that we would like to revisit if possible in the future as we are confident that we will be much more effective with our procurement of data—thereby allowing us to produce a model of extremely impressive quality. We also recognise that the texture of our both our artefact and box needs some work. It was our hope that we could have visualised our models using VRay, however due to constraints on time and a difficulty in communication (our group was, by the last week of this project, physically separated by an ocean and several time zones) we were unable to realise this level of quality with the remaining time. As such, the pursuit of visualising the scene using VRay is a top priority for our future work on this project. Another issue we want to tackle is making the animation a bit more complex. The level of detail on this artefact is stunning and we recognise that static lights do not do this detail justice. Thus we intend to tackle this issue by actively “painting” the model with light so that its movement across the surface shows the depth of detail as the highlights and shadows shift about.

This project has been an extremely useful test of our abilities and has functioned to teach us a great deal about one of the ways virtual reality in archaeology is applied outside of the classroom. We would like to sincerely thank Lovisa Dal for her assistance at the Gastelyckan storage facility, Jerry Rosengren for permitting us to work with this artefact, and Nicolo Dell’Unto for his guidance and tutelage throughout this project and in the ARKN10 course in general. It is with great eagerness that we look forward to continuing our work in order to produce a 3D model and animation of outstanding quality for the exhibit at the Historical Museum in Lund.

No comments:

Post a Comment