The study of prehistoric architectural remains generally includes consideration of how different areas of a structure were utilized. This type of analysis can be important in developing models of social organization of the site's former inhabitants. In the case presented here, we describe how computer graphics techniques make it possible to incorporate the display of artifact data distributions into the structural and environmental context of site architecture (specifically, a prehistoric pithouse) in order to assist in the modeling of socioeconomic organization. We have found that visualizing excavation data in this manner can be a valuable aid in the identification of correlations between the spatial distribution of artifacts and significant site features, as well as a useful tool for identifying architectural constraints on usage areas.
Computer graphics techniques are increasingly being used to visualize complex data in archaeological investigation. In recent years, several projects involving the creation of detailed virtual reconstructions of archaeological sites have been undertaken. In these applications, a three-dimensional (3D) computer graphics model of the site is constructed and viewed using standard modeling, rendering, and animation techniques.
Some of the most well-known examples of virtual reconstruction have involved the application of computer graphics techniques in the recreation of historical architecture. Initially applied to the modeling of the temple precinct of Roman Bath, these techniques were later refined and applied to the more ambitious modeling of the Saxon Minster of Winchester (Reilly, IBM Systems Journal, 28(4):569-579, 1989). A later project, the modeling of the Furness Abbey (Delooze and Wood, Computer and Quantitative Methods in Archaeology, pp. 141-148, Oxford, 1990), expanded on the techniques pioneered in the earlier efforts and offered interactive viewing capabilities, although at a fairly low degree of realism.
Recent efforts in virtual reconstruction have produced increasingly photorealistic results. A detailed model of the Dresden Frauenkirche, destroyed during the Second World War, has been used to create a high-resolution, computer-generated film (Collins, IEEE Computer Graphics and Applications, pp. 13-15, Nov. 1993). Unlike most archaeological sites where the only source of data is the site itself, original plans and photographs of the architecture existed on which to base the model. One of the outstanding features of this reconstruction is the attention paid to the recreation of interior lighting and surface detail, evoking in the viewer a sense of the architectural space. Attention to detail and accurate surface characteristics are also evident in the virtual reconstruction of the Visir tomb in Egypt (Palamidese et al., IEEE Proceedings of Visualization '93, pp. 420-423, Oct. 1993). In both of these cases, the visualization is not the ultimate goal; rather, it is a part of a process leading to the physical reconstruction of the site.
Computer graphics techniques have also been applied to the analysis of archaeological sites. In particular, the ability to visually relate the distribution of data to surface characteristics has proven useful in the identification of spatial correlations. This technique can be extended to include less tangible features such as viewsheds (the regions of a scene not obscured to the viewer by occluding features) and solar paths (the apparent path of the sun for a given date and time with respect to the viewer). While less physical than surface characteristics, these environmental factors are nonetheless important. Analyses of these features have been applied in architecture, and have potential application in archaeology.
The computer modeling approach that we have employed also opens up a new potential for analyzing artifact concentrations that may be related to uneven surfaces, such as wall slopes, shallow depressions in the floor, raised floor areas, and areas of partial roof collapse. A 3D model that includes main structural supports also facilitates the assessment of areas in which activity would be restricted and which might therefore be reserved for other purposes such as storage. Visual display of selected types of artifact concentrations in such a 3D setting make these organizational considerations very easy to recognize.
The Keatley Creek site in British Columbia, Canada, consists of over 100 cultural depressions, several of which have been fully excavated since 1986. The majority of these depressions are the remains of pithouse dwellings that were occupied from 1,100 to over 2,400 years ago. The case we present here describes the virtual reconstruction of one of the largest of these dwellings, with a rim diameter of approximately 19 m.
The tools and techniques used to create this visualization range from the
manual collection of data to the use of commercial software to render some of
the more detailed images. The interactive application described in this paper
is a custom program written by the authors using Silicon Graphics
OpenInventor(TM) toolkit. The pithouse model can be imported from this
application into commercial software available from Alias Research Inc.(TM) in
order to generate higher-quality images at noninteractive rates. Although the
software used to model the pithouse can be used on other platforms, it would
likely run more slowly and produce less realistic results. Archaeologists can
also use other less powerful, but still useful, modeling and rendering software
packages that are commercially available for most computer platforms at
increasingly lower prices.
In constructing the model, the focus was on creating a visual representation
that conveyed the major structural features, yet was simple enough to navigate
at interactive rates on a Silicon Graphics Indigo2 workstation.
Features that were not necessary to convey the basic architectural shape were
intentionally left out, including the intricate roof thatching and other
The floor was reconstructed from survey data by fitting a non-uniform rational b-spline (NURB) surface that interpolated known surface points (Figure 1a). Due to the irregular nature of this surface and its lack of symmetry, the reconstruction of the floor was an unusually complex task given the current software. Beams, posts, and joists were modeled as simple cylinders of varying thickness. Their orientation was based on archaeological data, such as evidence (earth coloration, preserved wood fibers, etc.) indicating the size and placement of both vertical posts on the pithouse floor and major and minor beams pinned to the edge of the structure (Figure 1b). Ethnohistoric descriptions of pithouses were used to amplify the details. Given the archaeological data, an architectural study provided further information indicating the most probable height of the posts and the incline of the beams necessary to maintain structural integrity, as well as the proper width, length and distribution of the joists crossing the beams that support the roof (Figure 1c).
The roof covering of the 3D model was constructed from simple polygonal patches. The vertices of the patches were simply the positions that the posts, beams and joists intersected, offset so that the roof sat upon the underlying structure. In order to achieve more realism, the patches were randomly perturbed and texture mapped; that is, a two-dimensional (2D) image of a thatch-like texture was mapped to the polygons (Figure 1d).
In addition to the major structural components of the pithouse, simple representations of significant interior features are included. In Figure 2, spheres on the pithouse floor indicate hearth locations and black disks denote storage pits.
The user can move in, out, and around the model in real time by steering a virtual camera using a mouse. The standard display is in the first-person (that is, the user's virtual point of view), and the interaction method used does not require the user to change focus away from the model. Allowing the user to move about the structure serves two purposes. First, it provides a convenient method for a user-selected view of the data. Second, it can provide a sense of space within the 3D model and aid in delimiting the bounds of the structure. In addition, an orthographic overview of the floor can be displayed simultaneously.
Spatial data display
Most artifact data at this site were recorded as discrete counts within square
regions, determined using standard archaeological methodology (Spafford,
Master's thesis, Simon Fraser University, Department of Archaeology, 1991). Two
meter-wide strips were drawn vertically and horizontally, sectioning the site.
Each square region was subdivided into 16 numbered subsquares (50 x 50 cm
each), for which artifact counts were recorded.
Conventional 2D distribution plots of the pithouse data display the artifact frequencies using a limited number of shades of gray (the darker the shade, the higher the number) and superimposed over significant interior features such as hearths and storage pits (Figure 3). Although the plots do provide an archaeologist with a tool for analyzing the data, the context is restricted to 2D spatial information. Furthermore, the superimposed shaded squares are opaque and obscure underlying features such as postholes, hearths, and pits. This, along with the limited shades of gray, makes overlaying multiple distributions difficult if not impossible without loss of information. It is also difficult to visualize how artifacts mapped on standard 2D plots may have varied across irregular surfaces such as lower or upper wall slopes (where some artifacts appear to have been stored) versus flat floor areas. In general, 2D plots must omit many types of relevant data and can take painstaking efforts to decipher.
As in the conventional plot, artifact distributions in our model are plotted as discrete counts associated with the excavated subsquares. However, each artifact type is given a unique user-defined color, and its frequency is reflected in the transparency of each colored square; the higher the frequency, the greater the opaqueness. The color of a square is determined by linearly interpolating (blending) the colors of all those artifacts and the pithouse features (terrain, interior, architecture) that coincide with that square, using their associated transparencies. Thus, it is possible to distinguish between several distributions on a single plot. (Unfortunately it is difficult to distinguish between the colors of each distinct artifact distribution once the color image is converted to a greyscale image. The reader is referred to the original article which contains the color images: Peterson et al., IEEE Computer Graphics and Applications, 15(4): 41-46, July 1995.)
Our approach moves even further beyond traditional methods by placing the data in a 3D perspective view of the reconstructed pithouse (Figure 4). This provides the user with a tool for examining the correlation between the artifacts and significant architectural features.
A characteristic feature of pithouse architecture is the central smoke-hole
(Figure 2). This opening at the apex of the roof structure is the only source
of natural light in the dwelling. This limitation suggests the hypothesis that
use of different areas of the floor was subject, at least partially, to the
availability of working light.
A single light source is modeled to approximate the illuminated areas of the house interior. Specifically, the model takes into account the beam of direct sunlight entering the smoke-hole as well as a small amount of sunlight scattered by the atmosphere (Figure 5a). The light source is modeled as a spotlight having a cone-shaped beam whose width, when entering the pithouse, approximates the area of the smoke-hole (Figure 5b). By changing the angle of the cone, the light distribution varies to approximate the direct and scattered sunlight. Figure 5c illustrates, with the aid of fog effects, the cone of light as it enters the 3D pithouse model and illuminates the floor.
The path of the sun is calculated for the site and can be specified to correspond to a given date. The user can then interactively model the daily motion of the light to see its relation to data distributions on the floor surface (Figure 4). While this method does not produce an accurate model of actual sunlight distribution, it does provide a useful approximation for data exploration.
In the case presented here, we have used 3D computer graphics techniques to
integrate the display of spatial data into a model of a prehistoric pithouse.
This has provided us with some valuable insight into the effects that the
structure of a pithouse can have on usage areas within the dwelling. By
viewing artifacts in their original context, we were able to more readily
identify potential relationships between artifact distribution and the pithouse
Artifact data from the Keatley Creek excavation are probably the most detailed and comprehensive ever collected from a site of this type. Various hypotheses exist relating the spatial distribution of these artifact types to areas of the pithouse floor. In order to visualize how particular data distributions relate to the site, distributions were layered onto the surface model of the floor (Figure 4). Using the 3D model has the advantage of making any relationships to surface slope immediately obvious. For example, 3D displays can show when a high density of artifacts is found on a wall slope, which may indicate a probable storage area.
While the analysis of spatial relationships between data and floor position has proved useful, we have found that the most insightful information from this visualization comes as a result of considering the constraints the structure imposes on usage areas. Analysis of an orthographic projection of bifaces, heavily retouched scrapers, and debitage revealed that the bifaces tended to be distributed in the central region close to the hearths, which supports the common assumption that they were associated with food preparation. The scrapers are more concentrated around approximately two-thirds of the perimeter of the floor and on wall slopes, possibly indicating that they were set aside in storage locations near the wall for safe-keeping and later re-use. Debitage, being the by-product of several processes, is scattered throughout the floor with the densest concentrations at the periphery. While not identifiable in the 2D plot, it is readily seen in Figures 2 and 4 that the ceiling height at the edge is quite low, and therefore this area could not have been useful for much more than sleeping, tasks performed while sitting, or storage. These constraints on movement can best be appreciated in a 3D model by the introduction of scale human figures.
Another form of architectural constraint inherent in the pithouse structure is the availability of working light. Since the only source of natural light is the smoke-hole, we have approximated how much of the interior would be lit at a given time of day. By examining both orthographic distributions and 3D views (Figure 4), it is apparent that heavily retouched scrapers are more dominant in the area lit by the midday sun. This would suggest that these artifacts are associated with work requiring some visual acuity.
While the simple lighting model implemented thus far has proven useful, the development of a more accurate model that also accounts for light from multiple hearths would provide a better basis for work area theories. Furthermore, the positioning of human figures of various statistical proportions within the structure would aid in determining whether certain hypothesized tasks could be performed (without colliding with the roof, for example). These figures may also provide a better sense of space within the model. A further enhancement would be to implement the model in a virtual reality (VR) environment. In addition to providing the user with a sense of the architectural space, VR techniques have great potential for integrating the display of data in the context of a computer model.
A much more difficult visualization problem is the perceptual limitations which arise from the combination of multiple artifact distributions, interior features, architectural structures, the terrain model, and sunlight. In our model, the ground elevation provides shading, the beams, posts and roof are texture-mapped, the pits and hearths are colored, the artifact distributions have color and transparency, and the lighting model affects intensity. The challenge is to determine methods of blending such features together such that the information provided by each component is not lost. For example, the lighting model can greatly affect the intensity, and thus perceived transparency, of the squares. One possible solution is the use of glyphs, such as a pie chart for each subsquare that is divided into the various types of artifacts. Another is to divide the subsquare into smaller subregions that are assigned to specific artifact types.
It would be of interest to integrate the visualization model with the underlying data such that it is not just a passive display device, but also a tool for querying data and performing statistical functions. For example, using the mouse to choose a particular subsquare could result in the textual or graphical display of the incident artifacts. Aside from greatly facilitating research and analysis, the creation of interactive, virtual 3D models of prehistoric structures has obvious pedagogical values both for classroom uses, distance educational programs, special presentations, and interpretive centers. These are significant advantages that can more than repay the initial high cost of developing the database.
This work was supported in part by Natural Sciences and Engineering Research
Council (NSERC) research and equipment grants and a post-graduate
Philip Peterson and F. David Fracchia are in the School of Computing Science and Brian Hayden is in the Department of Archaeology at Simon Fraser University, Burnaby, B.C.