Discovering Petra: Archaeological Analysis in VRvis.cs.brown.edu/docs/pdf/Vote-2002-DPA.pdf · a...

In field archaeology, analysis of an excavat-ed region is a meticulous process requiring

exploration at a variety of levels. While the act of exca-vation offers the best way for archaeologists to monitorclose-range details, they can use empirically basedanalysis to analyze only the findings to which they them-

selves have been exposed. With ateam of people excavating, synthe-sizing the wealth of observationfrom one year to the next is difficult.Archaeologists also use quantitativemethods to compare findingsthroughout the site over time to pin-point base trends among recordedartifact typologies.1 Quantitativeapproaches, unfortunately, areoften limited because they lack aspatial component—explicit infor-mation about the physical 3D rela-tionships among the excavatedobjects and the site.

In this article, we chronicle a col-laborative effort (from 1997 to thepresent) with Petra Great Templearchaeologists to augment tradi-tional analysis approaches. We

introduce new archaeological analysis tools that com-bine novel visualization and interaction techniqueswithin a Cave Automatic Virtual Environment (CAVE).2

These tools

� give archaeologists access to formerly inaccessibleparts of the archaeological record;

� support navigation and interaction with virtualtrenches, stratigraphy, artifacts, and architecture; and

� preserve and display the spatial relationships presentbefore excavation.

Using an iterative approach and working with archae-ologists from the Brown University-sponsored excava-tions at Petra, Jordan, we built four successive prototypes,

each of which refined tools from earlier prototypes andadded new tools to help users query, navigate, andexplore site findings in three dimensions and at differentscales. The tools incorporated new visual representationsfor the many data values associated with each artifact tohighlight patterns and anomalies in the record.

Evaluation of the fourth and final prototype showsthat our system provides an environment and tools thatfacilitate empirical analysis by familiarizing team mem-bers with data from many years. It also lets multipleteam members corroborate observations while exam-ining the record together. Finally, it provides a model ofthe excavation site where quantitative results can bevisualized in context with other aspects of the data col-lected on site. The “Related Work” sidebar discussesother researchers’ efforts to develop similar tools.

Petra Great TempleFigure 1 shows Jordan’s Petra Great Temple site,

which was the underlying focus of our archaeologicaltool development. Some of the most important researchquestions archaeologists at Petra ask during excavationand site analysis include these:

� What’s the chronology of the architectural phases atthe Petra Great Temple precinct?

� What was the temple’s function?� What did the Temple look like before it was

destroyed?� During what time period was the temple and its

precinct in use?

Archaeologists establish hypotheses and use a vari-ety of on- and off-site inquiry methods to answer thesequestions, as the “Archaeological Analysis: Current Prac-tice” sidebar (on p. 40) explains. Unfortunately, it isn’talways possible to answer them using traditionalapproaches alone. If given new methods to analyzeaspects of the archaeological record that are currentlydifficult to analyze, archaeologists’ ability to resolve cru-cial questions will be enhanced.

0272-1716/02/$17.00 © 2002 IEEE

Computer Graphics in Art History and Archaeology

38 September/October 2002

New tools give

archaeologists access to

formerly inaccessible parts

of the archaeological record.

The result is a demonstrably

improved model for inquiry

to pose, and answer,

important research

questions.

Eileen Vote, Daniel Acevedo Feliz, David H. Laidlaw, and Martha Sharp JoukowskyBrown University

Discovering Petra:ArchaeologicalAnalysis in VR

IEEE Computer Graphics and Applications 39

Related WorkThe search for new methods to analyze exca-

vation findings began around the time that arch-aeologists standardized data-collection processesfor cataloging field information in the latter part ofthe 20th century. Archaeologists quickly adoptedquantitative methods to analyze their findingsbecause they now had immense amounts ofphysical data. Yet most of these approaches failedto use the 3D components of the archaeologicalrecord for postexcavation analysis.

Early approachesIn the early 1990s, Paul Reilly began working on

techniques for archaeological data visualization toexamine survey data, provide virtual excavations fortraining and evaluation studies, and reconstructand exhibit archaeological findings. He developedGrafland, a simulated excavation tool consisting ofa series of topological layers with various featurescut into them.1 This tool was intended to demon-strate that archaeologists can produce realisticrecords of the data destroyed during excavationand employ improved methods for managing andinterrogating the data for postexcavation analysis.However, because of inaccuracies in how the datawere recorded, Grafland was used primarily as ateaching and simulation tool, not for the specializedresearch tasks archaeologists perform as describedin this article.

CAD and GISMore recently, archaeologists have begun to

integrate physical aspects of the excavation byemploying methods derived from geographicinformation system (GIS) and CAD-basedapplications. Unfortunately, because many ofthese application methods have been adaptedfrom those created for disciplines such asgeography, meteorology, and physics, forexample, they don’t give access to the toolsarchaeologists need to effectively manageconditions peculiar to archaeology.

A number of visualization systems employingimmersive virtual reality (IVR) have attempted touse GIS to handle large data sets such as thosepresented by climatological data and the urbanenvironment. Sandbox, for example, wasdeveloped as a VR tool to let an investigatorvisualize the contents of a climatologic databasewhile retrieving data.2 The tool is significantbecause it gives scientists a means to interact witha variety of data using visual clues. However, itfocuses on completing tasks with a 2D climatologydata set. Although this system gives users a way toexplore and interact with data, the researchproblems addressed differ markedly from the onesarchaeologists face in dealing with data having astrong 3D component.

Karma VI is a VR interface for ESRI’s Spatial

Database Engine that supports powerfulvisualization, manipulation, and editing ofstandard GIS data in a VR environment.3 Users ofthis interface can walk through 3D environments,see planned buildings, and view changes in thelandscape. The tool lets users experience the dataset at close range and access important statistics.However, these IVR applications weren’tdeveloped for archaeological inquiry and thereforedon’t consider the specific research tasksarchaeologists need to perform.

The research reported in this article builds onthe methods developed in GIS and CAD packages.It also incorporates the spatial paradigm employedin recent visualization projects to facilitate theinterrogation of 3D attributes from thearchaeological record.

References1. P. Reilly, “Data Visualization in Archaeology,” IBM Sys-

tems J., vol. 28, no. 4, 1989, pp. 569-570.2. A. Johnson and F. Fotouhi, “The SANDBOX: A Virtual

Reality Interface to Scientific Databases,” Proc. 7th Int’lWorking Conf. Scientific and Statistical Database Man-agement, IEEE Computer Soc. Press, Los Alamitos,Calif., 1994, pp. 12-21.

3. R. Germs et al., “A Multi-view VR Interface for 3D GIS,”Computers & Graphics, vol. 23, no. 4, pp. 497-506.

1 Archaeolo-gists working ina Petra excava-tion trench thatmeasuresapproximately10 meters deep.Findings fromeach sedimentlayer (locus)must be careful-ly indexed foroff-site analysis.Once away fromthe site, archae-ologists typical-ly have no wayto access thespatial relation-ships that areimportant foranalysis.

Phot

o co

urte

sy o

f Bro

wn

Uni

vers

ity/J

ohn

Fora

sté



Archaeological Analysis: CurrentPractice

Archaeologists are concerned withmaintaining accurate records of the materialunearthed during excavation because whenexcavation is completed, the site is essentiallydestroyed (see Figures A1 and A2). Thearchaeological record generally comprises itemsin diverse media such as paper-based records(maps, plans, elevations, sections, anddrawings), photographs of artifacts andarchitectural finds, digitally preserved surveydata, and detailed statistics on artifacts stored ina site database.1 During postexcavationanalyses, archaeologists attempt to synthesizetheir on-site observations with details containedin these records.

Much of the data collected on site isn’t usedduring the final analysis process because thespatial components of individual artifacts,architectural finds, and site features aren’tadequately synthesized. For example, althoughthere may be relationships among the artifactsfound in an adjoining trench, if the two trencheswere excavated separately by differentarchaeologists during successive field seasons,the relationships will likely be missed. Figure Bshows the different loci, or sediment layers, at anexcavation that could represent work done byhalf a dozen different archaeologists over asmany different years. Trends that connect thetrenches can only be identified if the samearchaeologist is involved to visually identify them.In the postexcavation process of analyzing artifacts, theirconnection can be correlated only if their in situ spatialrelationship is understood.

The difficulty in relating findings that are inherently three-dimensionally linked is hard to alleviate using currentmethods. Such methods would require an enormousamount of time to physically correlate artifacts andassociated finds inside a trench with those in adjoiningtrenches.2,3 Archaeologists who want to explore theserelationships must first search for the relevant objects fromspecific locations in the site database. Next, they mustinvestigate the physical information about the trench andtrench loci using the site notebooks. Finally, they must applystrategies to understand the spatial connections among theobjects and trenches with the existing site features recordedin the digital survey. Even if it were possible to understandthe data’s spatial aspect, it would still be difficult to considerthe other attributes (such as shape, color, and decorativemarkings) associated with the data that provide additionalclues.

References1. P. Reilly, “Towards a Virtual Archaeology, Computer Applications

in Archaeology,” Computer Applications and Quantitative Methodsin Archaeology, K. Lockyear and S. Rahtz, eds., British Archaeolo-

gy Reports, Int’l Series 565, Archaeopress, Oxford, UK, 1990, pp.133-139.

2. E. Vote, A New Methodology for Archaeological Analysis: Using Visu-alization and Interaction to Explore Spatial Links in Excavation Data,PhD dissertation, Departments of Old World Art and Archaeolo-gy and Computer Science, Brown Univ., Providence, R.I., 2001,pp. 92-93.

3. D. Acevedo et al., “Archaeological Data Visualization in VR: Analy-sis of Lamp Finds at the Great Temple of Petra, a Case Study,”Proc. IEEE Visualization 2001, IEEE Computer Soc. Press, Los Alami-tos, Calif., 2001, pp. 493-496.

A (1) Aerial view of Petra Great Temple precinct section, preexcavation 1993.After excavations conclude (right), the site is essentially destroyed becausephysical associations among sediment, architectural fragments, and artifactsare lost. Arrow view shows where the Petra Great Temple’s monumentalcolumns were found. The columns were left in place after excavations con-cluded (right). (2) Aerial view of Petra Great Temple precinct section, postex-cavation 2001. Excavations to date span eight field seasons (or years) andhave uncovered an enormous amount of physical evidence. The entire exca-vated region is roughly the size of three football fields and measures approxi-mately 7,560 square meters. The dark circle shows a human’s scale. (Figure A1courtesy of J. Wilson Myers. Figure A2 courtesy of Artemis A.W. Joukowsky.)

(1) (2)

B Section through a trench showing the sediment layers (loci)and in situ architectural findings that must be carefully indexedduring excavation. Artifacts must be recorded by the findlocation in a locus.

Prototype one: A conceptual modelWe created this first prototype to index many of

the architectural fragments unearthed in Petra toinvestigate the research question, What is thechronology of the building phases of the temple andits precinct? We designed the conceptual plan to con-sider a set of variables present in the site databases,along with 3D models of objects, drawings, and pho-tographs for new comparisons and to establish linksamong objects.

Our design goals were motivated by the need to estab-lish a chronology of architectural phases among thePetra Great Temple areas. This complex problem can besolved only by relating small finds such as coins and oillamps with monumental architectural fragments, in situremains, and site features. Currently, the archaeologicalteam attempts to establish building dates for each areaof the temple by comparing observations and physicalevidence collected during excavation.3 At the end ofeach season, the team meets to assign phases to eacharchitectural area by reviewing notes and comparingindividual observations. The process somewhat resem-bles forensics because each site detail is potentially animportant clue.

The fundamental relational aspect of the physical evi-dence challenges archaeologists because their currentmethods can’t rigorously compare evidence from suc-cessive excavation years. Therefore, it’s difficult for themto recall the many relationships in the site record they’veobserved during excavation. Also, because multiplearchaeologists excavate the site and witness importantaspects of the record, synthesizing their observations isinherently problematic.

Design goalsBased on these observations, we established the fol-

lowing design goals for our first prototype:

� Provide a way to relate smaller artifacts such as coinsand oil lamps with their associated architectural frag-ments (see Figure 2).

� Attempt to assign relative dates to three-dimension-ally linked objects.

� Relate evidence from datable stone-cutter markingson the architectural fragments at remote sites to helpassign dates to architectural fragments with similarmarkings at the Petra Great Temple.

System developmentIn 1997, we built a conceptual prototype guided by

our design goals. We posited that if each architecturalfeature were given an exact or relative date based onassociated datable evidence, we could ultimately eval-uate an architectural region by examining the resultingassociations. We specified a framework in which archae-ologists could index architectural fragments with dis-tinguishable stone-cutter markings with other objectssuch as coins (already assigned absolute and relativedates; see Figures 2 and 3 on the next page).4 We alsodeveloped ways to correlate the in situ find locations forobjects with physical characteristics that can provideclues for comparing the objects.

EvaluationWe spent six weeks on site at the Petra Great Temple

in 1998 and received feedback from the team that neces-sitated fundamental changes in our prototype. We deter-mined that automating the process of assigning datesto architectural regions by using datable objects exclu-sively would severely limit our results. Datable objectsare problematic because many of the Temple’s archi-tectural components—such as columns, walls, andfloors—are so badly eroded that they can’t yield ade-quate datable evidence for our relational model.

Prototype two: A test using ageographical information system

After evaluating the first prototype, we concludedthat we couldn’t use the system to synthesize the archi-tectural findings and provide a chronology for the tem-ple’s building phases. However, in the process, teamarchaeologists helped us analyze the temple in terms ofits function, use, and appearance.

Petra has been ravaged by several devastating earth-quakes, so many of the architectural components suchas column capitals and wall friezes have either beendestroyed or displaced.

Archaeologists investigate these questions withempirically based observations. They consult with theexcavation team to compare findings and analyze thestatistics derived from the site databases. However,because the record’s physical components are diverseand difficult to synthesize, archaeologists often rely ontheir memory for details and spatial relationships in for-mulating hypotheses.

Before implementing prototype two, we posited that


2 Examples ofthe artifactsarchaeologistsuse to assigndates for areasof the temple.Top: a jug, an oillamp, a coin,and a sculpturalfragment.Middle: a col-umn base withstone-cuttermarkings and asection of acolumn capitalwith sculpturaldetail. Bottom:small bulk-findpottery and oillampfragments.

Phot

os c

ourt

esy

of A

rtem

is A

.W. J

ouko

wsk

y

we could help archaeologists solve some of theseresearch questions by providing new methods to inves-tigate the parts of the record with which they aren’t wellacquainted. To do this, we used input from the excava-tion team to specify some simple tasks they would need,including these:

� The ability to reinvestigate and become betteracquainted with the range of on-site findings, all

excavators’ discoveries, and data from many years ofexcavating.

� Ways to understand the spatial relationships amongobjects and features within or among excavationtrenches.

� Methods to investigate, correlate, and provide evi-dence for sitewide artifact concentrations.

� The ability to generate evidence about the basic com-position of the sediment removed during excavation.



Artifact (coin)Found directly under Upper Temenos floor paver

FourAttributes:

DimensionsProportions (height/length/width)Stone cutter marks = Type ALocation = BMaterialDate givenResidue (plaster or paint)Color (Munsell value)

Architectural fragment (column base)In situ position on Upper TemenosSite A

OneAttributes:

DimensionsProportions (height/length/width)Stone cutter marks = Type ALocation = AMaterialResidue (plaster or paint)Color (Munsell value)

Site BArchitectural fragment

Attributes:

DimensionsProportions (height/length/width)Stone cutter marks = Type ALocationMaterialResidue (plaster or paint)Color (Munsell value)

Architectural fragment (column base)Fragment found 15 meters from Upper Temenos

TwoAttributes:

DimensionsProportions (height/length/width)Stone cutter marks = Type ALocation = X (off site)MaterialResidue (plaster or paint)Color (Munsell value)

Architectural fragment (floor paver)Upper Temenos

ThreeAttributes:

DimensionsProportions (height/length/width)Stone cutter marks = Type ALocation = BMaterialDate givenResidue (plaster or paint)Color (Munsell value)

Coin date

RomanCirca 116 CE

Stone cutter A

Image recognitionmatches linked artifacts

3 First concep-tual diagramdeveloped in1997. It shows aframework forarchaeologiststo index datableartifacts andother findsbased on theirrelative findlocation.

Design goalsTo give archaeologists a means to

perform these tasks, we establishedthe following design goals:

� Implement tools and features toperform specified research tasks.

� Test some preexisting features incommercial geographic informa-tion system (GIS) software to seeif it provides visualization or inter-action methods adaptable to ourpurposes.

System developmentWe began implementing a second

prototype on a desktop computerusing GIS software called ArcViewby Environmental Systems ResearchInstitute (ESRI). We also used the 3DAnalyst extension, which links theexcavation database to a 3D viewerprogram. We started by integratingthe 3D survey of architecture andsite features with 2D shape informa-tion for approximately 50 trenches.The result resembled a site map withblocks representing trenches, as inFigure 4. When we loaded the sitedatabases with more than 15 artifacttypologies, we were quickly able toindex bulk artifact find concentra-tions by trench. The result was a 2Dvisualization of an artifact typology (such as pottery) bya color scale (white to black). This wasn’t very helpfulbecause it displayed only the bulk concentration of pot-tery in the trench and gave no information about whereinside the trench the pottery was found.

Next, we used the 3D Analyst viewer to look at thepottery ranges inside each trench by locus. Because wedidn’t have digital 3D information about each trenchlayer, we had to approximate the layer thicknesses. Thistime-consuming process yielded a somewhat inaccu-rate representation of the excavation layers. Finally, weused the visualization tools to assign a color range thatrepresented the pottery concentrations in individualloci throughout the site.

EvaluationDuring evaluation, team archaeologists found the

prototype’s 3D visualization difficult to understand. Therendering of this attempt shows a relative spatial distri-bution of pottery in the site trenches that, as Figure 4shows, looks much like a 3D bar chart. The resultingdepiction of the site and trenches gives the misleadingimpression that each trench layer is equal. Additional-ly, when querying in the viewer, we could examine onlyone artifact type at a time (such as coin, pottery, or sculp-tural finds).

Although our test GIS offered a novel interface bywhich a variety of 2D data types could be correlated, itwasn’t sophisticated enough to give a thorough descrip-

tion of height relationships among the spatial entities,nor did it give us adequate tools to examine multipleartifact types together. Finally, since we had only a rota-tion wheel and zoom tool with which to examine thesite, we had trouble referencing occluded objects andremote features within the 3D viewer.

Prototype three: An inquiry modelOur test GIS prototype provided a 3D model with visu-

alization and interaction methods. Team archaeologistsappreciated the ability to visualize and explore the 3Dcomponents of the record but wanted additional fea-tures unavailable in GIS-based software. In response totheir tests, team members described their ideal inquirymodel for the analysis process. The sketch in Figure 5(next page) illustrates how a more complete 3D visual-ization and interaction paradigm might enable archae-ologists to interrogate important physical aspects oftheir site data.

We planned to build prototype three with some datavisualization methods and user interaction and naviga-tion tools to simulate the natural inquiry process occur-ring when the archaeologist is in an excavation trenchor moving about the site.

We wanted a 3D visualization interface because itwould let us add realistic visualization features andinteraction tools. Moreover, it would provide a way toimplement a range of inquiry methods in an intuitiveuser interface. We posited that the immersive VR inter-


4 Renderingfrom the 3DAnalyst viewer.The trenchlayers (loci) arerepresented asboxes. Colorranges indicatethe relativeconcentrationsof pottery findsin the variouslayers.

face offered by a CAVE would be an optimal environ-ment in which to perform these inquiries. An IVR inter-face is particularly well suited to archaeologicalvisualization and interaction tasks because it providesthe visual range that “... help(s) provide situationalawareness and context, aid(s) spatial judgments andenhance(s) navigation and locomotion.”5

Recently, researchers have attempted to compare theresults that can be achieved in immersive versus nonim-mersive systems.6 However, most studies admit that westill lack an adequate evaluation of IVR’s specific capa-bilities. To date, the consensus is that certain specific tasks(that is, navigation and object manipulation) can be com-pleted more comfortably in an IVR environment; othertasks are more effective in fishtank VR systems, using astereo-capable desktop display with head tracking.7

Design goalsWe began the design process for our third prototype

by specifying four categories of tools that archaeologistsneed:

� Specialized interface: to use immersive VR in a CAVEto facilitate exploration and analysis tasks.

� Visualization tools: to search for artifact concentra-tions, differentiate among artifact types, and isolateanomalous conditions in a visual field with multipleattributes.

� Navigation/interaction tools: to let users interact withthe 3D findings.

� Examination/inquiry tools: to observe the data at mul-tiple scales (both at close range inside a trench andwith overviews of the whole site).

System developmentBecause we intended this prototype to let archaeolo-

gists view and interact with a variety of entities simul-taneously, we presented the data entity types withgraphical rendering methods. We categorized the datatypes as follows:

� Site features and architecture: We provided a realistic3D reconstruction of the Petra Great Temple as a basecontext/interface in the CAVE at life-size scale, as Fig-ure 6 shows.

� Trenches and loci: We integrated some excavationtrenches in a concentrated region of the temple. Eachtrench is broken down into a sequence of layers (exca-vated loci), as Figure 7 shows.

� Artifacts: Users send queries to the site findings data-base. Local concentrations of specific artifact typesappear when each locus becomes a solid color (forexample, white indicates minimum concentration,dark red indicates maximum concentration). Whentwo artifact queries go to the database, the resultsappear in combination. Color indicates the first arti-fact concentration, and texture indicates the secondartifact concentration, as Figure 7 shows.

� Special artifacts: The most significant artifacts suchas a sculptural mask can be displayed in their 3D find


5 Sketch of aninquiry modelthat integratesnovel visualiza-tion methods tofacilitate theanalytic tasksarchaeologistswant toperform.


locations inside the trenches.Users access them by turning thetrenches off or dimming existingqueries to see inside loci.

We also developed two sets oftools in this prototype. First, wedeveloped tools for interaction andinquiry, with which new elements(trenches and special finds) can beadded to the scene and queries sentwith button interactions on the com-puter mouse. Second, we developedtools for navigation, with whichusers move in the IVR environmentby using a tracked wand equippedwith a trackball. The trackball letsusers “walk” (move freely on theground plane) or “fly” (move freely,unconnected to the ground plane).

EvaluationImplementing this prototype in a

CAVE provided archaeologists witha realistic inquiry environment sim-ilar to the one they establish on sitebut without physical impediments such as dirt andheavy objects. The team appreciated the ability to reex-perience the site at life-size scale and observe artifactsin their in situ locations. However, team members stillhad difficulty performing research tasks because theylacked refined, flexible visualization and interactiontechniques.

Over a six-month period, we collected feedback fromarchaeologists on the general problems they encounteredwith this prototype. They felt that the temple recon-struction wasn’t an accurate research context to exploreexcavation features because it didn’t represent their rawfind data and was visually distracting. They wanted toexperience the site at life-size scale but also wanted to“shrink” it so that they could look for features commonto different areas. Finally, artifact concentrations were

difficult to observe because the trenches occlude oneanother, and the method of integrating multiple artifacttypes with color and texture was too visually complex.

Prototype four: RefinementsIn our final prototype, we integrated essential refine-

ments so that two team archaeologists could evaluate itby investigating specific research questions.

Design goalsTo refine the visualization and interaction methods

in prototype three, we created a wish list of changes forour final prototype:

� Create an accurate research context with only in situsite findings, as Figures 8 and 9 (next page) show.


6 Users navigating in the life-size Petra Great Templereconstruction that serves as a base context for explor-ing site findings in prototype three. We built this modelwith data from a digital survey compiled from severalsources.

7 User interacting with an excavation trench in proto-type three. Each locus is expressed as a block of sedi-ment. Pottery concentrations are plotted as colorranges (white equals low, red equals high). Bone con-centrations are expressed as texture (loose equals low,dense equals high).

8 The coloredregions on theleft are theindividual loci(from trench 24at right) thatrepresent thedebris removedin a specificregion of thesite. Each locusrepresents alayer of sedi-ment, an archi-tectural feature(column, wall,rock, and so on)or an importantartifact.

� Develop ways to navigate through the model andchange scales.

� Find better ways to look at site features and synthesize

findings, for instance, to look at multiple artifacttypologies together and understand patterns oranomalous conditions.

In planning improvements, we incorporated these threeitems and applied perceptual rules for observing andinteracting with the site record.

System developmentOur development dealt with both new visualization

and interaction tools.

Visualization. The visual perception literature out-lines general rules for building visualizations thatexploit strengths of the human visual system. To miti-gate the complexity of our earlier visualization, we gaveusers some visual entities that they can recognize auto-matically prior to conscious attention:8

� Lightness: We approached the problem of visualizingthe site contents (such as architectural finds, trench-es, and loci) by composing all the elements in 3D lay-ers and then adjusting their individual properties(color, size, and shape) for users to isolate significantfeatures. For example, we removed the image-maptexture and rendered in situ findings in desaturatedgrayish tones. Then we chose fairly saturated colorswith high lightness values for the excavated features(trenches and loci) and artifact typologies (bulk con-centrations and special finds). As a result, the base insitu finds now contrast dramatically with excavatedfeatures, as Figure 10 shows.

� Shape and size: To facilitate discrimination betweenbulk and special finds, we modified their shape andsize. Modifications included, for example, large tetra-hedra (lamp finds) and hexagonal prisms—coin findsnow stand out visually from small tetrahedra (bulkconcentration finds).

� Hue: The symbolic use of color lets archaeologistsquickly identify bulk finds (red indicates pottery, greenindicates bone, and so on).9 Special finds are alsocolor-coded to reflect their cultural origin (blue indi-cates Roman, gold indicates Byzantine, for instance);they stand out in the context of bulk finds becausethey’re twice the size. A key to these values is project-ed on the left wall of the CAVE for easy reference.

New interaction tools. Along with the need torefine the system visually, we had to improve users’ phys-ical interaction with it. Research shows that an unde-tectable and unobtrusive user interface is important tousers in completing a task.10,11 As a result, we developedthe following tools:

� Introductory portal: We exploited the CAVE’s immer-sion to introduce the site and acquaint users with thecontext and tools. They enter the site through anintroductory portal, a room with rusticated walls anda map projected onto the floor that helps orient themto the excavated context, as Figure 11 shows.

� Miniature model: After looking at the site map, usersare introduced to the in situ and excavated remains in


9 Three-dimensional model with in situ site and architectural remains(gray) and 17 test trenches (multiple colors). This model provides an accu-rate context for archaeological inquiry.

10 User examining the entire Upper Temenos region of the temple withtrenches (semitransparent) and in situ finds. The user can easily pick outhigh concentrations of pottery (red), bone (green), stone (purple), andmetal (blue).


the context of a miniature 3Dmodel,12 as Figure 10 shows. Usersinitiate exploration by moving ared block representing the portalto the area of interest. They’reautomatically relocated to thatposition in a full-scale model formore detailed exploration. Ifthey’re performing queries, theycan return to the miniature modelat any time for a synthesis of glob-al site features (such as trench datafrom opposite sides of the site).

� Interactions: In this mode, userscan begin moving and interactingwith excavation data via a wandand a pinch glove. The wand isequipped with a trackball to move(walk or fly), select, and turnobjects on and off. Users can weara tracked-pinch glove to access a3D virtual widget.13 Various handgestures let users choose relevantartifact types—such as bone, shell, and pottery—byrotating the wheel and picking associated colors.

Evaluation of new methodsWe originally hypothesized that we could help

archaeologists answer key research questions by giv-ing them new methods for examining and synthesiz-ing the record. To test this hypothesis, we asked ourteam archaeologists to use prototype four to investi-gate specific research questions. We performed twouser evaluations to observe the archaeologists anddetermine what sorts of analyses they could performwith the prototype.

The first test, with two team archaeologists, focusedon evaluating the users’ general observations in look-ing at the site data. In the second test, two other archae-ologists (whom we will refer to as users A and B) usedthe prototype in the CAVE and were prompted to per-form tasks to explore their own research questions.

The following discussion focuses not on the testingprocess but on observations made in the latter test withusers A and B.

PreparationBefore beginning the test, users A and B were asked

general questions about their own research. We want-ed to identify the research questions they wouldexplore during the test and suggest that they focus onthose that could realistically be answered given thecurrent state of the prototype.

User A wanted to identify when the temple was in useby assigning dates to her lamp finds. To do this, sheplanned to examine relationships among lamp and coinfinds in different trenches. User B aimed to identify thetemple’s function and its relationship to the adjoiningsite, which she is excavating for her dissertationresearch. She had excavated a few significant trenchesat the temple but wanted to observe some trenches shedidn’t excavate personally.

Observations and discussionWhen we encouraged archaeologists A and B to use

the system, they performed three different types of tasks.

Perform queries with the site informationand formulate and explore hypotheses. In theprocess of making queries to observe the architecturalremains with special and bulk finds, users A and B beganto form personal hypotheses about what they saw.

User B became interested in some of the site’s metalfinds. While looking at bulk finds in combination, shehypothesized that the metal fragments correspondedwith the frame of a door close by. If the original wooddoor had disintegrated, its metal hardware would remain.

Although it’s fairly easy to find high concentrationsof metal using the site database alone, it isn’t easy toassociate a specific layer and its architectural compo-nent. When user B queried the database for metal findsand observed them in the IVR environment, she was sur-prised to find that all the metal in the western aisle wasat ground level. Additionally, she observed that themetal in the western aisle was aligned with the door-frame on the west side.

She also posited that another cache of metal found inthe lower levels of trench 47 (in front of the theatron ortheater) could have come from the hardware thatremained after old wooden banisters lining the theatroncirculation routes disintegrated. However, user B wasunable to confirm this hypothesis because the metalobjects she found didn’t have additional attribute infor-mation such as shape or function (currently accessiblefrom the site database but not represented in this visu-alization attempt). In this case, integrating more phys-ical attributes from the database for these objects willlet her investigate this hypothesis further.

Investigate site data and find patterns andanomalies. User A wanted to consider the oil lampfind locations in relation to other relevant and datable


11 Users enter the site through an introductory portal, a room with rusti-cated walls and a map projected onto the floor.

objects, such as coins and pottery. In her research, shehadn’t yet compared the lamp finds with spatially relat-ed coins because this would be too laborious with exist-ing methods.

User A used the visual query tools to view the lampsin their find locations (see Figure 12a). Next, she queriedthe coin finds to observe them together. After queryingthe lamp finds from the Upper Temenos region, she wasable to isolate a cache of Byzantine lamps in the middleloci of trench 29, bordering trench 45 in the western cor-ridor (see Figure 12b). This finding suggests that theremight have been activity in that area during the Byzan-tine occupation of the site. Because she hadn’t person-ally excavated trenches 29 and 45, she was neither

familiar with these lamp find locations nor aware thatByzantine lamps were the only kind found in that gen-eral area. This observation became a particularly strik-ing curiosity and perhaps a vital clue regarding site useduring the Byzantine period.

Confirm on-site findings and proof forhypotheses. User A found several areas with mixeddeposits by looking at the trenches and their associatedfinds along with the site features. (During excavations,it was suspected that the heavy annual rains and earth-quakes that ravaged the site disturbed the sediment lay-ers covering the building and its environs. Consequently,the stratigraphic sequence became mixed, making it dif-ficult to determine the relationships among varioustrenches, loci, and artifacts.) For example, after observ-ing the placement of coins and lamps together, user Aposited that the finds at the Petra Great Temple precinctmight be physically related to the Nabataean Az Zantur,a domestic excavation site to the immediate south. Thatis, heavy rains might have washed surface objects downinto the temple precinct. This supposition might explainthe seemingly random distribution of objects from dif-ferent cultural periods located on the surface layers ofmost trenches.

In interacting with a range of data from the PetraGreat Temple site, user B derived initial proof to supporther hypothesis regarding metal findings. The metalobjects she located close to the ground near a door framecould certainly have been used for door hardware. Thisresult would have been extremely difficult to investigateusing traditional approaches alone because the artifactswere likely removed from the ground without this obser-vation. This method of inquiry allows their relationshipwith the find context to be preserved.

These findings help provide tangible proof for empir-ically derived excavation results because archaeolo-gists now have additional documentation (in the formof screen shots and renderings from observations madein the CAVE) of results. They’re also significant becausethey confirm some of the archaeologists’ longstandingsuspicions about the sedimentary levels throughoutthe site.

DiscussionIn our evaluations of prototype four, we observed that

the two team archaeologists initiated queries with thenew interaction features and by navigating freely usingthe mouse and pinch glove. In addition, the improvedvisualization methods helped them identify importantpatterns and anomalies in a visual field with manydiverse elements.

The experience essentially changed their perceptionof the site by exposing them to a much wider range ofphysical data and acquainting them with site areas, arti-facts, and features previously unfamiliar to them orexcavated by other team members. Their observationslet them check existing hypotheses and formulate newones with physical evidence not otherwise accessible.They also commented that the application will helpthem share findings with colleagues from related sites byproviding tangible evidence for their hypotheses.



(a)

(b)

12 (a) User examining lamp and coin finds in the miniature model. Lampsare tetrahedra and coins are hexagonal prisms. These shapes are easy toidentify because they’re quite a bit larger than the clustered bulk finds inFigure 7. Here the colors indicate the cultural period of the objects (blueequals Roman, gold equals Byzantine, red equals Nabataean, white equalsunknown). (b) User examining a cache of Byzantine lamps (large yellowtetrahedra) in trench 29, in the western aisle. The small green (bone) andblue (metal) tetrahedra represent bulk artifact finds.

At the beginning of our collaboration, team archae-ologists had expressed interest in new techniques to helpanswer the difficult research questions about the PetraGreat Temple site. Traditional methods helped them inperforming some research tasks but prevented the col-lection of evidence sufficient to substantiate hypothe-ses about the larger research questions. We believe thatthey will soon be able to generate this evidence usingadditional excavation features, not present in the fourthprototype, to describe a greater number of attributesfrom the excavation record, for example, specific phys-ical details of smaller artifacts such as pottery and boneand more precise in situ location. We are currently devel-oping features to provide greater amounts of artifactdetail with additional visualization tools linked to thesite database; however, specific information about theartifacts’ in situ location must be collected on site. In thefuture, with slight adjustments to current field record-ing methods (to integrate more 3D attributes), we canimprove the capabilities of our current prototype.

We’re encouraged by the fact that, after evaluatingthe fourth prototype, team archaeologists became con-vinced of the value of modifying their data-recordingmethods and are working with us to integrate new stan-dards for digital recovery and record keeping.

ConclusionsIn this body of research, we created an archaeology-

specific application to augment the analysis methodsemployed by the Petra Great Temple excavation team.To accomplish this, we built four prototypes that addressarchaeologists’ important research questions and spec-ified some analysis tasks they need to perform in orderto answer them. Through an iterative process, we addeda series of tools and techniques to help them interrogateaspects of the site record and perform the research tasks.

Our final prototype four provides a greatly improvedmodel for inquiry. According to our team archaeologists,in evaluating prototype four they could accomplishmany of the research tasks we specified. For example,in exploring the site findings they had been exposed towhile excavating, they were able to acquire proof forhypotheses they established on site. They were also ableto explore site areas they had not personally excavatedand so were able to formulate new hypotheses usingdata derived from those areas.

The archaeologists used the tools in prototype four toquery, navigate, and explore the Petra Great Temple siteas if it had never been excavated and the sediment stillremained. The prototype application also provided avenue for collaboration among multiple archaeologistsand the possibility of sharing data among remote sites.Although many observations are best made with tangibleevidence derived from the on-site excavation, ourapproach can play a significant complementary role in theentire archaeological inquiry and analysis process. �

AcknowledgmentsThis work was partially supported by the National Sci-

ence Foundation (BCS-9980091, CCR-0086065, EIA-9724347, and EIA-8920219). The opinions expressed inthis article are those of the authors and don’t necessar-

ily reflect the opinions of the NSF. We would also like tothank the entire Petra Great Temple excavation team,the SHAPE Lab at Brown University, Frederic Leymarie,Katrina Avery, and Bill Kim.

References1. P. Delicado, “Statistics in Archaeology: New Directions,”

New Techniques for Old Times, Computer Applications andQuantitative Methods in Archaeology (CAA 98), J.A.Barceló, I. Briz, and A. Vila, eds., British ArchaeologyReports Int’l Series 757, Archaeopress, Oxford, UK, 1999,pp. 29-37.

2. C. Cruz-Neira, D.J. Sandin, and T. DeFanti, “Surround-Screen Projection-Based Virtual Reality: The Design andImplementation of the CAVE,” Computer Graphics (Proc.Siggraph 93), vol. 27, 1993, pp. 135-142.

3. M.S. Joukowsky, Petra Great Temple: Volume I: Brown Uni-versity Excavations 1993–1997, E.A. Johnson Co., East Prov-idence, R.I., 1998, p. 241.

4. J. McKenzie, The Architecture of Petra, Oxford UniversityPress, New York, 1990.

5. A. van Dam et al., “Immersive VR for Scientific Visualiza-tion: A Progress Report,” IEEE Computer Graphics andApplications, special issue on virtual reality, vol. 20, no. 6,Nov./Dec. 2000, p. 27.

6. R. Pausch, D.R. Proffitt, and G. Williams, “QuantifyingImmersion in Virtual Reality,” Proc. Siggraph 97, AnnualConf. Series, ACM Press, New York, 1997, pp. 13-18.

7. C. Ware, K. Arthur, and K.S. Booth, “Fish Tank Virtual Real-ity,” Proc. INTERCHI 93 Conf., S. Ashlund et al., eds., ACMPress, New York, 1993, pp. 37-42.

8. C.G. Healey and J.T. Enns, “Large Datasets at a Glance:Combining Textures and Colors in Scientific Visualization,”IEEE Trans. Visualization and Computer Graphics, vol. 5,no. 2, Apr. 1999, pp. 145-167.

9. M. D’Zmura, “Color in Visual Search,” Vision Research, vol.31, no. 6, 1991, pp. 951-966.

10. J. Raskin, The Humane Interface, Addison Wesley Long-man, Reading, Mass., 2000.

11. B. Schneiderman, Designing the User Interface: Strategiesfor Effective Computer Interaction, 2nd ed., Addison-Wesley,Reading, Mass., 1992.

12. R. Stoakley, M.J. Conway, and R. Pausch, “Virtual Realityon a WIM: Interactive Worlds in Miniature,” Proc. HumanFactors and Computing Systems (CHI 95), ACM Press, NewYork, 1995, pp. 265-272.

13. D. Conner et al., “Three-Dimensional Widgets,” ComputerGraphics (1992 Symp. Interactive 3D Graphics), vol. 25,no. 2, 1992, pp. 183-188.

Eileen Vote is a postdoctoral fellowin the Computer Science Departmentat Brown University. Her researchinterests include scientific visualiza-tion in VR environments, 3D userinterface design, and art-based visu-alization of complex data sets. She

received an interdisciplinary PhD with the Departmentsof Archaeology and Computer Science and an MA in the


Department of Art and Architectural History at BrownUniversity. She also holds a BA in architecture from RiceUniversity.

Daniel Acevedo Feliz is a PhDstudent in computer science at BrownUniversity. His research interestsinclude human–computer interac-tion in VR, scientific visualization,3D modeling, and intelligent inter-faces. He received a BS in civil engi-

neering from the University of A Coruña (Spain) and anMS in computer science from Brown University.

David H. Laidlaw is the StephenRoberts Assistant Professor in theComputer Science Department atBrown University. His research cen-ters on applications of visualization,modeling, computer graphics, andcomputer science to other scientific

disciplines. He received a PhD in computer science from the

California Institute of Technology, where he also did post-doctoral work in the Division of Biology.

Martha Sharp Joukowsky is aprofessor, Center for Old WorldArchaeology and Art and Departmentof Anthropology, at Brown Universi-ty and is a veteran of more than 30years as a field archaeologist special-izing in the Near East. Since 1992, she

has directed the Brown University excavations at the PetraGreat Temple in Petra in the Hashemite Kingdom of Jor-dan. She received a PhD in archaeology from the Sorbonneafter attending Brown University as an undergraduate.

Readers may contact Eileen Vote at the Dept. of Com-puter Science, Brown University, Box 1910, Providence, RI02912, email [email protected].

For further information on this or any other computingtopic, please visit our Digital Library at http://comput-er.org/publications/dlib.



January–March: Advances in MultimediaMultimedia means different things to differentcommunities. For researchers, it might bedatabases, search engines, or indexing tools,whereas content providers might be moreconcerned with streaming audio and video,compression techniques, and contentdistribution methods. This issue will offervarious viewpoints, practices, evolvingstandards, and innovative projects inmultimedia.

April–June: Content-Based MultimediaIndexing and Retrieval

Important research areas in multimediaindexing include audio, video, image, textual,and information retrieval. This special issue willcover the state of the art in multimediaindexing, especially image indexing, videoindexing, user access and annotation,description of semantic content, andapplications.

July–September: Multimedia R&DMultimedia systems and applications involve abroad range of topics, including hardware andsoftware for media compression, mediastorage/transport, data modeling, andabstractions to embedded multimedia inapplication programs. Even with this widecoverage, multimedia is still spreading itsinfluence to nontraditional professional sections.

October–December: Multimedia TrendsLearn more about the latest trends inmultimedia and explore what researchers aredeveloping for the next generation ofmultimedia applications. Find out whatpractitioners have learned in the field and whatthey plan to do next to improve the form andfunction of tomorrow’s multimedia.

2002 Editorial Calendar

Discovering Petra: Archaeological Analysis in VRvis.cs.brown.edu/docs/pdf/Vote-2002-DPA.pdf · a...

Documents

Transcript of Discovering Petra: Archaeological Analysis in VRvis.cs.brown.edu/docs/pdf/Vote-2002-DPA.pdf · a...