Stratigraphie Modeling And 3D Spatial Analysis Using Photogrammetry And .

Stratigraphie Modeling and 3D Spatial Analysis Using Photogrammetry and Octree Spatial Decompositionto what has always been the textbook standard of archaeological reporting.To qualify as "painless," a data collection technologymust not only be simple, reliably accurate, and efficient touse under field conditions, but also affordable within thebudgetary constraints faced by the average archaeologicalproject, including projects based in developing countriesand graduate student dissertation research. Otherwise,technology will not be of much help in counteracting thedangerous tendencies discussed in the introduction. Onpurely technical grounds, 3D laser scanning clearly hasgreat potential as a stratigraphie recording technology, butits sticker price puts it out of reach of all but the wealthiestarchaeological projects and institutions. Equipment priceswill have to come down by at least 90% before 3D laserscanning is ready to go mainstream in archaeology, andthat is not likely to happen any time soon. In the meantime,oblique photogrammetry offers a productive and costeffective alternative with a spatial resolution that, whilesignificantly lower than that of 3D laser scanning, issufficient for most archaeological purposes and far superiorto traditional profile drawings and spot elevations. Whilephotogrammetry has been used in archaeology for quitesome time (e.g., Anderson 1982; Fussell 1982; Grün et al.2003; Meyera et al. 2006; RSuuther 1998; Sauerbier andGrün 2003; and numerous CIPA papers (see http://cipa.icomos.org/)), to our knowledge it has not been previouslyemployed as a systematic and comprehensive stratigraphierecording device.features may require multiple outlines on each surface theypenetrate. A 3D model is subsequently composed from theseslices, using a "tomographic" approach not unlike computertomography in medical imaging.As a spatial data collection task, stratigraphie recording has high accuracy requirements, particularly inthe vertical dimension, which is the hardest to measurewith accuracy. Layers of less than 1 cm in thickness mayrepresent meaningful stratigraphie events and thereforeneed to be distinguished. At the same time, systematic andcomprehensive interface mapping of entire sites produces asubstantial volume of spatial data, typically in the hundredsof interfaces and hundreds or thousands of polygon featureseach season. For example, the modest-sized excavations(ca. 270 m ) at the Late Intermediate Period site of Huayurion the south coast of Peru (Siveroni et al. 2004) so far haveproduced over 1,300 interface maps (Figure 1). Huayuriis a dense residential site with a complex sequence ofsuperimposed occupations that have left a multitude of oftenintricately nested features posing a significant challenge tocomprehensive three-dimensional recording (Figure 2).2.2 Oblique Digital PhotogrammetryOblique digital photogrammetry has proven itself capableof coping with this challenge, both in terms of accuracyand fieldwork efficiency, generating 2.5D surface models,polygon features, and orthophotos in one quick and simpleprocess. 2.5D surface models are extracted from sets of2.1 Contact TopographyAlthough we ultimatelywant to build solidvolume models, the mosteconomical way to do so,both from the fieldworkand software perspectives,is surface-based contacttopography, i.e., microtopographic maps of theupper interfaces of all unitsof stratification. Standard2.50 GIS software isable to create and displaythe interfaces; specialtysoftwaresubsequentlycreates solid volumessandwiched in between.Cut features, such as pits,postholes, etc., are treatedthe same way as layers.Their top interfaces areoutlined and mapped onthe surface from whichthey were cut; their bottominterfaces are mapped afterthe features have beenexcavated. Deep and narrowFigure I. Stratigraphie features at the Late Intermediate Period site of Huayuri, Santa Cruz Valley,south coast of Peru. Some 1,300 stratigraphie units have heen recorded in an excavated area of ca.270 m'.259

Hartmut Tschauner and Viviana Siveroni Salinas. 'Figure 2. Abundance of cut features in C3R19 at Huayuri.Photogrammetric recording has proven quite capable of handlingsituations as complex as this one.Figure 3. Typical camera pos it ions used in recording a stratigraphieinterface. The eight positions virtually assure that all significantpoints will show up on at least two, usually on three or more,images with strong (i.e., near right) angles between them.ground-referenced, oblique digital photographs takenfromseveral angles. Photogrammetry software designedfor reverse engineering, forensics and accident-scene260reconstruction, and architectural applications is relativelyinexpensive and appropriate for the scale of most excavationunits. The following discussion is based on my experienceswith PhotoModeler Pro 5.2 from EOS Systems (cf. Greenet al. 2002).Photographs to be processed in PhotoModeler maybe taken with standard digital cameras. These need to beindividually calibrated, but calibration for small to mediumsized target areas is a simple process that can be performedin-house. A standardized dot pattern known to the softwareis photographed from various angles and by resolvingthe photogrammetric equations in reverse, any deviationsfrom the known measurements of the pattern are used todetermine the relevant parameters of the camera-lenssystem. This process needs to be repeated for every camera,lens, and zoom level/focal distance. It is also advisable tocreate separate calibrations for different sizes of areas to bemeasured in the field, as the size of the calibration pattemought to be roughly the same as that of the objects to bemeasured. This imposes a practical limit on the size of areathat can be measured. Although the calibration dot pattemmay be projected at any size against a screen or clean wall,reasonably controlled lighting conditions will only beavailable indoors and rich collections of graffiti may makeit hard to find a suitably clean wall on a university campus.However, this problem is more apparent than real. Givenavailable camera resolutions, the level of detail desired offield photography, and the limited distances from the targetattainable through ladders or photo towers, it is preferableto cover larger excavation units with multiple, partiallyoverlapping sets of photographs. These sets are easilystitched together when processing the imagery, and theextra time for field photography is minimal.In the field, photographs for photogrammetricstratigraphie recording are best taken at the end of astandard field photography session for a newly exposedinterface, adding just a couple of extra minutes to thesession. Additional special photogrammetry sessions maybe required as individual features are exposed. Taking thesepictures hardly differs from standard archaeological fieldphotography. The camera may be hand held or mounted ona tripod or pole. Photographs are taken at a vertical angleof approximately 45 ; precise control of this angle is notrequired. All photographs in a set need to be taken with thesame lens, focal distance/zoom level, and from roughly thesame distance. For this reason, fixed lenses are safer thanzoom lenses, and autofocus must be turned off. For maximumaccuracy, the area of interest should cover as much of thecamera's field of vision as possible. To accomplish opfimalcoverage, it may be necessary to switch between landscapeand portrait modes (or anything in between) within the sameset of photographs. This will not cause any problems inprocessing the image sets.Every significant point needs to show up on at leasttwo ideally epipolar images, i.e., photographs taken frompositions at right angles to each other with respect to thetarget, which provide the strongest basis for measuring 3Dpositions. Two photographs are the minimum to determinea point's 3D position. A third image will add a redundantmeasurement that is helpfiil in detecting blunders. Field

Stratigraphie Modeling and 3D Spatial Analysis Using Photogrammetry and Oetree Spatial Decompositionexperience has shown that taking pictures froin eightpositions around a unit is just the right level of redundancy(Figure 3). It takes only a couple of minutes and virtuallyguarantees that even at the bottom of an excavation unit allrelevant points will show up on at least two near-epipolarimages.2.3 TargetsThe key to a successfully employing photogrammetry as astratigraphie recording tool is the placement of standardized,high-contrast targets on the surfaces to be recorded. Artificialtargets arc crucial because archaeological interfaces tend tobe of irregular shapes and seldom offer any sharply definedpoints that might be easily and precisely matched on severalphotographs. Moreover, assuming a reasonable amountof background noise, PhotoModeler can automaticallyrecognize and reference such targets (particularly circularones) on sets of photographs of the same scene with anaccuracy of less than one pixel, which is not achievableby a human operator marking "natural" points. Since theidentification of circular targets is based on a sphericityindex, the targets need to have a minimum diameter of 3-5pixels on each image, thus requiring different target sizesfor different excavation areas.The density of target placement determines the spatialresolution of the resulting model. At Huayuri, we place about500 targets on each interface of 6 to 10 m in 2D surfacearea (Figure 4). Targets are arranged to form a dense, moreor less regular grid, with additional ones placed whereversurface detail needs to be worked out. This density isevidently a far cry from point clouds produced by 3D laserscanning, but the resulting models are quite detailed andeven aesthetically pleasing without any manual touch-up.Extremely accidented surfaces covered with rubble or wallfall, particularly if there are vertical rock faces, may be theone exception from this rule. However, it is possible to sticktargets onto the vertical faces using reusable adhesive orpins (for soft materials).In addition to providing surface mass points, targets alsoserve to outline features on a surface, such as the mouthof a pit, a lens, or an artifact concentration (Figure 4). Theresulting points are later easily connected—not unlikechildren's cormect-the-dots coloring books—to form 2.5Dpolygons that represent the features in the excavation'sspatial database. A volume model may subsequently beconstructed from several such slices through the sameFigure 4. Circular targets placed on a stratigraphie interface at Huayuri. Here, 516 targets have been placed in an area of about 6.5m . The density, and hence the resolution of the model, could he increased with little extra effort. Feature edges have been outlinedwith circular targets. The resulting 3D points are easily connected to form 2.5D polygons in a GIS database. A full 3D model may becomposedfrom multiple such polygon slices though a feature that may be difficult to capture otherwise, for example a narrow posthole.Some images in a set may he taken with circular targets removed and only coded targets left on the interface. These images are used toproduce orthophotos for publication (cf. Figure 9).261

Hartmut Tschauner and Viviana Siveroni Salinasfeature photographed on successive surfaces cut by thatfeature. This approach is particularly useful for narrowfeatures such as pestholes, as it may be impossible to placeand photograph targets on their bottom interfaces.The targets placed on the interfaces fall into twocategories: (1) at least 6-10, relatively large, coded targets,evenly distributed across the image area and used fororienting the photographs, i.e., determining the camerapositions from which the pictures were taken (Figure 4);and (2) targets for automatic marking that produce the masspoints defining the surface models (Figure 4). Coded targetsfor orienting the images need to show up on every image ina set. Referencing them is either a somewhat tedious manualprocess of marking matching targets on multiple imagesor, if standardized targets are used, may be automaticallyperformed by the software. For automatically marked masspoints, we use circular, retro-reflective targets mountedon chips of heavy plastic material or—at sites with strongwinds—heavy, magnetic rubber. Coded targets are mountedon similar, if larger chips. These may, of course, also be usedfor mass points, but this is not recommended since codedtargets are substantially more expensive than the circularones and their larger area effectively limits the density oftarget placement. We produce the target chips ourselvesfrom commercially supplied rolls of adhesive target tape.Depending on field conditions and care in handling, thesechips may survive multiple field seasons. A more elegantand 100% wind-proof alternative to plastic targets is a targetprojector, but these devices are costly and do not operate onbattery power; thus, a power outlet or generator is neededin the field.based on brightness differences (as well as target shape) andit will be impossible to find a contrast setting that will notblur the target boundaries in some part of an image. Noisybackgrounds with objects whose shapes mimic that of thetargets (e.g., pebbles) may have a similar effect.In these cases, automatic target marking may beperformed in separate runs for sections of similar contrastwithin an image, but sometimes manual target marking maybe inevitable. This can be somewhat tedious, but at leastit does not add to the field time required for stratigraphierecording, and manual marking unfailingly works. Evidently,this problem is easily prevented by using an awning (Figure6) or roof, which many excavations in hot areas will haveanyway. Excessive heat may also deform or even meltthe plastic material of the targets. In hot climates, targetexposure to the sun should therefore be kept to a minimum,and the targets must be stored in the shade.Wind may either move the targets or cover them withdust or sand. At the bottom of an excavation pit, however,this problem is much less acute than might be expected.Where the wind is too strong, targets mounted on magneticrubber or even heavier materials will be the answer. Despite2.4 GeoreferencingPhotogrammetric measurements are entirely relative; thus,it is necessary to shoot in at least three coded targets witha total station, preferably more for backup and accuracychecks. PhotoModeler will use these points to performa least-squares adjustment, properly scaling and rotatingthe model. The total-station is the most expensive piece ofequipment required for this stratigraphie recording method.However, total stations are in common use by archaeologiststoday. Moreover, since the number of points to be shot isfairly low and no linework is required, even a simple andcheap instrument without fancy data collection, graphicalmap display, and geometry editing capabilities will beadequate to do the job.2.5 Problem AreasAt Huayuri, this method of stratigraphie recording has provenrobust, productive, and accurate. A few minor problems arecaused by excessively bright sunlight, heat, and wind, butthere are simple solutions to all of them. Bright sunlight mayresult in extreme contrasts and the black plastic backgroundof the targets reflecting as much light as the reflective dot(Figure 5). Under these circumstances, automatic targetmarking will not be effective since target recognition is262Figure 5. Excessively bright sunlight may make the black plasticbackground ofsome targets reflect as much light as the central reflective dot and cause extreme brightness differences between different sections of the image. Multiple automatic marking runs oreven manual target marking may be required. This problem is bestavoided by using an awning or roof (see Figure 6).

Stratigraphie Modeling and 3D Spatial Analysis Using Photogrammetry and Octree Spatial Decompositionthese minor issues, not a single image set—including ourvery first experiments—has ever failed to process.2.6 ProductivityPhotogrammetry is not only a robust method of stratigraphierecording but also exceptionally productive. The entirerecording procedure—placing targets, photography, andshooting reference points—will take about 20-30 minutesper surface. Cleaning the unit prior to photography is onlynecessary if orthophotos for publication are to be produced,but this will cause no extra work if photogrammetry imagesare taken as part of regular photo sessions. Additional detailcomes virtually for free; placing another 100 or so targetswill take no more than a couple of minutes. Around 15-60minutes of office time are required to post-process eachsurface, depending on how much manual labor is required.In most cases, the actual time will be closer to the lower endof this range.Figure 6. A simple, improvised awning and photo tower go a longway toward avoiding potential problems with photogrammetricrecording in the field.Figure 8. Outputs of photogrammetric stratigraphie recording(2): 2.5D polygons.In return, we simultaneously obtain detailed 2.5D surfacemodels of layers and features (Figures 7, 1), 2.5D polygonsrepresenting features or slices through features (Figure 8),and orthophotos (Figure 9). Thus, this recording methodreplaces profile drafting, plan view drafting, and part of fieldphotography—the whole range of graphical documentation,eliminating the need for essentially all hand drafting. Theresulting records are far more complete than traditionalpaper records and immediately available in digital format,ready to enter a GIS database.2.7 Accuracy Under Field ConditionsThe accuracy attainable with photogrammetry under actualfield conditions is quite acceptable. From the Huayuri site,we have 81 coded targets whose position was determinedboth photogrammetrically and by total station and that werenot used in the PhotoModeler least-squares adjustments.Thus, these measurements are redundant and may be usedFigure 7. Outputs of photogrammetric stratigraphie recording(1): 2.5D surface projected onto a digital photograph.Figure 9. Outputs of photogrammetric stratigraphie recording(3): orthophotos. Circular targets for automatic marking may beremoved and only coded targets left in the scene iforthophotos areto be published. See Figure 6

of stratigraphie volume solids. This paper presents a stratigraphie recording strategy based on oblique digital photogrammetry that is simple and inexpensive, yet comprehensive and accurate, and discusses approaches to solid modeling and 3D spatial analysis that meet the needs of archaeological stratigraphie analysis.