Palaeontologia Electronica · GPIT 1 Plateosaurus engelhardti: dorsal 2, left ilium, left radius...

Palaeontologia Electronica http://palaeo-electronica.org

MECHANICAL DIGITIZING FOR PALEONTOLOGY- NEW AND IMPROVED TECHNIQUES

Heinrich Mallison, Alexander Hohloch, and Hans-Ulrich Pfretzschner

ABSTRACT

Three-dimensional digitized representations of bones offer several advantagesover real bones or casts. However, creation of 3D files can be time consuming andexpensive, and the resulting files are difficult to handle due to their size. Mechanicaldigitizing was hitherto limited to large bones. Here, new and improved data collectiontechniques for mechanical digitizers are described, facilitating file creation and editing.These include:

- Improvements to the in-program digitizing procedure, reducing time and financialdemands.

- Specifics for an easy to assemble and transportable holder for small fossils.- A significant increase in the size range of digitizable bones, allowing both exact

digitizing of bones only a few centimeters long and bones larger than the rangeof the digitizer. This increase allows the study of assemblages including bothsmall and large bones.

- Complex shapes such as costae and vertebrae can now be digitized with ease. - Step-by-step directions for digitizer and program use to facilitate easy acquisition

of the techniques.3D-files of fossils digitized with these methods can be added to online databases

easily, as small-scale preview and complete files. The file formats are common and thefile sizes relatively small in comparison to CT or laser-scan data. Pointcloud files canbe used interchangeably with laser-scan files of similar resolution. Other possible usesfor mechanical digitizing data are described.

Additionally, techniques to extract and edit comparable data from CT scans arebriefly described. CT-based data is used to check the accuracy of mechanically digi-tized data.

Heinrich Mallison. Institut für Geowissenschaften, Eberhard-Karls-Universität Tübingen, Sigwartstrasse 10, 72076 Tübingen, Germany. Current address: Museum für Naturkunde – Leibniz-Institut für Evolutions- und Biodiversitätsforschung an der Humboldt-Universität zu Berlin, Invalidenstrasse 43, 10115 Berlin, Germany. [email protected] Hohloch. Institut für Geowissenschaften, Eberhard-Karls-Universität Tübingen, Sigwartstrasse 10, 72076 Tübingen, Germany. [email protected] Pfretzschner. Institut für Geowissenschaften, Eberhard-Karls-Universität Tübingen, Sigwartstrasse 10, 72076 Tübingen, Germany. [email protected]

PE Article Number: 12.2.4TCopyright: Society for Vertebrate Paleontology August 2009Submission: 22 December 2008. Acceptance: 3 May 2009

Mallison, Heinrich, Hohloch, Alexander, and Pfretzschner, Hans-Ulrich, 2009. Mechanical Digitizing for Paleontology - New and Improved Techniques. Palaeontologia Electronica Vol. 12, Issue 2; 4T: 41p; http://palaeo-electronica.org/2009_2/185/index.html

MALLISON, HOHLOCH, & PFRETZSCHNER: NEW DIGITIZING TECHNIQUES

KEYWORDS: mechanical digitizing; methods; computer; vertebrates

INTRODUCTION

In recent year, digital files have increasinglybeen used for scientific research instead of realbones or casts. Currently, the most common way ofobtaining a digital representation of a specimen iscomputer assisted tomography (CT) (see e.g., Zuoand Jing 1995; Gould et al. 1996; Knoll et al. 1999;Stokstad 2000; Golder and Christian 2002; Ridgelyand Witmer 2004, 2006; Sereno et al. 2007; Wit-mer and Ridgely 2008). These digital images canconsist of cross sections, but usually are threedimensional models of internal shapes of anobject, e.g., in order to assess as yet unpreparedspecimen or depict internal structures without dam-aging the object (e.g., Witmer and Ridgely 2008).Models of external shapes can be used to rapidprototype (RP) scaled models or exhibition copies,because the high accuracy of CT scans justifiesthe high costs of CT scanning and RP. This tech-nique also allows mirroring of specimen or combin-ing several partial specimens into one completeindividual or bone. Neutron tomography (NT) hasalso been tested (Schwarz et al. 2005), with mixedresults.

Another method to obtain 3D files is laserscanning, either from three perpendicular views orwith a surround scan. Alternatively, repeated scanscan be taken at many angles and combined in thecomputer. An extensive project at the TechnischeUniversität Berlin used laser scanners to digitizecomplete mounted skeletons and skin mounts(http://www.cv.tu-berlin.de/menue/abgeschlossene_projekte/3d_rekonstruktion_von_dinosauriern/fruehere_arbeiten/brachiosaurus_brancai/, seealso Gunga et al. 1995; Gunga et al. 1999, Bell-mann et al. 2005; Suthau et al. 2005; Gunga et al.2007; Gunga et al. 2008). Bates et al. 2009 alsoemploy such laser scans, albeit apparently at alower accuracy. Also, some of the dinosaur skele-tons mounted in the MFN exhibition were high res-olution laser scanned as separate elements byResearch Casting International (www.rescast.ca)during the museum renovation in 2006/2007.

All three methods produce vast amounts ofdata, depicting the object in very high detail. Whensuch high resolution is not needed, the large filesize becomes cumbersome. As long as only exter-

nal surfaces are of concern, mechanical digitizingprovides a cheap and fast alternative (Wilhite2003a, 2003b), delivering small files of sufficientaccuracy for most applications. Mechanical digitiz-ing means creating a computer representation of aphysical object by means of using an apparatusthat samples 3D landmarks on the object’s surfacethrough touching it. Other techniques involving dig-itizing were used by Goswami (2004) and Bonnan(2004), who focused on specific bone landmarks.In contrast to our methods, these do not producecomplete 3D images of bones and will not beaddressed here. Similar in handling and data out-put to the methods described here is the sonic digi-tizer used by Hutchinson et al. (2005). It is limitedto collecting point data, but provides a large rangeof up to 14 feet, albeit at slightly lower accuracy.

Here we detail improvements for digitizingtechniques for dinosaur bones as described by Wil-hite (2003a, 2003b), expanding the size range ofsuitable bones for the method significantly, both forlarger and smaller specimens. New methods alsoallow complex shapes to be digitized with relativeease, and remove the need to edit the digitizingdata in other programs before use. Also, theextraction of surface data from CT data in AMIRA3.11® and the subsequent editing is describedbriefly. This CT based data is used to evaluate theaccuracy of mechanical digitizing data.

Fossils (vertebrate or invertebrate) digitizedwith the methods described here can easily beadded to online databases, instead of or alongsidewith photographic images. Most databases, suchas the database of the New Mexico Museum ofNatural History (Hester et al. 2004) or the Ameri-can Museum of Natural History (http://paleo.amnh.org./search.php) can easily accommo-date small-scale previews as well as completefiles, since the file formats are common and the filesizes relatively small in comparison to CT or laserscan data. Stevens and Parrish (2005a, 2005b,www.dinomorph.com) used several files createdduring this project for modeling Brachiosaurus inDinomorph™. The University of Texas runsanother digital library (http://www.digimorph.org/index.phtml) based on high-resolution CT scans.Objects digitized via dense point clouds asdescribed herein could conceivably be added to

2

PALAEO-ELECTRONICA.ORG

this database as stereolithographies (*.stl files),provided sufficient resolution is obtained. For mostapplications, pointcloud files created with theMicroscribe® can be used interchangeably withlaser scan files of (or reduced to) similar resolution.The digital files can also be used to rapidly testpossible skeletal assemblages, joint mobilityranges (Wilhite 2003a, 2003b), inter- and intraspe-cific variation (e.g., Wilhite 2005). Virtual skeletonscreated from them in CAD softwares such as Rhi-noceros® can be an aid in planning museummounts.

Another possible application is rapid prototyp-ing. Scale models of bones can be produced atalmost any scale, as well as molds for casting, ornegatives of the bones that can serve as storagecasts or as mounting racks for museum exhibition.High resolution rapid prototyping or 3D printing(600dpi) calls for CT or laser scan data, due to theability to exactly create surface textures, but atlower resolutions (300dpi), accurate NURBS orSTL objects from mechanical digitizing are of suffi-cient quality to create exhibition copies of fragilespecimens or mirror images to replace missing ele-ments in skeletal mounts. Research Casting Inter-national (www.rescast.ca) used full scale 3D printsof the exhibition skeleton of the MFN Kentrosaurusto construct the armature that was used for thenew mounting of the skeleton in 2007.

Our methods probably work well for a plethoraof disciplines aside from vertebrate paleontology,such as archeology. However, aside from a singletrial using a fossil vertebrate footprint, we devel-oped them solely on vertebrate body fossils.Researchers from other fields are encouraged toexperiment with mechanical digitizing, and to adaptand improve upon the methods described here.

MATERIALS

Institutional abbreviations

IFGT Institut für Geowissenschaften, Eberhard-Karls-Universität Tübingen, Tübingen (GER). Formerly Geolo-gisch-Paläontologisches Institut Tübingen(GPIT)

GPIT IFGT collection numbers

MB.R. collection numbers of MFN

MFN Museum für Naturkunde – Leibnitz-Institut fürEvolutions- und Biodiversitätsforschung an der Humboldt-Universität zu Berlin, Berlin (GER) (also

abbreviated HMNB, MN, or HMN in litera-ture)

JRDI Judith River Dinosaur Institute, Malta, MT(USA)

Computer software

(1) McNeel Associates ‘Rhinoceros© 3.0 NURBSmodeling for Windows®’

Rhinoceros 3.0® is a NURBS based CAD pro-gram. Versions 2.0, 3.0, and 3.0SR4 (ServiceRelease 4) were used to obtain and process digitaldata. Version 4.0 is available, but was not usedhere. All digitizing methods described here weretested and can also be used in Version 4.0.

To curb costs we tried to use Rhinoceros 3.0®exclusively when developing new methods. Somedata operations described here dealing with com-plex geometric shapes, however, require either avery high level of program knowledge, or areimpossible in Rhinoceros®, and are thus easierperformed in or require another software (position3 below).

(2) TGS Template Graphics Software Inc. ‘AMIRA3.11’ (time-limited evaluation version)

AMIRA 3.11 is a 3D visualizing and modelingsystem that allows creation of surfaces (3D bodies)from computer tomography (CT) data.

(3) Geomagic Corporation ‘Geomagic Qualify 8.0®’(time-limited evaluation version)

Geomagic 8.0® is a suit of CAQ (computeraided quality assurance) program components thatcomplement each other. Geomagic Studio 8.0®includes all parts, but the more limited GeomagicQualify 8.0® is also sufficient. It was used for edit-ing those files based on point cloud digitizing or CTscans.

Technical equipment

(1) Immersion™ ‘Microscribe 3D’

(2) Immersion™ ‘Microscribe 3GL’ (on loan fromthe Institut für Zoologie der Rheinischen Friedrich-Wilhelm-Universität Bonn)

The Immersion Microscribe3D© (‘Microscribe’,‘digitizer’) is a three-dimensional mechanical pointdigitizer. The digitizer is easily transportable, costeffective, and reliable. The GL version of the digi-tizer has a longer arm, allowing for a greater reachwith only a negligible loss in accuracy. The inputfrom the Microscribe® to the computer was con-trolled with the foot pedal provided together withthe digitizer. Various desktop and laptop PCs wereemployed, the least powerful being a Pentium II PCwith an 800MHz processor and 256 MB of RAM,

3


connected to the digitizer via a serial connectioncable, or a USB cable in case of the MicroscribeGL®.

Fossil material

HM and AH mechanically digitized over 100bones in various institutions. For the description ofthe methods given here only the following areused:

IFGT:

GPIT 1 Plateosaurus engelhardti: dorsal 2, leftilium, left radius

GPIT 2 Plateosaurus engelhardti: left humerus, leftpedal phalanx II-1

GPIT ?610 Diplodocus sp.: right metacarpal 3

JRDI:

JRDI 200 Brachylophosaurus canadensis: left dor-sal rib

MFN:

MB.R.2246 Giraffatitan (Brachiosaurus) brancai:left calcaneum

MB.R.2912 Giraffatitan (Brachiosaurus) brancai:left scapula

MB.R.2249.R9 - R17 Giraffatitan (Brachiosaurus)brancai: left metacarpus and manual phalan-ges I-1, I-2, II-1, III-1, V-1

unnumbered Dicraeosaurus sattleri.: left scapuloc-oracoid, tibia, fibula, astragalus

Further materials

Vertebrae were in some cases stabilized bywrapping one half in aluminum foil or a plastic filmand burying this half in a box of sand. Adhesivemasking tape was used to provide a base for mark-ings on the bones and as a visual aid during digitiz-ing, and a specially constructed variable holderwas used to stabilize most medium-sized andsmall bones. Some very small markings must bemade on the bone for digitizing, and extensivemarking can be necessary in some cases. Weused painter’s masking tape for this purpose,choosing high quality brands with small amounts oflow-power glue. These always came off the bonewithout damaging it, but sometimes removed paintcoats from plaster fillings of damaged bone areas.

GENERAL OUTLINE OF MECHANICAL DIGITIZING METHODS

Prerequisites

As digital representations of fossil bones willusually lack many features of the real specimen,such as surface rugosities and textures or discolor-ations indicative of breaks and deformation, maxi-mum care must be given to the process ofselecting specimens for digitizing. Especially thosewith deformations of the bone obvious on the realspecimen but invisible on a digital representationmust be avoided.

There are two possible aims when digitizing:

a) digitally constructing ‘ideal’, that is undeformedand complete bones from several partial ordamaged specimen, or

b) digitizing individual specimens exactly, e.g., toobtain a digital representation of one com-plete animal.

For (a), as an absolute minimum, a specimenmust either allow measuring of at least two charac-teristic dimensions and their relation to each other(preferably total length and proximal or distal width)or three distinctive landmarks that can be pin-pointed with millimeter accuracy. Additionally, thespecimen must possess a significant section ofnon-deformed and non-eroded bone surface to bedigitized in correct relation to said characteristicdimension. For example, a complete articular endthat has been shifted in relation to the long axis ofthe bone through compression is useless, as theexact orientation cannot be ascertained. Only if thecorrect three dimensional relations of the charac-teristic dimension and the area digitized can beascertained, can several pieces be combined cor-rectly. These requirements are far less strict thanthose commonly used for other studies (e.g., Wil-hite 2003b), as the methods described here allowcombining sections from several specimens toobtain artificial ‘ideal’ digital bones.

Regarding (b), any specimen can be used.However, for most applications using digital data,especially scientific study, it is advisable to usewell-preserved material. In case of elements thatexist as left and right copies, it is often better to mir-ror contralateral elements, than to use a badly pre-served bone. Similarly, in elements with a bilateralsymmetry, mirroring one well-preserved half maybe better than using a badly damaged part, as longas the symmetry plane is obvious on the bone.Such data operations must, however, be clearlymentioned, best as a text entry in the digital fileitself. Also, it is important to remember that dam-

4


age obvious on a real bone will usually not be eas-ily visible on the digital file. Because digital files canbe (and often are) widely shared betweenresearchers worldwide, it is important to selectwell-preserved and typical examples. Otherwise,there is a risk that imperfect or unusual specimenswill be accidentally treated as complete or normal.

General Overview of the Digitizing Procedure

Here, only a short description of the generalprocess is given. Various versions of the basic pro-cedure have different advantages and limitations,and are best suited for various kinds of fossils, asdetailed in Appendices A and B. Step-by-stepdirections for program and digitizer use are given inAppendices C through E.

For digitizing we used Immersion™ Micro-scribe 3D digitizers. A Microscribe consists of abase plate, on which a four part arm is mounted.The base plate contains sockets for cables con-necting the Microscribe to a PC. The position of thearm’s tip is measured through the displacement ofthe joints between the various parts compared tothe ‘neutral’ position, into which the machine mustbe put before it is switched on. By pressing a but-ton on a foot pedal, the operator can determinewhen data on the tip’s location is transferred to the

computer. Various commercially available Com-puter Aided Design (CAD) softwares can receivethis data and transfer it into data points. We onlyused Rhinoceros®, which has the additional abilityto automatically interpolate NURBS curvesbetween the data points delivered by the digitizer.Figure 1 shows a typical setup of the digitizer andlaptop along with a specimen (Diplodocus sp. GPIT?160). It is possible to digitize large objects whilesitting on the floor (Wilhite 2003b), often made nec-essary by the large weight and resulting immobilityof specimens such as sauropod longbones. Work-ing on a table as shown in Figure 1 is decidedlymore comfortable and reduces worker fatigue, as itis much easier to work the foot pedal in this posi-tion. While this seems trivial, we found that pro-longed digitizing in an uncomfortable positionincreases the likelihood of time-consuming errorssignificantly, and also increases the time requiredfor digitizing specimens. The quality of the digitizeddata also decreases when the operator assumesan uncomfortable working position such as squat-ting on the floor. At worst, this can lead to crampsor jittering of the hand, making data acquisitionimpossible.

Preparation of specimens: Some markings mustbe made on the specimens before digitizing cantake place. We placed masking tapes of differentmanufacture on them, on which we drew therequired markings with felt tip pens. Although wenever caused any damage to specimens this way,it is theoretically possible that the masking tapedamages the bone surface. Thus, curatorial per-sonnel should always be involved in the decisionwhat tape to use, and where and how to apply it.Extremely delicate specimens should therefore bemarked as little as possible. We recommendremoving any tape from the bones as soon as pos-sible.

If it is impossible to put masking tape on a speci-men, digitizing it with point clouds or NURBScurves is still possible. Point clouds that partlyoverlap can be aligned manually in Rhinoceros®after digitizing. However, recalibration is impossi-ble, unless there are at least three distinct and verysmall landmarks on the bone that can be usedinstead of markings. Not being able to recalibratethe digitizer creates a large risk of errors in the finalfile. Also, digitizing may take more time, and moreerroneous curves may be created, if the bone can-not be marked in places difficult to digitize. NURBSdigitizing without markings on the bone requiresmaking a mental mark of curve starts and paths, toavoid drawing curves that intersect, leave large

FIGURE 1. Typical setup for digitizing: laptop, digitizer,and fossil holder. Note the position of the digitizer closeto the fossil and away from the operator, so that the tipcan be pulled instead of pushed.

5


gaps, or otherwise results in errors in the final digi-tal surface. This is possible even for large speci-mens, but requires extreme concentration, whichincreases worker fatigue. Additionally, the likeli-hood of erroneous curves increases, requiringadditional time for correction both during digitizingand editing. We have successfully tested digitizinga sauropod metacarpal without markings.

Data acquisition: The easiest way to obtain 3Ddata of large bones with the Microscribe is by stor-ing curves, not points, as detailed by Wilhite(2003b). Both curves and surfaces in Rhinoceros®are created as NURBS object. NURBS stands fornon-uniform rational B-spline. Constructing a sur-face is easy when using a loft function on curves,which creates a NURBS surface. Even more com-fortable is lofting a ‘closed loft’, leading directly to aclosed 3D body, which is the method used mostextensively here. Point clouds cannot be surfacedwithout much effort in Rhinoceros 3.0®. The cur-rent version Rhinoceros 4.0® allows meshing pointclouds directly, but the process is error-prone andless accurate than in Geomagic®.

The process of digitizing with NURBS curvesis best described as the electronic equivalent ofwrapping sub-parallel wires around the bone, thenpulling a cloth tight around the wires. See Figure 2for an example of a digital bone (left radius ofDicraeosaurus sattleri MB.R.2622) and the curvesused to create it. The curves are obtained by enter-ing a simple command into Rhinoceros®, placingthe tip of the digitizer on the bone at the start pointof the intended curve, pressing down the foot

pedal, and moving the digitizer tip over the boneuntil the desired end point of the curve is reached.Then, the foot pedal must be released. Neighbor-ing curves must be of similar length and should beroughly parallel. Large differences in length or sep-aration tend to produce artifacts in the final surface.Also, curves may not cross each other.

Curves are placed at intervals at the opera-tors’ discretion and should be closely spacedwhere the morphology of the bone exhibits impor-tant features or where the topology changesabruptly, e.g., near cristae or at the articular ends.Relatively simple surface areas like shafts of long-bones or scapular blades require few curves. Theoperator’s judgment on the placement is one of thekey elements that determine the accuracy of thedigital bone.

If a bone cannot be represented by one set ofsub-parallel curves due to its shape, it can be digi-tized by joining several partial surfaces or bodiestogether. Separate curve sets must be digitized foreach part.

To reduce post-digitizing workload andachieve the most accurate results, closed curvesreaching 360° around the bone are best. If a bonecannot be digitized with closed curves, due to itssize or a fixed mounting that makes reaching allaround it impossible, partial curves can be drawnand joined into closed curves.

Alternatively, a point cloud can be collectedwith the digitizer. Figure 3 shows point clouds ofthe lower left hind limb of the mounted Dicraeosau-rus from the MFN (unnumbered) and the 3D filescreated from them. Point cloud digitizing is a

FIGURE 2. A digital bone (GPIT 1 Plateosaurus engelhardti left radius) and the curves and points used to create theloft. Note the sub-parallel arrangement of the curves. This digital bone is a NURBS body (closed surface), displayedin rendered view. Length of the bone 214 mm.

6


method usually more time consuming than digitiz-ing curves, as the full surface of the bone must bedensely sampled. On the other hand, hardly anyplanning ahead is required, and there is no need tomark the bone extensively, saving time especiallywhen a complex geometry renders curve-planningdifficult. It is best used for small bones of complexshapes, or for rough representations of largebones at low resolution.

Surface creation: A surface is created fromcurves as a NURBS surface, a ‘loft’, through a sin-gle command. If, which is most advisable, theentire surface is to be created in one piece (fromclosed curves), two points are also needed, one ateach end of the bone. This will result in a closedbody (resembling a deformed balloon) instead ofan open surface (resembling a deformed tube).

When digitizing point clouds, curves can behand-built from suitable points, but this method isusually not advisable due to the high amount ofwork involved. Instead, current versions of Rhinoc-eros® can produce polygon meshes directly frompoint clouds. These usually require a few minutes’to half an hour’s work of editing to remove artifactsand mesh errors, but this method allows accuratedigitizing of small and complex shapes, such as

small to mid-sized vertebrae. Figure 4 shows theleft humerus of GPIT 1 as a point cloud file, theresulting mesh in Geomagic®, and the final fileproduced by manual editing of the mesh. Both theinitial meshing and all editing are, however, bestaccomplished in Geomagic®. Unfortunately, whenRhinoceros® is used to create the mesh, theresulting 3D bodies are often smaller than the vol-ume covered by the original point cloud, producingsignificant errors in the surface shape. Also, Rhi-noceros® tends to produce more meshing errorsnear sharp bends in the surface geometry than theGeomagic programs suite (Figure 5). Additionally,as opposed to Geomagic®, Rhinoceros® does notoffer an option to preserve the edges of mesheswhen reducing their polygon number. This resultsin ‘digital erosion’ of sharp edges. Digitizing bonesvia point clouds may require more effort than viacurves, but is decidedly cheaper than CT or laserscanning.

Mobile fossil holder: Accuracy is paramountwhen digitizing fossils, as even slight aberrations ofthe digital curves can lead to significant shifts ofvolume or appearance. A slight unnoticed rotationof the specimen during digitizing may lead to a mis-interpretation of range of movement of joints thatinclude the articular ends of the bone. Mass esti-mates of complete animals may be off by signifi-cant amounts if bones of the pelvis girdle aremisshaped or longbones gain or lose length or vol-ume through errors during digitizing. More commonthan unnoticed errors are significant movements ofthe specimen due to unstable placement or physi-cal contact. Especially small bones will shift ateven the slightest touch while curves are beingdrawn, invalidating the last curve drawn and requir-ing time-consuming recalibration. A commonmethod to avoid this is placing the specimen eitherin sandboxes, where they are often still prone toshifting and the sand is likely to get into the com-puter and digitizer, or to fix them with Play-Doh® orsimilar deformable substances. Since variouschemicals damaging to fossil bone may leak fromthese materials, their use is problematic. Also,360° reach around the fossil is not possible. A con-struction drawing of the holder is available onrequest.

To solve these problems a variable holder wasdesigned. It can be separated into small piecesand quickly reassembled. Figure 6 shows theholder in the minimum configuration with anunnumbered GPIT Diplodocus metacarpal and theextension parts used for larger bones. This con-struction kit setup makes the holder easy to stow

FIGURE 3. Lower hind limb of Dicraeosaurus sattleri(MFN unnumbered). (1) pointcloud file. Each bone wasdigitized in a different layer, indicated by different colors.(2) Low-resolution polygon mesh file created from thepoint clouds in Geomagic®. These meshes were notsubjected to any editing. Length of the tibia 790 mm.

7


and transport. It consists of a base plate madefrom heavy polyurethane, custom made metalholders that can be placed at variable intervals onthe base plate as desired, and commercially avail-able plastic contour gauges supported by the metalholders. On these, the bones rest stably, are wellsupported, and resist shifting even when bumped.Using smoother plastic gauges instead of metalholders avoids the risk of scratching the bone. Thebase plate is split into four parts. These parts canbe stuck together as needed in order to accommo-date large bones, but are not cumbersomely largewhen used for small bones. The smallest possibleassemblage, sufficient for objects up to the size ofsauropod metatarsals or hadrosaur humeri (ca.

10x10x35 cm), weighs approximately 3 kg, thelargest tested assemblage, sufficient even for sau-ropod pubes and radii, weighs about 8 kg. Theoret-ically, the holder can hold even larger bones, if asufficient number of contour gauges are used tosupport the bones.

The fragility of the specimen strongly influ-ences the ideal setup. More gauges mean bettersupport, fewer gauges mean better access. Thelongest bone digitized during this project was aBrachylophosaurus rib from the JRDI. The excel-lent preservation and hardness of the bone allowedusing only four gauges (Figure 7). On the otherhand, the ribs of the Plateosaurus skeleton GPIT 1could not be supported on the holder due to their

FIGURE 4. GPIT 2 Plateosaurus engelhardti left humerus (length 351 mm) point cloud based 3D file creation exam-ple. Clockwise, starting top left: lateral, proximal, cranial, caudal, distal, and medial views (terms refer to standardizedin vivo position, assuming parasagittal posture). (1) Point cloud from mechanical digitizing. (2) Initial mesh as createdin Geomagic®. Note the large holes and many small surface errors. (3) Final edited mesh. Small surface errors weredeleted and the resulting holes as well as the large holes were closed with curvature-based filling.

8


fragility. Close enough spacing of the gaugeswould have made access to the lower side impos-sible. Curatorial personnel should always beincluded in the decision on what setup of the holderis employed.

Size range: The lower end of the size range ofobjects that can be digitized accurately is definedless by dimensions, but rather by the necessity tokeep the object perfectly immobile without obscur-ing a large part of its surface. The smallest bonesthat we were able to digitize were less than 5 cmlong, held in place on the table with two fingers,and digitized using point clouds. NURBS curvedigitizing usually works well for objects with a

length greater than 10 cm. Similarly, the maximumsize is not defined by greatest (or least) extension,but rather by the complexity and size of the surfacearea. The larger and/or more complex in shape abone is, the more recalibrations of the digitizer willbe necessary to sample it sufficiently. For example,a 50 cm long sauropod vertebra may require morerecalibrations than a 2 m long sauropod scapula.Each recalibration decreases the accuracy of anydigitizing taking place after it. While in theory thesize range is unlimited, our experience shows that

FIGURE 5. Dicraeosaurus sattleri (MFN unnumbered)left scapula. (1) point cloud file as digitized. (2) Polygonmesh created in Rhinoceros®. Note the massive mesh-ing errors along the edges of the bone and on thin sur-face parts. (3) Polygon mesh created in Geomagic®.Note the drastically lower number of meshing errors andthe smaller triangle size along the bone edges. Meshinghas also closed the coracoid foramen. (4) Finished 3Dsurface based on (3) after editing in Geomagic®. Lengthof the scapula 1067 mm.

FIGURE 6. Minimum configuration of the fossil holderwith a Diplodocus sp. Metacarpal 3 (GPIT ?610). Thebone has been marked for digitizing with coordinates (Yand O, X is hidden from view), seam line, some curvepaths, and end points. On the right and on top extensionparts and double-wide contour gauges are shown.

FIGURE 7. Digitizing a Brachylophosaurus canadensisrib (JRDI 200, length 1048 mm). Note the extensivemarkings on the bone. The finished 3D file is visible onthe laptop screen.

9


recalibrating more than five times should beavoided. However, this number is not a hard limit,and there is no single bone in any extant or extinctvertebrate that is too large for the methodsdescribed here. Digitizing very large objects usu-ally results in a reduced absolute, but not neces-sarily relative accuracy.

Manpower requirements: Normally, one personcan transport the equipment and digitize bonesalone. The digitizer and the fossil holder together fitinto a standard suitcase, and weigh less than 20kg. When digitizing very large bones it may be diffi-cult for one person to operate both the digitizer armand the foot pedal, especially if it is necessary tostep around a mounted bone during digitizing. Asecond person should then be employed to oper-ate the foot pedal. In this study, only a scapula ofGiraffatitan (Brachiosaurus) brancai (MB.R.2912),mounted vertically, made a helper necessary.

Digitizing time requirements: The time neededfor digitizing depends significantly on the expertiseof the person operating the digitizer. Generally,between 5 and 20 minutes suffice to digitize asmall or medium sized bone of simple geometry,such as a longbone, metatarsal, pelvic bone, or rib.Very large bones (over 1 m length) or complexshapes (vertebrae, skull elements) may take sev-eral hours, although usually 30 minutes are suffi-cient. Post-digitizing editing may require severalhours. Normally, though, simple bone shapes willnot require editing at all, while editing complexshapes can usually be completed within 15 min-utes. Polygon meshes based on point cloud dataare easier to edit in Geomagic®, which usuallytakes a few minutes only, than in Rhinoceros®,which may take up to half an hour. Overall, thetechniques allow digitizing and final editing of vir-tual bones as fast as or even faster than extractingand editing virtual bones from CT data. Consider-ing that CT scanning involves wrapping and trans-porting the specimens, plus time for the actualscanning, mechanical digitizing is decidedly faster.

EXTRACTING VIRTUAL BONES FROMCT DATA

One of the two most detailed and expensivetechniques of creating ‘virtual’ bones is high-reso-lution computer tomography (HRCT, short CT)scanning specimens. This allows maximum resolu-tion, far higher than required for most uses, similarto high resolution laser scans of individual bones.The former technique has the advantage of allow-ing the study of internal structures and does not

suffer from ‘blind spots’, as X-rays penetrate thematerial. Even surfaces completely blocked fromview such as deep cavities and recesses on skullsare faithfully reproduced in the virtual bones. Vir-tual bones from both methods can be assembledinto virtual skeletons either simply based on theirown shapes, much as it is possible for real bones.Drawings, photographs, or measurements ofmounted skeletons can be of help, but are rarelyrequired, since the high-resolution virtual bonesprovided by both methods contain all the informa-tion needed for assembly. One drawback of thesemethods is the relatively large file size. Both Rhi-noceros® and the Geomagic® program suite offeroptions for reducing the number of polygons ineach mesh, reducing the file size proportionately.The latter program offers the additional option ofpreserving the outside contours better and thusshould be preferred. While reducing the meshnumber decreases file size, the resulting virtualbones lose accuracy, and the reduction should notbe taken too far. On average, a reduction to 2.5 to10% is the maximum tolerable, depending on bonesize and shape. Delicate structures may start los-ing shape at 20% reduction already (i.e., 80% ofthe original size).

For data extraction, the files of one scan areloaded into AMIRA 3.11®. Then, a ‘LabelVoxel’module is created and applied to the data. Here, upto four different areas of density can be defined. Ahistogram is helpful for interpreting the data anddeciding where to set the borders. It is possible,e.g., to remove or include plaster fillings by choos-ing different settings. Now, an ‘OrthoSlice’ modulecan be created to view cross sections. In order tokeep the computing time and memory require-ments low, the re-labeled data should be croppedto contain no unnecessary space, e.g., emptyspace under or above the bones. Large bonesshould be cropped out so that each bone is treatedseparately. Since the original data is still present inunaltered form, after extraction of the first bone itcan simply be ‘labeled’ again and the next bonetreated. To each cropped set of labeled data, a‘SurfaceGen’ module is attached and executed.This creates a polymesh surface, which can besaved as a number of different formats, e.g., ASCIIstereolithography (*.stl). The resulting files arehighly detailed and accordingly huge. A longbonecan easily have 10 million polygons and exceed 1GB in file size. To reduce the size, it is useful toload the files into Rhinoceros® and re-save themas binary STL files, which have a significantlysmaller size without any data loss. Reducing the

10


number of polygons, on the other hand, results in aless accurate representation of the surface. Usu-ally a reduction to 20% is hardly noticeable to thehuman eye if a bone is displayed at full-screensize. Therefore, a slight to generous reduction maybe acceptable depending on the planned use of thedata. As mentioned above, this is best done inGeomagic®, as this program has an option to ‘pre-serve edges’, guaranteeing a minimum of shapechange during polygon reduction. AMIRA 3.11®also offers this option, here called ‘Simplifier’. ‘Pre-serve slice structure’ is the equivalent to the edgepreservation option in Geomagic®.

ACCURACY OFMECHANICAL DIGITIZING DATA

Any 3D file is only of use if it mirrors the origi-nal object accurately enough for the investigationat hand. As described above there exists aninverse relationship between accuracy and filesize. The smaller files produced from mechanicaldigitizing offer the benefit of easier handling overthe large files from laser or CT scanning, but istheir accuracy sufficient for e.g. skeletal recon-structions or rapid prototyping of scale models? Inorder to test this, a number of mechanical digitizingfiles were compared to the high-resolution CTbased files scaled down to the same final size asthe mechanical digitizing files. The following virtualbones of GPIT1 Plateosaurus engelhardti wereused in order to cover different sizes and shapes:left humerus, left ilium, second dorsal vertebra.Additionally, the pedal phalanx II-1 from GPIT 2Plateosaurus engelhardti was digitized. The fileswere imported into Geomagic® and aligned auto-matically with the ‘Best fit’ option. A 3D comparisonanalysis was performed, creating a 3D map of thesurface pair, in which distance between the sur-faces at any point is displayed by varying colors.The program also calculates the maximum, aver-age, and standard deviations. Note that the actualerrors of the mechanical digitizing files are proba-bly smaller than determined by the program,because the alignment was not optimized for mini-mal deviation. Geomagic® offers a variety of align-ment options, but we considered it unnecessary totest which one offers the smallest deviation values.

Test files

Ilium: The CT data of the left ilium consisted of1778 slices, of which 889 (every second) wereused to extract the file. The reduction was madenecessary by the fact that each slice of 0.5 mmthickness overlapped the neighboring files by half

that amount, which for unknown reasons createdmassive artifacts (wrinkling) in the finished sur-faces. The scan of the ilium also included the rightfibula, totaling a data volume of 894 MB at 516 kBper file. From it, an STL file in ASCII format with203 MB was extracted. This file, which still includedpieces of the fibula and internal 3D bodies in theilium, was edited to gain the maximum resolutionSTL file of the ilium in Rhinoceros®, having shrunkto 47 MB by removal of the excess data and savingin binary STL format. One deep pit stemming fromobvious damage was removed by manual editing.The file has 977244 triangles and was reduced inGeomagic® to 89816 (9,19%) to achieve the samefile size as the point cloud file.

The point cloud data consisted initially of44865 points, which were meshed into a surfacewith 89816 polygons. This was edited manually toremove some obvious artifacts along sharp edges,where curvature-based filling was applied. Also,various small artifacts on flat surfaces wereremoved. The files size is 4,33 MB.

For the 3D deviation comparison a displayscale was selected that details deviations ofbetween +/- 0.5 mm and +/- 5 mm. Deviationssmaller than half a millimeter we assume to beirrelevant. Deviations up to 2.5 mm are tolerable,and for values up to 5 mm (~ 1% of greatest lengthof the ilium) it is important where they occur. Ifstrong deviations are limited to damaged areas ofthe bone, they can be ignored outright. On theother hand, several millimeters of deviation overlarger areas are unacceptable. Any deviationgreater than half a centimeter can only be toleratedif it has no influence on the likely shape of theundamaged bones, i.e., if it occurs due to damageof the specimen. The likeliest scenario for such adeviation is a tiny, deep hole in the bone that mustbe passed over with the digitizer.

Figure 8.1 shows the 3D deviation maps ofthe pointcloud based file compared to the CT file.Deviation ranges from nearly +5 mm to almost -15mm, but values greater than 2.5 mm occur exclu-sively near cracks in the bone. The mechanical dig-itizing file preserves a different 3D body than theCT scan data in such places, because it is limitedto gathering data at the air/solid boundary, while inthe CT file some of the crack infill is missing,because the extraction threshold was set roughlyat the bone/sediment boundary. Geomagic detectsand ignores non-alignable areas, but the edges ofthese areas are assessed. Despite these, averagedeviation is below 0.5 mm.

11


12

FIGURE 8. 3D deviation maps from Geomagic® of the left ilium of GPIT 1 Plateosaurus engelhardti (maximum exten-sion 426 mm). (1) pointcloud-based file compared to unedited CT file. (2) pointcloud-based file compared to CT file inwhich all major cracks and holes were smoothed.


To test how significant the influence of theshape differences caused by the cracks is, the CTbased file was extensively edited to smooth thecracks over. This lead to the removal of furtherinternal surfaces and created 43 holes in the outersurface, all of which were automatically closed bycurvature-based filling. In all, the number of poly-gons dropped by 15.8% to 75658 polygons, reduc-ing file size to 3,6 MB. 3D comparing this file to thepoint cloud file (Figure 8.2) resulted in a significantreduction of the maximum deviations (+4.12 mm / -3.1 mm). The average and standard deviationswere little influenced, in contrast, due to the largeundamaged surface areas, which outweigh thecracks.

Humerus: The humerus based on NURBS curves,created in roughly seven minutes, was lofted inRhinoceros® with a loft rebuild option with 25 con-trol points, and exported for comparison inGeomagic® as a polymesh file with 13764 poly-gons. The original NURBS file has a size of 1.27MB.

The mechanical digitizing file with points con-sisted of 24640 points, with only a handful of obvi-

ously erroneous points. Digitizing time was roughly10 minutes. Meshing in Geomagic® produced asurface with various small and two large holes. Allcould be filled with curvature-based filling withoutproblems. The file was now manually smoothed,after which 49102 polygons remained. Figure 4shows the original point cloud, the wrapped sur-face, and the edited final surface. The file size is2.47 MB, nearly double that of the NURBS file.

The CT data of the left humerus stemmedfrom an earlier scanning opportunity and had alower resolution than all later CT scans. The STLfile extracted from 282 MB of raw data initially hada size of 359.202 MB (ASCII STL) and 1763876polygons. It was reduced to 2.81% to match the49524 polygons of the point cloud file. This opera-tion alone required over 12 minutes calculationtime on a 2.4 GHz PC with 2 GB of RAM and a256MB graphics card.

Figure 9 shows the 3D deviation maps for thepointcloud (Figure 9.1) and NURBS (Figure 9.2)files using the same scale as the ilium comparisons(+/- 0.5 mm to +/- 5 mm). Average deviation is ~0.2 mm for the point clouds file and ~0.4 mm for

FIGURE 9. 3D deviation maps from Geomagic® of the left humerus of GPIT 1 Plateosaurus engelhardti (length 351mm). (1) pointcloud-based file compared to CT file. (2) NURBS file compared to CT file.

13


the NURBS file. In contrast, maximum deviation issignificantly different between the two files. Whilethe point clouds file differs at most 1.61 mm in onesmall location on the humeral crest, the NURBS fileshows larger areas of strong deviation along prom-inent edges. The extreme value of over 15 mm,however, is limited to a small spot, where appar-ently a lofting artifact creates a deep indentation inthe loft file. Note that the large holes in the originalmesh (Figure 4.2) do not result in large errors inthe final surface due to the use of the curvature-based filling algorithm.

Pedal Phalanx II-1: The point cloud file, created inroughly four minutes, consisted of 9212 points afterremoval of erroneous points. The mesh createdfrom it required some editing due to internal poly-gons. They were apparently caused by small errorsduring recalibrations, leading to a suboptimal fitbetween the point clouds created before and afterrecalibrations. The file size is 906 kB with 18540polygons after smoothing. The NURBS file, arebuild loft with 100 control points, was created in10 minutes, most of which was spent taping andmarking the bone. It has a size of 933 kb as a STL.

FIGURE 10. 3D deviation maps from Geomagic® of the left pedal phalanx II-1 of GPIT 1 Plateosaurus engelhardti(length 73 mm). (1) pointcloud-based file compared to CT file using 5 mm scale. (2) as (1), but using 1 mm scale.

14


The left pedal phalanx II-1 was CT scannedalong with various other small elements. Originalsize was 170054 polygons and 33.1 MB. After sur-face extraction, it was reduced to 18540 polygons(10.9%) as well.

Because of the much smaller size of the pha-lanx compared to the ilium (roughly 18% if maxi-mum lengths are compared), it appearsunreasonable to demand the same degree ofabsolute accuracy for digital files of both. Accept-able error should be expressed not as an absolutevalue (e.g., 2.5 mm), but as a percentage of totalsize (e.g., 2.5% of largest dimension). Therefore,the phalanx files were 3D compared using two dif-ferent scales: the same scale that was used for theilium and humerus (0 – +/- 5 mm deviation, with the

minimum displayed deviation greater than +/- 0.5mm), and a scale with one fifth the values: 0 – +/-1mm, minimum displayed > +/- 0.1 mm.

The NURBS file 3D deviation maps for bothscales are given in Figure 10. Although the aver-age deviation is very small at ~ 0.1 mm, maximumdeviations of slightly over 2.5 mm occur at thesharp dorsal edge of the proximal articulation sur-face and on the medial distal condyle. However,these are three localized deviations that do notappear to alter the general shape of the edge if the5 mm scale is used to assess the differences (Fig-ure 10.1). Using the tighter 1 mm scale, in contrast,exposes deviations of 0.5% maximum specimenlength along all edges (Figure 10.2) and shows thatthis deviation is continuous along the dorsal margin

FIGURE 11. 3D deviation maps from Geomagic® of the left pedal phalanx II-1 of GPIT 1 Plateosaurus engelhardti(length 73 mm). (1) NURBS file compared to CT file using 5 mm scale. (2) as (1), but using 1 mm scale.

15


16

FIGURE 12. 3D deviation maps from Geomagic® of dorsal 2 of GPIT 1 Plateosaurus engelhardti (centrum length 79mm). (1) pointcloud-based file compared to CT file using 5 mm scale. (2) as (1), but using 1.25 mm scale.


and the distal condyle. This means that the com-pound error when measuring across these twopoints may exceed 1% of maximum length.

The point cloud file deviations are shown inFigure 11.1 (5 mm scale) and Figure 11.2 (1 mmscale). Here, average deviation is more than dou-ble that of the NURBS file, and the errors arewidely spread over the bone surfaces. Maximumdeviation, though, is much lower, expect for oneartifact on the ventral side near the proximal end.

Dorsal 2: The CT data, which had the same wrin-kling problems as the ilium file, was reduced to28266 triangles for use in the virtual skeleton. Itshall serve here as an example of an object with acomplex shape combined with a small file size. Themechanical digitizing file, with 51582 points (41592after removal of obviously erroneous points), wasmeshed in Geomagic® and required some filling ofholes. It took nearly 12 minutes to create. Spikeswere removed on an average setting. Now the filecontained 8611 polygons. It was now reduced to28266 polygons (32.83%), to fit the CT based file.The size is now 1.381 MB. Figure 12 shows the 3Ddeviation, both using the 5 mm scale (Figure 12.1)and a size adjusted scale running from +/- 0.125mm to +/- 1.25 mm (Figure 12.2). The standarddeviation at less than 0.4 mm is tolerable, butmany edges show strong deviation along theirentire lengths. The maximum deviation of over 7mm occurs along the sharp edges of the leftpostzygapophysis. These deviations indicate thatthe problem rests with the meshing of the pointcloud in places where interpoint distances are simi-lar between points on the same surface and pointson different surfaces (upper and lower face of thezygapophysal process). Note that the point cloudfile has an artificially constructed neural canal,while the canal of the fossil is filled with matrix.High deviations in this area are to be expected andmeaningless for digitizing accuracy.

Discussion of 3D deviation analyses

In general, the 3D deviation analyses show that –at the file size of mechanical digitizing with pointclouds or NURBS curves – the accuracy of bothCT and mechanical digitizing data is sufficientlysimilar to allow using the various formats inter-changeably for most applications. At the maximumresolution of CT data available to us, the deviationsare probably marginally larger. However, compar-ing the reduced CT files of all four bones to themaximum sized ones in 3D deviation analysesresulted in differences only in places where

there were extremely fine cracks,

there were very fine connections betweeninternal and external surfaces (irrelevant formechanical digitizing), or

there were ‘wrinkling’ artifacts present in thehigh-resolution surface, apparently caused bythe overlap of neighboring slices.

These differences all remained under 0.1% ofthe greatest length of the bone. Therefore, thereduced CT files can serve as an accurate modelof the high-resolution files.

Our sample number is low, but except for verylarge bones or extremely thin structures (e.g., sau-ropod vertebral laminae) all typical problems arerepresented by the sample. Generally, it is possibleto mechanically digitize mid-sized to large bones(>20 cm greatest length) with errors below 0.5% ofthe maximum length or 1 mm. While one of the fileswe used to assess the accuracy of our methods,the dorsal 2 file, is close to this size class andshows significantly higher errors, it is important tonote that this point cloud file was our first attempt atdigitizing a vertebra at all. The deviations, spreadout over nearly all the surface, and consistentlypositive or negative over relatively wide areas ofthe bone, are apparently caused by insufficientlyaccurate calibration of the digitizer between the dif-ferent point cloud parts. The complex shape of thespecimen and our inexperience in mechanical digi-tizing lead to a high number of recalibrations, alongwith the instable support of the specimen in asandbox. The deviations evident in Figure 12.2underscore the importance of both stable supportfor the specimen and as few recalibrations as pos-sible during the digitizing process. Files smallerthan 20 cm maximum length are probably betterdigitized using point clouds.

There is no general pattern of one of the twomechanical digitizing methods being more accu-rate or faster. NURBS digitizing suffers when thespecimen is small or has a very complex shape,but surprisingly the pedal phalanx NURBS file ismore accurate over large amounts of the surfacethan the point cloud file. However, in the NURBSfile errors concentrate in specific crucial areas,namely the extreme edges of articular surfaces.Which of the two methods is more suitable to agiven task depends on what that task is. Researchthat uses volumetric data is probably better servedby the NURBS file, while investigations concerningthe exact shape of articular surfaces, e.g., motionrange analyses, should be conducted using pointcloud based files.

17


BENEFITS AND LIMITATIONS OF MECHANICALLY DIGITIZED DATA

The obvious benefit of digital data is the easewith which it can be copied, shared and stored, andcompared to real bones and casts. Mechanical dig-itizing data has smaller file sizes than unreducedCT or laser scan based data, and is, therefore,even easier to email, upload, or use in a CAD soft-ware.

Due to the sizes and weights involved, manip-ulation of real bones, especially sauropod bones,to ascertain joint mobility, is problematic even withonly two elements. Trying to manually sort togetherpartial skeletons such as a sauropod manus with-

out sandbags or extensive custom-cut styrofoamsupports is impossible. Digital files, on the otherhand, can easily be used for this purpose, e.g., inRhinoceros® (see Figure 13, digitally mountedhand of Giraffatitan (Brachiosaurus) brancai[MB.R.2249 R9 through R17] and Wilhite 2003a,2003b, and 2005; Mallison 2007) or other CAD pro-grams (e.g., Allen 2008). Paper drawings also workwell, but are limited to two dimensions, while digitaldata can be freely rotated, sectioned, and rear-ranged as desired. Figure 14 shows a CAD mountof a complete Plateosaurus skeleton as it could beposed in a museum mount. Here, the correct artic-ulation of a large number of elements can be

FIGURE 13. Digital mount of NURBS files of the left hand of Giraffatitan (Brachiosaurus) brancai MB.R.2249 R9 –R17. Length of metacarpal 3 390 mm. Total file size 4.4 MB.

18


checked easily, and exact measurements from alldimensions can be taken with a mouse click beforeany work is done with the real bones. Exhibitdesign and arrangement can be accuratelyplanned and altered easily at any time. While thePlateosaurus skeleton in Figure 14 is derived fromCT data, the same work could also be done usingNURBS bodies from mechanical digitizing.

Sharing data with researchers abroad can beproblematic with conventional methods, too. Eitherexpensive travels are required, or casts or originalsmust be shipped at great cost. Digital files can besent via email or on CDs/DVDs instead, given suffi-cient resolution for the planned project. They alsodo not require storage room, in contrast to casts.

Digital files can also be accurately and quicklyscaled to produce proportionally correct compositeskeletons, while physical scale models must bemolded by hand, a process that requires consider-able time and resources. Skeleton drawings of thetype made famous by the work of Paul (1987,1996, 1997, 2003) usually also include only twodimensions – the width of the animal is not indi-cated. Even if a top view or a cross section drawingis available (Leahy 2003; Paul 2003), much inter-

pretation is needed. Often the operator has nochoice but to guess the third dimension in manyplaces, incurring significant inaccuracies in themodel. Sometimes, lateral and cross-section viewsof the same animal contradict each other (Paul1987: Plateosaurus engelhardti, see Mallison2007). Digital bones, in contrast, allow articulatinga digital three-dimensional skeleton. It can berotated to view it from any aspect, sectioned tofacilitate modeling sections otherwise hidden bybroader neighboring areas, and has the addedbenefit compared to measurements that errorsbecome easy to spot. Also, the exact articular sur-face geometries are depicted, whereas a drawingcan hardly detail a sloping or curving surface well.

Digital bones can also be used to produceexact casts of the original bones without subjectingthem to molding – a process that may damage thefossils even if great care is taken to minimize thephysical stresses exerted. For example, ResearchCasting International created rapid prototypingcopies of the MFN Kentrosaurus mount. Thesewere used instead of the real bones to build thearmature for the new mount in order to reduce therisk of damaging the original material. Obviously,

FIGURE 14. Digital mount of CT based files of the complete skeleton of GPIT 1 Plateosaurus engelhardti. The animalis posed running quickly, as might be done for a museum exhibition mount. Various dimensions are measured andmarked directly on the digital skeleton in cm.

19


specimens must be handled carefully, which is trueboth for moving them for CT or laser scanning andfor mechanical digitizing. Extremely fragile bonesare best scanned with touch-free methods and notsuited for mechanical digitizing. However, asidefrom minimal scratches on the surfaces of lacquer-covered specimens, we never damaged any bonesduring our digitizing.

The biggest limitation of digital data is the lackof detailed representation of surface features andcolorations. Also, the smaller the files are therougher the resolution will be, reducing detail.Hence, for delicate objects, CT data or high resolu-tion laser scans have a clear advantage overmechanical digitizing as described here and byWilhite (2003b). Additionally, all research thatrequires information not encoded in the digital filescan only be done by studying the real bones. Thisproblem can be somewhat amended by addingcolor photographs of surface colors and features tothe digital file.

For a detailed discussion of the benefits andproblems of digital data, also see Wilhite (2003a).

CONCLUSIONS

Mechanical digitizing with a Microscribe 3Ddigitizer provides a cheap and easy alternative tocomplicated high-resolution digitizing techniquessuch as CT scanning and laser scanning, at anaccuracy sufficient for most research and curatorialtasks. The accuracy of mechanical digitizing data iscomparable to CT data of similar file sizes.Mechanical digitizing provides a far superior data-base for digital 3D skeleton creation than photo-graphs, measurements, or drawings of bones.

We find that for medium accuracy or complextopographies, point cloud based digitizing worksbest, while very large objects can be rapidly digi-tized at slightly lower accuracy using NURBScurves. Costs are much lower than CT or laserscanning, especially if only NURBS elements areused, while point cloud digitizing requires one addi-tional computer program. The main cost factor,however, is the digitizer. The work time require-ments are comparable or below those of high-detailtechniques. A custom-made and adjustable holderfor specimens eases the workload of digitizing sig-nificantly, by allowing 360° access.

Mechanical digitizing data can easily beshared by email or on websites with otherresearchers around the world. Computing powerrequirements and post-digitizing workload are

comparatively low, when using our methods forNURBS digitizing, and all equipment is easilytransported in a single suitcase. Thus digitizing cantake place in collections worldwide. Transport ofspecimens to hospitals or other institutions with CTscanners is not required. The risk of loss and dam-age to specimens is reduced somewhat. However,the digitizing process itself increases the risk ofdamage more than CT or laser scanning, and thusexcludes the use of the techniques on fragile spec-imens.

The biggest drawbacks of mechanical digitiz-ing are the inability to acquire color data and thelimited resolution. However, we found that the res-olution is nearly comparable to CT scan-baseddata at similar overall file sizes.

Three-dimensional digital files can be used fora wide variety of research studies, including onto-genetic and biomechanical aspects, and are usefulfor museum display and curatorial aspects. How-ever, data from mechanical digitizing is limited toreproduction of the general shape of bones, nothigh resolution surface detail such as rugosities.Delicate structures, especially thin edges below 2mm thickness, may be significantly deformed in thedigital files, and internal surfaces can not bedepicted at all. Also, post-digitizing file editing canconsume additional time. These factors should bekept in mind before projects based on mechani-cally digitized data are planned.

ACKNOWLEDGEMENTS

This work was inspired by R. Wilhite, who pio-neered mechanical digitizing. In many discussions,he gave valuable advice and encouragement.Additional helpful discussions took place with K.Stevens, N. Murphy, A. Andersen, and S. Hart-mann. S. Perry and J. Codd loaned us a longer-range Microscribe GL. We would like to extend ourheartfelt appreciation to them all. W.-D. Heinrich,B. Schwarz, and S. Scheffel at the MFN, earnedHM’s gratefulness by helping with collectionaccess. HM is equally deep in debt to C. and D.Mackie and especially K. Krudwig of RCI foraccess to the MFN material temporarily in theircare. Two anonymous reviewer’s commentsgreatly helped to improve this manuscript.

This work was funded by the Germans Sci-ence Foundation as part of the Research GroupFOR 533 ‘Sauropod Biology’, and is contribution #54.

20


REFERENCES

Allen, V., Hertel, J., and Hutchinson, J.R. 2008. Centreof mass evolution in bird-line Dinosauria. Journal ofVertebrate Paleontology, 28 (Supplement to No.3):44A.

Bates, K.T., Manning, P.L., Hodgetts, D., and Sellers,W.I. 2009. Estimating mass properties of dinosaursusing laser imaging and 3D computer modelling.PloS One, 4(2): e4532. doi:10.1371/jour-nal.pone.0004532.

Bellmann, A., Suthau, T., Stoinski, S., Friedrich, A., Hell-wich, O., and Gunga, H.-C. 2005. 3D-Modelling ofDinosaurs, p. 3-10. In Grün, and Kahmen (eds.),Optical 3-D Measurement Techniques VII (Proceed-ings of the 7th Conference) Part 1, Vienna.

Bonnan, M.F. 2004. Morphometric analysis of humerusand femur shape in Morrison sauropods: implicationsfor functional morphology and paleobiology. Paleobi-ology, 30(3):444-470.

Golder, W., and Christian, A. 2002. Quantitative CT ofdinosaur bone. Journal of Computer AssistedTomography, 26(5):821-824.

Goswami, A. 2004. Cranial modularity across mamma-lia: morphometric analysis of phylogenetically andecologically-related variation. Journal of VertebratePaleontology, 24(Supplement to No. 3):65A.

Gould, G.C., Marcus, L.F., Gueziec, A.P., Daleo, G.F.,and Reddy, D.P. 1996. Medical CT scan visualiza-tion and its application for the study of fossils usingData Explorer and InterFormat. Journal of VertebratePaleontology, 16(Supplement to No. 3):38.

Gunga, H.-C., Kirsch, K.A., Baartz, F., Röcker, L., Hein-rich, W.-D., Lisowski, W., Wiedemann, A., andAlbertz, J. 1995. New data on the dimensions ofBrachiosaurus brancai and their physiological impli-cations. Naturwissenschaften, 82(4):190-192.

Gunga, H.-C., Suthau, T., Bellmann, A., Stoinski, S.,Friedrich, A., Trippel, T. Kirsch, K., and Helliwch, O.2008. A new body mass estimation of Brachiosaurusbrancai Janensch, 1914 mounted and exhibited atthe Museum of Natural History (Berlin, Germany).Fossil Record, 11:33-38.

Gunga, H.-C., Suthau, T., Bellmann, A., Friedrich, A.,Schwanebeck, T., Stoinski, S., Trippel, T., Kirsch, K.,and Hellwich, O. 2007. Body mass estimations forPlateosaurus engelhardti using laser scanning and3D reconstruction methods. Naturwissenschaften,94:623-630.

Gunga, H.-C., Kirsch, K., Rittweger, J., Röcker, L.,Clarke, A., Albertz, J., Wiedemann, A., Mokry, S.,Suthau, T., Wehr, A., Heinrich, W.-D., and Schultze,H.-P. 1999. Body Size and Body Volume Distributionin two Sauropods from the Upper Jurassic of Tendag-uru (Tanzania). Mitteilungen des Museums fürNaturkunde Berlin, Geowissenschaftliche Reihe,2:91-102.

Hester, P., Heckert, A., Berkhoudt, R., and Mathias, S.2004. 30,000 vertebrate fossils online: New MexicoMuseum of Natural History and Science's digitaldatabase, a research tool. Journal of VertebratePaleontology, 24(Supplement to No. 3):70A.

Hutchinson, J.R., Anderson, F.C., Blemker, S.S., andDelp, S.L. 2005. Analysis of hindlimb musclemoment arms in Tyrannosaurus rex using a three-dimensional muscoloskeletal computer model: impli-cations for stance, gait, and speed. Paleobiology,31(4):676-701.

Knoll, F., Buffetaut, E., and Buelow, M. 1999. A thero-pod braincase from the Jurassic of the VachesNoires cliffs (Normandy, France); osteology andpalaeoneurology. Bulletin de la Societe Geologiquede France, 170(1):103-109.

Leahy, G. 2003. Noses, lungs, and guts, p. 52-63. InPaul, G.S. (ed.), The Scientific American Book ofDinosaurs. Byron Press, New York.

Mallison, H. 2006. Plateosaurus was a biped - prooffrom kinematical computer modeling. Hantkeniana(Abstracts: 4th Annual Meeting of the EuropeanAssociation of Vertebrate Palaeontologists), 5:95.

Mallison, H. 2007. Virtual Dinosaurs - Developing Com-puter Aided Design and Computer Aided EngineeringModeling Methods for Vertebrate Paleontology. Doc-toral Thesis. Eberhard-Karls-Universität Tübingen,Tübingen, Germany. http://tobias-lib.ub.uni-tuebin-gen.de/volltexte/2007/2868/

Mallison, H., and Bachmann, E. 2006. Kinematicalmodeling in MSC.VisualNastran proves bipedality ofPlateosaurus. Journal of Vertebrate Paleontology, 26(Supplement to No. 3):94A.

Paul, G.S. 1987. The science and art of restoring the lifeappearance of dinosaurs and their relatives. A rigor-ous how-to guide, p. 4-49. In Czerkas, S.J., andOlson, E.C. (eds.), Dinosaurs Past and Present Vol.2. University of Washington Press, Seattle.

Paul, G.S. 1996. The complete Illustrated Guide to Dino-saur Skeletons. Gakken Mook. Japan.

Paul, G.S. 1997. Dinosaur models: the good, the bad,and using them to estimate the mass of dinosaurs, p.129-154. In Wolberg, D.L., Stump, E., and Rosen-berg, G. (eds.), Dinofest International: Proceedings ofa Symposium held at Arizona State University. Phila-delphia.

Paul, G.S. 2003. Restoring the life appearances of dino-saurs, p. 78-106. In Paul, G.S. (ed.), The ScientificAmerican Book of Dinosaurs. Byron Press, NewYork.

Ridgely, R.C., and Witmer, L.M. 2004. New applicationsof CT scanning and 3D modeling for dinosaur visual-ization. Journal of Vertebrate Paleontology, 24(Sup-plement to No. 3):103A-104A.

Ridgely, R.C., and Witmer, L.M. 2006. Dead on arrival:optimizing CT data acquisition of fossils using mod-ern hospital CT scanners. Journal of Vertebrate Pale-ontology, 26(Supplement to No. 3):115A.

21


Schwarz, D., Vontobel, P., Lehman, E.H., Meyer, C.A.,and Bongartz, G. 2005. Neutron tomography ofinternal structures of vertebrate remains: a compari-son with X-ray computed tomography. PaleontologiaElectronica, 8(2):http://palaeo-electronica.org/2005_2/neutron/issue2_05.htm.

Sereno, P.C., Wilson, J.A., Witmer, L.M., Whitlock, J.A.,Maga, A, Ide, O., and Rowe, T.A. 2007. StructuralExtremes in a Cretaceous Dinosaur. PLoS ONE,2(11): e1230 doi:10.1371/journal.pone.0001230.

Stevens, K.A., and Parrish, M.J. 2005a. Digital recon-structions of sauropod dinosaurs and implications forfeeding, p 178-200. In Curry-Rodgers, K.A., and Wil-son, J.A. (eds.), The Sauropods: Evolution and Pale-obiology. University of California Press, Berkeley.

Stevens, K.A., and Parrish, J.M. 2005b. Neck posture,dentition, and feeding strategies in Jurassic sauro-pod dinosaurs, p.212-232. In Carpenter, K., andTidwell, V. (eds.), Thunder Lizards: The Sauropodo-morph Dinosaurs. Indiana University Press, Bloom-ington.

Stokstad, E. 2000. Learning to dissect dinosaurs, digi-tally. Science, 288(5472):1728-1732.

Suthau, T., Bellmann, A., Stoinski, F., Friedrich, A., Hell-wich, O., and Gunga, H.-C. 2005. 3D- Rekonstruk-tion von Dinosauriern, p. 244. In Luhmann, T. (ed.),Photogrammetrie, Laserscanning, Optische 3DMesstechnik – Beiträge der Oldenburger 3D-Tage.Herbert Wichmann Verlag, Heidelberg.

Wilhite, R. 2003a. Biomechanical reconstruction of theappendicular skeleton in three North AmericanJurassic sauropods. Ph.D. dissertation. LouisianaState University, Baton Rouge, USA.

Wilhite, R. 2003b. Digitizing large fossil skeletal ele-ments for three-dimensional applications. Paleonto-logia Electronica, 5(2):http://palaeo-electronica.org/2002_2/scan/issue2_02.htm.

Wilhite, D.R. 2005. Variation in the appendicular skele-ton of North American sauropod dinosaurs: taxo-nomic implications, p. 268-301. In Tidwell, V., andCarpenter, K. (eds.), Thunder Lizards. Indiana Uni-versity Press, Bloomington, Indiana.

Witmer, L.M., and Ridgley, R.C. 2008. The paranasal airsinuses of predatory and armored dinosaurs (Archo-sauria: Theropoda and Ankylosauria) and their con-tribution to cephalic structure. Anatomical Record,192:1362-1388.

Zuo, O., and Jing, W. 1995. First report on studyingdinosaur egg fossils with method of medical CTscanning. Gushengwu Xuebao = Acta Palaeontolog-ica Sinica, 34(6):777-779.

22


Appendix A – The digitizing techniques in detail

Setting things up

Before digitizing begins, the digitizer, com-puter, foot pedal and the object to be digitized mustbe readied. Experience shows the following orderworks best:

1. Sort the specimens by size and stability.Determine which can be placed on the holderand which are too large or fragile.

2. Set up the holder (if used). Place the digitizerbehind it, as it is very hard to push the tipsteadily across the bone surface, but easy topull it.

3. Place the computer so that you can bothreach the keyboard and see the screen whiledigitizing. Make sure that you can reachacross and under the specimen if using theholder.

4. Start the Rhinoceros® program and load atemplate file. Using the ‘Centimeters.3dm’ fileis recommended. Save this file with the filename intended for the finished file, include thespecimen type (e.g., ‘dex radius’) and number(e.g., ‘MB.R.1664’) in the file name. Set thetolerances for the file according to the objectsize. Example values:

Absolute tolerance: 0.01 units (0.001 for smallbones)

Relative tolerance: 0.1 percent

Angle tolerance: 0.1 degree

Higher accuracy values lead to significantlylonger computation times, including the risk ofprogram crashes on less powerful computers,for little gain.

5. Prepare the first specimen for digitizing:Check the range of the digitizer arm anddecide on coordinate placement and curvedirections (see below). Usually, curves shouldbe roughly orthogonal to the long axis of thebone. Then decide on seam line and coordi-nate placement. Mark the bone accordingly.

6. Calibrate the digitizer to the first set of coordi-nates.

7. Start digitizing.

8. After data collection is completed, immedi-ately create a surface in Rhinoceros®(Geomagic® for point clouds) and check foraccuracy. Only if the surface is roughly satis-factory, remove markings from bone. Other-wise redigitize non-satisfactory areas.

We recommend digitizing several curves inquick succession, without consulting the laptopmonitor often, instead of checking each new curvefor accuracy immediately. A smooth, uninterruptedwork flow is key to short digitizing times.

Coordinate placement, recalibrationand seam line placement

Coordinates and recalibration (multiple coordi-nate sets): Before digitizing can begin, coordi-nates for recalibration must be marked on thespecimen as well as (when using closed curves,see below) a seam line (line through the contactsof all curves start and end points). In most cases,thorough planning of the placement of coordinatesand the seam line is necessary to avoid compli-cated recalibrations of the digitizer. Sometimes, itis not possible to avoid a recalibration, but reducingthe number of instances necessary will result infewer inaccuracies. Also, the fewer different sets ofcoordinates are used the smaller the inaccuracies.

For small and medium sized specimens,approximately up to a size of 80 cm greatest length(110 cm for the Microscribe® GL), a single set ofcoordinates located roughly halfway down thelength of the bone is sufficient. Three coordinateson the specimen are needed: an origin point (O1)for the origin of the coordinate system and twopoints (X1, Y1) to determine the direction of the x-and y-axis respectively. These can be placed inany relation to each other except for a straight line,because Rhinoceros® translates into a Cartesiancoordinate system internally. Thus there is no needto place the coordinates in a right triangle. It isadvisable to space them at least 5 cm apart in eas-ily accessible locations to reduce the influence ofthe unavoidable slight inaccuracies during recali-bration. Multiple coordinate sets allow digitizingvery large objects; theoretically there is no sizelimit.

Coordinates should usually be placed (seeAppendix C) so that one set (Cset1=O1, X1, Y1) isaccessible in all positions the specimen will have tobe placed in during digitizing. If this is not possible,a second set (Cset2=O2, X2, Y2) must be placed sothat it can be reached with the digitizer after cali-bration through Cset1. This means that two sets ofcoordinates should be placed at approximately 1/4and 3/4 of the length of the bone to allow maximumrange for the digitizer.

23


Complex bone shapes, or large flat bones(e.g., sauropod ilia) may require more sets of coor-dinates. Cset2, 3, etc. should all be accessible fromCset1 to minimize inaccuracies. Thus Cset1 shouldbe placed roughly halfway down the bone, withsets of higher number to both sides.

Small or ball-shaped flat bones (e.g. calcanei,dermal scutes) tend not to rest stably on the holderunless placed horizontally. Here it proved best touse one set of coordinates placed on the narrowedges, digitize curves as concentric rings on theupper surface, then flip the bone over onto theother side and digitize concentric curves there (Fig-ures 15), using the same coordinate set Cset1.

Seam line: The seam line is an imaginary line con-necting all curve starts and ends when digitizingusing closed curves (Figures 6 and 15). Properplacement of the seam line is equally important asthe placement of the coordinates. The seam lineneeds not be digitized, but should be marked onthe bone. It should run on a relatively flat area ofthe bone, where the lofted surface will show littlechange in topology. Also, the bone should rest sta-bly on the holder (or against other support) with theseam line positioned downwards (on the sideopposite to the digitizer and the operator whenother support is used); otherwise access to it fromboth directions will be difficult. It can be helpful todigitize a short open curve down part of the seamline to gain a reference in Rhinoceros®. This helpsselecting the curves for lofting properly if selectionby hand is necessary. When digitizing closedcurves, the seam line must always be placed onthe side of the bone away from the digitizer, other-wise the reach of the digitizer arm will not be suffi-cient to draw the curve completely.

Gathering data -open and closed curves, point clouds.

Open curves: Open curves run across one side ofthe specimen as subparallel lines, requiring accessto only one side of the specimen. Wilhite (2003b)used this technique exclusively. A loft over opencurves results in a surface. Open lofts may, butneed not, start and end with a point object. Joiningthese surfaces into closed bodies (solids) is oftendifficult, thus this technique is not recommended.

Closed curves: The most important improvementwe made compared to the technique of Wilhite(2003a, 2003b) is the use of closed curves. Thismeans that each curve reaches 360° around thebone as an infinite loop (Figure 2), allowing aclosed loft over the entire bone in one step. Thus,

there is no need to assemble two surfaces into onebody, a process very difficult in Rhinoceros®. Thissaves effort, increases accuracy, and reducescosts by making the purchase of a separate editingprogram unnecessary. Additionally, a closed loftdoes not possess a visible seam that has to bemanually smoothed over in Rhinoceros®. Itrequires, in addition to closed curves, a start andan end point at each end of the loft. These pointscan be digitized at any time before, after or inbetween curves. When the bone it too large to bedigitized without moving the digitizer, the pointsshould be digitized together with the neighboringcurves, to avoid recalibrations for just one point. Ifseveral separate lofts are combined to model com-plex shapes, surfaces open at one or both endscan be used. These require one or no points,respectively.

In order to achieve a surface with minimumartificial distortion, all curve ends must meet therespective curve starting points with minimumoverlap and shift along the seam line (Figure 16),and point in roughly the same direction (have simi-lar tangency). To achieve this it is useful to markstarting points on the bone by taping a strip ofadhesive tape (masking tape) along the intendedseam line (usually the long axis of the bone) andmark curve starts by a lengthwise line with crossmarks. This has the additional benefit of reducingwriggling of the seam line, avoiding a commonsource of massive lofting artifacts. To avoid overlapa small gap of 1 or 2 mm should be left betweencurve start and end, which Rhinoceros® closesautomatically when the foot pedal is released.Also, to minimize distortion at the bone ends, it isoften advisable to cover at least a circle with r=2.5cm at each end with masking tape and draw thefirst and last few curves onto the tape prior to digi-tizing (Figure 6). The end-points should also bemarked here.

Composite closed curves: Some bones are solarge that drawing closed curves around them isimpossible due to the constricted range of the digi-tizer arm, e.g. sauropod ilia, or bones that are heldin fixed mounts. Here, it is advisable to createclosed curves by digitizing them in parts. Each partis an open curve, and the parts are joined togetherusing the ‘match’ and ‘join’ commands. In theory,there is no size limit for this method! The onlydrawback is the need for extremely accurate digi-tizing at the contact points of partial curves. Thisrequires extensive marking of the bone prior to dig-itizing, as each separate contact point must bemarked. Also, it is often necessary to redistribute

24


the sampling points of the curve more regularlyafter joining the various parts. This can be done viathe ‘rebuild’ command and slightly decreases accu-racy. Note that in case of a bone mounted with ametal rod that closely follows the shaft longitudi-nally, it is also possible to digitize with closedcurves and edit the curve control points to removethe armature instead of using composite curves.

Points: Single points are collected using the ‘point’command. They are useful to mark coordinatesand as start and end points for closed lofts. The‘points’ command can also be used, but if the digi-tizer tip is not kept very still, a string of point objectswill be digitized. We recommend deleting surpluspoints, as they can lead to confusion and loftingerrors.

Point clouds: With the ‘digsketch’ command pointclouds (Figures 3 and 4) can be digitized continu-ously or in several parts, without having to worryabout slipping off the object with the digitizer tip.Complex shapes can be sampled better with pointclouds than with curves. Also, complete reacharound the object is not necessary, nor planning

partial curves for joining into closed ones. This isuseful when bones are mounted closely togetherand can not be taken off the mount for digitizing.

The object is placed on a stable support, e.g.placed in a sandbox. Very small objects can beheld in place on the table with two fingertips. Coor-dinates must be marked so that they are accessi-ble in all positions necessary for digitizing thecomplete bone. Now, point clouds are digitizedover the entire accessible surface. Usually, severalpercent of all points digitized are erroneous. Thesecan, however, usually be spotted easily, andquickly removed. Then the object is turned over,the digitizer recalibrated, and the remaining sur-faces are digitized. Experience tells that drawingthe digitizer tip along all edges repeatedly is advis-able; larger flat areas can be painted in roughlywith a to and fro movement of the digitizer. Notethat near sharp edges, such as cristae or theedges of transverse processes, artifacts willappear near the edges of the flat surfaces if thesampling distance on the surface is not signifi-cantly smaller than the thickness of the bone.

FIGURE 15. Digitizing small round or flat specimens. Giraffatitan (Brachiosaurus) brancai MB.R.2246 left calcaneum.(1) Curves from digitizing; top half shows first set digitized before the bone was turned upside down. Bottom halfshows complete set. (2) Photograph of the specimen readied for digitizing. Important markings are labeled. (3) Fourviews of the final 3D file from Rhinoceros®. Greatest length of the bone is 125 mm.

25


Meshing then erroneously connects points fromboth sides to each other instead of to the points atthe edge (Figure 5). Geomagic® produces fewermeshing errors that Rhinoceros®. The samplingdistance should be at most 0.2 times the distanceof the surfaces to avoid this. During digitizing, it isadvisable to create meshes from time to time inorder to judge which areas need further digitizing. Itis also possible to digitize with this preliminarymesh visible, best in ‘Shaded’ viewport mode,which facilitates the task. Alternative, the mesh canbe created in Geomagic® and viewed while digitiz-ing, as Rhinoceros® will accept digitizer input evenif it is running as a background process (seeAppendix E).

Although curves for lofting can be createdfrom these points by a variety of methods, usingthe ‘wrap’ function in Geomagic® usually is thebest option to create a polygon mesh. If someareas prove troublesome, separate meshes can becreated for parts of the point cloud and then com-bined. The gaps can be filled with the ‘FillHoles’function of Geomagic®.

Editing raw data

Editing curves: Curves can be edited to removeartifacts in them or to join several curves into one(commands: ‘controlPts’, ‘rebuild’, ‘match’, ‘join’).The concept of ‘control points’ and their use isexplained below. As this editing changes the origi-nal input data as few changes as possible shouldbe made.

Editing point clouds: Point clouds can be editedto remove points that are either incorrectly col-lected during digitizing or supernumerary. In Rhi-noceros®, point clouds are ‘groups’ of points. Eachconsists of all points collected during one period ofsampling (keeping the foot pedal pressed). In orderto edit single points in a cloud, the cloud must beselected and separated (command ‘explode’).Erroneous points are best spotted by rotating theview until it is nearly parallel with the bone surface.Erroneous points will now be visible above themain bulk and can be deleted (Figure 17.1).Remember to also check the inside of the pointcloud for stray points! To remove these it is advis-able to select and hide (‘hide’) the ends of the pointcloud in order to create a clear background againstwhich the erroneous points can be easily spotted.

If a digitizing error is detected during the digi-tizing of the point cloud (that is while the foot pedalis still pressed), the tip of the digitizer should beremoved a generous distance from the bonebefore the foot pedal is released. This way the

group of points that contains the faulty points has a‘trail’ of points leading away from the surface that isquite conspicuous and facilitates finding andremoving that specific group (Figure 17.2).

Creating bodies - lofting and joining surfaces

Surfaces from curves: Lofting a surface overclosed curves (command ‘loft’) is the easiest wayto create surfaces for digitized objects in Rhinoc-eros®. A loft that, in addition to the curves, con-tains a point object at one end will be closed at thatend, but still be a surface, not a body. A loft thatboth starts and ends with a point will create a body,not a surface. This method requires the least post-digitizing data editing. In order to loft, the respec-tive curves must be selected and the proper loftoption chosen. Here, different versions of Rhinoc-eros® differ markedly. Often, Rhinoceros3.0® doesnot reliably sort curves correctly, so each point/curve must be selected by hand in the properorder, starting with one endpoint, then the closestcurve, then the next etc. to the other end of thebone. In Rhinoceros 2.0® and earlier version, eachcurve has a direction that is not automaticallyadjusted during lofting. It is necessary to selecteach curve at the same side of the seam line; oth-erwise the surface will fold into itself. Rhinoceros3.0® sorts the directions automatically, usually cor-rectly. Rhinoceros 4.0® sorts curves correctly, butusually misplaces the endpoints, sorting each point

FIGURE 16. Example of NURBS curves in Rhinoc-eros® showing ‘shift’ and ‘overlap’ errors.

26


with the wrong end of the bone, so curves andpoints must be selected in the correct order byhand for closed lofts. The ’closed loft’ option cre-ates the same errors and should not be used.

Note that large bones digitized at high accu-racy will lead to long computation times for lofting.It may be advisable to increase the tolerances val-ues in the file preferences before lofting, as this willnot add a significant error but speed up lofting byup to 90%. Additionally, we have experienced pro-gram crashes at high accuracies, which can beavoided by downgrading the accuracy after datacollection and before lofting.

Surfaces from point clouds: Meshing is doneautomatically via the ‘Meshfrompoints’ command inRhinoceros®. Much more accurate results can beachieved in Geomagic®, using the ‘Wrap’ function.If the resulting mesh shows many inaccuracies,deleting it and editing the point cloud for a newmeshing is best. If there are few errors in the cre-ated surface, it is best to delete erroneous meshfacets (‘Deletemeshfaces’) and fill the resultingholes via the ‘fillhole’ command in Rhinoceros®. InGeomagic®, faulty areas can be selected directlyand removed by pressing ‘Del’. The holes can befilled using the ‘Fill Holes’ function.

In Rhinoceros®, tolerance settings shouldremain tight for meshing, as low values will result insignificant errors.

Joining and editing surfaces: Several NURBSsurfaces can be joined into one via the ‘joinsrf’command. Alternatively, for smoother contacts, it ispossible to use ‘blendsrf’. This requires that thesurfaces touch along all their common edge. Asthis cannot be reliably achieved when digitizingseveral surfaces on one object due to the invari-able drift of the digitizer and the inaccuracies of theoperator’s hand movements, edge contact must becreated artificially in Rhinoceros 3.0®. We couldnot evaluate the mesh edit tools in Rhinoceros 4.0because the program regularly crashed (updateduntil 9/2008).

Sometimes it is extremely difficult to join par-tial surfaces in Rhinoceros®, or close openings inpartial lofts. Also, cropped NURBS bodies can be

difficult to ‘cap’ properly. In this case, the NURBSsurface(s) should be transferred into an STLpolymesh. Further editing can then take place inGeomagic®.

Control points: Surface editing via the variousoptions in Rhinoceros® is usually too time consum-ing and difficult to be useful, except for the use ofcontrol points. Any ‘surface’ or ‘body’ (i.e. ‘closedsurface’) can be edited in Rhinoceros® via its ‘con-trol points’. These are point objects representingcoefficients of the NURBS function of a curve(called ‘nodes’ in many other programs). Movingthem in 3D space alters the curve function, chang-ing the shape of the curve and the associatedNURBS surface. In order to change the form of abody, the control points must be turned on(‘Cpoints’). Now, they can be selected and movedthe same way any point object can be moved inRhinoceros®. A lofted body has hundreds or thou-sands of control points along the surface iso-curves, which are essentially identical to the curvesthe body was lofted from. Editing larger areas byhand is a tedious process, but small deformationsor digitizing errors can be easily removed. Rebuild-ing the individual curves with a smaller number ofpoints before lofting makes editing of the lofted sur-face easier, but the accuracy can suffer signifi-cantly if the number of points is reduced too much.

Since a ‘control point’ is a coefficient of aNURBS function, the surface will not always passthrough the point object visual on the screen, andcare should be taken to always move neighboringpoints together. Otherwise, the surface topologywill show artifacts resulting from the rapid ‘swing-ing’ of the underlying spline function. If, e.g., onepoint is moved by 5 mm, then the two neighboringpoints should each be moved in the same directionby a few millimeters, too.

Polygon meshes are easy to combine inGeomagic®. The ‘FillHoles’ function allows build-ing ‘bridges’ between them. In order to do so, themeshes must first be merged using the ‘MergePolygon Objects’ command. The remaining open-ings can be filled with the ‘FillHoles’ command.

27


28

FIGURE 17. GPIT 1 Plateosaurus engelhardti (1) Pointcloud file of left pedal phalanx II-1 (length 73 mm) prior to edit-ing. Red circles mark obvious erroneous points. (2) Pointcloud file of dorsal 2 (centrum length 79 mm) during digitiz-ing. Yellow points: pointcloud with digitizing errors noted while the points were being digitized. The ‘trail’ of pointsleading to the top left was intentionally created to allow easy recognition of the flawed pointcloud during later editing.


Appendix B – Application of techniques to different types of specimens

Digitizing with closed curves – general remarks

Plan coordinate and seam line position priorto digitizing and mark them. Keep the limited rangeof the digitizer arm in mind! When aiming for aclosed loft, remember to digitize start and endpoints! Closed curves may lead to artifacts, usuallyalong the seam line or in places where neighboringcurves are of significantly different lengths or theirdistance and direction varies greatly. The same istrue if closed curves overlap or shift at the seamline. Here, using the ‘Simplify: Rebuild…’ option ofthe ‘loft’ command dialog box can reduce orremove the problem. Alternatively, the curves canbe edited prior to lofting via the ‘rebuild’ command,and/or by editing their ‘control points’.

Digitizing with closed curves – small bones (simple shapes)

Small bones up to 10 cm maximum lengthtend to be hard to digitize for a number of reasons.They shift position easily, drawing curves on oraround them at relatively regular intervals withoutoverlap alone is hard, and ensuring that thesecurves accurately reproduce small surface featuresmakes the task of drawing curves free-handed toocumbersome. Using the fossil holder reduces thefirst problem. Still, care must be taken not to putpressure on the specimen with the tip of the digi-tizer as this may shift the bone. As it is hard toplace a small bone on the holder stably while stillbeing able to reach both upper and lower surfaces,we recommend first digitizing only the upper halfwith concentric closed curves in a horizontal plane(Figure 15.1), then flipping the bone over andrepeating the procedure for the other half. Thisrequires a coordinate set Cset1 placed on the cir-cumference of the bone (Figure 15.2).

The problem of drawing curves properly -especially closed curves - can be solved by exten-sive marking of the bone. For this, much or all ofthe bone is covered in painter’s adhesive tape ormasking tape and the seam line and curves aredrawn onto it (Figure 15.2). It is also possible toonly mark partial curves or markers for curve pas-sage over radially placed strips of tape if the boneis too large for complete wrapping. It is of para-mount importance to select a tape that can beremoved completely without damaging the bone.The resulting curve set can be lofted normally (Fig-ure 15.3).

Digitizing with closed curves –medium sized bones

For medium sized elements (up to around 90cm maximum length), curves should be placedroughly perpendicularly to the long axis of thebone, with the seam line on the lower side. Thespecially constructed holder again proved best, asit keeps the specimen from shifting while allowingeasy access. Usually, it will not prevent access toimportant features and curves placed directly to theleft and right of the fittings are sufficient to give anaccurate surface. If not, it is necessary to shift thebone a few centimeters and recalibrate the digitizerbefore adding curves into the gaps. See Figure 6for an example of this setup.

Digitizing with closed curves – large bones

Drawing closed curves requires that it is pos-sible to reach around the specimen on both sidesdown to the seam line. For very large specimens,especially relatively slender longbones like tibiaeand radii of sauropods, it is advisable to place thespecimen close to vertical by leaning it against asolid structure, with a cut-to-fit styrofoam spacer tocreate a gap in between. Some bones, especiallysauropod longbones, rest stably in this position ontheir articular ends. If necessary, the lower end canbe secured with wooden wedges and small sand-bags or in a sandbox. Thus, bending moments onthe shaft are minimal, reducing the risk of damage.The digitizer may have to be set up on a woodenbox or a chair in front of the bone to achieve a suf-ficient vertical reach; care must be given to stabil-ity. For large specimens, a forklift or similaradjustable tool can be used. This allows digitizingthe top end and shaft with curves sub-parallel tothe ground; then the fossil must be turned upsidedown, the digitizer recalibrated and the bottom endadded. The seam line here runs down the ‘back’ ofthe bone, opposite the digitizer and operator duringthe first part, but facing the digitizer during the sec-ond part. When using early versions of Rhinoc-eros®, care must be given to either digitize thecurves in the opposite direction after rotating thespecimen (seam line now at front) or change curvedirection in Rhinoceros® before or during lofting.

Digitizing with composite closed curves (very large bones and/or complex bone shapes)

Bones that are too large in circumfence forclosed curves or encased in supports that makereaching around them impossible can be digitized

29


by drawing sets of open curves. Each curve in aset ends at the starting point of a curve fromanother set. The matching curves are then mergedinto one closed curve. Two or more curves togethermust fully surround the bone the way a closedcurve would. This requires extensive marking ofthe bone to minimize overlap and shift betweentouching curves (Figure 16). It is advisable to placethe contact line between curves at places with lowcurvature (i.e. flat bone surfaces) to minimize arti-facts and leave a very small gap between curves.The partial curves then must be joined into closedcurves using the ‘match’ and ‘join’ commands.Then, a closed loft can be made over the entirebone.

Usually, it is necessary to rebuild matchedcurves using the ‘rebuild‘ command to avoid arti-facts, as the control points will be unevenly distrib-uted along each curve. An additional advantage ofrebuilding curves is the ability to somewhat reducethe number of control points, making both the finalfile smaller and editing via control points simpler.This must be executed carefully, as too large areduction in control point numbers can induce sig-nificant shape changes in the curves.

Digitizing bones as composite bodies with separate curve sets - complex bone shapes

Some bone shapes cannot be represented byone loft over closed curves. In these cases, it ispossible to digitize parts of the bone, loft surfacesover these parts, and then combine them into onebody in Rhinoceros®. Examples are the articularheads of sauropod ribs, where one set of closedcurves cannot easily be drawn around both heads.Digitizing as large a part as possible with curves inone direction, then adding a second set of curvesfor the missing parts is usually the best option,although it is sometimes preferable to aim to mini-mize the suture length between the parts.

Several points must be kept in mind whenplanning a composite body:

- The separate parts must together contain theentire surface or volume of the specimenwhen using open surfaces or closed surfaces,respectively; otherwise there will be gaps orholes. If holes cannot be avoided, they shouldbe in flat areas of the bone, where filling themup in Rhinoceros® is easy and carries a lowrisk of error.

- Boolean operations on bodies (3D volumes)are much easier to conduct than joining oper-ations on surfaces.

- The separate lofts must overlap enough orcontact each other exactly to allow joining thebodies in Rhinoceros®.

- The overlap area should not contain repeatintersections between surfaces/bodies asthese tend to confuse Rhinoceros® and makemerging impossible. It is helpful to close loftedbodies by placing the end point inside thebone by manually adding it in Rhinoceros®.This guarantees overlap with only one sectioncontact.

- It is advisable to digitize a closed curvealong the contact line of the parts to be joined.This facilitates creation of a cutting plane,which allows cropping the parts to an exact fit.Ideally, the contact line allows the use of a flatplane for cutting.

- To create holes in a body (e.g., foramina inischia or coracoids) it can be easier to digitizeand loft an air volume and subtract it from themain body by a Boolean operation than split-ting the main body into parts and joining them.

- Remember to set the tolerances tight enough,otherwise cutting and joining surfaces may notwork properly.

Using open curves and multiple lofts: Aclosed curve should additionally be digitized alongeach suture line to facilitate closing each open loftinto a closed body before joining the adjacent bod-ies.

Using closed curves and multiple lofts: Aclosed curve should additionally be digitized alongeach suture line to facilitate cropping both adjacentbodies to avoid problems when joining them.

When it is not possible to digitize one closedcurve around the end of the surface it is possible todigitize it in parts, and then join them into one curveusing the ‘match’ command.

These methods often fail for joining surfaces,due to problems with joining cropped surfaces. Westrongly recommend using multiple closed lofts(bodies) and uniting them with Boolean operations.

Ideally, complex shapes should be digitizedusing point clouds. If that is not possible or desir-able, the above procedure can also be used in avariant that delivers a polygon mesh instead of aNURBS object as the final result: Digitizing takesplace with separate curve sets, then all surfacesare ‘meshed’, after which the meshes can bejoined in Geomagic®.

30


Digitizing with point clouds/meshes (small bones with complex shapes, partially

inaccessible bones)

Any bone too small to rest stably on the holderand/or with a shape too complex to be easily digi-tized by the above method for small bones usingclosed curves should be placed either on clay/PlayDoh® or in a sandbox. The ‘Digsketch’ com-mand allows collection of point clouds. Care mustbe taken to ensure that the distance between clos-est points on the same surface is always signifi-cantly smaller than the distance to points on anyother surface; otherwise artifacts will be producedduring meshing. ‘Meshfrompoints’ produces a poly-gon mesh that can be edited via the ‘deletemesh-faces’ and ‘fillhole’ commands. Reducing the meshcount in Rhinoceros® is usually not advisable, asthe errors incurred are significant. Both meshingand editing of the mesh are better carried out inGeomagic® (see Appendix D). This is especiallytrue for filling gaps in the mesh.

Theoretically, Geomagic® also supports theMicroscribe 3D digitizer. For unknown reasons, wewere not able to make the digitizer connect to theprogram correctly. Despite this technical problem,Geomagic® can be a helpful tool during digitizingof point clouds. It is possible to collect data with thedigitizer in Rhinoceros® while the other program isdisplayed on the screen (see Appendix E). There-fore, one can digitize points in Rhinoceros®, inter-mittently mesh them in Geomagic®, and then goon collecting data to fill the gaps shown in themesh in Rhinoceros® while the mesh inGeomagic® is actually visible on the screen. Dueto the far better ability to selectively collect datawhere it is needed, this process can save up to30% of the digitizing time. Also, errors will show upearlier, allowing faster correction.

Marking surface features

Rugosities, damaged or discolored areas,articular surfaces or muscle attachment scars canbe marked on the surface by drawing a closedcurve around them. This curve can later be pro-jected onto the lofted surface. While this data willusually be lost when exporting the 3D bodies intoother programs, some formats allow retention.

Also, a very thin body can be constructed from thecurve via the ‘extrude’ command; this will be pre-served in all formats. Rhinoceros® allows changingthe color of these markings to make them moreconspicuous. Alternatively, the area can be cov-ered with a point cloud, from which a polygon meshcan be created.

Composite bones from partially preserved specimens

Here, an explanation is given for combiningtwo parts; a higher number can be combined byrepetition. There are two methods for combiningdata from two specimens. It is possible to select atleast three landmarks, which are present on bothobjects, and digitize them as coordinates for acoordinate system, then digitize the specimensafter calibrating to these landmarks. This workswell if there are e.g. small foramina or sutural con-tacts with ontogenetically stable positions presenton the bones. Otherwise, the landmarks can bedigitized as points, point cloud or NURBS curvedata on both objects, and the data sets can bealigned later. Optimally, only one set of closedcurves is needed, with the last curve of one speci-men subparallel to the first of the second speci-men. If only a contralateral element is available tocomplete a partial specimen, it is possible to digi-tize one part including the landmarks and then mir-ror it before digitizing the other element.

If no landmarks are present, it is possible tojust digitize the specimen without calibrating fromone to the other, then copy and paste them intoone file and move one until it appears to fit theother. This process works well if there are cristaeor spinae present, or any other feature on the bonethat allows easy 3D orientation of the separateparts. For both methods it is possible to scale(‘scale’ command) one part isometrically to fit theother. If a large size difference exists, it may benecessary to scale the element differently for vari-ous axes to compensate for the allometric growthcommon in vertebrates, which is possible with the‘scale2D’ and ‘scale1D’ commands. For this, thebone should be aligned with the coordinate axes ofRhinoceros®. In any case, a note should be addedto the file indicating the composite nature of themodel.

31


Appendix C – Step-by-step instructions for Rhinoceros® for digitizingand model creation techniques

Abbreviations used herein:

TIP – tip of the digitizer arm

RFP – right foot pedal

LFP – left foot pedal

If asked to ‘at the command line enter…’,ensure that Rhinoceros® is ready for a new com-mand (not already processing a command) andtype the exact phrase given, then press theENTER or RETURN key on the keyboard. ‘Select’means click the appropriate box with the mouse sothat a tick appears in the box, ‘deselect’ means toclick the box so there is no tick in it.

‘Menu:’ refers to the pull-down menus at thetop of the Rhinoceros® screen. Select the menutitle given, and then select the (sub-)entry/-iesgiven after the arrow(s) -->.

For most commands, icons are available inRhinoceros®. In the standard workplace setup,only a few are visible. These are referred to in thetext. To make others available, refer to the in-pro-gram help.

Commands are not case sensitive in Rhinoc-eros®, thus ‘points’ and ‘Points’ are equivalent.

Note for users of earlier versions of Rhinocerosthan 3.0 SR4: As opposed to Rhinoceros 3.0®SR4 and later versions, earlier versions do notautomatically repeat the ‘digsketch’ command. It isnecessary, after digitizing a curve or point(s), topress the left foot pedal (LFP) once to repeat thecommand. Also, any version before ServiceRelease 4 (SR4) will not support polygon meshcreation via the ‘meshfrompoints’ command.

Note for users of more recent versions of Rhi-noceros (4.0 and updates): Rhinoceros 4.0 has aslightly different user interface, therefore somecommands are available from a different ‘Menu’than in earlier versions. The basic functions havenot changed, and all techniques described hereincan still be used.

Digitizing preparations

To begin digitizing it is recommended toopen a new file via the ‘New’ icon. This displays aselection of template files with varying units. Forthis study, the ‘centimeters.3dm’ template wasused for all objects. This sets the scale of the coor-dinate axes as centimeters. Digitizing distance forcurves should be selected accordingly. Open thefile properties page (Menu: File --> Properties),

select ‘Units’ and set the tolerances to values suit-able for the task.

Example values:

Absolute tolerance: 0.01 units (0.001 for smallbones)

Relative tolerance: 0.1 percent

Angle tolerance: 0.1 degree

Tolerances should be at least an order ofmagnitude higher than the smallest value used inany operation on the data. Joining surfaces, e.g.,requires distances smaller than 0.01 mm as thesurfaces themselves are a theoretical 0.01 mmthick. Thus, absolute tolerance should be set to0.001 mm. Angle tolerance should be equally low,as joining lines between surfaces otherwise willshow as kinks.

It is possible to create a new custom tem-plate file that includes appropriate tolerance set-tings. See the Rhinoceros® help file for details onhow to save a file as a template.

Digitizing coordinates, (re-)calibration with and without moving the digitizer or specimen

Initial setup and initial calibration: Before digitiz-ing, mark three coordinates on the specimen.

1. Place the digitizer so it can reach all threecoordinates.

2. Connect the digitizer to Rhinoceros® (Menu:Tools --> 3-D Digitizer --> Connect, or enter‘digcalibrate’ in the command line), select Dig-itizer type (Rhinoceros® suggests the FaroDigitizer; if using a different one, press ‘f’ andRETURN, then the first letter for your type ofdigitizer, confirm selections with RETURN).Rhinoceros® 4.0 offers a selection menuinstead.

3. When asked to ‘Enter origin with digitizer’ inthe command line, place the tip of the digitizer(TIP) on the mark for the origin (O1) on thespecimen and press the right foot pedal(RFP).

4. When asked to ‘Enter X-Axis with digitizer’ inthe command line, place TIP on the mark forthe x-axis (X1) on the specimen and pressRFP.

5. When asked to ‘Enter Y-Axis with digitizer’ inthe command line, place TIP on the mark for

32


the y-axis (Y1) on the specimen and pressRFP.

6. When asked to ‘Enter origin in Rhinoceros®(…)’, press RETURN

Now the digitizer is ready for digitizing curvesor points. The origin of the displayed coordinatesystem in Rhinoceros® coincides with O1 on thebone; the X-axis in Rhinoceros® runs in the direc-tion of X1. Note that the coordinate system is auto-matically transferred into a Cartesian system.Therefore, the Y-axis in Rhinoceros® runs onlyroughly in the direction of Y1.

Recalibration: To calibrate the digitizer back to theCset1 (O1,X1,Y1) proceed as follows:

1. In the command line, enter ‘digcalibrate’ orchoose ‘Menu: Tools --> 3-D Digitizer --> Cali-brate’

2. Proceed as above, starting with step 3.

The digitizer is now recalibrated to the originalcoordinate set Cset1.

Using several sets of coordinates: When usingseveral sets of coordinates, it is important to markeach set on the bone as well as in Rhinoceros® toavoid mixing them up. We found it easiest not toadd text or change object names, but connect eachset with a polyline and change the color of each Xpoint. X1 was left unchanged (black), X2 changedto white, X3 to red etc., following the default orderof colors in Rhinoceros® layers. If this simple con-vention is kept, recalibration takes less time thanwith any other method, while mix-ups can beavoided.

In order to use several sets, mark all sets onthe bone, calibrate the digitizer to Cset1 as above,and then proceed as follows:

1. In the command line, enter ‘point’.

2. Place the TIP on O2, press and release RFP.Watch the screen to note where the pointappears. If necessary, shift the view using the‘pan’ tool.

3. Press [space] or RFP to repeat the ‘point’command. Repeat 2. to digitize X2 and Y2,note where they are on the screen. PressESC after digitizing Y2.

4. At the command line enter ‘Polyline’.

5. On the screen, with the mouse select X2, O2,and Y2 in this order, press ENTER. To facili-tate this, make sure that ‘OSnap’ is on and setto ‘Points’. See Rhinoceros® help for the use

of ‘OSnap’. A polyline will appear with O2 atthe meeting point of the two sections.

6. On the screen, select X2 (make sure the otherpoints are deselected), press F3 and changethe color of X2.

7. Repeat 2. to 6. until all sets have been digi-tized.

To calibrate the digitizer to a different set(e.g., set 2), make sure the second set of coordi-nates has been digitized as points as describedabove. Place the digitizer so it can reach all threepoint of the set to calibrate to and proceed as fol-lows:

1. At the command line enter ‘digcalibrate’

2. When asked to ‘Enter origin with digitizer’ inthe command line, place TIP on the mark forthe origin (O2) of this set on the specimen andpress the right foot pedal (RFP).

3. When asked to ‘Enter X-Axis with digitizer’ inthe command line, place TIP on the mark forthe x-axis (X2) on the specimen and pressRFP.

4. When asked to ‘Enter Y-Axis with digitizer’ inthe command line, place TIP on the mark forthe y-axis (Y2) on the specimen and pressRFP.

5. When asked to ‘Enter origin in Rhinoceros®(…)’, use the mouse to select O2 in Rhinoc-eros®. To facilitate this, check that ‘Osnap(points)’ is on.

6. When asked to ‘Enter X-Axis in Rhinoceros®(…)’, use the mouse to select X2 in Rhinoc-eros®.

7. When asked to ‘Enter Y-axis in Rhinoceros®(…)’, use the mouse to select Y2 in Rhinoc-eros®.

The digitizer is now calibrated to Cset2.

Digitizing and editing curves, points and pointclouds

Neighboring curves must be of similar lengthand should be roughly parallel. Large differences inlength or distance produce artifacts. Also, curvesmay not cross each other.

Digitizing points: The digitizer must be properlyconnected and calibrated before data can be gath-ered.

1. At the command line enter ‘points’ or click theappropriate icon in the 3d digitizing toolbar.

33


2. Place TIP at the desired location for the pointand press RFP

3. To digitize more than one point, repeat 2.

4. To end data collection, press ‘Esc’.

Digitizing point clouds: A point cloud is a groupof points that can be digitized in one continuousmotion and is automatically grouped in Rhinoc-eros®. Ungrouping works best via the ‘explode’command.

The digitizer must be properly connected andcalibrated before data can be gathered.

1. At the command line enter ‘pointcloud’ or clickthe appropriate icon in the 3d digitizing tool-bar. It is also possible to enter ‘digsketch’ andselect the appropriate options.

2. Choose the desired sampling density bypressing ‘d’, [ENTER], and entering the num-ber followed by [ENTER].

3. Place TIP at the desired location for the pointand press and keep pressed RFP.

4. Move TIP across area for which points are tobe digitized.

5. Release RFP to end point data collection.

6. To digitize more than one point cloud, repeat2. to 4.


Digitizing open curves: The digitizer must beproperly connected and calibrated before data canbe gathered.

1. At the command line enter ‘digsketch’ orselect ‘Menu: Tool --> 3D Digitizer --> SketchCurve’.

2. In the command line deselect ‘Points’ bypressing ‘p’ followed by ENTER and similarlyselect ‘Curves’ and deselect ‘Closed Curves’.Choose the distance at which points are to beplaced on the curves by Rhinoceros®. Notethat this distance depends on the scale of thetemplate file you chose! Usually, 0.1 to 0.7proved to be acceptable values, for very largeobjects with hardly any sudden curvaturechanges even a value of 1 is acceptable(using the centimeter template; change valueaccordingly if using other templates). Enterthe value by changing the ‘Point Spacing’value.

3. Place TIP at the desired starting location forthe curve and press and keep pressed RFP.

4. Move TIP along the intended path of the curve

5. When TIP has reached the end locationrelease RFP.

6. To digitize more than one closed curve repeat3. to 5.


Digitizing closed curves: The digitizer must beproperly connected and calibrated before data canbe gathered.

1. At the command line enter ‘digsketch’ orselect ‘Menu: Tool --> 3D Digitizer --> SketchCurve’.

2. In the command line deselect ‘Points’ bypressing ‘p’ followed by ENTER and similarlyselect ‘Curves’ and select ‘Closed Curves’.Choose the distance at which points are to beplaced on the curves by Rhinoceros®. Notethat this distance depends on the scale of thetemplate file you chose! Usually, 0.1 to 0.7proved to be acceptable values, for very largeobjects with hardly any sudden curvaturechanges even a value of 1 is acceptable(using the centimeter template; change valueaccordingly if using other templates). Enterthe value by changing the ‘Point Spacing’value.

3. Place TIP at the desired starting location forthe curve (a point on the seam line) and pressand keep pressed RFP.

4. Move TIP along the intended path of the curve

5. When TIP has almost reached the startinglocation again (1 to 2 mm gap) release RFP.

6. To digitize more than one closed curve repeat3. to 5.


Note: A small gap should be left between the startand end of a curve to guarantee a smooth connec-tion at the seam. Failure to do so can lead to over-lap, which produces artifacts on the lofted surface.Also, the start and end points should not show shiftalong the seam line, as this also leads to artifacts.

Editing kinks and artifacts out of curves: Thisoption should be used sparingly and can producemassive artifacts!

1. Rotate a viewport so that the viewplane is asparallel to the main orientation of the curve aspossible. This means that the curve ideally willlie flat on the viewplane.

2. At the command prompt enter ‘CPlanetoView’to make the construction plane parallel withyour view.

34


3. Select the curve to edit

4. Turn edit points on by clicking the icon at thelower left in the toolbar

5. Carefully select points and drag them toadjust the curve. Watch the changes in sev-eral viewports with different directions to avoidintroducing deformations in the third axis.

Joining open curves into closed curves: Makesure the curves ends are close to each other. To‘match’ several curves sets, always select the firstand second curve in the same order to achieveequally directed closed curves with seam pointsaligned in one row.

1. At the command line enter ‘match’

2. Select the first curve near the end that is sup-posed to be connected to the other curve. Ifonly one of the curves is to be adjusted byRhinoceros®, this curve must be the one thatstays unchanged.

3. Select the second curve near the end that issupposed to be matched to the first curve. Ifonly one of the curves is to be adjusted byRhinoceros®, this second curve must be theone that gets changed.

4. In the match dialog box, select the appropriateoptions. We recommend selecting ‘curvature’,‘average curve’ and deselect ‘join’. ‘averagecurve’ adjusts both curves instead of onlycurve 2, leading to the distortion being morespread out but lesser in degree, reducing arti-facts.

5. Repeat this process until all contact pointshave been matched.

6. At the command line, enter ‘join’, then selectthe curves near the contact point.

Note: In Rhinoceros 2.0®, it is possible to use the‘join’ option in the ‘match’ options dialog. This doesnot work for Rhinoceros 3.0® as the curve can onlybe joined in one place with this option, while thesecond contact point will remain open.

Rebuilding curves: Especially curves closedusing the ‘match’ command have a very unevendistribution of their edit points. This tends to pro-duce artifacts on lofts using these curves. In orderto equally distribute the points on a curve the‘rebuild’ command can be used.

1. At the command line enter ‘rebuild’

2. Select the curve you want to rebuild. If severalcurves are to be rebuilt to the same standards

(same number of points), these curves can beselected at the same time.

3. Select the appropriate options. Usually, thesame number of points that the curve alreadypossesses should be chosen. However,reductions of 50% may be tolerable.

4. To rebuild more than one curve, repeat 1. to 3.

Curves of similar length can be selected andrebuilt in groups.

Lofting surfaces (open and closed)

A flat, ‘open’ surface resembles a sheet ofpaper, a closed surface open at the ends forms atube and a closed surface with closed ends resultsin a 3D body. It may be necessary to lower toler-ances in the file properties to shorten computingtime for lofting. Usually, this will not significantlyreduce the accuracy of the lofted surface.

Selecting curves automatically and manuallyfor a loft (for all lofting options): Rhinoceros3.0® may sorts curves into the proper order whena group of curves is selected at the same time(automatic selection). This can be done by drag-ging a box over the curves with the mouse. Drag-ging from left to right will select only those curvesthat are completely within the dragged box, drag-ging from right to left will select all curves that areentirely or partially in the box. Rhinoceros 3.0® willalso attempt to sort curve directions and adjust theseam line. Remember to de-select all other ele-ments such as coordinates.

If the loft contains both curves and points(start and end of closed lofts), the automatic sortingwill usually fail, especially in Rhinoceros 3.0® SR4and Rhinoceros 4.0®. If automatic sorting does notwork (i.e., the resulting surface curves back ontoitself or shows other defects), curves must beselected manually. Here, the potential for mistakesis quite large, and care must be given to the follow-ing points.

- All curves must be clicked with the mouse atthe same end. Otherwise, version of Rhinoc-eros® earlier than 3.0 may attempt to form asurface that connects two opposite ends.

- All curves must have the same direction(must have been digitized or later edited tohave the same direction). Rhinoceros 2.0®does not automatically sort curve directions,while later versions usually sorts them cor-rectly.

35


- Curves must be selected one by one fromone end of the bone to the other, otherwisethe surface will fold back onto itself.

- For closed lofts, the end points must beselected first and last, with the curves inbetween in proper order.

Lofting a flat surface: An open, flat loft requiresas a minimum two curves.

1. At the command line enter ‘Loft’

2. Select the appropriate curves

3. Press ENTER

4. Select the lofting options desired in theoptions box (see below)

5. Press ENTER

6. Click ‘Shade all viewports’ and check the sur-face for accuracy.

Lofting a closed surface open at one or bothends: A closed surface requires closed curves. Ifan end is supposed to be closed, a point isrequired at that end. This method lets Rhinoceros®automatically adjust the seam points and directionsof the curves; if the curves were digitized asadvised above this should be free of problems.

1. At the command line enter ‘Loft’

2. Select the appropriate curves and the end-point (if appropriate)

3. Press ENTER

4. Check that the selected curves all have thesame direction (arrow at seam line); switchdirections of odd curves


6. Press ENTER


Lofting a body (a closed surface closed at bothends): A closed surface requires closed curves. Apoint is required at each end. This method lets Rhi-noceros® automatically adjust the seam points anddirections of the curves; if the curves were digitizedas advised above this should be free of problems.

1. 1. At the command line enter ‘Loft’

2. Select the appropriate curves and both end-points, starting with one endpoints, thenselecting the curves in correct order, andselecting the second endpoint last.

3. Press ENTER

4. Check that the selected curves all have thesame direction (arrow at seam line); switchdirections of odd curves


6. Press ENTER


8. Select ‘Analyze --> Mass Properties --> Vol-ume’ and compare the given volume with arough estimate to ensure that the loft actuallyresulted in the correct 3D body. Sometimes,the body will be open at some point and thesurface folded back unto itself – the 3D bodyis hollowed out; this will result in a minusculevolume!

Using the various ‘Loft’ options: A variety ofoptions is available in the loft dialog box:

For ‘Style’, there are:

Normal - Uses chord-length parameterization in theloft direction. This is usually the best optionto choose.

Loose - results in a smother surface, but allows thesurface to move away from the curves. Thisoption should be used as rarely as possible,as it will significantly change the externalappearance of smaller bones.

Tight - Uses square root of chord-length parame-terization in the loft direction which results inthe surface sticking closely to the curves.This option tends to result in sharp changesof direction near the articular ends of bones.

For ‘Simplify’, there are:

Do not simplify - preferred option.

Rebuild with control points - Rebuilds the shapecurves before lofting. This is especially help-ful if the curves change direction in smallintervals a lot which can produce artifacts.

Refit within units - Refits the shape curves to a tol-erance value before lofting. Use same as theprevious option.

Troubleshooting: If the surface does not look cor-rect, check the following:

- Have all curves necessary been selected?Have all necessary end points been selected?If not, delete the surface and loft again.

- Do all curves have the same direction? Ifnot, delete the surface and loft again andchange the directions.

36


- Does any one curve show excessive differ-ences from its neighbors (erroneous curve)? Ifso, delete the surface and loft again withoutthe troublesome curve. Alternatively, rebuildcurves with the ‘rebuild’ command, each witha number of points intermediate between theoriginal number and that of the other curve.

- Do all curves look OK but the surface hasodd wrinkles and twists? If so, delete the sur-face and try various ‘Simplify’ options of theloft dialog in a new loft. Or try rebuildingcurves near the artifacts of the surface, andthen loft again. Also, try adjusting the ‘seamline’ visible during lofting setup by draggingthe points on all curves so that the seam linehas no kinks.

Editing and joining surfaces and bodies

Some helpful Rhinoceros® commands aregiven a short description here - for a tutorial onediting and combining surfaces please refer to thein-program help.

Joining bodies smoothly in Rhinoceros® canbe difficult, so we advise against using multiplebodies whenever possible. Check that the toler-ance levels are set to suitable values. Rememberthat you can copy and paste bodies, so that onebody can be deleted when used as a cutting objectand still become part of the final assembly.

Joining two bodies:

1. At the command line enter ‘BooleanUnion’ orselect the icon from the toolbar at the left ofthe screen.

2. Select both bodies

3. Press ENTER

Subtracting one body from another:

1. At the command line enter ‘BooleanDiffer-ence’ or select the icon from the toolbar at theleft of the screen (right-click BooleanUnionand select the second icon in the pop-up box)

2. Select the body from which to subtract theother body and press ENTER

3. Select the body to subtract and press ENTER.

Intersecting two bodies: results in only the over-lap being preserved.

1. At the command line enter ‘BooleanIntersec-tion’ or select the icon from the toolbar at theleft of the screen (right-click BooleanUnionand select the third icon in the pop-up box)

2. Select the bodies to intersect and pressENTER

Creating a surface from a curve:

1. Select the curve

2. At the command line enter ‘patch’ and pressenter.

3. As trimmed surfaces cannot by joined andtend to be difficult to use for cutting both sur-faces and bodies we recommend to deselect‘trim automatically’ in the dialog box.

4. Click OK.

Using a surface as a cutting object to trim sur-faces or bodies:

1. At the command line enter ‘Split’ (keeps thecut of pieces as separate objects) or ‘Trim’(deletes the cut off pieces).

2. Select the object to split/trim. If trimming, clickthe part that should remain.

3. Select the cutting/trimming object.

4. Press ENTER.

Creating a curve on a surface (for 3D model cre-ation, can also be used as a cutting object)

1. Select the surface.

2. Select ‘Curve-->Free-Form-->Interpolate onSurface’.

3. Click on the surface at the start of the curve.

4. Click close to this first point in order to gettight control of the direction of the curve start.

5. Add more points to describe the surface.

6. Place the last clicks near each other to controlthe curvature exactly at the end of the curve.

7. Press ENTER.

37


Appendix D - Guide to point cloud and polygon mesh editing in Geomagic Qualify®

Note: Many actions are not reversible in GeomagicQualify ®, or only until the next action has beencompleted. Take care to save your data often andwith altered filenames! Note: instructions here referto version 8.0 of Geomagic®. Other version mayrequire slightly different command inputs.

Transferring data from Geomagic® to Rhinoc-eros® and back. Both Geomagic® and Rhinoc-

eros® use program-specific file types. Neitherprogram can reliably import the other program’sspecific files. Therefore, data must be saved in for-mats that can be read by both programs, for whichstandardized file types are advisable. Point cloudscan be saved as IGES files (*.igs), while polygonmeshes are best saved as binary Stereolithogra-phy files (*.stl).

Creating polygon meshes from point clouds

1. Load the point cloud file. Acceptable formatsare e.g. *.obj and *.igs

2. Choose a selection tool (‘Edit-->SelectionTools-->) and select erroneous points.

3. Delete selected points by pressing [Del].Unmark the selection by pressing [Ctrl]+[C].

4. Select the wrapping tool (‘Points-->Wrap’) andchoose the ‘Surface’ wrap type.

5. Check the ‘advanced options’ box and set themeshing quality to maximum.

6. Click ‘OK’

Removing artifacts from and smoothing a poly-gon mesh. Smoothing via the ‘smooth’ commandcleans up the surface in one work step. Lessaggressive are ‘Remove Noise’ and ‘RemoveSpikes’. They allow the user to control the degreeof smoothing, and ‘Remove Noise’ also hasoptions to show the deviation between the originaland the smoothed version.

Smoothing a polygon mesh:

1. Select ‘Polygons-->Smooth’.

2. Click ‘OK’.

Removing spikes: This tool works well for littlepyramidal shapes on the surface caused by pointsthat lie slightly outwards or inwards of the plane ofthe neighboring points.

1. Select ‘Polygons-->Remove Spikes’.

2. Set the slider to the desired degree of smooth-ness.

3. Click ‘Apply’.

4. Check the result. Repeat 2. and 3. until satis-fied, then click ‘OK’.

Removing noise:

1. Select ‘Polygons-->Remove Noise’.

2. Set parameters as desired.

3. In order to see deviations, mark ‘Show Devia-tions’. Set colors and values as required.

4. Click ‘Apply’. If marked, the deviations arenow shown as color changes according to theoptions chosen.

5. Repeat 2. to 4. until satisfied, then click ‘OK’.

Reducing the polygon number in a meshThe shape of the bone is preserved better if

the reduction is done in small steps. Thus, reduc-ing to 80% several times is better than going downto 20% in one instance.

1. Select ‘Polygons-->Decimate’.

2. Set either the target number of polygons orthe target percentage desired.

3. Mark ‘Fix Boundaries’ to preserve the extremepoints and edges of the object. If this option isnot selected, edges will be smoothed over.

4. Click ‘Apply’.

5. Check the result. When satisfied, click ‘OK’.

Removing artifacts on a polygon meshRemoving artifacts and other erroneous trian-

gles will leave holes in the object. They can befilled using the ‘Fill Holes’ tool.

1. Choose a selection tool (‘Edit-->SelectionTools-->)

2. Choose the proper selection mode (‘Edit-->….). Selecting ‘Backface mode’ allowsselecting both triangle fronts and backs, dese-lecting it allows selecting polygons from theoutside only. ‘Select visible only’ selects visi-ble triangles only, while ‘Select through’selects any triangle that is in the direct line ofprojection, whether it is visible or hidden byother polygons.

CAUTION - It is easy to accidentally delete trian-gles on the reverse side of the model when using‘Select Through’

3. Select erroneous triangles. Holding down[CTRL] while clicking removes triangles fromthe selection.

4. Press [DEL].

38


Erroneous triangles can also be removedwhile using the ‘Fill Holes’ tool (see there). Espe-cially useful is the option removing floating data.

Filling holes: Note that it is usually advisable tofirst clean up a polygon mesh and then fill theholes. This is faster, as many small holes will befilled automatically by the ‘Smooth’, ‘RemoveSpikes’ and ‘Remove Noise’ commands.

1. Start the ‘Fill Holes’ mode (‘Polygons-->FillHoles’)

2. Select the appropriate mode:

‘Fill’ fills holes completely. Select the appropriateoptions and click on the

boundary of a hole to fill it.

‘Fill partial’ allows defining a stretch of boundaryalong which the hole will

be filled. The rest of the hole stays open. Clickon the boundary in two places, then click thatpart of the boundary were the fill should takeplace.

‘Create bridges’ creates strips of polygons betweentwo selected edges. Click

on one edge for the bridge start, then onanother for the bridge end.

‘Delete’ allows removing triangles based on vari-ous selection options. Checking the ‘floatingdata’ box allows removing triangles that haveno connection to the external surface.

‘Move’ allows changing the shape of a bridge cre-ated in ‘Create Bridge’ mode.

Click the place where you want to apply thechange, click and drag the arrow into the cor-rect direction and set the shape and extensionof the shape. Click ‘Next’ to confirm a moveand start a new one.

3. Decide whether you want to use curvaturebased or flat filling. Check or uncheck the cor-responding box accordingly. Flat filling createsa surface as planar as possible, while curva-ture based filling attempts to keep the tan-gency of the hole closing surface as constantas possible.

4. If satisfied with the fillings, click ‘OK’. Other-wise, discard them by clicking ‘Cancel’.

Sectioning polygon meshes. Cuts can be exe-cuted in two ways: with a plane or along a curve.The former option is not very useful if the resultingsections are used in kinetic modeling, as largegaps will appear when sections with straight cut-

offs rotate versus each other. A curve allows cre-ation of saddle-shaped cuts, so that rotary jointsare almost gap-free.

Using curves, there are two options: to createclosed objects (bodies) or open objects (surfaces).Creating closed bodies usually results in smallgaps between neighboring objects and is morework intensive. Creating open objects is faster, butcarries a high risk of artifacts at the edges if theopenings are closed using the ‘Fill Holes’ function.The higher the resolution of the file is, the weakerthe artifacts will be.

Using curves as cutting object to create closedsection object:

1. The import the STL file.

2. Rotate it so that the view axis is parallel to theintended cut.

3. Select 'Polygon-->Section by Curve'.

4. Draw a curve from control points by clickingon the model. It is important to imagine howthe two sections-to-be will be positioned afterhaving rotated versus each other in kineticmodeling and place the curve accordingly. Acircular arc would be ideal, but it is not easy todraw one given the tools offered byGeomagic®. [Ctrl-Z] removes the last point.This curve can NOT be saved!

5. Click 'Create' to see a preview of the curve onthe model

6. Click 'Intersect Extrusion. This extrudes thecurve to the depth given in the 'Profile Curve --> Depth' field. Usually, Geomagic® suggestsa sufficient depth, lower depths will not pene-trate the model fully.

7. Mark 'Delete Selection‘ and 'Fill Intersection'.

8. 'Reverse Selection' allows changing the partto be deleted and the part to be kept.

9. Click 'OK'

Now, one part has been cut away from therest of the model along the extruded curve. It isadvisable to rename and save it (right-click the partin the model manager window and choose theappropriate option). In order to create the contact-ing part, proceed as follows:

1. Into the same file, import the STL with the'File-->Import' command.

2. In the model manager, select both the sectioncreated before and the complete part (hold'Ctrl' to select both while clicking).

39


3. Rotate the viewport so that the cutting surfaceis parallel to the view axis.

4. Deselect the finished section by holding down'Ctrl' and clicking the part again. It is nowshown in ghosted view mode.


6. Draw a curve along the cut line. Try to followthe boundary of the already created part asexactly as possible.

7. Click 'Intersect Extrusion. This extrudes thecurve to the depth given in the 'Profile Curve --> Depth' field. Usually, Geomagic® suggestsa sufficient depth, lower depths will not pene-trate the full model.

8. Mark 'Delete Selection' and 'Fill Intersection'.

9. 'Reverse Selection' allows changing the partto be deleted and the part to be kept.

10. Click 'OK'

Now, the next part has been created. Depend-ing on how close the two cutting curves correlate,there will be a small gap or overlap between theparts.

Using curves as cutting object to create opensection object:

1. Import the STL file

2. Rotate it so that the view axis is parallel to theintended cut.


4. Draw a curve from control points by clickingon the model. It is important to imagine howthe two sections-to-be will be positioned afterhaving rotated versus each other in kineticmodeling and place the curve accordingly. Acircular arc would be ideal, but it is not easy todraw one in the fashion offered byGeomagic®. [Ctrl-Z] removes the last point.This curve can NOT be saved!

5. Click 'Create' to see a preview of the curve onthe model

6. Click 'Intersect Extrusion. This extrudes thecurve to the depth given in the 'Profile Curve --> Depth' field. Usually, Geomagic® suggestsa sufficient depth, lower depths will not pene-trate the full model.

7. Unmark 'Delete Selection‘

8. Click 'OK'

Now the object has been split along the curve.Each part can be separately selected, e.g. with thelasso selection tool, and turned into a separateobject (‘Edit-->Selection to Object…’). In order tofill the holes, use the ‘Polygons-->Fill Holes’ func-tion. Use ‘Flat fill’ and ‘Create bridges’ to build flatfills for both sides that correspond to each other.

40


Appendix E - Combined techniques for Rhinoceros® and Geomagic®

Note: instructions here refer to versions 3.0 of Rhi-noceros® and 8.0 of Geomagic®. Other versionmay require slightly different command inputs.

Collecting pointclouds in Rhinoceros® usingpartial mesh display in Geomagic®. While sam-pling a specimen with pointclouds it is often hard tojudge if there are any unsampled parts left, and ifthe sampling density is sufficient near suddentopology changes. Given sufficient computingpower the following procedure makes this processeasier.

Both Rhinoceros® and Geomagic® must berunning. The digitizer must be properly connectedin Rhinoceros® and calibrated before data can begathered.

1. Begin digitizing the surface normally

2. After a while, save the file in IGES format(*.igs)

3. Load this file in Geomagic® and mesh thepointcloud as described above.

4. Rotate this file so that the screen view coin-cides with the operator’s view of the real spec-imen.

5. Ready Rhinoceros® for continued digitizing(e.g.: if necessary, enter ‘digsketch’).

6. Shift the screen display to Geomagic® bypressing the MS Windows® hotkey[Alt]+[Tab]. Do not click anywhere in theGeomagic® window with the mouse!

7. Continue digitizing, using the partial surfacevisible in Geomagic® as a guide. Control theinput through the foot pedal exclusively.

8. Shift back to Rhinoceros® periodically usingthe MS Windows® hotkey [Alt]+[Tab].

9. Repeat 2. to 8. until digitizing is complete

10. Save the final file in any desired format in Rhi-noceros®.

41

Palaeontologia Electronica · GPIT 1 Plateosaurus engelhardti: dorsal 2, left ilium, left radius...

Documents

Transcript of Palaeontologia Electronica · GPIT 1 Plateosaurus engelhardti: dorsal 2, left ilium, left radius...