lable at ScienceDirect

Journal of Archaeological Science 43 (2014) 175e185

Contents lists avai

Journal of Archaeological Science

journal homepage: http : / /www.elsevier .com/locate/ jas

Multi-image photogrammetry as a practical tool for cultural heritagesurvey and community engagement

John McCarthyWessex Archaeology, United Kingdom

a r t i c l e i n f o

Article history:Received 9 September 2013Received in revised form3 January 2014Accepted 6 January 2014

Keywords:PhotogrammetryStructure from motionSurvey

E-mail addresses: johnkenningtonmccarthywessexarch.co.uk.

0305-4403/$ e see front matter � 2014 Elsevier Ltd.http://dx.doi.org/10.1016/j.jas.2014.01.010

a b s t r a c t

Multi-image photogrammetry is rapidly emerging as an important archaeological tool due in large partto the increasing level of automation in off the shelf software. The technique can offer significant re-ductions in the cost of archaeological survey and in the enhancement of survey results and is of particularvalue therefore to archaeologists working in contract-led context, which in many areas accounts for themajority of archaeological work (up to 80% in Scotland for example). Recent advances in multi-imagephotogrammetric software have resulted in highly automated workflows and significantly reduced theburden of technical knowledge required to produce survey results of an acceptable standard. Althoughthe majority of multi-image photogrammetry surveys are still undertaken in an academic context thetechnique is increasingly being used by a far wider proportion of heritage professionals, many of whomare not first and foremost specialists in photogrammetry. The adoption of such highly automatedworkflows presents certain risks with regard to accuracy and reliability of results as noted by Remondinoet al. (2012, 52). However the enormous potential of the technique for rapid and accurate survey and forreduced costs cannot be ignored and the challenge we face is to ensure that the highly automatedworkflows adopted by archaeologists in contract-led contexts are robust and reliable and underpinnedby guidance and knowledge exchange. This paper is not intended as a comprehensive technical review ofthe technical aspects of the technique or of its development but instead focusses on highlighting itspotential as a practical everyday tool for archaeological practitioners to apply in two of the main types ofcontract-led archaeological work, rapid survey and community engagement. A non-technical overview ofthe technique is given followed by case studies illustrating how the technique has been applied suc-cessfully in a non-academic contract-led and community engagement context. These surveys have beenundertaken with very limited budgets for both survey and post-processing of data and typically withvery limited time frames. In each case study, use of multi-image photogrammetry has allowed for better,faster and more cost-effective results than would otherwise have been possible. Case studies include asurvey of an Iron Age fort, a rapid survey of exposed segments of an intertidal wreck, both commissionedfor heritage management purposes and a community survey of a 17th century gravestone undertaken bychildren under the age of 16. Finally the obstacles to wider adoption in the contract-led sector are dis-cussed and it is argued that a concerted approach is required to create and disseminate simple andreliable workflows.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Multi-image photogrammetry or Structure from Motion is arelatively new technique for accurate digital capture of 3-dimensional objects and surfaces. The technique is practical andversatile and is increasingly being adopted for cultural heritage toreplace or enhance more established survey techniques such as

@gmail.com, j.mccarthy@

All rights reserved.

manual survey and laser scanning. In multi-image photogram-metry, some of the most technically challenging and time-consuming elements of traditional stereo and convergent photo-grammetry have now been automated and it is now possible tocombine large groups of images rather than pairs, making this a farmore cost-effective, user-friendly and powerful approach. As aresult there is an increasing number of published examples ofarchaeological surveys based wholly or partly on recording of 3-dimensional features using multi-image photogrammetry. Thevast majority of these have been undertaken in academic orresearch contexts despite the fact that in many areas the bulk of

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185176

archaeological survey and excavation, is undertaken by archaeol-ogists in commercial or non-profit heritage organisations fundedthrough contracted heritage projects, much of which can be char-acterised as pre-development mitigation survey and excavation.The slower adoption of the technique in this sector is due to theamount of time and technical expertise required to achieve accu-rate results, archaeologists working in commercial contexts havingrelatively little time to devote to research and development.Working under contractual obligations to produce results to apredefined standard can also discourage innovative or experi-mental methodologies which carry a higher risk of failure. As withthe early days of laser scanning there is a need for collaboration andskill sharing across the entire heritage sector in order to realise thefull potential of multi-image photogrammetry. Of particular inter-est, given the simplicity of the photogrammetric data capture, is thepotential for use of multi-image photogrammetry as a tool forcommunity outreach and engagement projects, an increasinglyimportant part of the contract-led heritage sector.

As recently as 2003 an assessment of the value of multi-imagephotogrammetry for recording of carved stones in Scotland foundthe technique to be unsuitable for detailed recording of archaeo-logical features, due mainly to the low level of detail in the modelsthat could be generated within a reasonable amount of time(Jeffrey, 2003, 112). However due to subsequent improvements insoftware, the technique has undergone a rapid evolution, making itmuch more useful and accessible than ever before. As a result ofthese developments it can now be said with confidence that undercertain conditions, multi-image photogrammetry offers archaeol-ogists a viable alternative in terms of technical complexity, accu-racy, cost and flexibility to established techniques such as manualsurvey and laser scanning. The wider adoption of the technique isbeing driven by an on-going convergence of related technologies,including advances in digital photography, in the development ofreliable Small Unmanned Aircraft (SUA) such as quadcopters andhexacopters, in more efficient and powerful software and byincreasing computing power both on desktops and in the cloud. Anumber of surveys undertaken by the author in the UK in 2012using this technique are presented in order to illustrate the value ofmulti-image photogrammetry as a practical and cost-effectivemethod for accurate survey and as a tool for community engage-ment with heritage.

2. Defining photogrammetry

As stated above the contract-led heritage sector has been con-servative in its use of multi-image photogrammetry and given thatone of the aims of this paper is to encourage the adoption of thetechnique by archaeologists with limited or no experience of thetechnique, it is appropriate to give a brief overview of the techniqueand its application in heritage. For a more comprehensive andtechnical discussion see Szeliski (2011).

In simple terms, photogrammetry is the process of makingmeasurements of features through analysis of overlapping photo-graphs, and is fundamentally based on trigonometry. As early as the1850s a French Army surveyor, Aimé Laussedat, realised that wherethe optical characteristics of a camera are known, multiple imagestaken with that camera from slightly different angles could becompared to obtain accurate measurements of the relative di-mensions of that subject (Laussedat, 1854, 1859). Laussedat suc-cessfully applied the technique to ground-level topographicalsurvey. However, the equipment available meant that great tech-nical skill and extensive manual calculation were required. Adop-tion of photogrammetric techniques was gradual and limited andtheodolites have remained the main tool for ground-based surveyto the present day. Since Laussedat’s time the most successful

application of the technique has been for large scale aerial survey oftopography by cartographers such as the Ordnance Survey of GreatBritain. This approach relies on overlapping vertical aerial photo-graphs analysed using various techniques such as stereo plotters.Because of the scale of the area covered this was one of the fewapplications of photogrammetry where the amount of timerequired for the technique was outweighed by the benefits derivedfrom it. In the last decades of the 20th century a significant effortwasmade to automate as much as possible of theworkflow in orderto reduce the burden of analysis. However it was not until theadvent of digital photogrammetry in the early 1990’s, where filmwas replaced by scanned digital surrogates, optical trains replacedby computers and left/right eye-pieces with 3D monitors, thatautomation in photogrammetry really took off allowing 3D data tobe automatically generated using pixel correlation algorithms.Despite this the technique remained a highly specialised andexpensive technique, both in terms of time and hardware.

3. Archaeological applications of stereo photogrammetry

One of the main applications of photogrammetry for a specif-ically archaeological purpose has historically been analysis of stereoaerial photography for accurate aerial mapping of archaeologicalfeatures such as crop marks. Despite the need for extensive manualinput, a number of attempts have also been made by archaeologiststo apply stereo photogrammetry to terrestrial survey. One of themost notable recent heritage photogrammetric surveys was a pio-neering survey at Stonehenge, resulting in the creation of 350digital models of the megaliths (Bryan and Clowes, 1997). However,as Jeffrey (2003) points out, due extensive manual editing requiredthe surveyors did not attempt to model fine surface detail such asfaint carvings on the stone but to aim for a point density of twocentimetres in order to capture an accurate geometric model of thestones. Even so the processing ran for over three years. Anothermore recent example of photogrammetry based on stereo pairs wasthe highly successful Northumberland and Durham Rock Art Proj-ect (NADRAP), an English Heritage-funded project led by North-umberland and Durham County Councils (Bryan and Chandler,2008). A total of around 1500 rock art sites were recorded usingstereo photogrammetry with a high degree of success, producingdense and accurate 3D models of the features. The project reliedheavily on volunteers, both for data capture and processing. How-ever, there were a number of limiting factors encountered.Although the workflow adopted was relatively simple, it relied onproprietary software and was limited to stereo-pairs captured withpre-calibrated cameras. The cameras were calibrated for a specifieddistance from the subject, thus limiting the size of area that couldbe captured (Bryan and Chandler, 2007, 213). Capture of larger rockart panels required multiple stereo-pair models to be combined.However this proved to be too time-consuming and the majority ofsites were recorded using stereo photogrammetry only. Despite thesuccess of these projects, it is clear that stereo photogrammetry hasbeen restricted to a very small and specialised proportion of thearchaeological community.

4. Defining multi-image photogrammetry

The term ‘Multi-image Photogrammetry’ (sometimes usedinterchangeably with ‘Structure fromMotion’) is used in this paperto describe to a more recently developed approach to photogram-metry, where stereo pairs are no longer the focus. Instead muchlarger datasets of overlapping digital images of a feature taken fromdifferent positions can be loaded in a single batch into softwarecapable of automatic camera calibration, feature matching andreconstruction of complex dense 3-Dimensional models, with

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185 177

minimal manual input. With the right combination of hardwareand software, multi-image photogrammetry can produce highlydetailed and accurate models of both topography and discrete ob-jects or monuments. As it is not limited to stereo pairs it is alsocapable of modelling more complex objects and it is not necessaryto maintain a known distance from the subject.

There has been an explosion in the range of multi-imagephotogrammetry software packages available in the last decadewith an attendant improvement in automation and interface. Cur-rent software packages range from free open-source programs toprofessional-grade packages costing thousands of pounds. Many ofthese have been released or significantly updated within the lastthree years. In this paper three software packages have been dis-cussed. These are VisualSfM (by Changchang Wu and others), 123DCatch (by Autodesk) and PhotoScan (by Agisoft). These surveyswere undertaken separately and do not constitute a direct com-parison of the various workflows although they do demonstrate arange of open source, commercial and cloud-based techniques,varying greatly in the level of technical knowledge and inputrequired. For an excellent comparative analysis of some of the ca-pabilities of some of the main software packages using a singledataset see Remondino et al. (2012). Archaeologists have beenquick to adopt the technology for a wide variety of purposes and itis sufficient to highlight some recent examples. It has been appliedto archaeological excavation in Belgium by the University of Ghent(De Reu et al., 2013; De Reu et al., 2014) and in Israel by the TellAkko Project (Pennsylvania State University, the University of Haifaand others) (Olson et al., 2013). Recent examples of aerial multi-image photogrammetry include projects by the University ofGhent in Belgium (Verhoeven, 2011) and Italy (Verhoeven et al.,2012). Rock art recording has been undertaken by the Universityof Ghent in Siberia (Plets et al., 2012a,b) and the universities ofLoughborough and Newcastle in Australia (Chandler et al., 2005).Artefact recording has been undertaken by the Institute for Lan-guage and Speech Processing and the University of Ljubljana(Koutsoudis et al., 2013) and the University of Haifa (Gilboa et al.,2013). Examples of applications for recording of buildings includestudies undertaken by Parma University at Italian sites (Roncellaet al., 2011). These few examples should serve to illustrate thatmuch of this work has been undertaken in an academic context.

5. How multi-image photogrammetry works

As discussed above there are numerous publications which havedescribed in technical detail single software packages and work-flows and a smaller number which attempt direct comparisons butgiven the relative novelty of the technology a generic and non-technical description is given below. Regardless of which soft-ware package has been chosen, the transformation of 2D digitalimages into 3D digital models is generally carried out in a similarway.

The first step is to acquire a set of images of the subject of in-terest. These images must have a large amount of overlap with eachpart of the surface to be modelled visible in at least three images.Almost any camera can be used although better results are morelikelywith a high end camera such as an SLR. For optimal results thesettings of the camera (particularly zoom) should not be changedduring the data capture. Choosing a subject which is not moving isalso crucial as moving elements will impair the ability of the soft-ware to successfully match features between images. In multi-image photogrammetry it is not necessary to record each cameralocation as this can be calculated from the images themselves andthis is undoubtedly one of the greatest advantages of the technique.Control points may be used if desired to tie the model into a knowncoordinate system and to provide a scale for the final model.

The second step is to load the images into the software foranalysis and automatic detection of matching correlated features atvarious scales between images. Many of the multi-image photo-grammetry packages currently available use an algorithm known asScale-Invariant Feature Transform or SIFT (Lowe, 1999) which canmatch features despite changes in the scale or orientation of theimages. Because multi-image photogrammetry software can anal-yse large numbers of images, this stage can require a great deal ofcomputer power and can be very time consuming. Each imagemustbe compared to every other image and the number of comparisonsnecessary increases exponentially with each additional image. Todeal with this issue some software packages are designed to uploadimages for processing in the cloud. In software such as Autodesk’s123D Catch or Arc 3D which rely on remote computers the need fora local high-spec computer is bypassed but broadband internetaccess is required. Where image processing is undertaken locally itmay be necessary to use a powerful computer, perhaps even onespecifically designed to cope with photogrammetric datasets.

In the third step these matches between points are used tocalculate the spatial relationship between them and in doing so, therelative positions of the cameras are indirectly derived from surveycontrol applied within the orientation process. Each multi-imagephotogrammetry software package does this in a slightly differentway but ultimately they all rely on trigonometry and iterativetesting and refining of a model of both the surface of the subjectand of the optical parameters of the camera. The end result of thesecond stage is a ‘sparse’ 3-Dimensional model which includes therelative camera locations and a small number of points on thesubject or feature.

In the fourth step, using this ‘sparse’ 3D model, the softwarereanalyses the images and generates a much denser model of thesubject. This takes the form of a point cloud similar to thosegenerated by laser scanners. Unlike basic Terrestrial Laser Scanning(TLS) the attributes of the points in this model will always includecolour values as they are derived from colour images. Themaximum number of vertices in the output model is limited onlyby the number of pixels in the source images, the total number ofthose images and the capacity of the processing hardware. As thereis no hard limit to the number of input images, there is theoreticallyno maximum size of point cloud and, given the high density ofpixels in modern cameras, photogrammetric point clouds can bevery dense. In practice the limiting factor is often computing power.Even where the system used has the capacity to process a modelfrom hundreds of images, such large datasets may not be desirableas it might take days or even weeks to process them, particularlywhere standard consumer-grade computers are used. Problems canalso arise where the resulting models are too large to work with, aproblem familiar to archaeologists dealing with laser scans. Themaximum number of points generated in a single model duringtesting by the author was just under a quarter of a billion points ona scan of a small carving less than 20 cm in diameter (Fig. 1). In thiscase it was found necessary to sub-sample the point cloud to pro-vide a more workable dataset.

6. Case studies

One of the most attractive aspects of multi-image photogram-metry to contract-led archaeologists should be its applicability toalmost every aspect of archaeological survey, from artefacts tolandscapes. To illustrate this, three recent surveys led by the authorare presented below. The first of these is a detailed survey of apromontory fort, commissioned by Forestry Commission Scotlandand carried out by a professional team of surveyors from WA Her-itage. This survey demonstrates how the technique can offer a cost-effective approach in a commercial context and a variety of

Fig. 1. Many multi-image photogrammetry software packages also have a fourth stage which involves interpolating the point cloud into a continuous surface or mesh, usually witha texture derived from the source images. This is not a crucial part of the process and although it often looks better, it can be unsuitable for archaeological purposes as meshing canlead to a decrease in accuracy. Additional operations such as georeferencing of models using external control, the capacity to output orthographic views or fly-throughs of the modelare also available as part of some multi-image photogrammetric software packages.

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185178

additional benefits. In this case photogrammetry was used togenerate a series of detailed sections and plans and as an aid tointerpretation of the phasing of the site. In the second example amuch simpler workflowwas used to model a pair of gravestones ina Scottish graveyard. In this case, image capture was carried outentirely by children between the ages of 10 and 16, all members ofthe Young Archaeologists’ Club (run by the Council for BritishArchaeology). This survey is used to illustrate the value of thetechnique for dissemination of archaeological features and forcommunity engagement, an increasingly important area forcontract-led archaeologists. In the final case study a survey of anintertidal wreck is used to illustrate how rapid recording usingmulti-image photogrammetry can be of use where time is a criticalfactor.

6.1. Survey at Rubha an Fhaing Dhuibh

In April 2012 Wessex Archaeology, a UK-wide heritagecontractor, was commissioned by Forestry Commission Scotland toundertake a detailed archaeological measured survey of Rubha anFhaing Dhuibh, a promontory fort of likely Iron Age date located onthe southern bank of Loch Shiel, near Glenfinnan (NM 8130 7185).The aim of the survey was to inform conservation managementplans for the monument and to provide an enhanced record of itscurrent condition. The survey was undertaken over a period ofthree days (McCarthy, 2012). As a commissioned survey relying onprofessional archaeologists, cost was a major consideration whenchoosing a survey methodology. It was felt that the complexirregular nature of the stone banks across the site would take toolong to survey manually using traditional techniques such as metrewide planning frames and the cost of laser scanning would be too

high due to hardware overheads. In consultation with the FCSArchaeologist Matt Ritchie, it was decided to use the opportunity totest a multi-image photogrammetry workflow, testing the resultsagainst an outline Total station EDM survey.

The promontory itself lies at the edge of Loch Shiel, at the base ofa large glacial valley andmeasures approximately 30m in diameter.It was discovered only thirty years ago (Kirby, 1983) under densevegetation. It faces on to the loch on its east, north and west sidesand is connected to the shore on the south side at an area of marshyground. The site is comprised of a jumble of boulders and rocks,many of which are arranged into recognisable walls including anenclosure divided into two cells, lying at the centre of the prom-ontory. All around the loch edge the site is defined by a collapsedrubble bank running below the water line. Five short linear stonefeatures of unknown purpose were observed radiating out from therubble bank, in each case fully or partially located just below thewater line. These stone lines or walls were all of a similar size andappear to be evenly spaced suggesting that they might becontemporary. They are of similar dry stone rubble constructionbuilt of stones around 30 cm in diameter and rest on the gravel lochfloor. A more detailed description and interpretation of the site isgiven in an unpublished survey report (McCarthy, 2012) in the ar-chives of the Royal Commission of Ancient and Historical Monu-ments of Scotland.

Much of the vegetationwhich had covered the site immediatelyprior to the survey was cleared in order to facilitate recording andinterpretation. In order to establish a basic record of the site a TotalStation survey was undertaken of the main upstanding features ofthe site and the outlines of rubble walls. Digital images were thengathered using a DSLR camera mounted on a 3 m long fibreglasspole. A rough grid was laid out across the site and vertical

Fig. 2. An orthographic plan of the promontory fort at Rubha derived from a photogrammetric survey.

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185 179

exposures were taken at approximate half metre intervals. Eachimage covered an area of ground approximately 4 m in length by3m transversely and this allowed for a considerably greater overlapthan the minimum required for photogrammetric modelling. Thismeant that it was not necessary to ensure the spacing betweenexposures was exact or even that the camera was exactly vertical.Oblique photos were also taken at various points around the site toprovide additional coverage, particularly for vertical or over-hanging surfaces. This approach resulted in over 2000 digitalphotographs which were successfully processed over the course ofa week to generate a point cloud of over 40 million points using themulti-image photogrammetry software VisualSFM (Wu et al.,2011). Of particular importance, given the commercial nature ofthe project was the fact that this processing stage was almost

entirely automatedwithminimal human intervention. VisualSFM isan open-source graphical user interface which bundles togetherseveral free programs including PMVS/CMVS (Furukawa and Ponce,2010). 2-dimensional orthographic renders of the point cloud werethen georeferenced in ArcGIS and used to create an accurate seriesof plans and elevations of the site including outlines of individualrocks which would have been impossible to capture in the samelength of time using traditional methods (Figs. 2and 3).

6.2. Survey at Aberlady

As a leader of the East Lothian branch of the Young Archaeolo-gists’ Club of East Lothian, the author undertook a community-based photogrammetric survey. A Young Archaeologists’ Club

Fig. 3. The accuracy of the point cloud was assessed through comparison with a series of 48 control points surveyed using a GNSS/Total Station System (Fig. 3). Despite the brackencover a total of 37 of the 48 control points could be easily identified from the point cloud data. A horizontal accuracy analysis undertaken using a rasterised image of the point cloudexported from Cyclone was conducted in ArcGIS. Comparison of the two datasets gave a Root Mean Squared error of 3.634 cm when compared to the surveyed control points.

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185180

meeting was arranged at Aberlady church yard in East Lothian,Scotland on the 14th November 2012, with participants betweenthe ages of 8 and 16. A simple multi-image photogrammetryworkflow using 123D Catch by Autodesk was chosen. This softwareis user-friendly but offers minimal user control over processing. Forexample although 123D Catch generates a point cloud in order tocreate the final textured 3D mesh, it cannot be accessed or edited,making it impossible to remove erroneous points that may thenreduce the fidelity of the final mesh. The software is therefore oflimited value for survey in a professional context but its simplicitygives it several advantages for use in outreach and communityengagement. Photogrammetric processing is undertaken on Auto-desk’s own servers rather than on a local computer and the soft-ware automatically generates visually appealing textured meshesand offers a simple interface for generating fly-through videos ofmodels.

The subjects of the survey, chosen within the graveyard by thechildren themselves, were two gravestones, one of 19th centurydate and one of 17th century date, located at opposite ends of thegraveyard. Only one hour was available for the survey includingtraining. The group of five children (all under the age of 16) weregiven a ten minute talk on how to use digital cameras and how totake images suitable for photogrammetry (in this case by movingin a circle around the gravestones and relying on automatic set-tings for exposure). They were then split into two groups, onegroup using traditional hand-recording methods and one grouprecording using photogrammetry, in each case having around

20 min to undertake the recording, before switching techniques.The resulting photographs were loaded into 123D Catch by theauthor and processed overnight. They were presented to thechildren at the next meeting of the club in the form of a fly-through video and orthographic projections output using Mesh-Lab, a 3D editing tool developed with the support of the 3D-CoForm project (http://www.3d-coform.eu/) and rendered withradiance scaling (Vergne et al., 2010) (Fig .4). This shader helps toenhance minor variations in depth across a model (in some casesincluding detail that is not visible to the naked eye) (Granier et al.,2012). The fly-through video was also uploaded to the internet(http://www.youtube.com/watch?v¼ZPxwC7mIToU). The resultsproved to be of excellent quality and it was noted that the surveysconducted through photogrammetry included far more detailthan it had been possible for the children to record on the com-plex gravestones by hand, including capture of the feature’scolour, depth and volume. This survey is particularly relevant forcontract-led archaeologists as it highlights how a simple multi-image photogrammetric workflow can facilitate 3-dimensionalarchaeological recording by members of the public with a mini-mum of training required to produce outputs that are not onlyaccurate but which communicate the essence of the archaeolog-ical site in a highly accessible format. Given the increasingnumbers of contract-led archaeologists working on communityheritage projects there is a huge and largely untapped potentialfor multi-image photogrammetric projects incorporating volun-teer survey.

Fig. 4. Orthographic elevations of the gravestone surveyed photogrammetrically by under-16 members of the East Lothian Young Archaeologist’s Club.

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185 181

6.3. Tetney Sea line replacement

The speed of 3D data capture is one of the most attractive fea-tures of photogrammetric survey for contract-led archaeologistsand this is well illustrated by a survey undertaken by the author in2013. A pre-development survey on Tetney Sands, to the south ofGrimsby ahead of construction of a pipeline discovered the recentlyexposed remains of a shipwreck of probably 19th/20th century datelying in the intertidal zone (Fig. 5).

The wreck lay on intertidal sands around a kilometre from thehigh water mark but much closer to the low water mark. A siteassessment was carried out in June 2013 by the author in order toplan thewreck and to try to carry out augering and small-scale test-pitting through homogenous sand deposits to establish its extent

Fig. 5. The ribs of an intertida

(McCarthy, 2013). At the time of survey the wreck remained almostentirely covered with sand and the only visible elements were aline of ribs (futtocks) defining one edge of the hull, protrudingabove the sand to a maximum height of 20 cm. The location of thewreck presented a challenge for the recording of the vessel.Although the visit had been planned to coincide with low tide,there was less than an hour available to carry out both test-pittingand survey before the incoming tide covered the vessel again. Asthe vessel lay on a slightly raised sandbank there was also a riskthat spending too much time carrying out survey of the site mightresult in the tide cutting the surveyors off from the beach. For thesereasons it was decided to employ multi-image photogrammetry asthe fastest method of making an accurate survey. A series of photoswere taken around the wreck in a circular pattern. These were

l wreck at Tetney Sands.

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185182

processed using Agisoft PhotoScan and used to create orthographicplans and elevations. The results of this survey were highly satis-factory and also capture the slight scour pit which had developedaround the remains while submerged at high tide. As with thesurvey at Aberlady, radiance scaling was used to emphasis thedetails in the renders (Fig. 6). The orthographic plan render wasthen georeferenced using pre-existing dGPS survey data. The use ofmulti-image photogrammetry on this occasion meant that an ac-curate survey of the remains could be carried out in a matter ofminutes allowing far more of the brief window of time available tobe allocated to other elements of the fieldwork such as interpre-tation and augering.

6.4. Value of photogrammetry as a practical survey tool forcontract-led heritage

These surveys demonstrate how valuable multi-image photo-grammetry could be in a contract-led heritage context. In all threecases recording time was very short compared to alternativetechniques, overheads were low and the outputs were highlydetailed and of sufficient accuracy to meet the objectives of theproject. Many multi-image photogrammetry software creatorshave focused their marketing at sectors other than heritage, forexample Agisoft have focused on themining industry and Autodeskhave targeted hobbyists interested in 3D printing. However, thevalue of a cost-effective, rapid and accurate method for measuredsurvey for cultural heritage is obvious and there are a small butgrowing number of published archaeological surveys which haveused the technique in a contract-led or community engagement.One of the most notable examples operating in the UK is the EUFP7-funded 3DCOFORM Project (Rodriguez-Echavarria et al., 2009)where members of the public have been encouraged to learn howto capture photographic datasets suitable for photogrammetricmodelling with the resulting models made available online (http://www.publicsculpturesofsussex.co.uk/). The software used in thiscase was Arc 3D, one of the earliest examples of an automatedphotogrammetry pipeline, developed by ESAT-PSI lab of K.U.

Fig. 6. An orthographic plan of the wrec

Leuven in Belgium and the Visual Computing Lab of CNR-ISTI inItaly (Vergauwen and Van Gool, 2006).

Around 80% of archaeological fieldwork in Scotland isdeveloper-led (Historic Scotland, 2012, 13) and at last count overhalf of the approximately 6000 archaeologists working in the UKwere employed in ‘commercial applied archaeology’ (LandwardResearch, 2011). In order for the heritage sector and the widercommunity to fully realise the benefits of this new tool for heritage,the technique must be adopted in the commercial as well as aca-demic communities. The practical value of multi-image photo-grammetry as a tool for developer-led archaeological survey mustbe measured against established techniques including as tradi-tional ‘manual survey’ and Terrestrial Laser Scanning (TLS). Thecrucial considerations are total cost, speed, reliability, accuracy,technical complexity and unique applications of photogrammetrywhere other techniques would be impossible or unsuitable.

Speed and cost of survey are critical and interconnected issuesfor all archaeological survey but are particularly important in acontract-led situation. For archaeological units working on a con-tract basis and competing to undertake contracts, there is aparticular emphasis on efficiency. It can be difficult or impossible todevote resources to experimentation with novel and potentiallyunreliable survey techniques. These units are legally committed toproduce specified deliverables to a set standard for their projects.Academic use of the technique is by its nature more experimentaland less constrained by pre-defined contractual and budgetaryconstraints. There are few published examples of multi-imagephotogrammetry being used in a commercial context in the UK.One of the most notable is the Oxford Archaeology survey of theViking massacre site at Weymouth (Ducke et al., 2011). As well asthe cost of survey the initial cost of purchase of hardware andsoftware for multi-image photogrammetry is an importantconsideration. There are some very expensive multi-image photo-grammetry software packages on the market which are powerfuland well-designed but have a limited take-up due to their cost.Tools like 123D Catch are free but cannot produce the kind of re-sults required bymost contracts. Other freeworkflows are availablebut at present tend to be much more technically demanding.

k generated using photogrammetry.

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185 183

Fortunately there has been a large investment by the heritagecommunity in Europe and beyond in the creation of free tools forphotogrammetric recording and the level of competition betweenthe numerous proprietary software providers suggests costs forthose are likely to fall. Projects such as the 3DCOFORM Project(Rodriguez-Echavarria et al., 2009) are helping to establish andpromote simple, free and accessible workflows for archaeologicaluse.

As well as cost and speed, the flexibility of photogrammetry isone of its biggest advantages. As long as a camera is available someattempt at a photogrammetric survey can be made. The techniqueis essentially scale-independent and can be applied to some of themost common types of contracted heritage survey, including his-toric building recording, small finds recording and topographicsurvey (using terrestrial or aerial platforms). Because of the mini-mal preparation required it is also useful for rapid ad hoc recordingof the type often required on walkovers and watching briefs. It isparticularly well suited for rapid recording of complex, irregular ordelicate surfaces which by their nature make traditional recordingdifficult. For example it can be an excellent way to produceorthographic plans of complex burials or rubble surfaces which areawkward or undesirable to stand on and draw accurately usingplanning frames as well as time-consuming to draw because oftheir complexity. In many cases the same or better results might beproduced using a laser scanner but not without incurring high andoften prohibitive costs associated with hardware. The survey atRubha also illustrates how the technique can be applied to complexsubjects with large amounts of self-occlusion. Attempting a laserscan of such sites would normally require a large number of sta-tions to be established. In multi-image photogrammetry eachcamera location acts like a survey station but requires with far lesseffort to set up. For heritage organisations who derive their incomethrough contracted work, investment in the technique can there-fore result in greater efficiency and enhanced outputs across manytypes of project.

Although most of the advantages of multi-image photogram-metry in comparison to other techniques must be expressed inrelative terms, there are a number of heritage survey applicationswhich might be considered exclusive to photogrammetry. Theseinclude the application of the technique to legacy datasets of his-torical images. There have been a number of successful examples ofthis using stereo photogrammetry such as the reconstructions ofthe Bamiyan Buddha statues destroyed by the Taliban inAfghanistan in 2001 (Gruen et al., 2004). In the same way, theoriginal photos of a contemporary multi-image survey retain thepotential to be reprocessed in the future for better results. Thesurvey at Rubha also demonstrates another advantage of having alarge number of data capture locations. Large numbers of verticalphotographs looking through the surface of the water were used togenerate 3D data for shallow submerged archaeological features.The use of vertical and near vertical capture locations minimisedthe effect of refraction and would be impractical to duplicate with alaser scanner. Another unique application is for completely sub-merged survey, i.e. where the camera and subject are both fullysubmerged (e.g. Drap, 2012). Although this is more challengingthan terrestrial multi-image photogrammetry it is possible toproduce excellent 3D data in an environment where laser scanningis not possible.

Accuracy is a major concern for archaeologists working in adeveloper-led context. Project briefs and industry guidelines usu-ally require that survey be carried out to a set of predefined stan-dards for the accuracy of data capture. One of the most detailed ofthese is Metric Survey Specifications for Cultural Heritage (Bryanet al., 2009) which largely predates multi-image photogrammetrybut which contains applicable standards on point density and

accuracy. This guidance suggests that when digitising from a pointcloud for illustration at 1:10 scale, the cloud should have a pointinterval density of not greater than 15mm. Less dense points cloudsare suggested for digitisation at greater scales, up to 750 mm fordigitisation for illustration at a 1:500 scale. English Heritageguidelines state that recorded points must be with the followingaccuracy ranges for standard scale output:

� for 1:50 output scale, 9 mm in reality� for 1:20 output scale, 4 mm in reality� for 1:10 output scale, 2 mm in reality

Whether these levels of accuracy can be achieved is largelydependent on theworkflowadopted. The accuracy of amulti-imagephotogrammetric survey can be affected by many variablesincluding the quality of the photos, the complexity of the lensdistortion, the depth of field, the texture of the surfaces and thelayout of the ‘stations’. There have been a number of studies whichhave demonstrated that multi-image photogrammetry can achieveresults approaching or even superior to those from laser scannersunder the right conditions (Pierrot-Deseilligny et al., 2011;Georgantas et al., 2012; Chandler and Fryer, 2013). As mentionedabove, the same equipment might be used to capture both alandscape and a pot sherd. In this case the accuracy of the land-scape is likely to be much lower in real terms than the pot sherd. Atthe micro scale, Adamtech, makers of the multi-image photo-grammetry software 3DM Analyst cite an example of a dentalsurvey which achieved accuracy of approximately 15 microns ‘us-ing inexpensive off-the-shelf digital cameras and lenses’ (http://www.adamtech.com.au/Blog/?p¼68 Accessed 16/01/2013). Duringthe recent laser scan of Stonehenge, multi-image photogrammetricmodels of carved faces of individual stones were made as the0.5 mm density laser scan was insufficiently detailed to capturesome of the fainter carvings (Abbott and Anderson-Whymark,2012). Ultimately the important question contract-led heritagebodies must consider is not how accurate the technique is in gen-eral but how accurate it can be. For now it is enough to state that, ifset up properly, multi-image photogrammetry is capable of resultswell within English Heritage guidelines and can be a meaningfulalternative to laser-scanning. Deciding whether multi-imagephotogrammetry is an appropriate approach with a low risk offailure in any given case is ultimately a judgement call and requiresexperience and investment.

The most valuable application of multi-image photogrammetrymay yet prove to be in facilitating community involvement inarchaeological recording as well as allowing for the easy creation ofinteractive 3D models and fly-throughs which allow the public toengage with their heritage sites in much more immersive way thanthat afforded by plans and sections. Important examples of projectswhich demonstrate this well include the aforementioned the EUFP7-funded 3DCOFORM Project (Rodriguez-Echavarria et al., 2009)and the Northumberland and Durham Rock Art Project (Bryan andChandler, 2008). Both of these projects centred on encouragingmembers of the public to record heritage sites using photogram-metry. The survey by the members of the Young Archaeologist’sClub at Aberlady demonstrates how straightforward this can be andit is likely that there will be many more large-scale projects of thistype in the future.

The photogrammetric surveys undertaken by the author withWA Heritage and the Young Archaeologists Club were presented inposter form at an Institute for Archaeologists (IfA) Workshop heldin Edinburgh in December 2012. The IfA is a professional organi-sationwhich sets standards and issues guidelines and which drawsa significant proportion of its membership from the contract-ledheritage sector. The seminar focused on how surveys are carried

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185184

out and presented and drew together some of Scotland’s leadingsurvey practitioners. The response of the professional archaeolog-ical surveyors present was cautiously positive and the attendeeshad an impromptu group discussion on the merits and challengesof the technique. One of the most significant issues raised in thegroup discussion was the risk that the increasing use of photo-grammetry, TLS and other remote sensing techniques could start toreplace more traditional manual survey methods. It was felt thatthis could lead to high resolution surveys of heritage featureswithout an appropriate level of on-site interpretation of phasingand context. This is certainly a real risk and one that heritagemanagers will need to keep in mind when undertaking orcommissioning photogrammetry/TLS surveys. However while it istrue that photogrammetry and other remote-sensing techniquesshould not replace professional interpretation during surveys,digital recording can be an aid to analysis rather than a replace-ment. For example the detailed 3-Dimensional models produced atRubha were found to be key in interpreting the phasing of the site,highlighting the variations in the size of stones used in differentparts of the site. When compared to traditional surveys, 3-Dimensional recording produces a more objective record of thesurface of a feature, making it easier for those who have not seenthe site in person to make their own judgements, and if necessary,re-interpretations.

7. Conclusion

De Reu et al. (2013) recently stated that multi-image photo-grammetry offers ‘better documentation of in-situ structures forfuture research and a higher public participation and awareness forthe archaeological heritage’. It is clear that within a few years thetechnique will make the leap from its current application in aca-demic contexts or as an ‘add-on’ for contract-led archaeology to astandard part of the archaeologist’s toolbox across the entire her-itage sector.

This will undoubtedly happen naturally over time but there area number of things which can be done to facilitate the process.Idea-sharing platforms such as conferences and workshops whichinclude both academics and the developer-led sector will promoteskill-sharing and understanding of data management, archivingand interpretation. In some cases there may also be a need to re-view current survey guidance which predates the release of currentmulti-image photogrammetry software (e.g. Bryan et al., 2009;Barnes, 2011). There is a particular need for technical guidance onrecommended pipelines/workflows, from data capture through toarchiving, based if possible, on free software. This will greatlyreduce the burden in terms of research and expense in the com-mercial sector and will ultimately lead to an enhanced record andunderstanding of our archaeological heritage.

Acknowledgements

The authorwould like to thankMatt Ritchie (Forestry CommissionScotland) for commissioning the survey at Rubha and its subsequentpresentation and publication, Changchang Wu of the University ofWashington for technical assistance andDan Slatcher of RPS PlanningandDevelopment forpermission touse theTetneycase study. Iwouldalso like to thank Paul Bryan of English Heritage for his review of thepaper and helpful comments prior to submission.

References

Abbott, M., Anderson-Whymark, H., 2012. Stonehenge Laser Scan: ArchaeologicalAnalysis Report. English Heritage Project 6457, English Heritage ResearchDepartment Report Series. no. 32e2012.

Barnes, A., 2011. Close-range Photogrammetry: a Guide to Good Practice. Archae-ology Data Service/Digital Antiquity Guides to Good Practice. Archaeology DataService, University of York, UK. http://guides.archaeologydataservice.ac.uk/g2gp/Photogram_Toc.

Bryan, P.G., Clowes, M., 1997. Surveying Stonehenge by photogrammetry. Photo-gramm. Rec. 15 (89), 739e751.

Bryan, P.G., Chandler, J.H., 2007. Cost-effective Rock Art Recording is there a WiderMessage?. In: The CIPA International Archives for Documentation of CulturalHeritage, vol. XXI, pp. 210e215. Athens.

Bryan, P.G., Chandler, J.H., 2008. Cost-Effective Rock-Art Recording within a Non-Specialist Environment. In: International Society for Photogrammetry andRemote, Sensing, ISPRS 2008, Beijing, China, July 3e11 2008.

Bryan, P., Blake, B., Bedford, J., 2009. Metric Survey Specifications for CulturalHeritage, second ed. English Heritage, Swindon http://www.english-heritage.org.uk/publications/metric-survey-specification/.

Chandler, J.H., Fryer, J.G., Kniest, H.T., 2005. Non-invasive three-dimensionalrecording of aboriginal rock art using cost-effective digital photogrammetry.Rock Art. Res. 22 (2), 119e130.

Chandler, J.H., Fryer, J., January/February 2013. 123D catch: how accurate is it?Geomat. World.

De Reu, J., Plets, G., Verhoeven, G., De Smedt, P., Bats, M., Cherretté, B., DeMaeyer, W., Deconynck, J., Herremans, D., Laloo, P., Van Meirvenne, M., DeClercq, W., 2013. Towards a three-dimensional cost-effective registration of thearchaeological heritage. J. Archaeol. Sci. 40 (2), 1108e1121.

De Reu, J., De Smedt, P., Herremans, D., Van Meirvenne, M., Laloo, P., De Clercq, W.,2014. On introducing an image-based 3D reconstruction method in archaeo-logical excavation practice. J. Archaeol. Sci. 41, 251e262.

Drap, P., 2012. Underwater photogrammetry for Archaeology. In: Da Silva, D.C. (Ed.),Special Applications of Photogrammetry. Available from: http://www.intechopen.com/books/special-applications-of-photogrammetry/underwater-photogrammetry-for-archaeology (accessed 18.05.13.).

Ducke, B., Score, D., Reeves, J., 2011. Multiview 3D reconstruction of the archaeo-logical site at weymouth from image series. Comput. Graph. 35 (2), 2375e2382.

Furukawa, Y., Ponce, J., 2010. Accurate, dense, and robust multi-view stereopsis.IEEE Trans. Pattern Anal. Mach. Intell. 32 (8), 1362e1376.

Georgantas, A., Brédif, M., Pierrot-Deseilligny, M., 2012. An accuracy assessment ofautomated photogrammetric techniques for 3DModeling of complex interiors.Int. Arch. Photogramm. Remote Sens. Spatial Inform. Sci. 39 (Part B3), 23e28.

Gilboa, A., Tal, A., Shimshoni, I., Kolomenkin, M., 2013. Computer-based, automaticrecording and illustration of complex archaeological artifacts. J. Archaeol. Sci.40, 1329e1339.

Granier, X., Vergne, R., Pacanowski, R., Barla, P., Reuter, P., 2012. Enhancing surfacefeatures with the radiance scaling Meshlab Plugin. CAA Conf. (Comput. Appl.Quant. Methods Archaeol.).

Gruen, A., Remondino, F., Zhang, L., 2004. Photogrammetric reconstruction of thegreat Buddha of Bamiyan, Afghanistan. Photogramm. Rec. 19 (107), 177e199.

Historic Scotland, 2012. Historic Scotland: A Review of the Archaeology Function.http://www.historic-scotland.gov.uk/archaeology-review-2012.pdf (accessed13.03.13.).

Jeffrey, S., 2003. Three Dimensional Modelling of Scottish Early Medieval SculptedStones. University of Glasgow (PhD thesis). http://ads.ahds.ac.uk/catalogue/adsdata/arch-387-1/dissemination/pdf/3demss.pdf (accessed 11.03.13.).

Kirby, J., 1983. Gorstanvorran, Loch Shiel (Ardnamurchan p) Promontory Fort. Dis-covery and Excavation in Scotland. Council for British Archaeology, pp. 13e14.

Koutsoudis, A., Vidmar, B., Arnaoutoglou, F., 2013. Performance evaluation of amulti-image 3D reconstruction software on a low-feature artefact. J. Archaeol.Sci. 40, 4450e4456.

Landward Research Ltd, 2011. State of the Archaeological Market e October 2011.Report prepared on behalf of the Institute for Archaeologists and the Federationof Archaeological Managers and Employers. http://www.archaeologists.net/sites/default/files/node-files/JoblossesOctober2011.pdf (accessed 13.03.13.).

Laussedat, A., 1854. Mémoire sur l’emploi de la chamber Claire dans les re-connaissances topographiques. Mémorial de l’Officier du Génie. 16. Mallet-Bachelier, Paris.

Laussedat, A., 1859. Mémoire sur l’emploi de la photographie dans la levée desplans; par M. Laussedat (Extrait par l’auteur). Comptes rendus séances l’Acad.Sci. 49, 732e734.

Lowe, David G., 1999. Object recognition from local scale-invariant features. In:Proceedings of the International Conference on Computer Vision, vol. 2,pp. 1150e1157.

McCarthy, J., 2012. Rubha an Fhaing Dhuibh, Lochaber Forest District Photogram-metric Archaeological Survey. Unpublished Report on behalf of the ForestryCommission, WA Heritage Ref 84240.01.

McCarthy, J., 2013. Tetney Sea Line Replacement, Tetney, Lincolnshire: Non-intrusive Archaeological Investigation of Wreck Remains. Unpublished Reporton behalf of RPS, WA Coastal & Marine Ref 100081.01.

Olson, B.R., Placchetti, R.A., Quartermaine, J., Killebrew, A.E., 2013. The Tel AkkoTotal Archaeology Project (Akko, Israel): assessing the suitability of multi scale3D field recording in archaeology. J. Field Archaeol. 38 (3), 244e262.

Pierrot-Deseilligny, M., De Luca, L., Remondino, F., 2011. Automated image-basedprocedures for accurate artifacts 3D Modeling and orthoimage generation. In:Proceedings of the 23rd CIPA Symposium. Czech Republic, Prague.

Plets, G., Gheyle, W., Verhoeven, G., De Reu, J., Bourgeois, J., Verhegge, J.,Stichelbaut, B., 2012a. Towards a three-dimensional registration of thearchaeological heritage of the Altai Mountains. Antiquity 86 (333), 884e897.

J. McCarthy / Journal of Archaeological Science 43 (2014) 175e185 185

Plets, G., Verhoeven, G., Cheremisin, D., Plets, R., Bourgeois, J., Stichelbaut, B.,Gheyle, W., De Reu, J., 2012b. The deteriorating preservation of the Altaian rockart e assessing three-dimensional image-based modelling in rock art researchand management. Rock Art. Res. 29 (2), 139e156.

Remondino, F., Del Pizzo, S., Kersten, P.T., Troisi, S., 2012. Low-cost and open-sourcesolutions for automated image orientation e a critical overview. Lect. NotesComput. Sci. 7616, 40e54.

Rodriguez-Echavarria, K., Morris, D., Arnold, D., 2009. Enabling users to create anddocument 3D content for heritage. In: The First International Conference onCreative Content Technologies, CONTENT 2009, Athens/Glyfada, Greece, 15e20November, 2009.

Roncella, R., Re, C., Forlani, G., 2011. Performance evaluation of a structure andmotion strategy in architecture and Cultural Heritage. Int. Arch. Photogramm.Remote Sens. Spat. Inf. Sci. 38(5/W16), on CD-ROM. ISPRS Int. Workshop 3D-ARCH 2011, Trento, Italy.

Szeliski, R., 2011. Computer Vision: Algorithms and Applications. Springer, NewYork.

Vergauwen, M., Van Gool, L., 2006. Web-based 3D reconstruction service. Mach. Vis.Appl. 17, 411e426.

Vergne, R., Pacanowski, R., Barla, P., Granier, X., Schlick, C., 2010. Radiance scaling forversatile surface enhancement. In: I3D ’10: Proceedings of the 2010 ACM SIG-GRAPH Symposium on Interactive 3D Graphics and Games. ACM, pp.1e8. http://hal.inria.fr/docs/00/44/98/28/PDF/RadianceScaling.pdf (accessed 11.03.13.).

Verhoeven, G., 2011. Taking computer vision Aloft e archaeological three-dimensional reconstructions from aerial photographs with PhotoScan.Archaeol. Prospect. 18, 67e73.

Verhoeven, G., Doneus, M., Briese, C., Vermeulen, F., 2012. Mapping by matching: acomputer vision-based approach to fast and accurate georeferencing ofarchaeological aerial photographs. J. Archaeol. Sci. 39 (7), 2060e2070.

Wu, C., Agarwal, S., Curless, B., Seitz, S.M., 2011. Multicore bundle adjustment. In:IEEE Conference on Computer Vision and Pattern Recognition (CVPR),pp. 3057e3064.

Contents lists available at ScienceDirect

Journal of Archaeological Science: Reports

journal homepage: www.elsevier.com/locate/jasrep

Photogrammetric texture mapping: A method for increasing the Fidelity of3D models of cultural heritage materials☆

Christopher Dostala,⁎, Kotaro Yamafuneb

a Texas A&M University, Department of Anthropology, MS 4352 TAMU, College Station, TX 77843, United StatesbAPPARATUS LLC, Tokyo, Japan

A B S T R A C T

Accurately recording information is the single most important stage of an archaeological project. The biggesttechnological improvement to documentation techniques in the last 15 years has been the spread of various 3Ddigitization technologies, such as computer vision photogrammetry and 3D laser scanning. These technologieshave allowed archaeologists to quickly and accurately capture and reconstruct the geometry and colors of thesubjects being studied. Each of the various methods for 3D digitization of cultural heritage materials has ad-vantages and disadvantages, such as processing speed, cost of equipment, and the accuracy of the captured data.Laser scanned data is among the most accurate geometrical data available in modern scanning techniques, but itlacks in its ability to accurately capture textures and diagnostic coloration information. Photogrammetric dataproduces highly detailed photographic textures on models, but the geometric data tends to be of a lower defi-nition than the laser scanned data. In this paper, the authors discuss a new methodology that combines theadvantages of computer vision photogrammetry with those of laser scanning by applying the photographictextures produced with photogrammetry to the geometric data obtained from laser scanning. This method allowsarchaeologists to achieve the best possible fidelity 3D models for interpretation and study.

1. Introduction

Three-dimensional digitization is fast becoming a standard practicefor cultural heritage documentation. Archaeological sites and artifactsare routinely digitized and curated in a variety of different ways, butthe ultimate goal of each method is to accurately represent the originalobject or place, and to convey reliable information. There is a level ofinterpretation and error inherent in each process of 3D modeling ar-chaeological artifacts, and the goal for archaeologists is to minimizethis error. Laser based scanning methods, such as the FARO laser scanarm, and others, have been used with great success, and in the pastseven years, computer vision photogrammetry (also known as multi-image photogrammetry, close-range photogrammetry, or structure-from-motion) has become nearly ubiquitous for certain aspects ofdocumentation. These technologies allow archaeologists to documenttheir study subjects accurately in 3D digital data formats. Availablescanning/documenting technologies use different methods to captureand reconstruct dimensions and colors of the study subjects in digitalformats. Each of these methods have advantages and disadvantages,such as processing speed, the cost of software and equipment, and the

accuracy of captured data.The intent of this study is to maximize the accuracy of scanned 3D

digital data by using both a laser scanner and computer vison photo-grammetry. Having processed many dozens of computer vision photo-grammetry models and fixed-base laser scanner models in the lastseveral years through the Nautical Archaeology Program at Texas A&MUniversity, the authors have noted that though photogrammetrymodels are very visually compelling due to the detail of the photo-graphic texture maps, the major drawback of the technique is the errorsfrequently found on the underlying mesh. When the texture map isremoved, the surfaces of smooth objects are frequently bumpy ormottled, and more complex surfaces are typically smoother and sim-plified compared to how they are in both the real world and comparablelaser scans (Fig. 1).

Laser scanned data tends to produce more consistently accuratedimensional data compared to other documentation methods, but theyfail to produce truly high-resolution textures that correctly convey di-agnostic coloration information (Gerbino et al., 2004; Radosevic,2010). Here a new methodology is proposed that combines the ad-vantages of laser scanners with those of computer vision

https://doi.org/10.1016/j.jasrep.2018.01.024Received 16 October 2017; Received in revised form 9 January 2018; Accepted 19 January 2018

☆ Neither author has any financial or personal conflicts of interests that could influence this paper. No external funding was required for this study.⁎ Corresponding author at: Texas A&M University, Department of Anthropology, MS 4352 TAMU, College Station, TX 77843, United States.E-mail address: dostalc@tamu.edu (C. Dostal).

Journal of Archaeological Science: Reports 18 (2018) 430–436

Available online 20 February 20182352-409X/ © 2018 Elsevier Ltd. All rights reserved.

T

photogrammetry using a technique called photogrammetric texturemapping (PTM). Using PTM, the texture map generated via photo-grammetry can be precisely mapped onto the more precise geometry oflaser scanned 3D digital models with relative ease. This methodologyallows archaeologists to have a 3D digital model that possesses laserscan quality geometry with photorealistic textures.

Past attempts to marry laser scanning and photogrammetry tech-nologies have largely focused on implementing measurements obtainedvia the highly precise laser scan data to improve the accuracy of thephoto-realistic photogrammetry models (see Grammatikopoulos et al.,2004; Lambers et al., 2007; Alshawabkeh and Haala, 2004). This ap-proach allows for the creation of visually accurate digital representa-tions of a site or artifact, but fails to address the inaccuracies inherent inthe mesh underlying the texture map. Other studies (Al-kheder et al.,2009) have focused on ways to improve texture mapping photographicimages to laser obtained 3D models, but the flaws inherent in projectinga 2D image on to a 3D surface prevent highly accurate texture mapping.

Instead of using laser scanning to improve the quality of the pho-togrammetry model, this methodology was designed to utilize thestrengths of each technology. PTM can be used to quickly and easilymap the photorealistic textures generated from computer vision pho-togrammetry on to 3D models obtained with a coordinate measuringmachine (CMM) laser scan arm, the result of which we believe to besome of the most realistic and accurate digital models of archaeologicalartifacts to date.

1.1. Materials and methods

For this study, a FARO Edge CMM arm with a laser line probe usingGeomagic's Design X 2016 software and Agisoft's PhotoScan professionalversion 1.3.4 (photogrammetry software) were utilized. Additionally,Adobe Photoshop CS6 was used to prepare photos prior to being loadedinto PhotoScan, and McNeel's Rhinoceros 5 was used to align meshmodels. To test the PTM procedure, three different artifacts werescanned and texture mapped: a historic period ceramic plate recoveredfrom Port Royal, Jamaica, a waterlogged timber from the bow of theship remains found below the World Trade Center, and a Clovis pro-jectile point from the Hogeye Clovis Cache near Bastrop, Texas. Thoughthe focus of this paper is to utilize the PTM methodology for

archaeological artifact modeling, it can be easily scaled up for recordingentire sites, buildings, or landscapes. PTM will also work with legacydata sets obtained with different types of hardware and software; if anobject has been laser scanned in the past, the mesh model can be ret-roactively texture mapped using this method, as shown below in thediscussion of the Clovis projectile point.

2. Advantages and disadvantages of computer visionphotogrammetry

Photogrammetry – a documentation technique by which accurate3D metric and descriptive object information from multiple images isderived (Al-Ruzouq, 2012) – was originally developed as a technologyto provide topographic information for map making (Burtch, 2008;Van, 2015; Konecny, 2003). The development of computers in the1980s created opportunities for the introduction of iterative trigono-metric software to triangulate tridimensional points; such softwarecould be applied to digital photography (Canciani et al., 2003; Drap,2012). In the beginning of the 21st century, surveying methods usingphotography underwent rapid development, and photogrammetry tocalculate coordinates of assigned points using an accurately calibratedcamera made its appearance in archaeology (Green et al., 2002). Off-the-shelf photogrammetric that software using digital images, such asPhotoModeler, was tested for accuracy by archaeologists. In 2010, a newsoftware package using computer vision photogrammetry (also knownas multi-image photogrammetry, or close-range photogrammetry) be-came available to archaeologists (Skarlatos and Rova, 2012; Doneuset al., 2011; Drap, 2012; Diamanti et al., 2011; Zhukvsky et al., 2013;Henderson et al., 2013; McCarthy and Benjamin, 2014). Under thecommercial designation PhotoScan, this package does not require pre-calibration of the cameras by the user. Using computer vision photo-grammetry, PhotoScan creates meshes, or geometry of 3D digitalmodels, using pixel information from digital images. The data filesproduced by computer vision photogrammetry are similar to thoseobtained from tridimensional laser scanners. Today, computer visionphotogrammetry is rapidly becoming one of the standard recordingmethods in archaeology, especially in underwater archaeology due tothe time and equipment constraints encountered when recording un-derwater (Yamafune, 2016).

Fig. 1. A mesh model created with photogrammetry of a ceramic plate from Port Royal Jamaica. The left images show the model with texture, and the right models show without. Notethe mottled surface of the mesh when the texture is removed.

C. Dostal, K. Yamafune Journal of Archaeological Science: Reports 18 (2018) 430–436

431

Since computer vision photogrammetry uses pixel information, orcolors, of digital 2D images to structure 3D models, the greatest ad-vantage of photogrammetric 3D modeling is photorealistic textures.The basic workflow of PhotoScan is: Align Photos, Build Dense Cloud,Build Mesh, and Build Texture. Since digital photos make up the sourcedata, the quality of the camera, the skill of the photographer, and theenvironmental constrains like lighting have a significant impact on thequality of the images captured and subsequently the models producedfrom those images. Because computer vision photogrammetry usesanywhere from dozens to hundreds of overlapping photos to structurethese models, slight changes in the pixel values across images that occurdue to variations in lighting and refractions tend to create noisy datathat results in poor quality mesh (Fig. 2 Image of PhotogrammetricModels).

3. Advantages and disadvantages of laser scanning

Laser scanning, at its core, is using a digital camera to photographlaser light and calculate the geometry of a surface based on the returnangle of the laser to the digital camera. Lasers scanners consist of apulsing laser emitter that projects a line of laser light, and a charge-coupled device (CCD) sensor, which is a grid of pixels that convertphotons of light into electrical charges that are then interpreted as di-gital images (FARO, 2012, 10). The CCD measures the reflection of thelaser photons emitted from the probe, and much like a camera, thesettings need to account for all of the things that can affect the way thatthe photons will return to the sensor. Though tightly focused andmonochromatic, lasers are still beams of photons, and as such, they aresusceptible to reflection, refraction, and absorption when they interactwith the objects being scanned. Different algorithms and settings canminimize the noisy data that is returned to the sensor, but there is al-ways a built in level of error to any 3D scanning technique. Even so,with practice, it is possible to obtain very accurate point cloud datausing a laser scanner, and these mesh models made from these pointclouds tend to be of higher accuracy than computer vision photo-grammetry packages like Photoscan.

The laser scanner used for this study was a FARO Edge CMM armwith a laser line probe attached. The accuracy and speed of data ac-quisition are the two main strong points of a CMM based laser scanner.The speed of data acquisition depends on the equipment and the con-figuration of the software, but as an example, when digitizing the largefloor timbers from the ship found below the World Trade Center in NewYork, the average scan took only 30min. The mesh models producedfrom these scans were so detailed that the wood grain could easily beseen even without any texture mapping.

The cost and portability of the laser scanner arm are its two majorweak points. The initial quote for the Faro Edge CMM arm and thelaser-line probe (LLP) were $49,900 and $29,900, respectively (Luza,2017). Coupled with the computing resources and software packagesneeded to gather and process the data, the overall cost of a system likethis can easily reach $100,000. This is a substantial investment, espe-cially for smaller CRM firms or universities. Compared to the $3500price tag of Agisoft's Photoscan software, this cost is a substantial dis-advantage. The size and portability of this system are also significantlymore cumbersome than what is needed for computer vision photo-grammetry. A tripod or other firm stand is needed to use the arm, andthe range of articulation means that it is bulky and difficult to travelwith. Given the cost, and the air travel industry's reputation for roughtreatment of luggage, it is also a nerve-wracking experience to travelwith a CMM arm based laser scanner. Despite the drawbacks, thequality of the 3D models produced is consistently excellent, and ifavailable, it is the best option for capturing accurate geometry of cul-tural heritage artifacts.

4. Workflow for photogrammetric texture mapping of laser-obtained geometries

As stated in the introduction, PTM aims to combine the advantagesof computer vision photogrammetry and laser scanning, allowing ar-chaeologists to create 3-D digital models that have geometries gener-ated through laser scanning and textures composed by computer visionphotogrammetry. To begin, a photogrammetric model and a laserscanned model of a single object must be obtained. Once both modelsare complete, the mesh of the photogrammetric model is replaced withthe laser scanned model after it has been properly aligned in a thirdparty 3D modeling program. Once replaced, the textures can be createdon the surface of the imported laser scanned 3D model directly inPhotoscan.

4.1. Photogrammetric modeling

Instead of outlining each stage of Photoscan's workflow, for thepurposes of this paper, the focus will be the generation of the texturemap, which occurs after importing the photos, aligning the photos,building the dense cloud, and building the mesh. During the generationof a texture map, PhotoScan creates a photomosaic and places it on thesurface of the mesh geometry. For a full explanation of each settingduring the previous stages, refer to Yamafune, 2016.

The “Build Texture” stage in the workflow has several variablesettings that change how the software processes the source images to

Fig. 2. The laser-scanned model (white) being aligned to the position of the PhotoScan model (the darker areas).

C. Dostal, K. Yamafune Journal of Archaeological Science: Reports 18 (2018) 430–436

432

create the texture map. The first of these settings is ‘Mapping Mode’,which alters the way Photoscan orients the texture on the model.Options for the Mapping Mode setting are ‘Generic,’ ‘AdaptiveOrthophoto,’ ‘Orthophoto,’ ‘Spherical,’ ‘Single Camera,’ and ‘Keep UV.’(UV mapping is a 2D atlas for textures, or a canvas on which texturescan be projected). The default setting is “Generic”; with this option,PhotoScan automatically chooses the best photos (based on cubic pro-jection which creates a cube that encompasses the subject) and com-poses a UV atlas of meshes to convert 3D structural meshes into a 2Dimage canvas for the created textures. This mode is preferred for modelsthat have 3D rich structures. The ‘Orthophoto’ setting creates a pho-tomosaic from a projection plane (generally, a top view plane) andprojects a created photomosaic onto the meshes from one direction(usually from a plane which is parallel to the mesh surfaces). Differingslightly from Orthophoto, ‘Adaptive Orthophoto’ detects vertical sur-faces and creates textures independently for those vertical meshes. The‘Spherical’ mode creates a UV map ideal for a ball-like sphericalstructures; given that a texture atlas is a 2D plane, textures of sphericalshapes require a unique mapping projection (similar to projection stylesfor world maps). This mode is only suitable for spherical structures.‘Single Photo’ creates a texture based on one single photo; a photo canbe chosen from any photo that was used for the photogrammetricprocess. ‘Keep UV’ imports a UV map that is created in different soft-ware; many computer modeling software suites can create textures on3D digital models, or meshes. If a UV map had been previously beencreated in a different program, PhotoScan can import UV maps in thisprocess (Agisoft, 2017).

Another sub-setting of “Build Texture” is the “Blending Mode”.While the mapping mode options determine how to project the texturemap onto the mesh, the blending mode options determine how the colorvalues of the texture map are calculated. The ‘Blending Mode’ has fiveoptions: ‘Mosaic,’ ‘Average,’ ‘Max Intensity,’ ‘Min Intensity,’ and‘Disabled.’ The default blending mode is “Mosaic”, which selects pixelsfrom the source photos where the pixels are the closest to the center ofthe photos. The ‘Average’ setting calculates an average pixel value, orRGB value of colors, of assigned photos, and uses those pixel values astextures. The ‘Max Intensity’ setting applies images that have themaximum intensity, or brightest pixels, to compose textures. The ‘MinIntensity’ setting is the opposite of the Max Intensity setting; it usesminimum intensity images to compose texture. The ‘Disable’ setting isused for an imported model that already has suitable textures. The threemodels shown in this paper used the ‘Generic’ option for the mappingmode and the ‘Mosaic’ option for the blending mode.

Once the photogrammetric mesh and texture are complete, themesh model for each artifact should be exported as a .stl or .obj file.Though Photoscan can export the mesh in a plethora of file formats, .stland .obj file types are recommended due to their wide implementationand compatibility across nearly all 3D modeling software packages.With the mesh model exported, the next step is to create a 3D model ofthe same object with a laser scanner. In this case, the authors used theFARO Edge ScanArm, but as mentioned in the introduction, any scan-ning method that produces a meshable point cloud can be used.

4.2. Obtaining laser data

The laser scans of the ceramic plate and the water-logged woodentimber from the World Trade center ship were completed at theConservation Research Laboratory at Texas A&M University using aFARO Edge CMM arm with an attached laser line probe, usingGeomagic's Design X program to capture and process the data. The re-solution of these scans was set to a 0.2mm grid, with a close intervalbetween lines (1.0 mm spacing). The scan of the Hogeye cache Clovispoint was done in 2011 with a FARO Fusion CMM arm with the 2009model laser line probe. The original point cloud data obtained was notsaved and therefore the resolution settings are not known, but uponinspection of the mesh, the settings must have been similar.

Each scan was completed in two parts, one of the obverse side andone of the reverse side. The scans of each side were overlapped to fa-cilitate alignment between the scans. Once both sides were scanned,both point cloud sets were edited to remove the surface of the table onwhich the objects were set during the scanning process, as well as anyother extraneous data that was acquired. The point cloud for each sideof the object was then aligned to the other and combined to make asingle point cloud that represented all sides of each artifact. Once thepoint clouds were complete, they were converted to meshes and ex-ported as .stl files to be aligned to the meshes created in Photoscan.

4.3. Model Replacement in rhinoceros 3D CAD Modeling Software

With the models obtained from the laser scanner and computer vi-sion photogrammetry, the next step is to align the laser scanned modelto the position of the photogrammetry model. When a 3D model iscreated, its orientation and positioning in 3D space does not correspondto its position or orientation in the real world, unless those data arespecified by the user. When a mesh model is created in PhotoScan forinstance, it is assigned an arbitrary orientation and position, and thoughit does not correspond to a real-world orientation or position, it is tiedto the calculated camera positions for the photos used to create thepoint cloud. Maintaining the precise relationship between the positionof the mesh model and the calculated camera positions lies at the heartof successful implementation of this methodology.

The next step is to replace the mesh model created with PhotoScanwith the mesh model created with laser scanned data. For this to work,the laser mesh model needs to be exactly the same scale and correctlyoriented at the same coordinates as the PhotoScan mesh model. For thisstudy, the models were aligned using the digital modeling softwareRhinoceros 5, because of its relative ubiquity in the archaeologicalcommunity, its user-friendly interface, and its ability to manipulatemodels with very high precision.

For this methodology to be successful, the original position andorientation of the model as it was exported from Photoscan must beexactly matched by the laser scanned model. Once both the photo-grammetric model and the laser scanned model are imported intoRhinoceros 5, it is important to ensure that the photogrammetric modelis not accidently moved or scaled. To do this, the photogrammetricmodel should be assigned a separate layer from the laser scannedmodel, and that layer should be locked. With that done, the laserscanned model can be aligned to the location and scale of the PhotoScanmodel by using various commands such as ‘move,’ ‘scale3D,’ ‘rotate,’and ‘align’(Fig. 2). Automatic alignment between models can be donewith alternative programs, such as Cloud Compare, Geomagic Design X,and others. Whichever method of alignment is used, as long as the laser-obtained model is identically scaled and positioned to the photogram-metric model, the next step will work.

Once the position and orientation of the laser scanned model mat-ches that of the PhotoScan model exactly, the laser scanned modelshould be exported, again as a .stl file. The original model should thenbe re-opened in Photoscan, and the 3D model (the mesh model) in theWorkspace should be deleted. Next, the laser scan mesh should beimported, using the ‘Import Mesh’ function in the tools menu. With themesh replaced, the ‘Build texture’ step can be run, and the texture mapwill be placed on the laser scanned model with the same precision itwould have been placed on the Photoscan model (Fig. 3). Fig. 4 shows aworkflow diagram summarizing the steps listed above.

The methodology proposed here produces a model with the pho-tographic accuracy of photogrammetic 3D models and the geometricprecision of laser scanned 3D models. Measurements taken from visualcues of the photographic texture are more accurate due to the increasedfidelity of the underlying mesh, and the incidence of inadequate modeldetail being mistaken for an actual characteristic of an artifact insteadof an erroneous model should be reduced.

C. Dostal, K. Yamafune Journal of Archaeological Science: Reports 18 (2018) 430–436

433

5. Results and discussion

The models of all three artifacts were substantially improved afterthe PTM procedure was carried out. Even processed on the highestquality settings in PhotoScan, photogrammetric models do not have thelevel of detail achievable through laser scanning, at least for the timebeing. PTM allows for as close to an archival-quality digital model as isachievable with current technologies.

The first artifact used to develop the methodology of PTM was atransfer print ceramic plate from Port Royal, Jamaica. This artifact wasused as a teaching aide for students learning computer vision photo-grammetry, and it was over the course of two years with consistenterrors in the mesh that the initial problem with photogrammetricmeshing was identified. The laser scanned model is quite obviouslymuch cleaner and more representative of the actual artifact, and Fig. 5shows the obvious disparity between the two. The quality of the modelafter the PTM process was complete was well beyond initial expecta-tions, and is what led to this paper.

Adding photographic texture maps to laser-scanned meshes isnothing new, and had been done extensively by Dr. Joshua Keene tomodel projectile points and bifaces excavated through the Center forthe Study of First Americans at Texas A&M University. One of thedrawbacks to manually mapping a 2D photograph to a 3D model is theerrors that occur where photographs from different views meet. Toverify that a PTM would handle the texture mapping more successfullythat this method, a photogrammetric model was made of one of theClovis points that had been modeled, and then the texture was pro-jected on his original laser scan. Fig. 6 shows the difference between thePTM model and the model manually texture mapped with two photo-graphs. (Figure three in Section 4.3 shows the three iterations of thisartifact).

Finally, to check the utility of the PTM method with organic ma-terial, a random piece of timber was selected from the assemblage oftimbers from the 18th-century ship remains discovered below theWorld Trade Center in Manhattan, currently undergoing conservationat the Conservation Research Laboratory at Texas A&M University.Unlike the previous two examples, the mesh detail missing from thephotogrammetric model of this timber is less obvious, due to relativelyrough texture of the wood and its larger size. Close examination doesshow that the photogrammetric mesh does not show the grain structureof the wood that can be seen in the laser-scanned mesh, and is alsomuch smoother around the concreted bolt seen on the left of the modelin Fig. 6. The addition of the photogrammetric texture map to thismodel allows for easier identification of Teredo navalis damage to thewood, as well as the evidence of iron corrosion (Fig. 7).

Fig. 3. Once aligned, the photogrammetric mesh (left) isreplaced by the laser scanned mesh (middle) and the tex-ture map is re-mapped to the laser-scanned mesh (right).

Fig. 4. Workflow diagram for PTM.

C. Dostal, K. Yamafune Journal of Archaeological Science: Reports 18 (2018) 430–436

434

Computer-aided 3D modeling has been the ‘next big thing’ in ar-chaeology for over two decades now. Even with the vast troves of vi-sually appealing archaeological models populating the world now, veryfew –if any- archaeologists would suggest a 3D model of an artifact bestudied in lieu of the actual artifact. Errors in processing, incon-sistencies in quality, and a general distrust of virtual data prevent theadoption of using the digital forms of an object to form the backbone ofany archaeological study. This is probably for the best; as seen in figuresx and x, there are typically underlying errors in the data for any 3Dmodel produced.

However, it is likely that the distrust of ‘virtual’ archaeology willease in the coming years. Many students of archaeology today havenever known life without computers, and to many of them it is naturalto consume information in a virtual format. If that is indeed the case,then the standards by which we critique the fidelity of 3D models ofcultural heritage materials need to be extremely high. This study aims

to provide a methodology by which so-called ‘archival-grade’ 3Dmodels can be produced; 3D models that mimic the texture and geo-metry of their real-world counterparts as accurately and precisely asachievable with existing technologies.

6. Conclusion

The PTM method outlined in this paper allows archaeologist toimprove the quality of the digital 3D models currently being acquiredand archived. Due to the open-ended nature of the types of objects canbe digitized via laser scanning and computer vision photogrammetry,the PTM method can easily be extended into other fields of study thatdesire the same level of accuracy as archaeology. Additionally, othermethods for obtaining accurate geometric data beyond laser scanninglike LIDAR or multibeam sonar can be used as the underlying point datato build a mesh. Photogrammetric models produced by extracting stills

Fig. 5. The final post-PTM model of the ceramic plate from PortRoyal, Jamaica.

Fig. 6. The PTM model (left) is much more evenly colored andconsistent than the model that was texture mapped using pho-tographs of the two faces of the projectile point (right).

C. Dostal, K. Yamafune Journal of Archaeological Science: Reports 18 (2018) 430–436

435

of high-definition video captured by aerial drones or by underwaterremotely operated vehicles (ROVs) could be paired with the geometricdata obtained by LIDAR or multibeam sonar respectively to produceever-higher quality 3D models.

References

Agisoft, L.L.C., 2017. Agisoft Photoscan User Manual: Professional Edition. http://www.agisoft.com/downloads/user-manuals/, Version 1.3 (Accessed 9/27/2017).

Al-kheder, S., Al-shawabkeh, Y., Haala, N., 2009. Developing a documentation system fordesert palaces in Jordan using 3D laser scanning and digital photogrammetry. J.Archaeol. Sci. 36 (2), 537–546.

Al-Ruzouq, 2012. Photogrammetry for Archaeological Documentation and CulturalHeritage Conservation. INTECH Open Access Publisher, Rijeka.

Alshawabkeh, Y., Haala, N., 2004. Integration of digital photogrammetry and laserscanning for heritage documentation. In: International Archives of Photogrammetryand Remote Sensing, Jul;35:B5.

Burtch, 2008. History of Photogrammetry, (Based on Lecture Notes on the History ofPhotogrammetry at Ferris State University). https://spatial.curtin.edu.au/local/docs/HistoryOfPhotogrammetry.pdf Last Updated 2008 (Accessed 11/24/2015).

Canciani, M., Gambogi, P., Romano, F.G., Cannata, G., Drap, P., 2003. Low cost digitalphotogrammetry for underwater archaeological site survey and artifact isertion. Thecase study of the Dolia wreck in Secche della Meloria-Livorno-Italia. In: InISPRSWorkshop Vision Tecniques for Digital Architectural and Archaeological Archives,Jul.

Diamanti, Georgopoulos, Vlachaki, 2011. Geometric Documentation of UnderwaterArchaeological Sites, XIII CIPA International Symposium, Prague.

Doneus, M., Verhoeven, G., Fera, M., Briese, Ch., Kucera, M., Neubauer, W., 2011. FromDeposit to Point Cloud: A Study of Low-cost Computer Vision Approaches for theStraightforward Documentation of Archaeological Excavations, Geoinformatics.Faculty of Civil Engineering, Czech Technical University In Prague.

Drap, 2012. Underwater Photogrammetry for Archaeology. INTECH Open AccessPublisher, Rijeka.

FARO, 2012. Scanning FAQ. In: Technical Guide. FARO.Gerbino, S., Martorelli, F., Speranza, D., 2004. Cheap Photogrammetry Versus Expensive

Reverse Engineering Techniques in 3D Model Acquisition and Shape Reconstruction.

In: International Design Conference 2004, Dubrovnik, Croatia.Grammatikopoulos, L., Kalisperakis, I., Karras, G., Kokkinos, T., Petsa, E., 2004, October.

Automatic multi-image photo-texturing of 3D surface models obtained with laserscanning. In: CIPA International Workshop on "Vision Techniques Applied to theRehabilitation of City Centres", Lisbon, pp. 25–27.

Green, Matthews, Turanli, 2002. Underwater archaeological surveying usingPhotoModeler, VirtualMapper: different applications for different problems. Int. J.Naut. Archaeol. 31, 283–292.

Henderson, Pizarro, Johnson-Roberson, Mahon, 2013. Mapping submerged archae-ological sites using stereo-vision photogrammetry. Int. J. Naut. Archaeol. 42,243–256.

Konecny, 2003. Geoinformation Remote Sensing, Photogrammetry and GeographicInformation Systems. Taylor & Francis, London.

Lambers, K., Eisenbeiss, H., Sauerbier, M., Kupferschmidt, D., Gaisecker, T., Sotoodeh, S.,Hanusch, T., 2007. Combining photogrammetry and laser scanning for the recordingand modelling of the Late Intermediate Period site of Pinchango Alto, Palpa, Peru. J.Archaeol. Sci. 34 (10), 1702–1712.

Luza, Rebekah, 2017. Geomagic software quote. In: Personal Correspondance, CollegeStation, TX, February 7.

McCarthy, Benjamin, 2014. Multi-image photogrammetry for underwater archaeologicalsite recording: an accessible, diver-based approach. J. Marit. Archaeol. 9, 95–114.

Radosevic, 2010. Laser scanning versus photogrammetry combined with post-modeling inStecak digitization. In: Proceedings of the 14th Central European Seminar onComputer Graphics.

Skarlatos, D., Rova, S., 2012. Comparison of laser scanning, photogrammetry and SFM-MVS pipeline applied in structures and artificial surfaces. In: ISPRS Annals of thePhotogrammetry, Remote Sensing and Spatial Information Sciences. Vol. I-3.

Van Damme, 2015. Computer Vison Photogrammetry for Underwater Archaeological SiteRecording: A Critical Assessment. M.A. Thesis. University of Southern Denmark,Esbjerg.

Yamafune, K., 2016. Using Computer Vision Photogrammetry (Agisoft PhotoScan) toRecord and Analyze Underwater Shipwreck Sites. Doctoral Dissertation. Texas A&MUniversity.

Zhukvsky, M., Kuznetsov, V., Olkhovsky, S., 2013. Photogrammetric techniques for 3Dunderwater record of the antique time ship from Phanagoria. In: InternationalArchives of the Photogrammetry, Remote Sensing and Spatial Information Sciences.Vol. XL-5/W2. pp. 717–772.

Fig. 7. Models of a timber from the World Trade Center shipremains. The photogrammetric mesh (left) lacks the detail of thelaser scanned mesh (middle), and the final PTM model (right)combines the strengths of each method.

C. Dostal, K. Yamafune Journal of Archaeological Science: Reports 18 (2018) 430–436

436