Efficiently capturing large, complex cultural heritage sites with a handheld mobile 3D laser mapping...

E�ciently Capturing Large, Complex Cultural Heritage

Sites with a Handheld Mobile 3D Laser Mapping System

Robert Zlot1, Michael Bosse1, Kelly Greenop2,Zbigniew Jarzab2,1, Emily Juckes2, and Jonathan Roberts1

1Autonomous Systems, CSIRO Computational Informatics, Brisbane, Australia2School of Architecture, The University of Queensland, Brisbane, Australia

Abstract

Accurate three-dimensional representations of cultural heritage sites are highly valuable

for scientific study, conservation, and educational purposes. In addition to their use for

archival purposes, 3D models enable e�cient and precise measurement of relevant natural

and architectural features. Many cultural heritage sites are large and complex, consisting

of multiple structures spatially distributed over tens of thousands of square metres. The

process of e↵ectively digitising such geometrically complex locations requires measurements

to be acquired from a variety of viewpoints. While several technologies exist for capturing

the 3D structure of objects and environments, none are ideally suited to complex, large-

scale sites, mainly due to their limited coverage or acquisition e�ciency. We explore the

use of a recently developed handheld mobile mapping system called Zebedee in cultural

heritage applications. The Zebedee system is capable of e�ciently mapping an environ-

ment in three dimensions by continually acquiring data as an operator holding the device

traverses through the site. The system was deployed at the former Peel Island Lazaret,

a culturally significant site in Queensland, Australia, consisting of dozens of buildings of

various sizes spread across an area of approximately 400⇥250m. With the Zebedee sys-

tem, the site was scanned in half a day, and a detailed 3D point cloud model (with over

520 million points) was generated from the 3.6 hours of acquired data in 2.6 hours. We

present results demonstrating that Zebedee was able to accurately capture both site con-

text and building detail comparable in accuracy to manual measurement techniques, and

at a greatly increased level of e�ciency and scope. The scan allowed us to record derelict

buildings that previously could not be measured because of the scale and complexity of

the site. The resulting 3D model captures both interior and exterior features of buildings,

including structure, materials, and the contents of rooms.

Keywords: 3D digitisation; laser scanning; mobile mapping; heritage sites; documentation; large-

scale scenes; Peel Island

1 Research Aims

Cultural heritage sites exhibit a wide diversity of physical complexity and configuration, rangingfrom sparse, open areas containing multiple structures or ruins, to compact and dense buildings,monuments, or natural areas. Given this variability, technologies for 3D digitisation of large-scale sites must be versatile, reliable, and e�cient.

This study investigates the application of a recently developed handheld mobile mappingsystem called Zebedee to cultural heritage sites. Zebedee can generate 3D point cloud maps

1

Journal of Cultural Heritage, DOI: 10.1016/j.culher.2013.11.009 2

(a) Standard hardware (b) System with camera (c) Mapping on Peel Island

Figure 1: (a) Components of the Zebedee handheld 3D mapping system include the handheld sensing device(foreground), acquisition laptop, and battery (back left). (b) A modified Zebedee system containing a GoProcamera mounted in front of the handle. (c) The Zebedee system in use on Peel Island.

from laser range data collected as the operator walks around a site. A primary research aim isto explore the suitability of the system for complex, large-scale cultural heritage sites in termsof accuracy, e�ciency, scalability, ease of use, and budget considerations. We also introduce anenhancement of the system in which a video camera is added to colourise the point cloud, whichis expected to make the data more informative for cultural heritage applications.

2 Introduction

Technologies and methodologies for digitisation of cultural heritage sites are wide-ranging, frommanual measurements to high-resolution laser scanning and computer vision techniques. Eachapproach has distinct characteristics appropriate for particular scales and complexities of theobjects or environments captured [1, 2]. A significant number of cultural heritage sites consistof multiple structures spatially distributed over large areas (thousands of square metres). Theprocess of e↵ectively digitising such geometrically complex locations in detail requires measure-ments to be acquired from a variety of viewpoints.

Currently, the most e�cient way to cover a large site is from above, using airborne scanningand imaging (e.g., [3]) or satellite-based sensing. A 2011 report by English Heritage [1] comparesa number of technologies, and suggests that these are the only methods capable of surveyinglarge-scale complex areas. Sensing from above, however, cannot capture all of the detail thatis available at ground level and generally only captures—at limited resolution—the tops ofstructures, treetops, and exposed ground.

Terrestrial laser scanning (TLS) can acquire millions of range measurements with high preci-sion from ground level, and has been employed at a number of cultural heritage sites [4, 5, 6, 7].These systems are typically mounted on a tripod and require several minutes to generate a 3Dscan from a static location. As a terrestrial scanner can only measure surfaces visible from itscurrent position, shadows occur in the data due to occlusions. Therefore, in order to achievereasonable coverage of a complex site, a scanner must be placed at multiple (typically carefullyselected) stations. Multiple scans can be aligned using software post-acquisition; however, thisstep requires su�cient overlap between the scans, and often (depending on the software used),accurately surveying the tripod location or placing reflective targets in the environment.

Despite the high data-capture rates, when considering large-scale complex heritage sites,TLS is still relatively ine�cient and expensive in terms of equipment cost, time in the field,and data post-processing e↵ort. As a result, relatively few examples appear in the literature


where TLS is used for digitising large sites, though such campaigns are becoming increasinglymore common [5, 6, 7]. One of the most impressive recent examples is the scanning of Lalibela,Ethiopia, in which a 700⇥400m site containing rock-hewn churches was digitised using 1150terrestrial scans [7]. The authors report that data acquisition required six weeks in the field,which was followed by four months of e↵ort to register the laser scans into a common coordinateframe.

Mobile mapping technology positions the scanning equipment on a moving platform duringdata acquisition. The platform’s motion ensures that the sensors continuously view the envi-ronment from a variety of viewpoints, thereby significantly reducing shadowing from occlusions.Most existing systems are mounted on vehicles [8] and require reliable GPS coverage; there-fore, they are not suitable for delicate, vehicle-inaccessible, restricted, indoor, underground, orconfined cultural heritage spaces [1].

Members of our team have developed a handheld mobile mapping system, called Zebedee,that generates 3D point cloud maps of an environment—indoors, outdoors, and underground—as the operator holding the device walks through it [9]. Large-scale and complex environmentsare mapped relatively quickly, requiring minutes or hours to capture sites that might require daysor weeks with terrestrial scanners. We demonstrate the system’s capabilities and advantages,applied to modelling a historic lazaret on Peel Island, near Brisbane, Australia, consisting ofdozens of small buildings spread across an area of approximately 400⇥250m. The site, which weestimate would have taken multiple days to scan with TLS, was scanned with Zebedee in underhalf a day, producing a 3D point cloud map consisting of over 520 million points. We have alsoenhanced the Zebedee system by adding a video camera, which allows us to generate true-colourmodels by fusing the image data with the laser data. The raw data is processed in less timethan the acquisition and automatically, with no human intervention required to produce thefinal point cloud model. In terms of cultural heritage practice, the point cloud models producedenable remote 3D visualisation of sites for cultural heritage interpretation and documentationthat have been previously too expensive and/or time consuming to generate.

We propose that the introduction of a lightweight handheld 3D mobile mapping system, andin particular, Zebedee, can deliver multiple benefits to cultural heritage applications stemmingfrom several key properties.

E�ciency. The reduction in both data acquisition and data processing time can translateinto significantly lower project costs. Notably, e�cient operation presents the capabilityto generate results in the field, thereby providing feedback that can guide further datacollection while still on site.

Accessibility. Zebedee can be used to scan areas where it may be di�cult or impossible to useother systems; for example, on rough or steep terrain. The Zebedee system can also bemore practical for accessing sites with limited availability.

Automation. Since data processing is fully automated, the complete workflow for acquiringdata and generating registered point clouds can be performed by users with minimaltraining.

Compatibility. The maps produced by Zebedee are standard point clouds that can be inte-grated with data collected from other devices, and post-processed with existing softwarepackages.


3 Materials and Methodology

Zebedee is a handheld mapping system that can generate a 3D model of a site that it is carriedthrough by a human operator [9] (Figure 1). Its primary exteroceptive sensor is a 2D laserscanner, which captures 43 200 measurements per second of ranges to visible surfaces in theenvironment based on the time-of-flight of an infrared laser pulse. In order to extend the laserscanner’s 270� two-dimensional field of view into a three-dimensional field of view, a flexiblespring connects the scanner to the device’s handle. The spring allows the scanner to pivotrelatively freely (sweeping up to 170� in amplitude and tuned to a desired frequency of approxi-mately 0.5Hz) as a result of the operator’s natural walking or arm motion. An industrial-gradeMEMS inertial measurement unit (IMU) is also mounted on the laser scanner side of the springto provide rough estimates of the scanner orientation. The IMU provides three-axis rotationalrate, linear acceleration, and magnetometer readings with a 100Hz update rate. The systemalso includes a small laptop computer for recording the data, and a lithium-ion battery whichcan power the system continuously for several hours. The handheld device has a mass of ap-proximately 650 g, and the entire system can be transported in a small pack with a total massof 3.8 kg (including device, laptop, cables, battery, and the bag itself).

The Zebedee system uses a continuous data registration procedure to convert the raw rangemeasurements into a consistent 3D point cloud. In order to do so, an accurate measurementof the scanner trajectory (i.e., its position and orientation at all times) must be available. Thesystem does not rely on any external positioning systems (such as GPS)—the device is self-localising, in that specialised software computes the scanner trajectory based on the sensormeasurements (laser and IMU). The solution concept is largely based on the observation thatas the scanner is moved through the environment, static surfaces in the local view appear tomove. By continually tracking this apparent ‘motion’, the scanner trajectory can be inferred. Ingeneral, estimating both a sensor platform’s motion and a map based on the sensors’ measure-ments of the environment is a well-studied problem in the field of Robotics called SimultaneousLocalisation and Mapping (SLAM) [10]. The solution developed for processing Zebedee data isunique in that it must accurately correct for significant distortions initially present in the rawmeasurements that arise from the continuous, highly non-deterministic motion of the sensorsduring acquisition.

Over the long-term, small errors in the trajectory estimate can accumulate, resulting ininaccuracies or misalignments in the map data. A global registration algorithm can eliminatethese errors, applying corrections to the trajectory by identifying and aligning common surfacesthat were measured at di↵erent times. Therefore, to maximise the ability of this optimisationstep to improve the trajectory, it is recommended to include loops in the operator’s traverse toensure that a su�cient amount of overlap exists.1 Over very long timescales where the drifterror may reach several tens of metres, an intermediate place recognition step is used to identifyregions that were mapped more than once in the dataset and apply coarse corrections to thetrajectory to seed the global registration algorithm [11]. The place recognition algorithm is alsoused to automatically match and align multiple datasets of a site that are collected at di↵erenttimes or simultaneously by multiple operators, provided there is su�cient overlap between thedatasets.

The accuracy of the computed trajectory depends on the amount of structure in the environ-ment and the distribution of surface normal directions in each view: if there are some directionswhich are not spanned by surface normals (e.g., in a very long, smooth tunnel, or an open field),then there is not a unique local trajectory solution and the result may be unreliable. While

1The concept is similar to the idea of closed or compound traverses in traditional surveying, though here theloop closures are detected automatically and supported by many measurements on the scanned surfaces ratherthan by returning to a control point.


it is challenging to obtain an independent, accurate ground truth for the trajectory, in limitedindoor experiments the system has been demonstrated to have a positional root mean squareerror well under a centimetre [9].

Though there are many valid ways to scan a location using Zebedee, to make data acquisitionat a large site more manageable, coverage is achieved by collecting multiple datasets around thesite—typically ten to forty minutes per dataset—rather than a single, large dataset. Eachdataset tends to focus in detail on a region of the site, or it may be a quick traverse arounda large section of the site with sparser detail captured. The processing software automaticallycalculates the relative alignment of the datasets and optimises the joint trajectory to produce aglobally consistent map of all the scanned areas.

3.1 Point Cloud Colourisation

By default, the laser scanners only record geometric information and the point clouds are oftenfalse-coloured for visualisation. However, with the addition of a video camera, it is possible toassign a realistic RGB value for each 3D point. A prototype system was developed in which aGoPro camera was mounted rigidly to the Zebedee handle (Figure 1b). Colours are determinedby projecting each 3D point into a subset of the video frames around the timestamp of the pointand sampling the median red, green, and blue colour values at the valid pixel locations. Eventhough the camera is not rigidly attached to the laser scanner, it is possible to compute itstrajectory using an additional IMU attached to the camera. The yaw drift of the camera’s IMUwith respect to the scanner’s IMU can be filtered out by matching the low frequency rotationalmotion components, and the relative position of the camera with respect to the scanner iscomputed by modelling the spring as a universal hinge joint.

3.2 Mobile Scanning

It is worth noting that there have been other reported examples of mobile mapping systems ap-plied to cultural heritage applications. Ikeuchi et al. [12] report on a campaign at Bayon Templein Cambodia which included laser scanners climbing along rails or tethered to balloons, amongother technologies. While their system was able to successfully digitise a very complex temple, adisadvantage of such a design is that it requires infrastructure to be placed around the buildingwhich likely involves significant setup time and cost, and may pose risk of damage to fragilesites. Bok et al. [13] describe a wearable system consisting of two laser scanners and four colourcameras. Their approach to motion estimation is based on a computer vision technique calledStructure From Motion, primarily using the camera data to estimate the system’s trajectory.Their system has been applied to digitise large-scale heritage sites, including a 600⇥350m his-toric village in South Korea traversed following a 3.1 km trajectory. Loop closures in their dataare not identified automatically: instead, these are determined either manually, or using GPSwhere available. Another key di↵erence between Bok et al.’s system and Zebedee is that Zebedeedata can be processed relatively quickly (even while in the field), while Bok et al.’s system isreported to require about a day of processing for each hour of data acquired (or slightly less atreduced resolution) [13].

4 Experiments

4.1 Peel Island

Peel Island is located within Moreton Bay, four kilometres o↵ the coast of Brisbane, in Queens-land, Australia. Peel Island has importance in both Aboriginal and colonial cultural heritage [15]


Figure 2: Site map of Peel Island Lazaret residential, medical and communal buildings from 1940–1959. Diagrambased on Riddel’s [14].

(a) 1907 (b) 2012

Figure 3: (a) Photograph of huts and kitchen buildings on Peel Island circa 1907 (courtesy John Oxley Library,State Library of Queensland. Neg: 36680). (b) Photograph of one of the dilapidated lazaret buildings, 2012.

.

and was the site of Australia’s only multi-racial lazaret, or leper colony (1907–1959) [16]. Thelazaret site is large and open, approximately 400⇥250m (Figure 2) and comprised of dozensof small accommodation huts of di↵erent types designed for individuals and groups to inhabit


Table 1: Descriptions of the thirteen datasets collected on Peel Island. Processing times provided are froma mid-2012 MacBook Pro with a 2.6GHz Intel Core i7 processor running a MATLAB/MEX implementationof the processing software. The ‘Total’ processing time includes individual dataset processing times plus thetime required for aligning all datasets using the place recognition software and optimising the combined result.Datasets for which colour video is available are indicated with an asterisk.

Dataset Data Processing Trajectory Number

description duration time length of points

[(h:)mm:ss] [(h:)mm:ss] [m] [millions]Sparse overview loop 24:15 14:14 1039 34Male compound 1⇤ 17:26 11:32 854 30Male compound 2 14:23 9:23 505 25Male huts (recreation area) 15:38 9:23 413 24Male hut⇤ 1:55 1:30 77 3Coloured compound⇤ 39:45 25:09 1998 64Female and male compounds⇤ 29:52 18:37 1215 55Female hut and surrounds 9:45 6:00 289 16Recreation hall 9:53 6:39 235 17Path to doctor’s residence 10:55 6:54 376 15Doctor’s residence 19:22 11:40 522 32Nurses’ residence 15:41 8:52 811 21Superintendent’s quarters 8:26 4:57 178 15Total 3:37:17 2:35:52 8511 520

(Figure 3). Other buildings include sta↵ quarters, medical treatment rooms, and recreationfacilities [14, 15, 17].

The distances between each building were designed according to perceived correct spacingfor the hygienic containment of disease through the circulation of ‘fresh air’ [18]. The heritagesignificance of the Peel Island Lazaret site lies not only in the unique architecture, but alsoin the spatial segregation of ‘white’ and so-called ‘coloured’ patients, and within those zonesinto gender-specific areas [17, 18, 14]. This history, reflected in the lazaret site is recognised asbeing of state-level heritage significance and is protected by cultural heritage legislation [19].The site’s buildings are in various states of repair ranging from dangerously dilapidated to fullyrestored. The fragile material qualities of timber and tin predispose the structures to relativelyrapid deterioration, creating an aesthetic quality of decay valued in the heritage citation. Theorganic nature of this decay creates complex physical attributes that demand a significant levelof skill and time to document. Existing scanning technologies have proved impractical or tooexpensive for use at the remote and complex site. To date, drawings of a few buildings havebeen made using the traditional system of taking manual site measurements [15]. Documents arequickly made obsolete by the rate of deterioration in the structures, requiring regular updatingof measurements if records are to be of any real value. Peel Island is accessible only by boat,weather permitting, and lacks a landing jetty, requiring a wet-landing from a dinghy. A permitfrom the Aboriginal traditional owners is required prior to visiting the island beyond the beach.These factors of a large site with many small complex buildings, in a remote and culturallysensitive location, provide an ideal location for testing the capabilities of Zebedee.

4.2 Data Acquisition and Processing

Two Zebedee systems were available for scanning, one of which included a camera for colourisingthe captured model. Over approximately four hours, operators covered various sections of


(a) Trajectories (b) Scan coverage

Figure 4: (a) Overhead view of the trajectories followed during data collection. Each dataset is rendered in adi↵erent colour. Small loops in the trajectory typically indicate the operator circumnavigating around a buildingor other significant structure (in some examples, paths into the buildings can also be seen within the smallloops). Overlap between the datasets is generally apparent as places where trajectories of di↵erent colours crossor approach one another. (b) Overhead view of the site with the coverage area of Zebedee scan data indicatedby the shaded area. (Aerial imagery from nearmap.com.)

the lazaret simultaneously, collecting thirteen overlapping datasets for a combined 3.6 hours ofdata and 8.5 km path length. Due to the sparseness of some areas in the lazaret, some carewas needed around open areas to maximise the likelihood that the trajectory could always beaccurately estimated by the software. In particular, the operators attempted to maintain pathsthat kept within about ten metres of the nearest structures (e.g., buildings, vegetation) and,where possible, ensured that a good distribution of surface normals was visible to the scanner.Although it was important to keep these considerations in mind, it was fairly straightforwardto reactively establish trajectories to satisfy these conditions.

Each dataset was first processed individually—the majority of them in the field—to producea trajectory estimate and point cloud model of the scanned area. The individual models werethen aligned and merged together to generate an overall 3D model of the site consisting of 520million points. Datasets for which video is available were colourised following the approachdescribed in Section 3.1.

4.3 Results

Descriptions and key statistics of the thirteen datasets collected are presented in Table 1. Dataprocessing times for all the datasets (with the exception of one very short one) are between 56%to 67% of the amount of time required to collect the data, which suggests the possibility thatthe data could be processed in real-time during acquisition. Combining and registering all ofthe datasets in a common coordinate frame requires just over twenty minutes, and overall theprocessing time in going from raw data to a single coherent 3D model requires approximately72% of the acquisition time (2.6 hours to process 3.6 hours of data). The combined trajectoryfor the full data collection is illustrated in Figure 4a and the overall coverage of the site is


Figure 5: Point cloud of the Peel Island Lazaret generated from Zebedee data. A number of small and largebuildings are visible, while others are present but occluded in this view by trees.

shown in Figure 4b. The complexity of the paths taken is indicative of the complexity of theenvironment, and suggests that a considerable number of measurements from static locationswould have been required to capture the site with similar coverage. We estimate in excess of 150terrestrial scanner stations would be needed to even approach the amount of coverage achievedwith Zebedee; such an e↵ort would likely require two to three days to achieve.

The point cloud model produced by the Zebedee system provides a model of the site (Fig-ure 5) that allows for detailed examination of individual buildings including their internal andstructural properties (Figures 6 and 7). The density of the point cloud depends on the measure-ment rate of the scanner (which is a constant) and the operator’s walking pace. For the scannedstructures at Peel Island, the median point spacing is about 1 cm. The walls, floors, openings,and sub-floor structure were all able to be captured, as were the internal fittings and furniture,adding to the utility of the model for cultural heritage interpretation. In the case of di�cultto access buildings, the Zebedee operators were able to reach in through windows, doorways, orother openings to scan internal features of the small huts without damaging the deterioratingstructure.

Previous surveying of the site involved manual measurements of specific buildings, each of


Figure 6: Aerial view of rendered Zebedee point cloud, showing the Peel Island Lazaret men’s huts and recreationhall in the foreground. The point cloud seen here has been subsampled to 9% of the original data.

which required several days to measure and draw in CAD software packages. While the man-ually measured model is accurate at the few points where the measurements were taken, themethod assumes right angles and planar surfaces, which leads to an approximate convention-alised drawing. A key example that highlights the di↵erence between the point cloud data andthe manually measured data is the comparison of the bath house illustrated in Figure 8. Here,the CAD drawing does not reveal the mild deviations from the orthogonality assumptions thatare apparent in the point cloud. Building dimensions can be obtained from the bath housepoint cloud by fitting planes to the walls and measuring the mean distance of the points onthe opposing wall in the direction of the plane’s surface normal. The resulting building lengthsare less than 11mm from the hand-measured quantities, which is within the tolerance expectedgiven the wall deformations not accounted for in the hand measurements. In our experience,and based on previous experiments [9], relative measurement errors for buildings of similar sizesto those in the lazaret (3–30m) are typically on the order of one centimetre.

Overall accuracy of the Zebedee point cloud at the site level is illustrated in Figure 9, wherethe Zebedee data is overlaid on aerial imagery. In this comparison, we observe that the locationsand dimensions of the buildings roughly match between the maps, though some drift error thathas built up in the Zebedee trajectory is apparent in terms of slight misalignments in the relativelocations of buildings. These errors are typically on the order of tens of centimetres to a metrebetween structures that are hundreds of metres apart.

Some of the Peel Island data is available as colourised points (Figures 7 and 10), which allowfor more meaningful interpretation of data and clarification of information that might otherwisebe ambiguous. A key example is shown in Figure 10 which shows a photograph of the bathhouse together with both greyscale and colourised point clouds. Viewing the scan in colour(Figure 10b) assists with proper interpretation of the point cloud; for example, identifying thehorizontal line on the verandah as red-and-white-striped ‘caution’ tape rather than a handrail.Given the dynamic laser-camera calibration problem inherent to the device configuration, thereare often small misalignments between the individual colour images and the local point cloud;


Figure 7: Colourised perspective view of the point cloud model showing details in the callouts. Top left: hospitalbuilding; Bottom left: Typical male hut; Right: Superintendent’s quarters. The point cloud seen here has beensubsampled to 9% of the original data.

however, the technique of using the median colour value from multiple frames results in a morereliable colour in the final result.

5 Discussion

The introduction of mobile mapping technology such as Zebedee to cultural heritage applicationshas the potential to change the way in which large-scale site documentation is practised. Com-pared to existing technology, Zebedee enables sites to be scanned faster, data processing to beperformed quickly and automatically, and otherwise inaccessible environments to be accuratelymapped. The example of Peel Island demonstrates that a large site can be mapped in hours,rather than days as would be expected with TLS. The ability to generate registered maps in thefield provides immediate feedback to the operators on which areas have been adequately cov-ered, and which require further scanning. In comparison, large-scale TLS scanning projects havereported spending months of e↵ort for scan registration [7, 6], and therefore obtaining similarfeedback in the field would be impossible. Mobility also allows the operators to think di↵erentlyabout how to capture a site, as there is no need to set up a tripod or strategise about scanlocations to ensure su�cient overlap between scans, maximise coverage, and minimise shadowsfrom occlusions. Because a site can be mapped quickly with Zebedee, there is often only minoradditional e↵ort to scan areas on site that are of lower priority (but still of importance), and mayhave been omitted from the campaign had other equipment been used. Usage of the Zebedee

system does not require extensive training or expertise, and in fact one of the three operatorsscanning Peel Island was using the system for the first time.

While the majority of the Peel Island Lazaret site could have been accessed by TLS, thereare several dilapidated structures that would have been too delicate to enter or place tripodequipment within. The interiors of these structures were easily scanned with Zebedee by reachingthe device through windows and doorways. The ability to dynamically raise and lower Zebedeeo↵ers another advantage over TLS scanners, which are locally limited to a small vertical range,often making it impossible to see above and below features and furniture. Other types of


Figure 8: Comparison of plans. Left: Plan extracted from the Zebedee point cloud. Centre: Overlay of point cloudand hand-measured drawings showing deformation of walls revealed by the point cloud data is not captured inthe hand-measured drawings. Right: CAD drawing based on hand measurements (Janeke Wissekerke, ZbigniewJarzab, and Garth Hsu, University of Queensland).

(a) Comparison with aerial imagery (b) Closeup of central region

Figure 9: (a) Comparison of Zebedee data with aerial imagery. The black edges represent the outlines of structuresin the Zebedee point cloud which are overlaid on aerial imagery. (b) Close-up of a portion of the site includingparts of the male compound, recreation, and administration zones. (Aerial imagery from nearmap.com.)

environments encountered in cultural heritage applications have properties that would renderthem impossible or impractical for existing technology, but straightforward with Zebedee. Forexample, Zebedee has been used to scan the interior of the Tower of Pisa, much of which consistsof a 35m-high spiral stairway encased in a tunnel. Previous TLS surveys have only capturedthe exterior of the structure, as scanning the stairway would have been a tedious exercise giventhe limited visibility at any one location (and would require closing o↵ the site to visitors forextensive periods). However, with Zebedee we were able to capture the stairway in under tenminutes. This interior data has been merged with existing TLS data covering the tower exterior.

While the Zebedee system was able to e�ciently capture the Peel Island site, there are sometradeo↵s and limitations with respect to existing systems. The precision of the measurements inthe point clouds produced with Zebedee is governed by the range precision of the Hokuyo scanner,


(a) Greyscale point cloud (2012) (b) Colourised point cloud (2012) (c) Photograph (2011)

Figure 10: (a) Greyscale image of the bath house point cloud, with intensities coloured using the ambientocclusion method. (b) Point cloud of the bathhouse colourised using the video data. (c) Photograph of thebathhouse. (Note that the photograph is from the previous year and the red and white caution tape is notpresent.) The point clouds seen here have been subsampled to 9% of the original data.

which is in the order of 2–3 cm. High-end terrestrial laser scanners typically have millimetre-scalenoise, and as a result they are often better able to capture fine details. Although Zebedee pointclouds are noisier (a comparison of Zebedee and TLS point cloud data has been presented in arecent study [20]), it is still possible to smooth or convert them to surface meshes. Environmentsthat lack a good distribution of 3D structure, such as open fields or long, smooth tunnels, can bechallenging areas for Zebedee as there is no information to constrain the trajectory estimate insome directions when registering the data. Additional sensors can be integrated into the systemto improve performance in such circumstances, and we are currently investigating several suchpossibilities. The Zebedee system is also limited in terms of scanning tall objects or buildingsdue to the maximum range of the scanner, typically 15–20m outdoors (although this limitationwas not of particular concern at Peel Island). Most of the upper surfaces of building rooftopsat Peel Island were not captured with Zebedee as they are not visible at ground level (TLS hasa similar limitation). In some cases on other sites we have been able to use stepladders to scanthe tops of structures.

Precision and range limitations could also be overcome by integrating Zebedee point cloudswith TLS, photogrammetry, or triangulation-based scanner data. In such a scenario, Zebedeewould primarily be used to capture the overall structure of a site, while one of the more tradi-tional acquisition systems could be employed in areas of interest where high-resolution detail isrequired. Given su�cient overlap, the Zebedee place recognition and registration software canautomatically combine the data from these various sources. It is also expected that small formfactor laser scanners will improve over time, and therefore the range and precision of the Zebedeesystem will likewise improve.

We see three key cultural heritage practice applications for the digital model of the PeelIsland Lazaret. These applications have broad applicability across many types of cultural her-itage sites that share similar characteristics of remoteness, complexity, and limited resources formaintenance, interpretation, and visualisation.

Conservation monitoring and documentation. The e�ciency and ease of Zebedee’s oper-ation can provide a cost-e↵ective regular monitoring of the site, allowing regular reacqui-sition of data for comparison, without a highly skilled operator. The data can be used forremote analysis and generation of construction documentation for conservation work.

Continued research. Documentation has previously been limited to photographic recording


in conjunction with site plan drawings, mainly a consequence of the resource requirementsof documenting the site in detail. The point cloud data is an accurate record of the form,scale, and colour of the structures and site that can be analysed remotely by practitionersproposing maintenance, restoration, or new works.

Virtual tourism and education. The remoteness and lack of marine infrastructure severelylimits the site’s accessibility for heritage tourism and education at any viable scale. Thepotential to enhance the point cloud with additional information such as historical pho-tographs, texts, and audio-visuals to create a digital interactive tool for remote site inter-pretation could open up the site to the public.

Zebedee provides a key step forward towards more accessible technology for 3D data acqui-sition. While every site is unique, the e↵ectiveness of Zebedee at a large and complex site thatwould otherwise be very challenging and time consuming to capture demonstrates its potentialas an additional tool for cultural heritage practitioners to consider. In addition to the lazaretat Peel Island, we have recently deployed the technology to capture a nineteenth century fort,a former penal settlement, historical European buildings, a WWI tank, and archaeologicallysignificant caves.

Acknowledgements

Many thanks to the following individuals and organisations: The Quandamooka YoolooburrabeeAboriginal Corporation representing the Traditional Owners of Turkrooar (Peel Island); PaulFlick of CSIRO for Zebedee hardware design and support; Queensland Parks and Wildlife Ser-vice, in particular Roland Dowling, for facilitating access to and providing support and infor-mation about Peel Island.

References

[1] D. Barber, J. Mills, and D. Andrews, “3D laser scanning for cultural heritage (second edition): Advice andguidance to users on laser scanning in archaeology and architecture,” tech. rep., English Heritage, Swindon,England, 2011.

[2] G. Pavlidis, A. Koutsoudis, F. Arnaoutoglou, V. Tsioukas, and C. Chamzas, “Methods for 3D digitizationof Cultural Heritage,” Journal of Cultural Heritage, vol. 8, no. 1, pp. 93–98, 2007.

[3] R. Lasaponaraa and N. Masinib, “Full-waveform airborne laser scanning for the detection of medievalarchaeological microtopographic relief,” Journal of Cultural Heritage, vol. 10, pp. e78–e82, December 2009.Supplement 1.

[4] N. Yastikli, “Documentation of cultural heritage using digital photogrammetry and laser scanning,” Journalof Cultural Heritage, vol. 8, pp. 423–427, 2007.

[5] S. Reardon, “A digital ark, come fire or flood,” New Scientist, vol. 216, pp. 22–23, November 2012.

[6] G. Guidi, F. Remondino, M. Russo, F. Menna, A. Rizzi, and S. Ercoli, “A multi-resolution methodologyfor the 3D modeling of large and complex archeological areas,” International Journal of ArchitecturalComputing, vol. 7, no. 1, pp. 39–55, 2009.

[7] H. Ruther and G. Palumbo, “3D laser scanning for site monitoring and conservation in Lalibela WorldHeritage Site, Ethiopia,” International Journal of Heritage in the Digital Era, vol. 1, pp. 217–232, June2012.

[8] G. Petrie, “Mobile mapping systems: An introduction to the technology,” GeoInformatics, vol. 13, pp. 32–43, January/February 2010.

[9] M. Bosse, R. Zlot, and P. Flick, “Zebedee: Design of a spring-mounted 3D range sensor with application tomobile mapping,” IEEE Transactions on Robotics, vol. 28, pp. 1104–1119, October 2012.

[10] H. Durrant-Whyte and T. Bailey, “Simultaneous localization and mapping (SLAM): Part I the essentialalgorithms,” IEEE Robotics & Automation Magazine, vol. 13, no. 2, pp. 99–110, 2006.


[11] M. Bosse and R. Zlot, “Place recognition using keypoint voting in large 3D lidar datasets,” in IEEE Inter-national Conference on Robotics and Automation, 2013.

[12] K. Ikeuchi, T. Oishi, and J. Takamatsu, “Digital Bayon Temple – e-monumentalization of large-scalecultural-heritage objects–,” in Proceedings of ASIAGRAPH, vol. 1, pp. 99–106, 2007.

[13] Y. Bok, Y. Jeong, D.-G. Choi, and I. S. Kweon, “Capturing village-level heritages with a hand-held camera-laser fusion sensor,” International Journal of Computer Vision, vol. 94, pp. 36–53, 2011.

[14] T. Blake, “The leper shall dwell alone: Peel Island Lazaret conservation plan: a report for the Departmentof Environment and Heritage,” tech. rep., Robert Riddel Architect, Brisbane, Qld, 1993.

[15] E. Juckes, K. Greenop, and Z. Jarzab, “Isolation and segregation: An intercultural analysis of the PeelIsland Lazaret,” in Proceedings of the Society of Architectural Historians, Australia and New Zealand: 30,Open (A. Brown and A. Leech, eds.), vol. 2, (Gold Coast), pp. 529–542, Society of Architectural Historians,Australia and New Zealand, SAHANZ, July 2013.

[16] J. Prangnell, “The Archaeology of the Peel Island Lazaret: Part I: Survey,” Recent Archaeological Researchin Southeast Queensland, vol. 13, pp. 31–38, 2002.

[17] A. Youngberry and J. Prangnell, “Fences, boats and teas: Engendering patient lives at peel island lazaret,”International Journal of Historical Archaeology, vol. 17, no. 3, pp. 445–464, 2013.

[18] M. Parsons, “‘Defining disease, segregating race’: Sir Raphael Cilento, Aboriginal health and leprosy man-agement in twentieth century Queensland’,” Aboriginal History, vol. 34, pp. 85–114, 2010.

[19] Queensland Government, “Queensland Heritage Register, ‘Peel Island’, Place ID 601091,” March 2013.

[20] M. R. James and J. N. Quinton, “Ultra-rapid topographic surveying for complex environments: the hand-held mobile laser scanner (HMLS,” Earth Surface Processes and Landforms, 2013.

Efficiently capturing large, complex cultural heritage sites with a handheld mobile 3D laser mapping...

Documents

Transcript of Efficiently capturing large, complex cultural heritage sites with a handheld mobile 3D laser mapping...