Using digital photogrammetry for pipe-works progress tracking

Using digital photogrammetry for pipe-worksprogress tracking1

Mahmoud Ahmed, C.T. Haas, and Ralph Haas

Abstract: Pipe-works are among the most complicated items to be tracked in the course of monitoring construction projectprogress. Traditionally, the tracking of pipe-works progress is conducted either manually or using laser scanning technology.While laser scanning is a 3D imaging technique, and commercial software exists to construct 3D CAD models of pipingbased on such images, it suffers from portability, purchase cost, and other disadvantages. This paper describes digital photo-grammetry technology as an alternative for pipe-works reconstruction and as a cost effective tracking tool. For validation,data was collected using a handheld digital camera to acquire images inside a new building under construction. Progress ofthe pipe-work networks of different types and sizes in the new building was monitored during the construction phases. Inaddition to the known accuracy and robustness of photogrammetry, it was found that the use of digital photogrammetry pro-vided a practical and low-cost approach.

Key words: construction, buildings, construction management.

Résumé : Les ouvrages impliquant des tuyaux sont parmi les plus compliqués à suivre pour déterminer le progrès des pro-jets de construction. De manière conventionnelle, le suivi du progrès des ouvrages impliquant des tuyaux est réalisé soit ma-nuellement ou soit par la technologie de balayage laser. Bien que le balayage laser soit une technologie d’imagerietridimensionnelle et qu’il existe un logiciel disponible sur le marché pour bâtir des modèles CAD tridimensionnels destuyaux basés sur de telles images, le système de balayage est peu portatif, le coût d’achat est désavantageux et il présenteaussi d’autres inconvénients. Le présent article décrit la technologie de photogrammétrie numérique utilisée comme alterna-tive aux ouvrages de reconstruction des tuyaux et comme outil de suivi rentable. Aux fins de validation, des données ontété recueillies en utilisant une caméra numérique à main pour acquérir des images dans un nouvel immeuble en construc-tion. Le progrès des réseaux de tuyaux de divers types et dimensions dans le nouvel immeuble a été suivi durant les phasesde construction. En plus de la précision et de la robustesse reconnues de la photogrammétrie, il a été trouvé que la photo-grammétrie numérique fournissait une approche pratique et à faible coût.

Mots‐clés : construction, immeubles, gestion.

[Traduit par la Rédaction]

IntroductionSeveral indoor localization technologies, in addition to

manual technology, are in use as tools for counting and mon-itoring progress of construction components and installationsof utility networks such as pipe-works. These technologiesinclude: (1) radio frequency identification (RFID) systemsincluding active and passive RFID tags and readers, (2) net-work-based systems such as wireless LANs, Bluetooth andultra-wideband, (3) infrared systems, (4) ultrasound systemsincluding transmitters and receivers, (5) inertial navigationsystems (INS) including accelerometer, gyroscope and mo-tion-sensing devices, and (6) hybrid systems of the abovetechnologies. A comparison among these technologies was re-ported in Ahmed and Hegazy (2008). Other researchers inves-tigated RFID tags for project progress tracking (Song et al.2004). The most common and practical technology for indoorpipe-works tracking is laser scanning due to its advantage of

3D imaging (Bosche and Haas 2008). However, the output oflaser scanning is mainly a sampled point-cloud, which lacksimaging data of texture, color components, and shades.In this research, the use of photogrammetry is reported as

a robust, economic, and accurate 3D measurement technol-ogy based on digital images. A low-cost handheld camerawas used to collect pipe-works progress data during severalconstruction phases. The data acquisition approach was prac-tical. No special arrangements were necessary, and no inter-ference with machines or workers occurred, because takingimages is a simple and very quick operation.Free movement with the handheld camera allows imaging

of the same pipe-work installation activity from different an-gles to avoid any occlusion. It provides a lot of visual dataredundancy which is useful for producing high quality out-puts. The image records provide visual information about thecomplex pipe networks and facilitate decisions which need

Received 15 October 2011. Revision accepted 20 April 2012. Published at www.nrcresearchpress.com/cjce on 17 August 2012.

M. Ahmed, C.T. Haas, and R. Haas. Department of Civil Engineering, University of Waterloo, 200 University Avenue West, Waterloo,ON N2L 3G1, Canada.

Corresponding author: Mahmoud Ahmed (e-mail: [email protected]).1This paper is one of a selection of papers in this Special Issue on Construction Engineering and Management.

1062

Can. J. Civ. Eng. 39: 1062–1071 (2012) doi:10.1139/L2012-055 Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

qualitative data. For accurate quantitative results, photogram-metric data processing of the images resulted in reconstruc-tion of two different and important types of 3D CADmodels. This is explained in the next sections.

Motivation for the researchPipe-works represent a particular challenge for photogram-

metry, because large size pipes are featureless; others mayhave highly reflective surfaces that reflect light in random di-rections. As well, pipe-works designs can be changed duringthe construction phases several times, and much pipe-work isfield-run in practice. For this reason, recent techniques forobject recognition and automated progress tracking based ona priori knowledge such as Bosche and Haas (2008) havelimitations. In practice, frequent data collection will berequired to track pipe-works in the field and to produce thefinal as-built model. Even the most recent laser based 3Dimaging technologies present logistics challenges due to tem-perature sensitivity, weight and volume of a laser based 3Dimaging field kit, and the need to plan scans carefully.With the recent advances in photogrammetric technology,

it has become possible to assemble a low-cost system towork with off-the-shelf high resolution cameras and off-the-shelf software that handles camera calibration, interior andexterior orientation, bundle adjustment, image normalization,and epi-polar stereo matching. With dramatic advances incomputing speed, photogrammetric software has becomeable to generate a high density point-cloud automatically aswell (Jiang et al. 2008; Mikhail et al. 2001). Low-cost digitalphotogrammetry is therefore an attractive alternative for pipe-works tracking (Ahmed et al. 2011a, 2011b, 2011c).

Related researchThe literature contains several publications on automatic

project progress measurement using cameras. In early re-search, a web based system for automatic real time projectmonitoring was developed that linked time-lapse digital mov-ies of construction activities, critical path method schedules,and progress control (Abeid et al. 2003). Another interestingweb-based system for performance monitoring was developedto assist project managers through the use of the internet andcentralized database technology (Cheung et al. 2004). How-ever, these systems did not provide 3D quantitative measure-ments. An interactive system integrating 3D CAD drawingsand digital images was developed at Construction Technologyand Management Center (CTMC), University TechnologyMalaysia (UTM). While this system enables 3D visualiza-tion of as-built collected data, it offers only a connection toa photogrammetric engine for accurate reconstruction of as-built data (Memon et al. 2006a, 2006b). This emphasizesthe need for such an engine. Despite the availability of pho-togrammetry as a robust and accurate image based approachfor some years (Ahmed and Haas 2010), publications re-lated to its use for construction progress tracking are lim-ited. Photogrammetry, which has deep roots in history(Ahmed and Haas 2010), can be defined as the art, science,and technology of obtaining reliable information aboutphysical objects and the environment through the processof recording, measuring, and interpreting photographic im-ages and patterns of electromagnetic radiant energy and

other phenomena (Slama 1980). Close-range photogramme-try has been applied in diverse applications in industry, ar-chaeology, architecture, automotive, aerospace, forensic, caraccident reconstruction, biomechanics, chemistry, and biol-ogy (Jiang et al. 2005, 2008). Applications have also beenexplored by researchers in automated construction progresstracking (El-Omari and Moselhi 2008; Jiang et al. 2005),pavement distress surveying and crack measurement(Ahmed and Haas 2010), and pavement rutting and faults3D modeling (Ahmed et al. 2011a, 2011b). In most of theliterature, either expensive metric cameras were assumed tobe used, or specific camera configurations are required;moreover, specific technical data must be known in advancebefore any computations. For example, the camera calibra-tion data in terms of the coefficients of a polynomial to de-scribe lens distortion and three interior orientationparameters are assumed to be known to users. Additionally,a minimum number of control points are assumed to bemeasurable in two or more overlapping images, and the co-ordinates of control points are assumed to be measured ac-curately using traditional surveying techniques. Then,exterior orientation parameters are calculated to enable ro-bust reconstruction of a 3D CAD model out of the 2D im-ages. Because of such requirements, the exploitation ofphotogrammetric techniques in construction sites is notcommon. However, recent advances in computing process-ors, parallel processing, digital camera speed and resolutionand the availability of several off-the-shelf software pack-ages have led to feasible close-range photogrammetry(Ahmed and Haas 2010; Ahmed et al. 2011a, 2011b).

Background

Close-range photogrammetry is mainly based on a numberof rigorous mathematical and geometrical models and con-straints. The two main mathematical conditions used in mostphotogrammetric applications, including close-range, are thecoplanarity and collinearity condition equations (Ahmed andHaas 2010; Ahmed et al. 2011a, 2011b). Universally, allphotogrammetric technologies use a technique called bundleadjustment, of which the main mathematical condition is theCollinearity condition equation. Data processing deals mainlywith the mathematical transformation between an image pointin one 2D rectangular coordinate system (image space) andan object point in another 3D rectangular coordinate system(object space). Bundle adjustment allows the enforcement ofmany other mathematical and statistical constraints. Thisflexibility combined with robustness of the mathematicalconstraints provides a powerful technique for 3D model re-construction from 2D images (Kraus 1993, 2007; Slama1980; Mikhail et al. 2001). The collinearity condition as-sumes that the light ray is a straight line at the moment ofexposure; hence, the centre of the camera lens, the imagepoint, and the object point all must lie on one and the samestraight line (ray). This geometrical condition can be repre-sented mathematically as a collinearity of two vectors, onevector from lens centre to image point and one vector fromlens point to object point in real world space or collinearityof three points; object point, image point and centre of thelens. The mathematical form of this condition can be repre-sented for each image point as

Ahmed et al. 1063

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

½1�x� xo

y� yo

0� f

264

375 ¼ l

m11 m12 m13

m21 m22 m23

m31 m32 m33

264

375

X� Xc

Y� Yc

Z� Zc

264

375

The above equation is a seven transformation parametermodel that scales, translates and rotates one vector fromreal-world space to image space where x, y, 0 = image coor-dinates of any point; xo, yo, f = interior orientation parameters(coordinates of the principal point); m11,…m33 = elements ofrotation matrix; Xc, Yc, Zc = coordinates of exposure sta-tion; X, Y, Z = ground coordinates of imaged point; l =scale factor.The nonlinearity of the equations requires the use of a Tay-

lor series approximation and an iterative solution. For furtherdetails about the derivation of the mathematical model andthe formulation of iterative solution using least squares ad-justment technique, see for example Mikhail et al. (2001)and Slama (1980).Reconstruction of 3D models requires that each point sat-

isfy the coplanarity condition (epi-polar constraint). This con-dition forces any two conjugate rays on two overlappedimages coming from the same object point and the base-linebetween these two images to be coplanar in the analyticalsolution, see Fig. 1. This condition can be expressed byequating the scalar triple product of three vectors: base lineB, left vector pL, and right vector pR to zero, where two vec-tors are the conjugate rays and the third is the base line, seeeq. [2]. For further details about the derivation of the mathe-matical model and the formulation of iterative solution usingleast squares adjustment technique see Mikhail et al. (2001)and Slama (1980).

½2� Fi ¼ B � ðpL � pRÞ ¼ 0

Proposed photogrammetric approachGenerating point-clouds requires the computation of 3D

point coordinates of a large number of points on the surfaceof the investigated object using two or more images. Threedimensional coordinates can be computed from the intersec-tion of two vectors (rays) in space (Fig. 1 (top)), one fromthe left image and one from the right image. The automationof this process requires the automation of image matching, i.e.,for any point on one image, the system will automaticallyfind the coordinates of the conjugate point on the other im-age. The matching procedure must be done systematically,since the images have different positions and different angu-lar orientations in space; thus, it is required to know the di-rect relationship between any two conjugate points in anytwo images. This relationship is well known as the epi-polarrelationship as described briefly in next section.In Fig. 1; L1 and L2 are image planes, with C1 and C2 the

optical centrer of the first and second image. If the two im-ages are tilted; the line C1-C2 intersects the first image plane(or its extension in space) in a point e1, and intersects thesecond image plane (or its extension) in a point e2, (seeFig. 1 (bottom)). These two points are called the epi-poles ofthe stereo pair. Any plane containing the line C1-C2 is calledan epi-polar plane. This plane intersects the first image alonga line passing through the epi-pole, e1, and intersects the sec-

ond image along a line passing through the epi-pole, e2.These two intersection lines are called corresponding epi-polarlines and have special importance in stereovision. If the twoimages are truly vertical the epi-polar lines are parallellines. Converting the original image to an equivalent verti-cal one can be done through image normalization as de-tailed in Ahmed et al. (2011a).Normalized image generation is an important transforma-

tion procedure to enable automation for stereo image match-ing so that the detection of conjugate image points or featuresin two overlapping images becomes simpler. In essence, it isa systematic one dimensional search problem along oneknown epi-polar line as explained above. This procedurealso overcomes matching ambiguities in the image space. Im-age normalization can optionally be conducted after the stepof idealizing the images; i.e., after removing the eccentricityof camera axis so that the principal point coordinate valuesbecame (0,0). This removes the lens distortion effect, and

Fig. 1. Robust geometric constraints automatically satisfied at eachpoint during processing: images orientation and space intersection(top), and epi-polar geometry (epi-polar plane and lines) (bottom).

1064 Can. J. Civ. Eng. Vol. 39, 2012

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

corrects the shape of the pixels to squared in shape, i.e., hav-ing two equal dimensions with orthogonal angle betweenpixel sides. However, the image normalization can be con-ducted directly without image idealization using the same in-terior orientation parameters of the original image. To re-sample the idealized images requires generating new imageswith planes oriented parallel to the base line connecting thetwo perspective centres of the overlapped images. This makesthe epi-polar lines in both images as two groups of parallellines. Actually the two new images are generated to fall inone and the same plane parallel to the base line (Ahmed etal. 2011a). For further details see Kraus (1993, 2007); Choet al. (1992); and Schenk (1990a, 1990b). Fortunately, low-cost photogrammetry software, e.g., PMScannerTM, can gen-erate the idealized and the normalized images as well basedon the user preferences for data processing workflow. Theflowchart in Fig. 2 shows the procedures applied in this re-search and the optional functions available for automaticprocessing based on the user preferences.

Field study and image data acquisitionValidation of the approach described in this paper is based

on data obtained from the construction of the E6 (Engineer-ing 6) building on the UW campus from 2010 to 2011. Com-plicated networks of different types and sizes of pipe-workswere investigated using a low-cost consumer-grade camera,the Canon XSi 450D. To investigate the potential of the sys-tem, a large number of images were taken during differentphases of pipe-works installations. Their positions, orienta-tions and diameters were monitored and accurately recon-structed using two photogrammetric techniques described inthe following sections. Photogrammetric data acquisition re-quires an overlap to be maintained between any two consec-utive images so that one point-cloud can be generated usingthese two sequential images. Additional point-clouds can begenerated as a by-product between any third or fourth over-lapping images. The nominal shutter speed for a single imageis between 30/400 and 1/4000 s. In this research most of theimages were taken with shutter speeds between 1/60 and 1/120 s.

Fig. 2. Steps of data collection onsite and data processing at office including optional functions.

Ahmed et al. 1065

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

Practically, a single image may consume between five andten seconds to allow time for adjusting camera settings ifrequired plus 3 to 5 s more for moving between two shoots.In this research data acquisition time of a complete floorwas between 15 and 30 min. The camera calibration re-quired is 1 h, calibration can be conducted once per projector every few months. Establishing the layout plan to photo-graph takes 5 to 10 min. Point-cloud generation using twoimages ranges between 2 and 10 min.The camera calibration was conducted as a fundamental

step; see Table 1, a major challenge being calibration of anoff-the-shelf consumer-grade zoom lens due to continuousvariation of the focal length and other interior orientation el-ements, plus its geometrical instability. The lens was a CanonEF-S 18–55 mm 1:3.5–5.6 IS, nominal focal length: 18–55 mm, diagonal angle of view: 74°–27°, aperture: F3.5–5.6to F22–38 (at 18 and 55 mm). It is constructed of 12 opticalelements. Two calibration sessions were conducted to esti-mate the inner orientation parameters at the two nominalzoom-ends, i.e., 18 and 55 mm, using a technique describedin Ahmed et al. (2011a, 2011b).

Data processing

Two photogrammetric data processing techniques were in-vestigated. The first technique produced automatically adense point-cloud very-similar to the known output of laserscanners; see for example Fig. 3. The second technique uti-lized a very effective photogrammetric approach which hasthe capability to directly produce 3D CAD elements, espe-cially cylinders of monitored pipes, see Fig. 4 for example.This second approach overcomes the traditional need forpost-processing of scanner point-clouds, which is not a trivialor standardized task and may require special algorithms (Bo-sche and Haas 2008) to detect the geometric elements withinthe point-cloud. Both photogrammetric techniques use geo-

Table 1. Calibration of zoom lens at its two extreme ends and samples of coded targets (downleft), and calibration grid with four coded-targets (down right).

Calibration element Minimum zoom Maximum zoomFocal length 18.0131 52.6752Format size 22.2192×14.8336 22.2637×14.8336Resolution 4272×2848 4272×2848Principal point xo = 11.2369, yo = 7.4849 xo = 10.8724, yo = 7.3956p1, p2 –2.628e-005, 9.668e-005 6.625e-005, 2.478e-005k1, k2 5.915e-004, –1.186e-006 –2.702e-005, –2.781e-008

Fig. 3. Samples of zoom in to rendered point-clouds at variabledensities.

1066 Can. J. Civ. Eng. Vol. 39, 2012

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

metrical and mathematical constraints, the constraints includethe coplanarity condition and the collinearity condition thatwere explained in the background section.Computing the exterior orientation of overlapping images

is a standard step in the work-flow of generating 3D digitalsurface model (DSM) with a point-cloud or irregular triangu-lar network (TIN). This step can be conducted in an auto-matic or semi-automatic way. Also, it can be done as aglobal exterior orientation, so that the position and orienta-tion of all images are computed with reference to one andthe same arbitrarily selected coordinate system and originpoint, or alternatively for each group of two or more images.This last approach requires only eight common points to beidentified in overlapping images if the camera is un-calibrated(Ahmed 2007). The automatic identification of commonpoints in overlapping images can be done using coded-targetsas will be explained later in this section; alternatively, theuser can do it manually. The decision of using fully-automaticor semi-automatic depends on several factors, such as theapplication, the nature of the surface, the surrounding envi-ronment, the activities running in parallel in the same zoneduring shooting the images. The technology of coded-targets

has enabled the fully-automatic computation of exterior orien-tation elements. This technology is based on coding a numberof points, dots, circles, or geometric shapes, so that each onehas a unique code that can be identified automatically in allimages. Accordingly, the software, e.g., PMScannerTM, willbe able to read the coordinates of the conjugate targets in allimages (Ahmed et al. 2011a, 2011b).Despite the availability of coded-target technology; the

need for a semi-automatic approach is still important as analternative for such cases where the identification of sometargets does not meet the required or accepted accuracy level.Also, based on the application in hand, the user may not beable to maintain a frame of coded-points available in front ofthe camera(s) all the time. Without losing generality, fixingthe orientation of one image, and computing the orientationof other images as differences with respect to the fixed one,is known as relative orientation. Manual marking of eightconjugate points is enough to compute the relative orientationelements between any two un-calibrated images (Ahmed2007). In the following sections the experimental work willbe presented in detail.After calibration and orientation, several rendered 3D

point-clouds of pipe-works are generated to represent the sta-tus of the project at each date in which the images weretaken. Figures 5 and 6 show samples of the produced models,Fig. 7 shows a sample of two layers of generated pipes at dif-ferent dates.Generating point-clouds is almost a fully automatic process

in modern digital photogrammetry work flow. After generat-ing the point-clouds of several stereo-pairs and registering all

Fig. 5. Generating large model by registering several stereo modelstogether: images and part of compiled model (top), relationship be-tween registered stereo models (bottom).

Fig. 4. Generating CAD pipe entities: on top of point-cloud layer(top), as a raw CAD format (middle), as shaded entities (bottom).

Ahmed et al. 1067

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

point-clouds to one and the same coordinate system, it ispossible to scale the generated model based on any knownmeasurement. Point-clouds can be saved in one or more sep-arate layers according to the preferences of the user. Generat-ing the CAD elements is completely independent ofgenerating the point-cloud; actually, the point-cloud is not apre-requisite to generate CAD elements. Since the imagescontain by nature more information than the required ele-

ments, the CAD elements can be generated interactivelyfrom the images and can be saved in one or more layers sep-arately. Displaying the textured point-cloud overlaid by CADelements provide the advantage of both worlds as a powerfulcommunication way of project status. Figure 4 demonstrateshow CAD elements can be generated and displayed in vari-ous modes.Having a large number of overlapping images enables the

compilation and registration of a large number of stereo pairsto form an as-built model. Figure 5 shows the relationship be-tween cameras, images, rays to matched point and the com-piled model extend across several stereo pairs. The modelhas different densities, different types of pipes at differentheights including large ducts and parts of the wall and ceilingin two perpendicular directions. Figure 6 shows a larger as-built model from two viewing angles. The model showspipe-works having different sizes, types, heights, colors, andorientations, some pipes are extended vertically and horizon-tally in the ceiling while others are extended from the groundto the ceiling in the third dimension, or hang in the space be-tween ceiling and ground. Because different layers of pipe-work installations were carried out including installations oflarger size air-condition ducts, it was possible to distinguishthe newly generated CAD layer by giving its elements a dif-ferent color, as in Fig. 7. It is possible to distinguish betweenold and new layers of installations, in the same time by de-creasing the point size of the newly generated point-cloud.Several parts of pipe-works are visible from outside the

building. Taking overlapping pictures from a remote distanceenabled the generation of both point-cloud and CAD ele-ments without entering the building. Figures 8 to 10 showsamples of remotely generated point-clouds. The only disad-vantage is the relatively lower density of point-cloud due tothe limited resolution of the camera used. However, the lowerdensity point-cloud does not affect the measurements orCAD elements generation for visible pipe-works. This is ex-tremely useful in cases where access is limited due to safetyprecautions; e.g., project containing working boilers and (or)nuclear subsystems, or inaccessible zones.Using indoor and outdoor photography of pipe-works and

saving accomplished work in different CAD layers with dif-ferent colors; it was possible to visualize and generate reportsof pipe-work progress, mainly as a percentage of accom-plished work vs. scheduled work, and as total installation vs.total planned work, saving the newly accomplished work in anew CAD layer having the date as part of its name and usingdifferent colors to distinguish the new work when overlayedon older layer facilitated the computation, see Figs. 2 and 7.

DiscussionIn contrast to an expensive metric camera, the non-metric

camera has un-calibrated inner orientation parameters whichare necessary data for most photogrammetric mathematicaltreatments. As well there is instability of inner orientation pa-rameters, in addition to the lack of fiducial marks. Low-costconsumer cameras with changeable lenses add challenges;zoom lenses added more challenges for robust engineeringwork because of the changes in inner orientation parametersalong the whole zoom range. The use of the photogrammetricpoint-cloud generation approach may include simple data

Fig. 6. Point cloud of pipes from different perspectives: model re-sulting from registering several stereo-pairs together (top), the point-cloud from different angle to show more pipe-works of differenttypes, heights and directions (bottom).

Fig. 7. New layer of installations vs. older layer.

1068 Can. J. Civ. Eng. Vol. 39, 2012

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

collection with the following advantages: (1) lower cost,(2) cameras can work from an unstable platform using afford-able lenses with anti-shake image stabilization technology,(3) cameras can image almost any tiny size textured objectsusing suitable lenses for the task, (4) cameras can scan ob-jects at any distance ranging, from centimetres to hundredsof metres, (5) no eye safety or health issues, (6) can work ina wider range of temperatures than laser scanner, does not re-quire any interference with workers or machines, (7) imagingcan be repeated as many times as required, and (8) seamlessupgrading of the system capability over time as technologiesadvance and as budgets allow. The quality and specs of thelens had some impacts on the final output, not only thesharpness of the image but also the number of the images,for example less images could be taken with higher qualitywide angle lenses.It was found that in certain cases, some caution is required

for accurate point-cloud generation, (a) the closeness of alarge reflective surface, e.g., large air-condition ducts, fromthe camera itself may cause a random dispersion of light,(b) the large size of such ducts compared to the other pipe-works will increase the depth of field if the camera is veryclose, and (c) the featureless reflecting surface ducts maywork as an irregular mirror surface and decrease the numberof distinguished points on its surface. Accordingly, it is rec-ommended in similar cases to do the following: (1) shootingthe images from a larger distance, (2) using a lens of higherquality wide angle, (3) using higher resolution camera tocompensate for the larger distance effect, and (4) using apolarizing filter to minimize the mirror effect of reflectingsurfaces. While these cautions are required in the case of thepoint-cloud generation, however, using another photogram-metric technique enabled the production of as-built pipes andair-condition ducts hanged to the ceiling directly from the im-ages because it works independent of the process of point-cloud generation.The fine details of the model can be investigated by zoom-

ing-in to the required parts or measuring their depth, coordi-nates or dimensions. There is a flexibility to generate thesame model with different densities and (or) generate differ-ent densities for different stereo-pairs. The selection of thedensity of the generated point-cloud can vary according to re-quirements of the user to save processing time and storagespace. Figure 3 demonstrates how different densities can begenerated for different parts of the whole model. Examiningthe depth or the variation of heights is simple. It can eitherbe done by changing the viewing angle or by using the meas-

Fig. 9. Zoom in to point-cloud generated from remote images.

Fig. 8. 3D point-cloud of the whole E6 building generated from aremote distance.

Fig. 10. Remotely generated CAD elements fused on remotely gen-erated point-cloud.

Ahmed et al. 1069

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

uring tools to get the coordinates of any point in the modelas in Fig. 11. However, dealing with too much details ofCAD, point-clouds, textured surface, colored elements and(or) fusion of multi types of different formats and graphicdata representations can cause confusion to the user due to asimple fact that the screen is actually two dimensional. What-ever the perspective angle the result is a projection of the 3Dmodel to a 2D monitor model. Zooming-in may help but canalso cause more confusion. Photogrammetry overcomes thislimitation using the stereoscopic viewing capability which ismore natural to human vision. It is a standard function cur-rently available via several technologies, among them low-cost anaglyphic glasses. Figure 12 shows samples of stereomodels generated for use with anaglyphic glasses. Accuracyof the reconstruction process is not a simple task. In this re-search, the accuracy was evaluated using 30 control pointsadopting the technique used in Ahmed et al. (2011a). Threeranges of accuracy were found and can be summarized as fol-lows: (1) using coded targets, the accuracy of 3D coordinatesof check targets computed were in the range of plus or minus0.3 and 0.6 mm, (2) using autonomous orientation, seeFig. 2, with 80 to 400 matched points, accuracy ranges be-tween plus or minus 0.2 and 1.7 mm, as well it can be no-ticed that increasing the camera distance decreases theaccuracy, and the surface texture affects the number of auton-omously matched points, and (3) using manual matching theaccuracy ranges were between plus or minus 0.7 and 4 mm.In all cases the largest error was less than one centimetre.

ConclusionsThis paper describes a rapid and efficient monitoring sys-

tem composed of handheld camera and photogrammetry soft-ware as a powerful tool for constructing a 3D model of as-built pipe-works. It demonstrates high potential to overcomeproblems and delays in project monitoring, and avoid boththe accumulation of errors resulting from manual techniquesand the need for larger number of human resources to accom-plish project monitoring. The data collection is practical asone handheld camera moving in the site can register the sta-tus of the project in a short time without interference withother ongoing activities. A handheld camera for data collec-

tion at site and a modern close-range photogrammetry tech-nology at the office together provided a robust, practical andlow-cost progress tracking of pipe-works. The investigatedsystem shows that important and unique advantages can beexploited from a practical and economic point of view. Itproduced multiple output formats, e.g., raw CAD elements,

Fig. 11. Recognizing the difference in heights visually by variation of viewing perspective.

Fig. 12. Stereoscopic model of as-built CAD elements and point-clouds at different data scales.

1070 Can. J. Civ. Eng. Vol. 39, 2012

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.

3D points, lines, and surfaces, coloured and shaded elements,and textured elements using real world natural look; which ismore human-vision friendly. In addition, the unique 3D ster-eoscopic vision output can be viewed using low-cost anagly-phic glasses. These output formats will be useful as aseamless input to most if not all CAD and building informa-tion modeling technologies. Based on the above results, itcan be concluded that the investigated technique can trackmore than pipe-works and open the gate to a lot of applica-tions in a very cost effective way. Future work will investi-gate fully automatic recognition and construction of pipe-works and generating CAD pipes without manual interaction.

AcknowledgementThe authors would like to thank the graduate students

Arash Shahi and Afrooz Aryan for facilitating communica-tion and logistics during onsite data collection.

ReferencesAbeid, J.N., Allouche, E., Arditi, D., and Hayman, M. 2003.

PHOTO-NET-II: a computer based monitoring system applied to aproject Management. Journal of Automation in Construction,12(5): 603–616. doi:10.1016/S0926-5805(03)00042-6.

Ahmed, M. 2007. Reconstruction of partially damaged human facesusing un-calibrated images. Proceedings of the 11th World Multi-Conference on, Systemics, Cybernetics and Informatics, WMSCI,Florida, USA. 2007.

Ahmed, M., and Haas, C. 2010. The Potential of Low Cost CloseRange Photogrammetry towards Unified Automatic PavementDistress Surveying. CD_ROM. Proceedings of The 89th AnnualMeeting Of The Transportation Research Board, Washington,D.C., 11–15 January 2010.

Ahmed, M., and Hegazy, T. 2008. Comparison among indoorlocation-based technologies for construction and infrastructureapplications. CSCE 2008 Annual Conference, Québec, Que.,Canada, 10–13 June 2008.

Ahmed, M., Haas, C., and Haas, R. 2011a. Toward low-cost 3Dautomatic pavement distress surveying: the close range photo-grammetry approach. Canadian Journal of Civil Engineering,38(12): 1301–1313. doi:10.1139/l11-088.

Ahmed, M., Haas, C., and Haas, R. 2011b. Accurate and LessExpensive Pavement Distress Surveying Using Multi-Photogram-metric-Output Fusion. CD_ROM. Proceedings of The 90thAnnual Meeting Of The Transportation Research Board, Wa-shington, D.C., 23–27 January 2011.

Ahmed, M., Guillemet, A., Shahi, A., Haas, C., West, J., and Haas, R.2011c. Comparison of Point-Cloud Acquisition from Laser-Scanning and Photogrammetry Based on Field Experimentation,3rd International/9th Construction Specialty Conference, Ottawa,Ontario, Canada, 14–17 June 2011.

Bosche, F., and Haas C.T. 2008. Automated 3D Data Collection

(A3dDC) for 3D Building Information Modeling, 25th Interna-tional Symposium on Automation and Robotics in Construction(ISARC), Vilnius, Lithuania, 27–29 June 2008, pp. 279–285.

Cheung, S.O., Suen, H.C.H., and Cheung, K.K.W. 2004. PPMS: aweb based construction Project Performance Monitoring System.Automation in Construction, 13(3): 361–376. doi:10.1016/j.autcon.2003.12.001.

Cho, W., Schenk, T., and Madani, M. 1992. Resampling digitalimagery to epipolar geometry, IAPRS International Archives ofPhotogrammetry and Remote Sensing, 29(B3): 404–408.

El-Omari, S., and Moselhi, O. 2008. Integrating 3D laser scanningand photogrammetry for progress measurement of constructionwork. Automation in Construction, 18(1): 1–9. doi:10.1016/j.autcon.2008.05.006.

Jiang, R., Jauregui, D.V., and White, K.R. 2005. Review of Close-Range Photogrammetry Applications in Bridge Engineering. Apaper submitted to the Transportation Research Board (Washing-ton, D.C.) November 2005.

Jiang, R., Jauregui, D.V., and White, K.R. 2008. Close-rangephotogrammetry applications in bridge measurement: literaturereview. Journal of The International Measurement Confederation.Measurement, 41(8): 823–834. doi:10.1016/j.measurement.2007.12.005.

Kraus, K. 1993. Photogrammetry (volume 1). Dümmler Verlag,Bonn. 1993.

Kraus, K. 2007. Photogrammetry geometry from images and laserscans, 2nd Ed. translated by Ian Harley Stephen Kyle, Walter DeGruyter Berlin. New York. 2007.

Memon, Z.A., Abd.Majid, M.Z., and Mustaffar, M. 2006a.Digitalizing Construction Monitoring (Dcm):An Overview OfMalaysian Construction Industry And Proposing Prototype Soft-ware, Proceedings of the 6th Asia-Pacific Structural Engineeringand Construction Conference, (APSEC 2006), 5–6 September2006, Kuala Lumpur, Malaysia.

Memon, Z.A., Abd.Majid, M.Z., and Mustaffar, M. 2006b. ASystematic Approach For Monitoring And Evaluating TheConstruction Project Progress, Journal of The Institution ofEngineers, Malaysia (Vol. 67, No. 3, September 2006), Malaysia.

Mikhail, E.M., Bethel, J.S., and Mcglone, J.C. 2001. Introduction tomodern photogrammetry, John Wiley & Sons, Inc., New York.

Schenk, T. 1990a. Computation of epipolar geometry, technical notesin photogrammetry, No. 5, Department Of Civil and Environ-mental Engineering and Geodetic Science, The Ohio StateUniversity.

Schenk, T. 1990b. Digital Photogrammetry - Volume I, Terrascience,Laurelville, Ohio, USA, 428 p.

Slama, C. 1980. Manual of photogrammetry fourth edition. AmericanSociety For Photogrammetry and Remote Sensing, Virginia, USA.

Song, J., Caldas, C., Ergen, E., Haas, C., and Akinci, B. 2004. Fieldtrials of RFID technology for tracking pre-fabricated pipe spools.Proceedings of the 21st Annual International Symposium forAutomation and Robotics in Construction (ISARC), Jeju. Korea &World Affairs, (September). pp. 21–25.

Ahmed et al. 1071

Published by NRC Research Press

Can

. J. C

iv. E

ng. D

ownl

oade

d fr

om w

ww

.nrc

rese

arch

pres

s.co

m b

y U

nive

rsity

of

Wat

erlo

o on

06/

06/1

3Fo

r pe

rson

al u

se o

nly.