Assessing the Value of 3DReconstruction in BuildingConstruction

Uma MurthyURC Ventures Inc.8201 164th Ave NESuite 200Redmond, WA 98052 USAuma.murthy@urcventures.com

David BoardmanURC Ventures Inc.8201 164th Ave NESuite 200Redmond, WA 98052 USAdavid.boardman@urcventures.com

Chirag GargDept. of BuildingConstructionPurdue Univ.West Lafayette, IN 47907USAcgarg@purdue.edu

Copyright is held by the author/owner(s).

Abstract3-dimensional (3D) reconstruction is an emerging field inimage processing and computer vision that aims to create3D visualizations/ models of objects/ scenes from imagesets. However, its commercial applications and benefitsare yet to be fully explored. In this paper, we describeongoing work towards assessing the value of 3Dreconstruction in the building construction domain. Wepresent preliminary results from a user study, where ourobjective is to understand the use of visual information inbuilding construction in order to determine problems withthe use of visual information and identify potentialbenefits and scenarios for the use of 3D reconstruction.

Keywords3D reconstruction, images, user study, buildingconstruction

ACM Classification KeywordsI.4.5 [Image Processing and Computer Vision]:Reconstruction.

General TermsHuman Factors

arX

iv:1

201.

3172

v2 [

cs.H

C]

2 F

eb 2

012

Introduction

A

B

C

D

Figure 1: 3D reconstruction of a curb: A & B) Input imagesof the curb; C) 3D reconstruction model (10 million points) ofthe curb, zoomed out to reveal the context of the scene; andD) 3D view of the curb, zoomed in to show detail of thecrevice between the two stones on the curb.

3-dimensional (3D) reconstruction is an emerging field inimage processing and computer vision that aims to create3D visualizations/ models of objects/ scenes from imagesets [6, 7]. A set of images is processed to generate apoint cloud depicting the real-world scene (as present inthe images). This interactive point cloud can be manuallyprocessed and manipulated as a 3D model. Figure 1shows the 3D reconstruction (C and D) of a curb fromtwo images (A and B). 3D reconstruction technology hasbeen demonstrated to be effective in military training,tele-operation of vehicles [3], tourism [7], creating andusing story narratives [1], and street navigation [4]. Yet,its commercial applications are yet to be fully explored. Inthis paper, we describe ongoing work towards assessing

the value of 3D reconstruction in the building constructiondomain. Specifically, we are examining two questions:

1. Can 3D reconstruction from images increase the valueof images used in existing construction use cases?2. Can 3D reconstruction from photos provide acomparable alternative to laser scanning (LIDAR) inbuilding construction?To understand the domain and our findings better, weprovide a brief overview of the main people involved in atypical construction project. A construction project mightinvolve the following people 1: owner provides therequirements of the construction project; architectdevelops a design from the requirements, includingdrawings; general contractor is in charge of the overallconstruction of the building and is responsible forconverting the design to a constructed building; andsubcontractors, are responsible for specialized part of theconstruction, such as masonry or MEP systems engineer.

There has been prior work researching the use of photos inbuilding construction. PhotoScope [8] providesspatiotemporal visualization of photos to support typicalphoto seeking tasks in construction, such as for claimsand document management. The authors found thatpresenting the context of time and space facilitatesefficient photo searching. Liu and Jones’ study on the useof digital photos in the construction industry [5] wasaimed at understanding how photos are acquired, stored,edited, viewed, managed, and retrieved. One outcome oftheir study was a digital image shooting guide to help

1This set of people might vary based on the type of constructionproject and might include other people, such as a soil engineer or astructural engineer. Also, a group of people might work in the sameteam – e.g., the owner and the architect together might be considereda client.

with better storage, management, and retrieval of photosin building construction activities.Guiding questions for the

semi-structured interview:

1. Describe your job, yourresponsibilities, and theactivities that you are in-volved in.

2. Give examples of con-texts/scenarios in whichyou use visual informa-tion.

3. What problems do youface when you interactwith visual information?How do you deal withthem?

4. Give examples of in-stances when the visualinformation you had wasnot sufficient. How didyou address this situa-tion?

5. What contexts/ scenariosdo you make site visits?

6. Have you used LASERtechnology to develop a3D model in your job?Can you give examples ofthose situations.

The focus of our studies is to understand the value of 3Dreconstruction models (to the user in a construction task)in comparison with other “visual information” in buildingconstruction. We use the term “visual information” torepresent image-specific information, such as that in aphotograph. Visual information in a construction projectmight include: 1) paper-based and electronic drawings(e.g., plans, detailing, isometric views, and elevations); 2)photos, including those of the construction site, materials,structures, and parts; 3) BIM (Building InformationModeling) models (e.g. REVIT c©2) and other 3D models;4) videos of construction site; and 5) textual description(of a visual), which might be present in constructionproject documentation, such as a Request-for-information(RFI).

Most prior work on evaluation of 3D reconstructiontechnology focused on the evaluation of systemperformance (efficiency and accuracy), usability of systemand method (and not its value), or user performance on adomain-specific task. In their study that compared the(user) performance of a 3D tele-operation approach (a 3Dvisualization) with video-based tele-operation and directdriving, Huber, Kelly, et al. [3]. found that their 3Dtele-operation approach significantly improvedperformance, both in terms of driving speed and reducednumber of errors when compared to video-basedtele-operation. Users reported that they preferred the3D-based tele-operation mode to the video-based mode asit provided a wider field of view and had the ability toview a scene from arbitrary viewpoints. Overall workload

2http://usa.autodesk.com/revit-architecture/

was measured least for the 3D video interface versus thelive video and manual drive.

MethodsOur first step was to understand how people in buildingconstruction interact with and use visual information. Ourobjectives were to: 1) identify problems in thisinteraction/ use and 2) understand potential benefits andscenarios for the use of 3D reconstruction technology. Tothis end, we have been conducting semi-structuredinterviews with building construction personnel to learnabout their professional activities, especially those thatinvolve the use of visual information. The guidingquestions of the semi-structured interviews are listed inthe margin. At the time of writing the paper, weconducted and analyzed two interviews (P1 and P2).Each interview lasted about an hour. Also, participantsshared materials to support their responses. We analyzedthe interviews and materials using open coding whilekeeping in mind the aforementioned objectives. Sectionpresents the preliminary findings of the study.

Following this study, our plan is to test selected scenarioswith 3D reconstruction models and determine potentialbenefits. Finally, we would like to validate the benefits of3D reconstruction by comparing its use with other visualinformation in the building construction domain as well asidentify emergent use/ behavior.

Preliminary findingsParticipant P1 is an estimator with a constructioncompany that specializes in renovation projects. His job isto review the requirements, design, and the site of aproposed project and estimate the detailed tasks, budget,and timeline for the project. Through this process, heconstantly communicates with the project owner (or

architect) and with subcontractors and engineers todevelop an as accurate as possible budget and scheduleestimate. Participant P2 is a project manager in a largeconstruction company that does a variety of projects. Hisresponsibilities include evaluating the design proposed bythe architect/ owner, developing a budget and schedule,managing construction personnel on the project,overseeing the construction through completion,managing RFI’s and claims. P2’s job involves extensiveinteraction with the owner/ architect, subcontractors,engineers, and other personnel. We present preliminaryfindings considering our study objectives.

Architect GeneralContractor

Owner Sub-Contractor

we want to renovate this room & get a new ceiling

Everything above the ceiling needs to go

what is above the ceiling?

- what parts are there?- how are they connected?- how much do they measure?

Figure 2: Communication among people in a constructionproject.

Problems with the use of visual informationWe analyzed participant responses and for problems andfound at least three causes:

Miscommunication of visual information. Every personinvolved in a construction project has their ownunderstanding/perspective of the project design and site.This difference in understanding can lead tomiscommunication of specifications, which in turn canlead to increased workload, increased cost, and delays.One example of this communication dynamic is providedin Figure 2.

Incomplete or incorrect visual information. Since eachstakeholder in a construction project has their ownunderstanding/ perspective, they might not have all theinformation as required by the others. In Figure 2, notehow the need for detail keeps increasing as informationflows from the owner through the sub-contractors. Thefollowing example from P1 is another instance ofincomplete information.

“We had to renovate a historic building, [which was]constructed in the 1950s. We received the original 30design drawings and photos as part of the projectdocumentation. If the building had been constructedtoday, we would have had 30,000 drawings”

Both participants agreed that in many cases, the designdoes not reflect the field conditions accurately, leading toincomplete or incorrect visual information. In one examplefrom P2, an incorrect detail of 20 large skylights provedcostly for the contractor, resulting in hours of pre-planningand preparation going waste.

“The size of the skylight was shorter than the actualconcrete opening. ... At the design time, they probablyhad the skylight hanging outside beyond the concreteopening. But, it ended up being a shorter dimensionskylight and getting the water inside the building. Wereplaced all those skylights with new skylights”

Lack of a sense of orientation in current visualinformation. Photographs and drawings are not adequateto understand the the site conditions and orient oneself tothe surroundings. For example, a contractor might wantto know the locations and dimensions of trees on the siteor of neighboring plots and buildings. She might useGoogle Earth c©3 to get a view of the site. However, inmost cases that information is not updated.

REQUEST:Reference approved submittal #111 gutter at pre-cast roof of tug ramp and attached sketch.Please confirm submittal #222 roofing at sector-8 is approved per attached sketch and per submittal #111.RESPONSE:The contractor proposed detail, sketch-1 with TPO membrane over the SST gutter and sloped insluation, is acceptable provided that the TPO is fully adhered and any penetrations required are limited to the top edge of the SST gutter. GC to coordinate all required interface, openings for drain flashings and scuppers, etc. as noted in submittals.

6 days

Figure 3: The request, response,and response time in an exampleRFI document, involving therequest for approval of changesmade in construction.

Construction personnel work around these problems in anumber of ways. They might engage in frequent formalcorrespondence to clarify issues, such as exchanging RFIs(see Figure 3). New visual information anddocumentation, such as detailed photos, drawings, and 3Dmodels, would need to be developed for parts discoveredand to clarify issues. In some cases, parts of the projectmight need to be re-designed to match field conditions.Related to this is the re-fabrication of parts to fit actualspecifications of the site. In most cases, designers,owners, and construction personnel would need to makemultiple site visits to orient themselves to the site, toconfirm details of parts, to measure (and re-measure)areas for off-site fabrication of parts and to get a completeunderstanding of the project. All this additional workoften would, in turn, result in inefficiency (wasted time),increased costs and workload, and reduced productivity.

Potential benefits and scenariosConsidering prior work in 3D visualization and modeling,we believe 3D reconstruction has several visualizationbenefits, most notably being: the ability to experiencereal-world presence through a sense of orientation andimmersion; the ability to view a scene from arbitrarypoints to help understand various perspectives of thescene; and interactive contextual browsing to understand

3http://www.google.com/earth/index.html

the context, details, and scale (via geo-referenced points)of a scene. We now outline select scenarios where thesecapabilities will be beneficial.

Renovation work on an acoustic ceiling (such as that inFigure 2): A 3D reconstruction model (among other uses)could be used to communicate and understand the sitedetails, avoid measurement errors, document progress andcompare stages, and access context and details ofindividual parts.Orientation on a construction site: A 3D reconstructionmodel would facilitate site orientation and understanding(provide a sense of immersion and ability to understandand measure details) to help prepare with constructionactivities (what is the level is the soil, is there a stonestrata to support the foundation), thus avoiding severalsite visits.Help in pre-fabrication of parts (such as that mentioned in“skylight” example): With a 3D reconstruction model,one would have actual measurements of parts of theconstruction site. By referencing the model, one can avoidmistakes in fabrication (see skylight example above) andprepare parts to match the site specifications accurately.

Discussion and next stepsA 3D reconstruction model can provide rich informationfor many applications where spatial contexts andinteractivity are relevant. In this paper, we presentedpreliminary findings from the first of a set of user studiesto assess the value of 3D reconstruction in the buildingconstruction domain. Problems in the use of use of visualinformation might be caused by miscommunication,incorrect and incomplete information, and lack oforientation.

In the next phase of the project, we will compare the use

of 3D reconstruction models with other visual informationin construction activities. The scenarios developed fromthe current study will help in designing our nextevaluation and ground the evaluation/experiment tasks inreal-world activities. Eventually, our goal is to be able tomeasure the benefits of 3D reconstruction, irrespective ofthe domain in which it is deployed. Bowman andMacmahan have a similar discussion on the benefits ofimmersion [2]. They describe studies to evaluate thebenefits of immersion by isolating its components, such asfield-of-view and field-of-regard and examine how eachcomponents effects a user’s task performance. We wouldlike to explore similar questions in the context of 3Dreconstruction, such as what are the objectivelymeasurable components of 3D reconstruction model andhow do they impact the performance of the user? Whatlevels of zooming yield optimal visibility andunderstanding to support the user’s task?

AcknowledgementsWe thank our participants for their valuable inputs.Thanks to Dr. Hazar Dib, for his support in this work.Photo sources for Figure 2 are from Flickr: Kitchenrenovation by gmclean, Dollis Hill bunker, false ceilingtiles by RachelH , and DSC 4446 by gordan mullan; TheRFI information in Figure 3 has been provided by theparticipants.

References[1] N. Adabala, N. Datha, J. Joy, C. Kulkarni,

A. Manchepalli, A. Sankar, and R. Walton. An

interactive multimedia framework for digital heritagenarratives. In Proc. of MM ’10, pages 1445–1448,New York, NY, USA, 2010. ACM.

[2] D. Bowman and R. McMahan. Virtual reality: Howmuch immersion is enough? Computer, 40(7):36 –43,july 2007.

[3] A. Kelly, N. Chan, H. Herman, D. Huber, R. Meyers,P. Rander, R. Warner, J. Ziglar, and E. Capstick.Real-time photorealistic virtualized reality interface forremote mobile robot control. The InternationalJournal of Robotics Research, 30(3):384–404, 2011.

[4] J. Kopf, B. Chen, R. Szeliski, and M. Cohen. Streetslide: browsing street level imagery. ACM Trans.Graph., 29:96:1–96:8, July 2010.

[5] J. Liu and S. Jones. A study on utilization of digitalphotographs in construction industry. In AnnualAssociated Schools of Construction InternationalConference, Auburn, Alabama, 2008.

[6] M. Pollefeys, R. Koch, M. Vergauwen, and L. V. Gool.Automated reconstruction of 3d scenes from sequencesof images. ISPRS Journal of Photogrammetry andRemote Sensing, 55(4):251 – 267, 2000.

[7] N. Snavely, S. M. Seitz, and R. Szeliski. Phototourism: exploring photo collections in 3d. In ACMSIGGRAPH 2006 Papers, SIGGRAPH ’06, pages835–846, New York, NY, USA, 2006. ACM.

[8] F. Wu and M. Tory. Photoscope: visualizingspatiotemporal coverage of photos for constructionmanagement. In Proc. of CHI ’09, pages 1103–1112,New York, NY, USA, 2009. ACM.