Professional Documents
Culture Documents
Miroslaw J. Skibniewski
Dept. of Civil & Environmental Engineering
A.J. Clark School of Engineering;
University of Maryland College Park
mirek@umd.edu
miroslaw j. skibniewski onstruction project monitoring with site photographs and 4d project models pp 1106-1114 1107
Data
Information Action
Representation Information Environment Perspective
Perspectivev
Perspective
As-planned Progress
AR Model As-built Progresss Model
Model
Schedule and
work break- Actual schedule
down Detect deviation
Collect + Assess Process Model structure (Product components
+ Process Activities)
A...................................B....................................C
comparing the actual cost to accomplish the non-physical activities). After the on time and on budget, the red color
a task with the cost planned for that deviations in the physical activities represents activities that are behind
task. The 3D as-planned model, sched- and cost are detected and quantified, schedule and over budget. Any occlu-
ule and work breakdown structure are they have to be visualized (Pea-Mora sion or blockage that is caused by
taken as baseline measurements for et al., 2009). For visualizing the prog- superimposing the image is removed
monitoring the progress. Figure 1 ress accurately, the as-planned model at the end. Figure 2
A detailed schedule of the construc- is color-coded and superimposed on
tion activities should be prepared the site photograph. The visualiza- SfM technique for reconstructing
for correct progress measurements tion technique is described in a sub- 3D model
because the 4D as-planned model is sequent section in this paper. Although For reconstruction of the as-built
prepared according to the schedule and a quadrangle color scheme has been scene, the technique called Structure
work-breakdown structure. The 4D as- proposed, the image in Fig. 2 shows from Motion (SfM) can be used. For
planned model visualizes all the activi- only green and red colors. The green reconstructing the scene automati-
ties mentioned in the schedule (except color represents the activities that are cally, one needs to know the relative
locations, orientation, zoom or focal track are recovered at this stage. of its physical location can be made by
length of the photograph. The process Initially, this is done just for our first triangulating the point. All the pairs
of reconstruction consists of four steps image pair, because it is important to of rays that could be used for trian-
(Golparvar-Fard et al., 2009): get good initial estimates. For estimat- gulating that point are considered for
1. Detecting features in all the images ing the parameters of all tracks and estimating the location. The angle of
2. Matching these features across cameras, we follow an incremental separation between these pair of rays
image pairs approach. For robustly estimating the should be more than threshold limit
3. Combining the matches into cor- two-frame reconstruction, the initial (2.0 in this case), for the point to be tri-
respondence tracks across several pair should have a large amount of angulated. After the addition of all the
photos matches. For this purpose, we choose points, the global bundle adjustment is
4. Estimating the geometry of the scene an image pairs, which have a maximum run to refine and reconstruct the entire
and location of the cameras using number of matches (at least 100) to scene. This procedure is continued in
these correspondences be the initial pair. Nisters five-point all of the cameras. The data structure
The first step is to detect distinct fea- algorithm (Nister 2004) is used for esti- used for visualizing the as-built recon-
tures in each image that could be used mating the extrinsic parameters of the structed scene is as follows (Golparvar-
to prepare the initial arrangement of camera, and then the tracks visible in Fard et. al, 2009):
the scene. In this paper, we will dis- this image pair are triangulated. After a. A set of key points (with its 3D loca-
cuss the SIFT (Scale Invariant Feature this, another camera is added to the tion and an average color determined
Transforms) key point detector (Lowe optimization. Direct linear transform by the site images from which this
2004). This technique has showed (DLT) technique is used for finding the key point has been observed)
good invariance to image transforma- new cameras extrinsic parameters. b. A set of camera (with known extrinsic
tions in the experiments performed The intrinsic parameters of the new and extrinsic parameters)
by (Mikolajczyk et al., 2005). There camera are estimated with the help c. A mapping between key point and
are thousands of SIFT key points in a of DLT technique and EXIF-Exchange all the cameras that observe that
typical image. After detecting the key image file format-tags of JPEG images. key point
points, we matched key point descrip- After this, the bundle adjustment algo- Along with this information the
tors in an image pair. Once a set of geo- rithm is run, which allows only the number of cameras observing that
metrically consistent matches is found new camera and points it observes to point, local coordinates of the point
in an image pair, the matches are orga- change keeping the rest of the model in the image and SIFT key point index
nized into tracks. A set of connected constant. Finally, the points observed is also stored for all the key points.
matching key points across several by the new camera are added to the Figure 3
images is called tracks. The extrin- optimization algorithm. The points are For providing better resolution of the
sic (translation, rotation) and intrinsic added only when it is observed by at reconstructed scene, the camera frus-
properties (focal length, distortion) of least one of the already recovered cam- tum is texture mapped. This allows the
the camera and 3D location for each eras and if a well-conditioned estimate user to zoom in and get the information
miroslaw j. skibniewski onstruction project monitoring with site photographs and 4d project models pp 1106-1114 1109
information. In this paper, we will dis-
cuss the object-based project moni-
toring approach. In this approach, the
as-built construction photographs are
compared with the as-planned project
goals. The object-based model with
BIM is used to represent geometric
information of construction compo-
nents into meaning object. This is done
through an automated analysis using
computer vision techniques. The inte-
rior construction progress monitoring
using BIM can be explained as follows.
Decomposition of interior
construction objects
Figure 3 A sparsely reconstructed scene of the project site (Source: Golparvar- For creating a relationship to measure
Fard et al. [1], 2009) For providing better resolution of the reconstructed the interior construction progress, we
scene, the camera frustum is texture mapped. This allows the user to zoom in need that level of detailed objects and
and get the information about the productivity, construction progress from activities to be the same as defined in
the reconstructed scene (Golparvar-Fard et al., 2009) the as-planned model. The construc-
tion schedule for interior construc-
tion is not as detailed as the exterior
about the productivity, construction of specifying the set of correspon- construction schedule. For effective
progress from the reconstructed scene dences between the 3D as-planned interior construction progress moni-
(Golparvar-Fard et al., 2009) and points in an image can be done toring, the interior construction pro-
manually (Golparvar-Fard et al., 2009). cesses should be described in more
Geo-registration The same registration will be held for detail. As can be seen in Figure 4, HVAC
Geo-registration is the process of all the images. The geo-registration ductwork in ceiling spaces is sched-
aligning coordinates of one image/ process allows the as-built site images uled for 60 days. It is not possible in
model to image/model with known to exactly superimpose on the 4D as- this case to monitor the progress on a
coordinates. By the SfM technique, we planned model. weekly basis. The activities should be
get the relative position of the camera. treated as objects, so that it can be rep-
We use the geo-registration process Interior construction progress resented in a 3D model according to the
to determine the exact camera coor- monitoring percentage of work done. The activi-
dinates. This is done by aligning the The schedule for interior work is not ties can be divided down further on
as-built reconstructed model with the as detailed as it should be for proper the basis of location information, e.g.,
as-planned model. We need the exact tracking of the progress of interior zone or floor in the building. Thus, for
coordinates of the as-built scene to get construction. There is a significant representing all interior construction
it aligned with the as-planned model. schedule discrepancy in the interior progress, we need to detect objects
Theoretically, the estimated coordi- construction, which makes it difficult and location information of their activi-
nates of the reconstructed scene are for project managers to predict accu- ties. For instance, there are four HVAC
related to the absolute coordinates rate subcontractors progress and pro- duct units to be installed on second
(exact location) by a rotation, uni- ductivity (Seungjun et al., 2011). Many floor and 3 have been installed already;
form scale transformation and global cost overruns and schedule delays are then, the percentage of progress can be
translation (Snavely et al., 2007). For caused, due to the limited visual under- calculated. This progress can then be
correctly determining the transforma- standing, manually intensive process compared by the as-planned model and
tion, the sum of squared residual errors and complicated nature of the inte- then be represented on the 3D model
between the model and set has to be rior construction progress. Thus, the by a using color-coding metaphor. By
minimized. This is done by registering interior construction progress reports preparing a detailed schedule, the
a set of as-built 3D model points to as- should have object-oriented aspects efficiency of interior construction is
planned 3D set of points. The process and visual data for in-depth progress improved because the deviations in
Visualization of construction
progress
as-built and as-planned model can be as-built construction photographs and Proper visualization is very important
assessed quickly. BIM. The as-planned BIM is prepared for tracking the projects progress. By
For monitoring the progress of non- by monitoring thousands of construc- just representing whether the project
physical activities of interior construc- tion elements. The interior construc- is ahead of schedule or behind the
tion (e.g., installing water pipes) should tion elements are decomposed from schedule does not really convey the
be controlled and monitored by using the as-built photographs. Only the progress. The project can be ahead of
detailed physical objects from the 3D elements that can be seen in the as- schedule, but could have a lot more
model (Seungjun et al., 2011). built photographs have to be taken in money spent than planned. The con-
account in the as planned model and cept of EVM solves this problem. EVM
Conceptual model for interior their schedules for monitoring the is a project management technique that
construction progress monitoring progress. For example, in Fig. 5 one accurately conveys the project perfor-
using BIM only considers drywall components mance. It combines the measurement
Besides showing 3D geometric data and for calculating the construction prog- of scope, schedule and cost. By the
material properties, BIM also includes ress because it is only visible in the
schedule, cost, safety information, as-built photographs. First of all, the
equipment warranties, product data as-built construction elements whose
sheets, etc. BIM facilitates automati- progress needs to be monitored are
cally track the construction progress decomposed. This is done by using the
according to the as-planned schedule as-built photographs. After decomposi-
for interior components. In actual con- tion, these components are selected
ditions, the construction progress is in Industry Foundation Classes (IFC
not adjusted exactly according to the 2X3) format for matching with the
work being done because the schedule as-planned model. IFC is a file format
does not exactly describe the activity that enables interoperability in AEC
that is being performed. For example, industry (Renauld et al., 2008). This
in Fig. 6, installation of a duct on the file format is commonly used for BIM.
second floor takes 60 days, and this Here, this format allows the data to be
is represented in the schedule only by exchanged between the as-built pho-
one activity, but in actuality there are tographs and the as-planned model.
many building components included for Comparing the as-built data with the
this work. With the use of BIM, these as-planned data provides the prog-
HVAC duct elements are considered ress of the construction. Deviations
and a better visualization of interior in the schedule and cost are detected Figure 5 Using BIM for interior
construction progress is achieved. by comparing the actual data with the construction progress monitoring
Figure 5 shows how the interior con- as-planned data. This data can be used by as-built photographs
struction progress is monitored using for measuring the percentage of work (Source: Seungjun et al., 2011).
miroslaw j. skibniewski onstruction project monitoring with site photographs and 4d project models pp 1106-1114 1111
minimize future problems, the project 3. Causes of deviation
manager should analyze the data and Knowing the cause of deviation in the
forecast the future conditions. There schedule and cost is critical information
are three types of information that for project managers. By knowing the
are important for project managers to cause of deviation, project managers
properly monitor the progress (Song can take proper steps to overcome it.
et al., 2005).
Critical
3D
Figure 6 Colored representation of Performance Face model Border Criticality Non Critical
the interior construction progress.
(Source: Seungjun et al., 2011). Near Critical
miroslaw j. skibniewski onstruction project monitoring with site photographs and 4d project models pp 1106-1114 1113
overlaid on Time-Lapsed Photographs." ASCE
Journal of Computing in Civil Engineering,
2009: 391-404.