Automation in Construction 91 (2018) 44–52

Contents lists available at ScienceDirect

Automation in Construction
journal homepage: [Link]/locate/autcon

Comparison of point cloud data and 3D CAD data for on-site dimensional T
inspection of industrial plant piping systems
⁎
Cong Hong Phong Nguyena, Young Choib,
a
Department of Mechanical Engineering, Graduate School, Chung-Ang University, Seoul, Republic of Korea
b
School of Mechanical Engineering, College of Engineering, Chung-Ang University, Seoul, Republic of Korea

A R T I C L E I N F O A B S T R A C T

Keywords: Inspection is vital in industrial plant construction and management. However, traditional inspection methods
Point cloud that rely on human involvement and paper documentation are becoming untenable as modern industrial plants
As-built inspection are becoming larger and more complex than legacy facilities. Hence, an efficient and robust method is required
CAD-vs-scan to support the inspection of modern industrial plants. In this paper, an improved technique relying on terrestrial
Distance-based deviation analysis
laser scanning (TLS) for data acquisition and normal-based region growing and efficient random sample con-
Geometric parameter comparison
sensus (RANSAC) for point cloud data processing is proposed for the on-site dimensional inspection of the piping
systems of an industrial plant. Consequently, the as-built condition of the plant is assessed via a distance-based
deviation analysis and a comparison of geometric parameters between the as-designed and as-built models. The
method is validated using a dataset acquired from a compartment of a ship has verified the robustness and
reliability of the proposed approach.

1. Introduction from the compartment of a ship, which is a type of industrial plant, was
used to validate the efficiency and robustness of the proposed method.
Piping systems play an important role in the operation of industrial
plants as they appear widely in heating, ventilation, and air con- 2. Literature review
ditioning (HVAC) systems in terrestrial industrial factories and used for
supporting marine facilities. The inspection of piping systems in such Since the proposed approach is a combination of existing and effi-
plants is usually performed during the following three stages in the life cient methods, a literature review is required. Two following sections
cycle of a facility: production, post-production, and on-site inspection include a review of the existing literature on the laser scan data pro-
during construction [1]. Among these, on-site inspection poses the most cessing for the as-built modeling of mechanical, electrical, and
significant challenge in achieving operational efficiency and robust plumbing (MEP) constructions and utilization of the laser scan data for
results due to the large scale of the facilities. As mentioned in [2], it is the inspection of large-scale objects.
difficult to monitor and record the inspection results as doing so re-
quires significant manual operation, and therefore, needs tremendous 2.1. Point cloud processing for the as-built modeling of MEP constructions
human resources. Hence, a new method is needed to improve the au-
tomated monitoring and quality assurance of industrial plants. Various studies have been conducted on the data processing of point
As any dimensional deviation in the construction of the piping clouds obtained from cylindrical objects. Chaperon et al. [3] proposed a
system could lead to a serious problem in the operation of the entire method based on a Gaussian image and random sampling to detect and
plant, a robust and efficient method for on-site dimensional inspection extract the data points of cylindrical objects. The result was reliable and
is required. In this paper, a method to support the decision-making for robust, but the computational efficiency when applying the method to
the inspection of industrial piping systems is proposed. The proposed large data sets was not validated. Son et al. developed two related
method involves point cloud data processing for object segmentation techniques, relying on the curvature computation [4] and prior
and extraction to aid a comparative assessment via a distance-based knowledge, using a three-dimensional (3D) CAD database [5], whereas
deviation analysis of the geometric parameters of the as-built piping Lee et al. [6] used a skeleton method to reconstruct an as-built model of
components required for the inspection. A sample dataset acquired a piping system from the point cloud data. Kawashima et al. [7,8]

⁎
Corresponding author at: School of Mechanical Engineering, College of Engineering, Chung-Ang University, 84 Heukseok-ro, Dongjak-gu, Seoul 06974, Republic of Korea.
E-mail address: yychoi@[Link] (Y. Choi).

[Link]
Received 12 September 2017; Received in revised form 13 February 2018; Accepted 2 March 2018
0926-5805/ © 2018 Elsevier B.V. All rights reserved.
C.H.P. Nguyen, Y. Choi Automation in Construction 91 (2018) 44–52

suggested a method using normal-based region growing for the seg- proposed. The proposed method utilizes the advantages of as-built
mentation and cylinder fitting for the segmentation and recognition of modeling in providing the geometric parameters, as well as corre-
piping components. The recognition rate in the above studies was high, sponding data points of the piping components to support inspections.
at > 80%. Other notable studies are those performed by Patil et al. [9], Although only a simple comparison was performed between as-built
Liu et al. [10], Czerniawski et al. [11], Bosche et al. [12], and Dimitrov and as-designed models, its role in the on-site inconsistency inspection
et al. [13]. Although various effective methods for point cloud data is highly important because of its ability to balance the quality and cost
processing exist, the complexity of the industrial plants renders that the of the inspection.
use of a single-algorithm method that works perfectly in every case is
infeasible. Also, the use of a multi-step method may increase the effi-
ciency and reduce any of the disadvantages associated with each al- 3. Methodology
gorithm.
The proposed method is divided into two main stages: point cloud
data processing and as-built condition assessment based on the point
2.2. Utilization of laser scan data for the inspection of large-scale of objects cloud processing result (Fig. 1). As the scan data obtained from an as-
built industrial plant are expected to be highly complex, including the
Terrestrial laser scanning (TLS) has been employed as an optimal noise and data points corresponding to the non-piping components, a
and available method to capture the geometric data of the as-built in- combined method is developed for a more robust and efficient pro-
dustrial plants as it can rapidly capture the current configuration and cessing. Details of the method are described in Section 3.1. To assess the
layout of the plants with a high accuracy [12]. The idea of utilizing TLS dimensional quality of the as-built plant, the geometric parameters of
as a tool for supporting construction inspection is not new. In fact, it has the recognized components are extracted and compared against their
been applied for the inspection of other types of large-scale construc- as-designed specifications as derived from the CAD data in stereo-
tions such as tunnels and structural constructions [14–18]. The ad- lithography (STL) format. The derived quality metrics include the mean
vantage of the TLS technology has been demonstrated in the application distance error (MDE), point distance error (PDE), and other geometric
of progress monitoring [2,19–22]. However, the extant research on parameter deviations between the as-designed CAD and as-built scan
using TLS for the dimensional inspection of MEP constructions is lim- data. These metrics provide useful information for aiding decision-
ited [1,23,24]. Safa et al. in [1] proposed an automated workflow for making during the inspection of a plant. This process is described in
the dimensional quality control of manufactured pipe spools. In [1], the Section 3.2.
authors highlighted that the three important stages in the construction
of piping systems in which the quality must be controlled are: pro-
duction, post-production of the piping spools, and on-site construction. 3.1. Point cloud data processing
However, it is noted that little attention is paid to the on-site dimen-
sional inspection of MEP constructions, which plays a crucial role in 3.1.1. Related works
plant construction and management. Both Nahangi et al. in [23] and As mentioned above, the proposed approach is based on the com-
Anil et al. in [25] concluded deviation analysis as the most effective bination of existing robust and efficient methods. In this study, the two
dimensional inspection method. However, in large-scale constructions, most efficient algorithms in point cloud data processing for as-built
such as an industrial plant, performing only a deviation analysis is not modeling, namely, normal-based region growing [26] for point cloud
typically effective. This is possibly due to the complexity and size of the segmentation and an efficient random sample consensus (RANSAC)
corresponding object. Also, when the scale of the system is very large, [27] for feature recognition and parameter extraction were combined to
including a vast number of components, performing a direct deviation enhance the efficiency and productivity. Also, common preprocessing
analysis is not suitable. Therefore, it is not sufficient for checking the algorithms including octree-based downsampling and point-to-mesh
consistency of an as-built construction. iterative closest points (ICP) for registration were used. A comprehen-
In this paper, a combined method based on a comparison of the sive description of the point cloud processing method is discussed in the
geometric parameters and distance-based deviation analysis is following three sections.

Fig. 1. Overall outline of the proposed method.

45
C.H.P. Nguyen, Y. Choi Automation in Construction 91 (2018) 44–52

Fig. 2. Point cloud registration. (a). Point cloud data and CAD data. (b). Point cloud data registration. (c). Distance analysis for validation.

3.1.2. Preprocessing to the octree cell are replaced by the center of the cell; i.e., the above
[Link]. Point cloud alignment and downsampling. A key initial step in points are compressed into an octree cell. Fig. 3 depicts the utilization
point cloud processing is the alignment of the as-built point cloud of an octree for downsampling.
obtained from a terrestrial laser scanner, to the corresponding as-
designed CAD data in an STL-formatted mesh representation. The
registration process consists of two steps. First, a rough alignment is [Link]. Planar object removal. As mention in [28], planar objects such
performed manually, and then a point-to-mesh ICP algorithm is utilized as floors, walls, and ribs may contain numerous data points, making the
for a fine registration of both the models. Fig. 2 depicts an example of model complex. As a result, the computation time increased and the
the alignment procedure. processing result becomes unstable. Therefore, a removal process
Here, a downsampling is necessary to reduce the size of the data and reduces the size and complexity of the input data to enable further
computational cost because the size of the original scan data is massive. processing. In the proposed approach, the efficient RANSAC described
In this work, octree-based downsampling is performed. First, the input in [27] was used to detect and remove the planar objects. The
point cloud is spatially portioned, and a sufficient octree is computed. convergence error value should be chosen as larger than the tolerance
Based on the choice of the depth level, all the data points corresponding of the laser scanner. In this case, based on the experiments, the
convergence error was set as 0.035 m because the scanner had a

Fig. 3. Octree-based downsampling.

46
C.H.P. Nguyen, Y. Choi Automation in Construction 91 (2018) 44–52

Fig. 4. Planar object removal process. (a). Input point cloud. (b). Planar object detection. (c). Planar object removal.

tolerance of 0.006 m at 100 m. Also, to avoid over-removal, a constraint depends on the estimation of the normal of the points. Typically, two
ensuring that a single planar object is removed only if the point cloud methods are employed, k-nearest neighbors (KNN) or fixed distance
size is larger than a pre-defined threshold is added. The magnitude of neighbors (FDN). The input values of both the KNN and FDN methods
the threshold can be defined based on experiments. Fig. 4 illustrates the are chosen based on the density of the point cloud. In this case, the KNN
planar object removal process. method was applied. Because the data points after downsampling by
the octree compression are organized, it is suitable to select the number
of KNN based on the structure of the square block, as illustrated in
3.1.3. Normal-based region growing segmentation
Fig. 5(b). In this work, the KNN number is set as 24 (a 5 × 5 block), a
After the preprocessing, the input point cloud was segmented using
value obtained from the experiments.
the normal-based region growing algorithm. In this work, this process
The region growing algorithm requires two types of constraints of
was executed following the work described in detail in [26], being re-
local connectivity and smoothness. In this case, a k-dimensional (k-D)
implemented and provided in [29]. In brief, the process is as described
tree was constructed to connect the points, and angle threshold θth,
below:
which stands for the smoothness constraint, was set as 30° to avoid
over-segmentation.
• Step 1: From input point cloud {P}, the point with the lowest cur-
vature is picked, removed from {P},and add to seed point set {S}
• Step 2: For each point in {S}, 3.1.4. Cylinder feature recognition and parameter extraction
○ The k nearest neighbors available in {P} are examined
To detect and extract the parameters of cylindrical objects, the ef-
○ If the angle difference between the neighbor and pi, θ, is smaller
ficient RANSAC method proposed by Schnabel et al. [27] was used due
than θth, this neighbor is removed from {P} and added to region
to its efficient shape detection. Since most piping components, in-
{Ri}
cluding pipelines and joints, are cylindrical, the above algorithm is
○ If the curvature is smaller than cth, is added it to seed point set
applied to obtain only the cylindrical components. To define a cylinder
{S}, and step 2 is repeated
in the Cartesian space, the following seven geometry parameters are
○ If no new point exists that could be selected in {S}, then {Ri} is
used: center axis vector components (dx, dy, dz), coordinate of a single
saved to result segment set {R}.
point on the center axis (p1, p2, p3), and radius of the cylinder (rc).
• Step 3: If all the points in {P} have been tested, result segment set
Applying the efficient RANSAC algorithm proposed in [27], a best-fit
{R} is returned.
cylinder model is obtained, and its parameters are extracted. Distance
(τth) and angle (θth) thresholds are required for the algorithm. Because
Fig. 5(a) displays an illustration of the region growing segmentation
there is no common rule for setting these thresholds, their optimal
method.
values are found from the experiments using the input data.
The robustness of the normal-based region growing algorithm

Fig. 5. Normal-based region growing. (a). Region growing. (b). k-Nearest neighbor selection based on a square block.

47
C.H.P. Nguyen, Y. Choi Automation in Construction 91 (2018) 44–52

Fig. 6. Geometric parameter comparison of the as-built and as-designed models.

3.2. As-built condition assessment 3.2.2. Geometric parameter comparison

In addition to computing the distance deviations, a comparison of
After the recognition process is completed, the corresponding data the geometric parameters of the as-built and as-designed models should
points and parameters of the piping components could be used for the be executed to support the on-site dimensional inspection of industrial
as-built condition assessment via distance-based deviation analysis and plants. To investigate this process, the cylindrical parameters of each
geometric parameter comparison. Since it could be very difficult to detected piping component (direction axis, radius) that were extracted
identify an entire pipe along its axis due to the appearance of occlusion from the point cloud are compared to the as-designed model. The
due to the line of sight, the most practical solution is to perform the output aids in detecting the orientation deviations of the pipes from the
analysis on each recognized piping segment. analysis of the direction axis vector (orientation error) and radius dif-
ferences (radius error) relative to the corresponding as-designed data.
Fig. 6 depicts the comparison of the geometric parameters of the as-
3.2.1. Distance-based deviation analysis built and as-designed models.
Deviation analysis is considered as the most effective method for
dimensional inspection [1,25]. Subsequent to the completion of the 4. Experimental study
recognition process, the point cloud corresponding to each piping
component is segmented. To conduct the distance deviation analysis, 4.1. Experimental data
the computation of the MDE, which is the average distance deviation of
all the data points in the as-built model relative to the as-designed The accuracy and robustness of the proposed method were eval-
model, is performed. This index can be considered as the “global” dis- uated using two sets of data. The CAD data were taken from a design of
tance deviation between the as-built model and as-designed model. In a ship in STL format, whereas the point cloud data were obtained by a
addition to the MDE, the PDE, which represents the “local” distance Leica® P20 Laser Scanner in PTX format, and then converted to PCD
deviation of each data point, is computed. The global distance deviation format, captured from the same as-built object. The detailed specifi-
analysis from the MDE provides an overview of the current geometry cations of the laser sensor are presented in Table 1. In addition, to
match of each pipe, whereas the local distance deviation information of validate the ability of the proposed method to detect the errors in
each data point from the computation of the PDE yields more specific geometry, the position between the as-designed and as-built models
information on the dimensional accuracy of the piping system at a were varied and the resulting error values were observed. The experi-
specific location. The experimental results of the deviation analysis are mental data are listed in Table 2 and presented in Fig. 7.
discussed in detail in Section 4.2. To compute the MDE and PDE, the
algorithm for computing the distance between points and triangles in
three dimensions as originally described in [30], was used. The details Table 1
Laser scanner specifications used in the experiment.
of the process are outlined in Algorithm 1.
Algorithm 1. Mean distance error and point distance error No. Properties Value

computation. 1 Measurement type Time-of-flight

Mean Distance Error and Point Distance Error Computation 2 Accuracy 6 mm at 100 m
Given: Point Cloud {P}, Mesh {M} 3 Linearity error ≤ 1 mm
Initialization: Mean Distance Error = 0, Point Distance Error {PDE} 4 Angular accuracy 8″ (horizontal and vertical)
5 Beam divergence 0.2 mrad
For each ∈
6 Point density 6.3 mm at 10 m
=∞
For each ∈
Compute the closest distance from to :
Table 2
If < : = Details of the experimental data.
End If
End for CAD data (STL) Scan data (PCD)
Save
No. of Size (kB) Before downsampling After downsampling
End for
triangles
∑
No. of points Size (kB) No. of points Size (kB)

Return 163,311 7975 ~60,000,000 ~1351000 ~1600000 ~38,000

48
C.H.P. Nguyen, Y. Choi Automation in Construction 91 (2018) 44–52

Fig. 7. Experimental data. (a). CAD data in STL format. (b). Point cloud data.

Fig. 8. Point cloud data processing result. (a). Input point cloud. (b). Planar object removal. (c) Normal-based region growing. (d). RANSAC cylinder detection.

Table 3 However, some components remain unrecognized despite the high

Execution time for point cloud data processing. recognition rate. The two primary reasons for this are the missing input
data due to occlusion and the noisy data generated by the diffusion of
Hardware capacity Intel®Core™i7–6700 CPU @3.40 GHz, 4 cores, 8 threads,
16.0 GB RAM
the laser light of the components such as a thermal piping system. Fig. 9
displays the two cases that lead to unrecognized or incorrectly-re-
Process Time (s) cognized components.
The advantage of the proposed method is clearly demonstrated.
Planar removal 316.447
Region growing 546.489
Here, the region growing step performs as a ‘divide’ step, whereas the
Cylinder detection 796.769 cylinder fitting by RANSAC fulfills the ‘conquering’ role. The region
growing method could perform ‘under-segmentation,’ which would not
provide ‘relevant’ point cloud segments but simplify them sufficiently
The algorithms described in this study were implemented with the for the RANSAC cylinder fitting algorithm to exhibit an efficient per-
C++ programming language, using three supporting libraries: Point formance. Hence, subsequently, a geometric comparison could be per-
Cloud Library (PCL) [29], Computational Geometry Algorithm Library formed for each recognized segment that not be affected by the oc-
(CGAL) [31], and Geometric Tool [30]. Open source software package clusion. Fig. 10 depicts an example case of the occurrence of occlusion
CloudCompare® [32] was used to support the downsampling and vi- appears, and its solution by the above-mentioned approach. It can be
sualization of the results. seen that the input data for the RANSAC fitting algorithm become much
more simplistic, and more precise and robust results are achieved. As
4.2. Results and discussion mentioned above, there is no specific method that can perform effi-
ciently in the case of unstructured data; however, by combining the
4.2.1. Point cloud processing different methods into an appropriate process, the disadvantages of
Fig. 8 displays the results of the point cloud data processing. The each method could be diminished and their performances be enhanced.
figure shows that most of the piping objects are recognized and seg- Also, the noisy data problem could be ameliorated traditionally by in-
mented. The evaluation of the robustness and efficiency of the point creasing the distance and angle thresholds of the RANSAC algorithm.
cloud data processing method is based on the pipe recognition rate and The threshold values are carefully chosen to avoid incorrect detection
with respect to time. Overall, the recognition rate is high and stable, or instability of the algorithm. To deal with the occlusion problem, a
with a 91.60% recognition rate (i.e., 33/36 objects are recognized) and multi-position scanning technique could be chosen as a solution to
a reasonable execution time (Table 3). capture the entire size of the piping components. The inclusion of an

49
C.H.P. Nguyen, Y. Choi Automation in Construction 91 (2018) 44–52

Fig. 9. Two reasons causing components to not be recognized. (a). Occlusion due to the line of sight. (b). Diffusion of laser light.

Fig. 10. Combined method in the case of missing data. (a). Input point cloud. (b). Normal-based region growing. (d). RANSAC cylinder detection.

Fig. 11. Experimental region. (a) Point cloud scan data. (b) CAD data in STL format.

analysis of the accuracy of the point cloud registration process could depicts an example of the PDE computation for a piping component
also help achieve a more robust result. chosen from the sample list. The chosen pipes are marked by rectan-
In general, except in certain unrecognized cases, the point cloud gular boxes and numbered from 1 to 6. Both Table 4 and Fig. 12 present
processing technique described here is sufficiently robust and reliable the error of the as-built model when it is correctly constructed; clearly,
for use in a facility inspection. the errors are small and acceptable. The PDE analysis, as shown in
Fig. 12, also indicates the occlusion regions, providing useful in-
4.2.2. As-built condition assessment formation to inspectors about the status and quality of the input data. If
Figs. 11 and 12 and Table 4 present an example of the as-built di- there are missing data in the regions of interest for the inspection, the
mensional assessment process. Fig. 11 displays the sample region that is inspectors would be advised to adjust the scanner position to obtain the
assessed in the experiment and reported in Table 4, whereas Fig. 12 required data.

50
C.H.P. Nguyen, Y. Choi Automation in Construction 91 (2018) 44–52

Fig. 12. Point distance error analysis. (a) Point cloud data and CAD data. (b) PDE analysis result.

Table 4 herein for the post-construction inspection or subsequently, for periodic

MDE and parameter comparison result. inspection during maintenance. In addition, most steps in the proposed
approach are autonomous. Thus, the approach should decrease the time
No. Orientation error (°) Radius error (m) MDE (m)
and effort spent for the inspection in comparison with the traditional
1 0.0438 0.00170 0.0144 inspection methods.
2 0.3840 0.00010 0.0148 However, there are several factors that can affect the inspection
3 0.1999 0.00372 0.0137 result when using the proposed method. First, the robustness of the
4 0.1477 0.01134 0.0155
5 0.2296 0.00309 0.0143
result strongly depends on the quality of the input point cloud, which
6 0.2157 0.00160 0.0153 may include noisy and missing data. As mentioned above, these could
lead to unrecognition or incorrect recognition of the piping compo-
nents, which could affect the subsequent comparison process. Second,
as the accuracy of the comparison step also depends on the accuracy of
the point cloud processing step, errors could be found in the alignment
step using the ICP algorithm or during the parameter extraction process
using the RANSAC algorithm. In practical applications, an allowable
error should be established that includes the noise level of the input
data as well as the possible uncertainties in the point cloud processing
algorithms.

5. Conclusion

In this paper, an improved method was proposed for the on-site

dimensional inspection of as-built industrial plant piping systems by
using a terrestrial laser scan was proposed. The method includes two
stages: point cloud processing and a comparison of the CAD design data
with the as-built scan data. In conclusion, a combined method has de-
monstrated the benefit when the input data is uncertain, since it en-
hances the advantages of each algorithm and diminishes their dis-
Fig. 13. Variation in the MDE with the variation in the relative position of the as-built
advantages. As any deviation in geometry between the as-built and as-
and as-designed models.
designed models, such as in the position, orientation, or radius, could
indicate an inconsistency of the piping system in industrial plants,
Although the proposed method performs well when the construction leading to improper operation. Analysis of such geometric deviations
is correct, the ability to capture the deviations from the design of the as- provides useful information for their inspection. Distance-based de-
built geometry using the proposed method also requires validation. viation analysis and geometric parameter comparison can be used to
Therefore, further experiments were conducted by varying the angle support decision-making in plant inspection in both post-construction,
difference, α, (Fig. 13) between the direction axis vectors of the as-built as well as during periodic maintenance periods. Since the proposed
and as-designed models, from 0° to 7.5° in steps of 2.5°, to generate the approach is a point-cloud-based method, the robustness of the inspec-
incorrect cases. Fig. 13 shows the variation in the MDE when the angle tion result depends strongly on the quality of the input point cloud data
between the as-built and as-designed model varies. Clearly, the MDE obtained from the TLS devices as well as the accuracy of the point cloud
increases along with the increase in the angle difference, indicating that data processing computations. Future studies will detail the im-
the MDE index could be used to detect the difference between the as- plementation of a completely automatic inspection framework as well
built and as-design models. This simple test has demonstrated the as expand upon the current work to cover more variants of the struc-
ability of our proposed method to capture the relevant deviation suc- tural and facility components of an industrial plant.
cessfully.
Any difference in the geometry between the as-built and as-designed Acknowledgment
models would alert the inspector to the inconsistency of the as-built
industrial piping system, caused perhaps by deformation due to over- This work was in part supported by the Asian Young Scientist
load or poor workmanship. Therefore, it is key to utilize the methods Scholarship (CAYSS) program of Chung-Ang University in 2017, the

51
C.H.P. Nguyen, Y. Choi Automation in Construction 91 (2018) 44–52

Basic Science Research Program through the National Research [15] C. Wang, Y.K. Cho, C. Kim, Automatic Bim component extraction from point clouds
Foundation of Korea (NRF), funded by the Ministry of Education (NRF- of existing buildings for sustainability applications, Autom. Constr. 56 (2015) 1–13,
[Link]
2015R1D1A1A01060486), and the Korea Institute for Advancement of [16] F. de Asís López, C. Ordóñez, J. Roca-Pardiñas, S. García-Cortés, A statistical
Technology (KIAT) grant funded by the Ministry of Trade, Industry and method for geometry inspection from point clouds, Appl. Math. Comput. 242
Energy (MOTIE) of Korea (No. N0001075). (2014) 562–568, [Link]
[17] S.J. Gordon, D.D. Lichti, Modeling terrestrial laser scanner data for precise struc-
tural deformation measurement, J. Surv. Eng. 133 (2) (2007) 72–80, [Link]
References org/10.1061/(ASCE)0733-9453(2007)133:2(72).
[18] W. Wang, W. Zhao, L. Huang, V. Vimarlund, Z. Wang, Applications of terrestrial
laser scanning for tunnels: a review, J. Traffic Transport. Eng. (English Edition) 1
[1] M. Safa, A. Shahi, M. Nahangi, C. Haas, H. Noori, Automating measurement process
(5) (2014) 325–337, [Link]
to improve quality management for piping fabrication, Structure 3 (2015) 71–80
[19] F. Bosche, C.T. Haas, B. Akinci, Automated recognition of 3d cad objects in site laser
[Link]
scans for project 3d status visualization and performance control, J. Comput. Civ.
[2] H. Son, F. Bosche, C. Kim, As-built data acquisition and its use in production
Eng. 23 (6) (2009) 311–318, [Link]
monitoring and automated layout of civil infrastructure: a survey, Adv. Eng. Inform.
23:6(311).
29 (2) (2015) 172–183, [Link]
[20] J. Chai, H.-L. Chi, X. Wang, C. Wu, K.H. Jung, J.M. Lee, Automatic as-built modeling
[3] T. Chaperon, F. Goulette, Extracting Cylinders in Full 3d Data Using a Random
for concurrent progress tracking of plant construction based on laser scanning,
Sampling Method and the Gaussian Image, Vision Modeling and Visualization
Concurr. Eng. 24 (4) (2016) 369–380, [Link]
Conference 2001 (VMV-01), Stuttgart, Germany, [Link]
1063293X16670449.
hal-01259641, (2001).
[21] C. Kim, H. Son, C. Kim, Automated construction progress measurement using a 4d
[4] H. Son, C. Kim, C. Kim, Fully automated as-built 3d pipeline extraction method
building information model and 3d data, Autom. Constr. 31 (2013) 75–82, http://
from laser-scanned data based on curvature computation, J. Comput. Civ. Eng. 29
[Link]/10.1016/[Link].2012.11.041.
(4) (2015), [Link]
[22] N.J. Shih, S.T. Huang, 3d scan information management system for construction
[5] H. Son, C. Kim, C. Kim, 3d reconstruction of as-built industrial instrumentation
management, J. Constr. Eng. Manag. 132 (2) (2006) 134–142, [Link]
models from laser-scan data and a 3d cad database based on prior knowledge,
10.1061/(ASCE)0733-9364(2006)132:2(134).
Autom. Constr. 49 ( (2015) 193–200, [Link]
[23] M. Nahangi, C.T. Haas, Automated 3d compliance checking in pipe spool fabrica-
007.
tion, Adv. Eng. Inform. 28 (4) (2014) 360–369, [Link]
[6] J. Lee, H. Son, C. Kim, C. Kim, Skeleton-based 3d reconstruction of as-built pipelines
2014.04.001.
from laser-scan data, Autom. Constr. 35 (2013) 199–207, [Link]
[24] F. Bosche, Automated recognition of 3d cad model objects in laser scans and cal-
1016/[Link].2013.05.009.
culation of as-built dimensions for dimensional compliance control in construction,
[7] K. Kawashima, S. Kanai, H. Date, As-built modeling of piping system from terrestrial
Adv. Eng. Inform. 24 (1) (2010) 107–118, [Link]
laser-scanned point clouds using normal-based region growing, J. Comput. Des.
08.006.
Eng. 1 (1) (2014) 13–26, [Link]
[25] E.B. Anil, P. Tang, B. Akinci, D. Huber, Deviation analysis method for the assess-
[8] K. Kawashima, S. Kanai, H. Date, Automatic recognition of piping system from laser
ment of the quality of the as-is building information models generated from point
scanned point clouds using normal-based region growing, ISPRS Ann. Photogramm.
cloud data, Autom. Constr. 35 (2013) 507–516, [Link]
Remote Sens. Spatial Inf. Sci. II-5/W2 (2013) 121–126 [Link]
autcon.2013.06.003.
isprsannals-II-5-W2-121-2013.
[26] T. Rabbani, F.A. van den Heuvel, G. Vosselmann, Segmentation of Point Clouds
[9] A.K. Patil, P. Holi, S.K. Lee, Y.H. Chai, An adaptive approach for the reconstruction
Using Smoothness Constraint, IEVM06 (2006) [Link]
and modeling of as-built 3d pipelines from point clouds, Autom. Constr. 75 (2017)
XXXVI/part5/paper/RABB_639.pdf.
65–78, [Link]
[27] R. Schnabel, R. Wahl, R. Klein, Efficient Ransac for point-cloud shape detection,
[10] Y.J. Liu, J.B. Zhang, J.C. Hou, J.C. Ren, W.Q. Tang, Cylinder detection in large-scale
Comput. Graphics Forum 26 (2) (2007) 214–226, [Link]
point cloud of pipeline plant, IEEE Trans. Vis. Comput. Graph. 19 (10) (2013)
1467-8659.2007.01016.x.
1700–1707, [Link]
[28] T. Czerniawski, M. Nahangi, S. Walbridge, C. Haas, Automated removal of planar
[11] T. Czerniawski, M. Nahangi, C. Haas, S. Walbridge, Pipe spool recognition in
clutter from 3d point clouds for improving industrial object recognition,
cluttered point clouds using a curvature-based shape descriptor, Autom. Constr. 71
Proceedings of the 33rd International Symposium on Automation and Robotics in
(Part 2) (2016) 346–358, [Link]
Construction (ISARC), Auburn, AL, USA, 2016, pp. 357–365 [Link]
[12] F. Bosché, M. Ahmed, Y. Turkan, C.T. Haas, R. Haas, The value of integrating scan-
22260/ISARC2016/0044.
to-Bim and scan-Vs-Bim techniques for construction monitoring using laser scan-
[29] The Point Cloud Library Documentation, [Link]
ning and Bim: the case of cylindrical Mep components, Autom. Constr. 49 ( (2015)
documentation/, (2018) (Accessed: Feb 10, 2018).
201–213, [Link]
[30] The Geometric Tool Documentation, [Link]
[13] A. Dimitrov, M. Golparvar-Fard, Segmentation of building point cloud models in-
Documentation/[Link], (2018) (Accessed: Feb 10, 2018).
cluding detailed architectural/structural features and Mep systems, Autom. Constr.
[31] The Computational Geometry Algorithm Library Documentation, [Link]
51 (2015) 32–45, [Link]
org/latest/Manual/[Link], (2018) (Accessed: Feb 10, 2018).
[14] H.S. Park, H.M. Lee, H. Adeli, I. Lee, A new approach for health monitoring of
[32] The Cloudcompare Documentation, [Link]
structures: terrestrial laser scanning, J. Comput. Aided Civ. Infrastruct. Eng. 22 (1)
[Link]?title=Main_Page, (2018) (Accessed: Feb 18, 2018).
(2007) 19–30, [Link]

52