remote sensing

Article
Structure from Motion Point Clouds for
Structural Monitoring
Fausto Mistretta, Giannina Sanna , Flavio Stochino * and Giuseppina Vacca
Department of Civil, Environmental Engineering and Architecture, University of Cagliari, 09123 Cagliari, Italy
* Correspondence: fstochino@unica.it; Tel.: +39-070-675-5115




Received: 3 June 2019; Accepted: 15 August 2019; Published: 20 August 2019


Abstract: Dense point clouds acquired from Terrestrial Laser Scanners (TLS) have proved to be
effective for structural deformation assessment. In the last decade, many researchers have defined
methodology and workflow in order to compare different point clouds, with respect to each other
or to a known model, assessing the potentialities and limits of this technique. Currently, dense
point clouds can be obtained by Close-Range Photogrammetry (CRP) based on a Structure from
Motion (SfM) algorithm. This work reports on a comparison between the TLS technique and the
Close-Range Photogrammetry using the Structure from Motion algorithm. The analysis of two
Reinforced Concrete (RC) beams tested under four-points bending loading is presented. In order
to measure displacement distributions, point clouds at different beam loading states were acquired
and compared. A description of the instrumentation used and the experimental environment, along
with a comprehensive report on the calculations and results obtained is reported. Two kinds of point
clouds comparison were investigated: Mesh to mesh and modeling with geometric primitives. The
comparison between the mesh to mesh (m2m) approach and the modeling (m) one showed that the
latter leads to significantly better results for both TLS and CRP. The results obtained with the TLS for
both m2m and m methodologies present a Root Mean Square (RMS) levels below 1 mm, while the
CRP method yields to an RMS level of a few millimeters for m2m, and of 1 mm for m.

Keywords: structural health monitoring; structure from motion; terrestrial laser scanner; close-range
photogrammetry; point clouds modeling

1. Introduction
Structural Health Monitoring is one of the main objectives of engineering surveys [1] and it is
particularly important for buildings, bridges, dams, or other infrastructures subjected to extreme
events such as earthquakes, hurricane, flooding, fire [2,3], or due to natural aging [4,5]. For this reason,
periodic monitoring of the structural condition is necessary in order to ensure and maintain the safety
and functionality of buildings and infrastructure [6–14]. Sometimes it is sufficient to measure the
eigenfrequency variations along the lifecycle of the structure. In other cases, it is important to control
and record displacements and deformations even if it requires quite complex monitoring systems.
Geomatic techniques often represent an effective approach and the changes in time of target point
coordinates is the core of the method. Thus, knowledge of the typology, characteristics, and scale of
the structural deformations is essential for an accurate measurement of permanent damage [15].
Currently, the geometrical survey of objects (buildings, bridges, and/or other structures) can
be easily obtained with dense point clouds. Also, structural monitoring may be developed using
this approach. Until a few years ago, Terrestrial Laser Scanner (TLS) was the only technique widely
used to generate dense point clouds. Interesting reports on point clouds from TLS assessing changes
and deformations can be found in [16–18]. Deformation analysis can be developed at the level of

Remote Sens. 2019, 11, 1940; doi:10.3390/rs11161940 www.mdpi.com/journal/remotesensing

Remote Sens. 2019, 11, 1940 2 of 20

single-point location (point-wise [19]) or at the object level (see [20,21]). In [22], in order to identify
seismic-induced building deformations, the notion of morphological maps is introduced.
The measure of deformation is often obtained through the fusion of data collected by different
sensors. In [23–25], the laser scanner data are merged with data obtained through other sensors;
in [26,27], laser scanner data, acquired from different platforms, are combined together.
Structural monitoring based on TLS has been studied in detail in the literature. In [28], damage
detection and volume change analysis for a full-scale structural test are obtained using a terrestrial
laser scanner. In [22] and [29], the deformations of several ancient towers are analyzed by TLS and the
Finite Element Method (FEM). The structural health condition of historical arch bridges is assessed
by FEM and a combined use of TLS and Close-Range Photogrammetry (CRP) in [30]. Additionally,
monitoring of moving rotor blades of windmills is reported in [31]. In [32], dam deformations are
measured using TLS.
The use of point clouds to monitor or control structural deformations involves three challenges.
First, the difficulty in managing big data. Secondly, it is impossible to scan the same point at different
epochs and finally, the data itself is affected by the instrument’s noise [33]. The first problem is solved
by segmentation, the second through the insertion of specific targets or cloud-to-cloud comparison
solutions, while the third problem may be mitigated through interpolation models [9]. In this work,
in order to compare point clouds at different epochs, the latter two solutions were tested: Cloud to
cloud comparisons and interpolation models. These two methodologies have already been discussed
in different papers, especially with the use of TLS [10,11,22,34,35]. The “point to mesh” and “mesh
to mesh” methods have been applied and studied in [36]. Other authors have developed a model
of the surface of the deformed object. They proved that this approach is capable of assessing the
object’s behavior with a lower influence of measurement noise: [22,37]. In [38,39], a summary of the
TLS performance for several objects is also presented.
The behavior of a slender beam is often described by the Euler–Bernoulli model. In this approach,
the static deformed shape can be obtained from a 2nd order linear ordinary differential equation, which
enforces the equilibrium condition, obtaining a simple relationship between the bending moment and
the curvature. The solution of this equation is a polynomial function whose degree depends on the
loading conditions.
In [38], Tsakiri presents an overview of the different functions modeling several kinds of objects
measured with TLS. In particular, he reports the case of a wooden beam where it is proved that the
best function representing beam deflection is a third-degree polynomial. Also, in [39], the case of a
wooden beam modeled with a third-degree polynomial is reported, while in [35], the model of a steel
beam is obtained using a second-degree polynomial.
Dense point clouds obtained by photogrammetry [40] have become useful in monitoring thanks
to the development of Structure from Motion (SfM) algorithms (see [41–44]).
The state-of-the-art of dense image-matching is discussed in [45], where a comparison of the
available software is also presented. Most of the published studies (see [45–48]) estimate the accuracy
of the point cloud through the use of targets or through sections extracted from the cloud and compared
with the design ones. Some interesting examples of archeology and cultural heritage preservation
based on point cloud surveys are presented in [49–51]. In [52], a critical review of the developments
in automated image processing is discussed. It is also possible to track 2D displacements in rectified
images with acceptable accuracy, without going through 3D reconstruction, see [53,54].
In this work, the models traditionally used for deformation monitoring with TLS point clouds
were applied to dense point clouds obtained from CRP.
The main target of this research was to evaluate the accuracy and precision that can be obtained
by CRP point clouds measuring structural deformations. The case study involved two Reinforced
Concrete (RC) beams, tested under four-points bending loading conditions. The displacements were
measured using a standard Displacement Transducer (DT), TLS, and CRP. Preliminary results obtained
with CRP on the first RC beam were presented in [55].
 In this paper, measurement techniques and data processing of the structural deformations are
shown, and a comparison between the TLS technique and the CRP is also presented.
The validation of the results was performed using two metrological instruments: The Leica
AT402 Laser Tracker (LT) and the AICON MoveInspect DPA system with a Nikon D3x metric camera
Remote Sens. 2019, 11, 1940 3 of 20
(DPA).
The paper is organized as follows: Section 2 presents the instruments and technologies used in
the load tests
In this of the
paper, beams. Section
measurement 3 reportsand
techniques thedata
results of the measurements
processing of the structural with TLS, CRP, and
deformations are
metrological
shown, instruments.between
and a comparison The comparisons between
the TLS technique andthe thetwo
CRPtechniques and the metrological
is also presented.
measurements
The validation are reported in Section
of the results 3. Finally,using
was performed the discussion of the results
two metrological is developed
instruments: in Section
The Leica AT402
4, while
Laser conclusions
Tracker (LT) and and
theremarks
AICONare drawn in Section
MoveInspect 5.
DPA system with a Nikon D3x metric camera (DPA).
The paper is organized as follows: Section 2 presents the instruments and technologies used
2. Materials and Methods
in the load tests of the beams. Section 3 reports the results of the measurements with TLS, CRP,
and metrological
In order to test instruments.
the geomatic The comparisons
measurement between
methods the two techniques
presented and the metrological
in the introduction, several four-
measurements
points bendingare reported
tests in Section 3.
were developed onFinally, the discussion
two Reinforced of the(RC)
Concrete results is developed
beams in Section
(labeled beam#1 and4,
while conclusions and remarks are drawn in Section 5.
beam#2) in the Materials Laboratory of the University of Cagliari Department of Civil, Environmental
Engineering and Architecture. The tests were developed in three different steps: beam#1 first load
2. Materials and Methods
condition; beam#1 second load condition, beam#2. The adopted concrete was characterized by a cubic
In order strength
compressive to test the geomatic
of 30 measurement
MPa, while methods
reinforcement steel presented
presented ainnominal
the introduction, several
450 MPa yielding
four-points
strength. Thebending
beamstests were developed
cross-section, on two Reinforced
reinforcement Concrete
distribution, (RC)
and test beamscan
scheme (labeled beam#1
be seen and
in Figure
beam#2)
1. in the Materials Laboratory of the University of Cagliari Department of Civil, Environmental
Engineering
The loadand wasArchitecture.
applied by aThe tests were
hydraulic jackdeveloped in threeincremental
with quasi-static different steps:
stepsbeam#1
from 0 first
to 68load
kN
condition;
and for eachbeam#1
step, second
the beam load condition, beam#2.
displacements The adopted
were measured usingconcrete wasand
both TLS characterized by a cubic
CRP. The maximum
compressive
load of 68 kN strength of 30 MPa,
was selected while
in order toreinforcement steel presented
have crack openings withoutareaching
nominal the450 beam
MPa yielding
collapse
strength.
condition.The beams
Details of cross-section, reinforcement
the load conditions distribution,
1–2 for the two beams and
aretest scheme in
presented canTables
be seen in Figure 1.
1–3.

35 35
Hydraulic Jack
Loading Beam

RC Beam 20

15
DT 1 DT 2 Stirrups
1Ø8/10 15x10 L=74
125 84
25 250 25

Longitudinal bars 296

14 2Ø14 L=324
14

14 296 14
3Ø14 L=324

Figure 1. Reinforcement distribution, beam dimensions, and test setup (bar diameters are in mm, all
Figure 1. all
other
other measures
measures are
are in
in cm).
cm).

The load was

Usually, applieddisplacements
structural by a hydraulic are jack measured
with quasi-static
using incremental
displacement steps from 0 to 68
transducers kN and
(DT) that
for each step, the beam displacements were measured using both TLS and CRP. The maximum
convert rectilinear mechanical motion into a variable electric signal that can be digitally recorded, load of
68 kN the
while wasTLS
selected in order
and CRP to have
geomatic crack openings
techniques producewithout reaching
coordinates thereference
in a 3D beam collapse
system.condition.
Details of the load conditions 1–2 for the two beams are presented in Tables 1–3.
A DT was applied under the beam midspan section. Its nominal displacement is 100 mm,
Usually,
nominal structural
sensitivity displacements
2 mV/V, sensitivityare measured
tolerance usingmeasure
± 0.1%, displacement transducers
resolution (DT)
1 µm. Its that convert
position is also
rectilinear
represented mechanical
in Figure 1.motion into a variable electric signal that can be digitally recorded, while the
TLS and CRP geomatic techniques produce coordinates in a 3D reference system.
A DT was applied under the beam midspan section. Its nominal displacement is 100 mm, nominal
sensitivity 2 mV/V, sensitivity tolerance ± 0.1%, measure resolution 1 µm. Its position is also represented
in Figure 1.
Remote Sens. 2019, 11, 1940 4 of 20

Table 1. Plan of the performed measurements beam#1 first load condition (LC1).

Test Load (kN) Measurement Time (day 1) DT LT DPA TLS CRP

√ √ √ √ √
1 0 10.53
√ √ √ √ √
2 11 11.11
√ √ √ √ √
3 26 11.36
√ √ √ √ √
4 40 12.01
√ √ √ √ √
5 54 12.32
√ √ √ √
6 68 13.08

Table 2. Plan of the performed measurements beam#1 second load condition (LC2).

Test Load (kN) Measurement Time (day 2) DT LT DPA TLS CRP

√ √ √ √
1 0 10.39
√ √ √ √
2 22 11.27
√ √ √ √
3 48 11.50

Table 3. Plan of the performed measurements beam#2.

Test Load (kN) Measurement Time (day 3) DT LT DPA TLS CRP

√ √ √ √
1 0 10.26
√ √ √ √
2 19 11.02
√ √ √ √
3 39 11.29
√ √ √
4 0 11.48
√ √ √
5 19 12.07
√ √ √
6 39 12.15
√ √ √ √ √
7 68 12.26

For the TLS technique, the Faro Focus 3D Terrestrial Laser Scanner was used. It is a compact
scanner characterized by an operative range that varies between 0.6 and 120 m, with a ranging error
of ±2 mm for scanner–object distances between 10 and 25 m. The scans were processed using the
JRC Reconstructor software v. 3.1.0 by Gexcel Ltd. For the CRP, two digital cameras were used. The
first camera was a Canon Powershot S110 with a 1/1.70 ’ sensor (1.9 mm pixel dimension), Field of
View (FoV) 94.2g , and a 12 Megapixel resolution; zoom 5×: 5.2 (G) - 26.0 (T) mm (equivalent to 35 mm:
24 (G) - 120 (T) mm on a 24 mm film); the output data formats are Exif 2.3 (JPEG) and RAW (CR2
Canon original). The second digital camera was a Canon EOS M3 with a sensor CMOS 22.3 × 14.9 mm
(3.7 mm pixel dimension), Field of View (FoV) 81.5g , and a 24.2 Megapixel resolution; objective EF-S
18–55 mm; the output data formats are Exif 2.3 (JPEG) and RAW (CR2 Canon original).
The images were processed using the Photoscan software by Agisoft [56,57], which implements
the SfM method [58–60].
In order to validate the results and fix a vertical reference frame for photogrammetric measurements,
two metrological instrumentations were used:

• The Leica AT402 Laser Tracker (LT);

• The AICON MoveInspect DPA system with a Nikon D3x metric camera (DPA).

The first one is a metrological instrument characterized by an uncertainty in the measurement of

coordinates (Uxyz ) of +/− 15 µm + 6 µm/m (Maximum Permissible Error). The second has a nominal
measurement accuracy of 2 µm + 5 µm/m (RMS), or 3 µm + 7 µm/m (3σ).
Both systems obtain the coordinates of targeted points with a very high precision and all the
comparisons between the examined techniques (TLS and CRP) have been performed with respect to
those targets.
Figure 2 presents a view of the experimental setup.
Remote Sens. 2019, 11, 1940 5 of 20
Remote Sens. 2018, 10, x FOR PEER REVIEW 5 of 20
Remote Sens. 2018, 10, x FOR PEER REVIEW 5 of 20

Figure
Figure 2. 2.Experimental
Experimentalsetup:
setup: Beam
Beam and
and back-up
back-upframe,
frame,Laser
LaserTracker
Tracker(LT)
(LT)pointing
pointingat at
targets.
targets.
Figure 2. Experimental setup: Beam and back-up frame, Laser Tracker (LT) pointing at targets.
2.1.2.1.
Measurements
MeasurementsMethods
Methods
2.1. Measurements Methods
Although
Although bothTLS
both TLSand
andCRP
CRPmeasurement
measurement techniques
techniquesprovide
provide3D 3Dpositions
positionsofof points,
points,structural
structural
Although
displacements
displacements both
forfor
thetheTLS and CRP
considered
considered measurement
casecase techniques
are expected
are expected provide
within
within 3D positions
the vertical
the vertical of
direction,
direction, points, structural
yielding
yielding a 1D
a 1D problem.
displacements
Thus, for
an the considered
important issue case
is the are expected
materialization within
of thethe vertical
direction of direction,
Thus, an important issue is the materialization of the direction of interest. In the TLS, with the with
problem. interest. Inyielding
the TLS, a 1D
biaxial
problem.
the biaxial
compensator Thus, an important
compensator
switched on, the issue
switched is on,
the materialization
instrumental thereference
instrumentalof the
system direction
reference
includes of interest.
system
the In
includes
vertical the
theTLS,
direction, withthe
vertical
but
the biaxialbut
direction, compensator
the CRP switched
method needson,scaling
the instrumental
and an reference
external system
reference includes
system with the
the vertical
CRP method needs scaling and an external reference system with the vertical direction as one of the
direction,
direction asbut
onethe CRPaxes.
offor
the method needs scaling
The problem and
for because
CRP an external
remains a 3D onereference
becausesystem
the does withdisplacement
sought the vertical
axes. The problem CRP remains a 3D one the sought displacement not necessarily lie
direction as one of the axes. The problem for CRP remains a 3D one because the
does not necessarily lie on the longitudinal beam plane. In order to fix a common reference frame sought displacement
on the longitudinal beam plane. In order to fix a common reference frame containing one vertical axis,
does not necessarily
containing one vertical lieaxis,
on the
six longitudinal
special aluminumbeam targets
plane. (Figures
In order 3toandfix 4),
a common
named FT#,reference
treatedframe
with
six special aluminum targets (Figures 3 and 4), named FT#, treated with retro-reflective paint, and
containing one vertical
retro-reflective paint, and axis, six special
whose centeraluminum
coordinates targets
could(Figures 3 and 4),
be measured withnamed FT#, treated
LT, DPA system,withand
whose center coordinates
retro-reflective could be center
measured with LT,could
DPA system, and CRP,
with were fixedsystem,
in the area
and of
CRP, were fixedpaint,in theandareawhose
of the laboratorycoordinates be measured
that was not subjected to variable LT, DPA
loads.
theCRP,
laboratory thatinwas
were fixed thenot
areasubjected to variable
of the laboratory thatloads.
was not subjected to variable loads.

Figure 3. Circular targets whose center coordinates were measured with both LT, inserting the prism
Figure
Figure 3. 3.
into the Circular
central
Circular targetsand
housing,
targets whose center coordinates
coordinates
photogrammetric
whose center were
systems measured
DPA
were (AICON
measured with both LT,
LT,inserting
MoveInspect
with both DPA thethe
system
inserting prism
with
prism
into
thethe
a Nikon
into central housing,
D3x metric
central andphotogrammetric
camera)
housing, and photogrammetric
and systems
CRP (Close-Range DPA (AICON
(AICON MoveInspect
Photogrammetry).
systems DPA MoveInspectDPA DPAsystem
systemwith
with a
a Nikon
Nikon D3xD3x metric
metric camera)
camera) andand
CRP CRP (Close-Range
(Close-Range Photogrammetry).
Photogrammetry).
Remote Sens. 2019, 11, 1940 6 of 20
Remote Sens. 2018, 10, x FOR PEER REVIEW 6 of 20

Figure
Figure 4.
4. FT
FTTargets
Targets and
and LT
LT prism (P) location in the laboratory.
laboratory.

targets were measured multiple times to check the stability of both the
The six FT targets the reference
reference frame
and the LT instrument.
instrument.
In order
order to tomeasure
measurethe thebeam
beam displacements,
displacements, fivefive
laser tracker
laser prisms
tracker (P#)(P#)
prisms werewere
fixedfixed
on theonfront
the
side—corresponding
front side—corresponding to thetocompressed part ofpart
the compressed the cross-section—to avoid triggering
of the cross-section—to cracks during
avoid triggering cracks
the test.the
during Thetest.
prisms
The(P#) were(P#)
prisms placed
wereatplaced
both ends of the
at both beam
ends of (P1
the and
beamP5),
(P1onandthe P5),
midspan
on thesection
midspan (P3)
and at 1/3 and 2/3 of the beam span (P2 and P4). Their positions are also depicted
section (P3) and at 1/3 and 2/3 of the beam span (P2 and P4). Their positions are also depicted in in Figure 4.
FigureA large
4. set of photogrammetric retro-reflecting targets were spread all over the measurement area,
according
A large to the following
set of list:
photogrammetric retro-reflecting targets were spread all over the measurement
area, according to the following list:
• 78 CTs (Coded Targets), nine of which are part of the AICON reference system (calibrated cross
• 78 CTs
and (Coded Targets), nine of which are part of the AICON reference system (calibrated cross
bars);
• and bars);
36 UTs (Uncoded Targets), four of which are part of the AICON reference system.
• 36 UTs (Uncoded Targets), four of which are part of the AICON reference system.
Out of the 114 points, 41 were positioned outside the load area in order to establish and verify the
Out of
reliability ofthe
the114 points, Another
outcomes. 41 were positioned outside
33 targets were the on
placed loadthearea in order
beam, five oftowhich
establish
wereand verify
along the
the reliability of the outcomes. Another 33 targets were placed on the beam, five
cross-section corresponding to the Laser Tracker P# points. Figure 5 shows the spatial distribution of of which were along
the
the cross-section
targets in the corresponding
whole test area.to the Laser Tracker P# points. Figure 5 shows the spatial distribution
of theThe
targets in the whole in
LT was stabilized test area.position during the measurement sessions and remotely controlled
a fixed
usingThe
the LTSpatialwasAnalyzer
stabilized(SA)in software
a fixed position
by New Riverduring the measurement sessions and remotely
Kinematics.
controlled using the Spatial Analyzer (SA) software
In each load condition of the beam, the DPA survey was by New River Kinematics.
performed and processed with the
In each load condition of the beam, the DPA survey was
AICON 3D Studio software. In each survey, the coordinates of all the targets in performed and processed
AICON’s ownwith the
internal
AICON
reference3D Studiowere
system software. In each
calculated, survey,
along withthe coordinates
their precisions. ofAfterwards,
all the targetsallinthe
AICON’s
coordinateownsets
internal
were
reference system were calculated, along with their
transformed into the LT coordinate system through the FT# points. precisions. Afterwards, all the coordinate sets were
transformed into the LT coordinate
TLS measurements were performedsystemunder
through the FT#
logistic points. which prevented the chance of
constraints
moving the instrument and covering the hidden areas with multiplewhich
TLS measurements were performed under logistic constraints scans. prevented
Thus, afterthe chance
several of
tests,
moving the instrument and covering the hidden areas with multiple scans. Thus,
the TLS was located about 3 m away from the beam, in a position where the contour of the front face of after several tests,
the
the TLS
beamwas couldlocated about
be fully 3 m away from
reconstructed (see the beam,
Figure in aestimate
5). An position ofwhere the contourdisplacement
the downward of the front face
can
of
be obtained by comparing each set of measurements to the reference one. Figure 6 showsdisplacement
the beam could be fully reconstructed (see Figure 5). An estimate of the downward the positions
can beTLS
of the obtained
and ofby thecomparing each set
LT with respect of measurements
to the beam. to the reference one. Figure 6 shows the
positions of the TLS and of the LT with respect to the beam.
Remote Sens. 2019, 11, 1940 7 of 20
Remote Sens. 2018, 10, x FOR PEER REVIEW 7 of 20
Remote Sens. 2018, 10, x FOR PEER REVIEW 7 of 20

Figure 5. Spatial distribution of the targets in the test area.

Figure 5. Spatial
Spatial distribution
distribution of the targets in the test area.

Figure6.6.Plan
Figure Planview
viewof
ofthe
theexperimental
experimentalinstruments,
instruments,distances
distancesare
areininmeters.
meters.
Figure 6. Plan view of the experimental instruments, distances are in meters.
During the measurement process, it was assumed that the TLS remained stationary in measuring
During the measurement process, it was assumed that the TLS remained stationary in measuring
During
position. the measurement
Nevertheless, before process, it was
each series assumed that the
of measurements, a scan at 360◦ to stationary
TLS remained in measuring
capture walls and other
position. Nevertheless, before each series of measurements, a scan at 360° to capture walls and other
position.
objects ofNevertheless,
the laboratorybefore eachstable
that were seriesover
of measurements,
time was run. a scan at 360° to capture walls and other
objects of the laboratory that were stable over time was run.
objects of thewere
Images laboratory
taken that were stable
at distances over time1 was
of between and run.
2 m with the optical axis orthogonal to the
Images were taken at distances of between 1 and 2 m with the optical axis orthogonal to the
Images were
longitudinal planetaken
of theatbeam,
distances of the
so that between 1 anddisplacements
downward 2 m with the optical axistoorthogonal
were close the image to the
plane.
longitudinal plane of the beam, so that the downward displacements were close to the image plane.
longitudinal
The plane ofDistance
Ground Sample the beam, so that
(GSD) wasthe downward
0.71 mm usingdisplacements
the Powershot were close tocamera
S100 digital the image
and plane.
it was
The Ground Sample Distance (GSD) was 0.71 mm using the Powershot S100 digital camera and it
The
0.41 Ground
mm using Sample
the EOSDistance
M3. (GSD) was 0.71 mm using the Powershot S100 digital camera and it
was 0.41 mm using the EOS M3.
was 0.41
The mm using
images theprocessed
were EOS M3. on an HP Z420 workstation with 64 GB RAM, Intel Xeon E5-16200
The images were processed on an HP Z420 workstation with 64 GB RAM, Intel Xeon E5-16200
The images
3.60 GHz CPU, andwere processed
NVIDIA on an
Quadro HP Z420
K2000 videoworkstation
card. with 64 GB RAM, Intel Xeon E5-16200
3.60 GHz CPU, and NVIDIA Quadro K2000 video card.
3.60 GHz CPU, and NVIDIA
The image-processing Quadro K2000
workflow, accordingvideoto card.
Agisoft Photoscan software, follows the standard
The image-processing workflow, according to Agisoft Photoscan software, follows the standard
stepsThe image-processing
of SfM software withworkflow,
the Bundle according
Adjustment to Agisoft
on keyPhotoscan software,afollows
points generating sparse the standard
point cloud,
steps of SfM software with the Bundle Adjustment on key points generating a sparse point cloud,
steps of SfM
external software
orientation, with
and the Bundle
generation Adjustment
of the dense point oncloud.
key points generating a sparse point cloud,
external orientation, and generation of the dense point cloud.
external
Theorientation,
images wereand generation of
georeferenced the dense
using the samepoint
sixcloud.
FT# targets used for the DPA measurements.
The images were georeferenced using the same six FT# targets used for the DPA measurements.
The images were georeferenced using the same six FT# targets used for the DPA measurements.
2.2. Methodologies Comparison
In order to evaluate the accuracy of TLS and CRP, the displacements estimated with these
methods were compared with the ones obtained from the LT and DPA.
RemotePreliminarily,
Sens. 2019, 11, 1940the vertical displacements from DT, LT, and DPA were compared. 8 of 20
The
comparison was performed on the P1 through P5 points placed on the beam (see Figure 5). These
comparisons
2.2. Methodologiesare referred
Comparisonto as LT-DPA and LT-DT hereafter in the paper.
Regarding TLS and CRP, starting from point clouds, two kinds of processing methods were
In order to evaluate the accuracy of TLS and CRP, the displacements estimated with these methods
investigated:
were compared with the ones obtained from the LT and DPA.
- mesh to mesh the
Preliminarily, comparison (m2m);
vertical displacements from DT, LT, and DPA were compared. The comparison
-
was modeling with
performed on the polynomials
P1 through(m) and comparison
P5 points of the
placed on the beam
beam (seedeflection
Figure 5).atThese
each comparisons
load step. are
referred to as
While LT-DPA
the andisLT-DT
first one hereafter
usually in the paper.
implemented in commercial software, the second one is not
commonRegarding TLS and
and depends on CRP, starting
the shape from
of the point object.
scanned clouds,For
two
thekinds
m2m of processingthe
comparison, methods were
z coordinate
investigated:
of the points corresponding to those of the DPA photogrammetry were obtained, reconstructing the
cross sections of each mesh and extracting the z value of the surface.
- mesh to mesh comparison (m2m);
In the modeling strategy, since the front face of the beam was very similar to a regular rectangle,
- modeling with polynomials (m) and comparison of the beam deflection at each load step.
the superior edge of the longitudinal section was modeled with a second-order polynomial. This
approach
Whilewas usedone
the first for isboth TLS implemented
usually point clouds inand CRP pointsoftware,
commercial clouds. To
the reduce
second each
one isdataset (DPA
not common
included)
and depends on the shape of the scanned object. For the m2m comparison, the z coordinate ofwas
in two dimensions, a Beam Reference System (BRS) with the following constraints the
built: corresponding to those of the DPA photogrammetry were obtained, reconstructing the cross
points
•sections of each
origin mesh in
contained and
theextracting
left-end ofthethez value
beam;of the surface.
• In the modeling strategy, since
z axis vertical and pointing down; the front face of the beam was very similar to a regular rectangle,
•the superior edge of the longitudinal section
x axis lying on the front face of the beam (parallel was modeled
to the with
beamalongitudinal
second-order polynomial. This
axis);
•approach
y axiswas usedafor
forming both TLS point
right-handed system clouds
withand
the CRP
otherpoint
two. clouds. To reduce each dataset (DPA
included) in two dimensions, a Beam Reference System (BRS) with the following constraints was built:
Figure 7 presents the position of the reference system on the beam.
• origin
In order contained in the the
to compare left-end
TLS ofandthe CRP
beam;displacements of the point clouds with the DPA
• z axis vertical
displacements, the and
latterpointing
have beendown;
transformed in the BRS. After the extraction of the coordinates of
the
• beam
x axistop-edge
lying on points,
the fronta face
least-squares
of the beam fitting with to
(parallel a second-order polynomial
the beam longitudinal was estimated to
axis);
obtain
• the beam
y axis deflection
forming along the
a right-handed beamwith
system span.the other two.
Given the x coordinate of the AICON DPA targets (see Figure 4), it was possible to calculate the
Figure 7using
z coordinate presents
the the position ofrepresenting
polynomials the reference system
beam on the beam.
deflection.

Figure
Figure 7.
7. The
The reference
reference system
system on
on the
the beam.
beam.

The downward
In order displacements
to compare the TLS and were calculated as the
CRP displacements difference
of the between
point clouds with the zDPA
coordinates at the
displacements,
initial zero load and the z coordinates under the considered load step.
the latter have been transformed in the BRS. After the extraction of the coordinates of the beam top-edge
points, a least-squares fitting with a second-order polynomial was estimated to obtain the beam
deflection along the beam span.
Given the x coordinate of the AICON DPA targets (see Figure 4), it was possible to calculate the z
coordinate using the polynomials representing beam deflection.
Remote Sens. 2019, 11, 1940 9 of 20

The downward displacements were calculated as the difference between the z coordinates at the
Remote Sens. 2018, 10, x FOR PEER REVIEW 9 of 20
initial zero load and the z coordinates under the considered load step.
The
The comparisons
comparisons of of the
the CRP
CRP displacements measured on
displacements measured on the
the meshes
meshes with with the
the DPA
DPA ones
ones are
are
labeled CRPCloud
labeled CRPCloud(m2m) -DPA. Similarly, the comparison between CRP displacements
(m2m)-DPA. Similarly, the comparison between CRP displacements obtained after
obtained after
polynomial
polynomial modeling
modelingwith withthe
theDPA
DPA displacements
displacementsis labeled CRPCloud
is labeled CRPCloud (m) -DPA. Likewise, for the TLS
(m)-DPA. Likewise, for the
displacements,
TLS displacements, comparisons are labeled
comparisons as TLS
are labeled as(m2m) -DPA
TLS(m2m) and and
-DPA TLSTLS
(m) -DPA.
(m)-DPA.
The
The overall data processing follows the workflow depicted in Figure
overall data processing follows the workflow depicted in Figure8. 8.

Figure 8.
Figure 8. Processing
Processing strategy
strategy for
for Close-Range
Close-Range Photogrammetry
Photogrammetry (CRP)
(CRP) and
and Terrestrial Laser Scanners
Terrestrial Laser Scanners
(TLS).
(TLS).

3. Results
3. Results
Measurements
Measurements started, first for
started, first for beam#1
beam#1 (first
(first load
load condition
condition (LC1)
(LC1) and
and second
second load
load condition
condition
(LC2)) and then for beam#2, with the instrumental observations at zero load
(LC2)) and then for beam#2, with the instrumental observations at zero load and progressed and progressed following
the load sequence
following the loadof Tables 1–3.
sequence The displacement
of Tables transducertransducer
1–3. The displacement remained active
remainedall the time,allconstantly
active the time,
recording displacements.
constantly recording displacements.
Figure
Figure 99 presents
presents thethecomparison
comparisonbetween betweenmidspan
midspandisplacements
displacements of of beam#1
beam#1 LC1
LC1 measured
measured by
by DT, LT, and DPA. Figure 10 compares midspan displacements of beam#1
DT, LT, and DPA. Figure 10 compares midspan displacements of beam#1 LC2 measured by DT and LC2 measured by DT
and
DPA,DPA,
whilewhile
FigureFigure 11 represents
11 represents the of
the case case of beam
beam 2 considering
2 considering DT,and
DT, LT, LT,DPA.
and DPA. Forlatter
For the the latter
case,
case,
only only the second
the second load load
cyclecycle is presented.
is presented. Different
Different structural
structural performances
performances of the
of the twotwo beams
beams cancanbe
be explained, considering that beam 1 was not completely straight along its
explained, considering that beam 1 was not completely straight along its longitudinal axis; this longitudinal axis; this
produced
produced different
differentdisplacements
displacements along
alongthethe
vertical direction
vertical in comparison
direction withwith
in comparison beambeam2. On2.the Onother
the
hand, this is this
other hand, negligible for thefor
is negligible aims
theof the of
aims considered analysis.
the considered analysis.
The small differences between DT and DPA that can be seen in Figure 9 are probably due to a
slight modification of the correct position of the DT due to beam deflection. Instead, the differences
between LT and DPA are negligible.
 Remote Sens. 2018, 10, x FOR PEER REVIEW 10 of 20
Remote Sens. 2018, 10, x FOR PEER REVIEW 10 of 20
Remote
RemoteSens. 2018,11,
Sens.2019, 10,1940
x FOR PEER REVIEW 10
10ofof2020
70
70
70
60
60
60
50
50 DT

(kN)
50
40 DT
LT

(kN)
40 DT
LT
DPA

(kN)
LT
Load
40 DPA
30
Load
Load 30
DPA
30
20
20
20
10
10
100
00 5 10 15 20 25
00 5 10 15
Displacement (mm) 20 25
0 5 10
Displacement 15
(mm) 20 25
Displacement
Figure 9. Displacements (mm)
comparison for beam#1 LC1.
Figure 9. Displacements comparison for beam#1 LC1.
Figure9.9.Displacements
Figure Displacementscomparison
comparisonfor
forbeam#1
beam#1LC1.
LC1.
70
70 DT
70
60 DT
DPA
60 DT
DPA
60 DPA
50
50
(kN)

50
40
(kN)

40
(kN)
Load

40
30
Load

30
Load

30
20
20
20
10
10
100
00 5 10 15 20 25
00 5 10
Displacement 15
(mm) 20 25
0 5 10
Displacement 15
(mm) 20 25
Displacement
Figure 10. Displacements (mm)for beam#1 LC2.
comparison
Figure10.
Figure 10.Displacements
Displacementscomparison
comparisonfor
forbeam#1
beam#1LC2.
LC2.
Figure 10. Displacements comparison for beam#1 LC2.
70
70
70
60
60
60
50
50 DT
(kN)

50
40 DT
LT
(kN)

40 DT
LT
DPA
(kN)
Load

40 LT
DPA
30
Load

DPA
30
Load

30
20
20
20
10
10
100
00 5 10 15 20 25
00 5 10
Displacement 15
(mm) 20 25
0 5 10 15
Displacement (mm) 20 25
Figure11.
Figure Displacement
11.Displacements
Displacements (mm)for
comparison
comparison forbeam#2.
beam#2.
Figure 11. Displacements comparison for beam#2.
Figure 11. Displacements comparison for beam#2.
Remote Sens. 2019, 11, 1940 11 of 20

The small differences between DT and DPA that can be seen in Figure 9 are probably due to a
slight modification of the correct position of the DT due to beam deflection. Instead, the differences
between LT and DPA are negligible.
Tables 4 and 5 report the comparison between the vertical displacement vectors measured using
LT on the five points placed on the beam and the closest DPA targets (placed in the same beam
cross-sections). During the test beam#1 LC2, it was not possible to use the LT and consequently, the
comparison, in that case, is not reported. The overall RMS level is obtained using Equation (1), where
xM1
i
is the coordinate of point i obtained with the method M1 (which can be DPA, TLS, or CRP) and
xM2
i
is the coordinate of the same point obtained with method M2 (which can be DPA, LT).
v
u
N
t
1 X M1 2
xRMS = xi − xM2
i
(1)
N
i=1

Table 4. DPA-LT (mm) beam#1 LC1.

Load (kN) P1 P2 P3 P4 P5
11 −0.04 −0.04 0.03 0.04 0.03
26 0.04 0.17 0.27 0.29 0.19
40 −0.03 0.00 0.11 0.08 0.07
54 −0.12 −0.01 0.29 0.38 0.09
68 −0.30 −0.25 −0.26 −0.23 0.00
Global RMS 0.15 0.14 0.22 0.24 0.10

Table 5. DPA-LT (mm) beam#2.

Load (kN) P1 P2 P3 P4 P5
19 −0.11 −0.19 −0.2 −0.18 −0.17
39 −0.11 −0.21 −0.25 −0.18 −0.15
68 0.10 0.60 0.78 0.48 −0.06
Global RMS 0.11 0.33 0.43 0.28 0.12

The overall RMS of beam#1 LC1 comparison is 0.18 mm.

The overall RMS of beam#2 comparison is 0.32 mm. Laser tracker (LT) and DPA measurements
are in good agreement; differences are small, as shown in Tables 4 and 5. Only the last measurement of
beam#2 shows a larger difference, which is possibly due to a less accurate set of measurements for
the DPA in that case. For the sake of simplicity, from now on, the accuracy of TLS and CRP will be
evaluated only versus DPA.

3.1. Terrestrial Laser Scanner Measurements

As stated in Section 2, for each load, both a 360◦ scan and a high-resolution scan in a smaller
window covering just the beam, were performed. During the processing, at each load-step, after a
stability check of the instruments, the meshes were built and the top edges of the beam were extracted.
Using a least-squares approach, these edges (composed of about 9600 points) were fitted with a
second-degree polynomial. Then, the values of this fitting curve were calculated at the x coordinate
corresponding to each of the 33 DPA targets on the beam, thereby obtaining the TLS z coordinate.
Tables 6–8 show the RMS values, at each load condition, for the mesh to mesh (m2m) and
polynomial (m) methodologies.
Figure 12 shows the polynomials fitted at each load step and for beam#1 second load condition
and for beam#2, see Tables 7 and 8.
 RemoteSens.
Remote Sens.2019,
2018,11,
10,1940
x FOR PEER REVIEW 12 of
12 of20
20

Table 6. RMS (mm) of each TLS methodology with DPA beam#1 LC1.
Table 6. RMS (mm) of each TLS methodology with DPA beam#1 LC1.
Load (kN) TLS(m2m)-DPA TLS(m)-DPA
11
Load (kN) 1.01
TLS(m2m)-DPA 0.38
TLS(m)-DPA
11 26 0.98
1.01 0.420.38
26 40 0.97
0.98 0.600.42
40 54 0.97
0.99 0.630.60
54 0.99
68 0.92 0.620.63
68 0.92 0.62
Global
Global RMS RMS 0.97
0.97 0.540.54

Table 7. RMS (mm) of each TLS methodology with DPA beam#1 LC2.
Table 7. RMS (mm) of each TLS methodology with DPA beam#1 LC2.
Load (kN) TLS(m2m)-DPA TLS(m)-DPA
Load (kN) TLS(m2m)-DPA TLS(m)-DPA
22 1.14 1.08
22 1.14 1.08
48 0.78 0.77
48 0.78 0.77
Global
Global RMS RMS 0.98
0.98 0.940.94

Table8.8.RMS
Table RMS(mm)
(mm)ofofeach
eachTLS
TLSmethodology
methodologywith
withDPA
DPAbeam#2.
beam#2.

Load (kN) TLS(m2m)-DPA TLS(m)-DPA

Load (kN) TLS(m2m)-DPA TLS(m)-DPA
19 0.54 0.58
19 0.54 0.58
39
39 0.50
0.50
0.560.56
68 68 0.95
0.95 0.760.76
Global
Global RMS RMS 0.69
0.69 0.640.64

0
Deflection (m)

0.05 Polynomial
Scanned edge 0 kN
Scanned edge 22 kN
Scanned edge 48 kN
0.1
0 0.5 1 1.5 2 2.5 3
Beam length (m)

0
Deflection (m)

Poynomial
0.05 Scanned edge 0 kN
Scanned edge 19 kN
Scanned edge 39 kN
Scanned edge 68 kN
0.1
0 0.5 1 1.5 2 2.5 3
Beam length (m)

Figure12.
Figure 12.Terrestrial
TerrestrialLaser
LaserScanners
Scanners(TLS)
(TLS)polynomials
polynomialsfitted
fittedalong
alongthe
thebeam
beamspan
spanfor
forbeam#1
beam#1LC2LC2
(top)and
(top) andbeam#2
beam#2(bottom).
(bottom).The
Therounded
roundedmarkers
markersrepresent
representaaselection
selectionofofthe
themesh
meshpoints
pointsused
usedininthe
the
fittingprocess.
fitting process.

3.2.
3.2.CRP
CRPMeasurements
Measurements
The
Thedigital
digitalcamera
cameraPowershot
PowershotS110S110was
wasused
usedforforbeam#1
beam#1LC1,
LC1,while
whilethe
thedigital
digitalcamera
cameraEOS
EOSM3 M3
was
was used for the beam#1 LC2 and for beam#2. With the Powershot, the images were takenin
used for the beam#1 LC2 and for beam#2. With the Powershot, the images were taken inraw
raw
format
formatand
andthen
thentransformed
transformedinto
intoTIFF,
TIFF,with
withaaresolution
resolutionequal 2816××2112
equaltoto2816 2112pixels.
pixels.With
Withthe
theEOS
EOS
M3,
M3, images were captured in raw and Jpeg format and then compressed at a resolution of 50% ofthe
images were captured in raw and Jpeg format and then compressed at a resolution of 50% of the
original.
original. At
At each
each load
load condition,
condition, aaphotogrammetric
photogrammetric survey
survey was
wasperformed,
performed, and
andthetheimages
imageswere
were
processed
processedaccording
accordingtotothe
thework-flow
work-flowdepicted
depictedininFigure
Figure8.8.The
Thedata
dataand
andprocessing
processingparameters
parametersofof
each
eachsurvey
surveyare
arereported
reportedininTables
Tables9–11.
9–11.
Remote Sens. 2019, 11, 1940 13 of 20

Table 9. Data and processing parameters of the CRP surveys beam#1 LC1 Powershot.

Parameters of the Dense

Load (kN) No. of Images Resolution MB Beam Points
Point Cloud
0 98 2816 × 2112 2.5 Ultra-High 362,429
11 112 2816 × 2112 2.5 Ultra-High 345,251
26 114 2816 × 2112 2.5 Ultra-High 251,002
40 108 2816 × 2112 2.5 Ultra-High 316,118
54 105 2816 × 2112 2.5 Ultra-High 363,539

Table 10. Data and processing parameters of the CRP surveys beam#1 LC2 EOS M3.

Parameters of the Dense

Load (kN) No. of Images Resolution MB Beam Points
Point Cloud
0 65 6000 × 4000 1 Ultra-High 1,460,945
22 55 6000 × 4000 1 Ultra-High 1,460,230
48 59 6000 × 4000 1 Ultra-High 1,459,589

Table 11. Data and processing parameters of the CRP surveys beam#2 EOS M3.

Parameters of the Dense

Load (kN) No. of Images Resolution MB Beam Points
Point Cloud
0 61 6000 × 4000 1 Ultra-High 1,686,267
19 63 6000 × 4000 1 Ultra-High 1,705,513
39 53 6000 × 4000 1 Ultra-High 1,627,828
68 57 6000 × 4000 1 Ultra-High 1,733,299

The absolute orientation was established by manual collimation of the six FT# targets, thus forcing
the verticality of the z axis. Tables 12–14 report the residuals of the absolute orientation on the Ground
Control Points (GCPs).

Table 12. Residuals (mm) on the Ground Control Points (GCPs) beam#1 LC1 Powershot.

Load (kN) X Y Z
0 1.01 1.36 0.71
11 2.62 1.34 0.69
26 0.64 0.83 0.68
40 0.70 0.60 0.33
54 3.25 5.03 2.27

Table 13. Residuals (mm) on the GCPs beam#1 LC2 condition EOS M3.

Load (kN) X Y Z
0 1.44 2.87 1.07
22 0.85 1.06 0.91
48 1.30 1.18 1.11

Table 14. Residuals (mm) on the GCPs beam#2 EOS M3.

Load (kN) X Y Z
0 1.61 0.97 0.64
19 1.30 1.39 0.60
39 0.83 2.01 0.55
68 1.15 1.95 1.88
Remote Sens. 2019, 11, 1940 14 of 20

Remote Sens. 2018, 10, x FOR PEER REVIEW 14 of 20

The meshes were created using the Reconstructor software. The upper edges were extracted,
The meshes were created using the Reconstructor software. The upper edges were extracted,
obtaining about 1600 points for each edge, which were then used as input data for a least-squares
obtaining about 1600 points for each edge, which were then used as input data for a least-squares
fitting to the second-order polynomials, as was done for the TLS.
fitting to the second-order polynomials, as was done for the TLS.
The polynomials were evaluated at the 33 sections corresponding to the target points under each
The polynomials were evaluated at the 33 sections corresponding to the target points under each
load condition. Thus, the displacements were calculated by comparing the coordinates with the ones
load condition. Thus, the displacements were calculated by comparing the coordinates with the ones
obtained at zero load.
obtained at zero load.
ForFor
both calculation
both modes
calculation modes(m2m
(m2mandandm),
m),the
thedisplacement
displacement values were compared
values were comparedwithwiththe
the
corresponding
corresponding onesones
obtained fromfrom
obtained the DPA
the and
DPAthe overall
and RMS was
the overall assessed
RMS for each for
was assessed loadeach
condition.
load
Tables
condition.15–17 show the RMS values for each beam and load condition and for each methodology.
Tables 15–17 show the RMS values for each beam and load condition and for each methodology.
Table 15. RMS (mm) in the CRP-DPA comparison beam#1 Powershot LC1.
Table 15. RMS (mm) in the CRP-DPA comparison beam#1 Powershot LC1.
Load (kN) CRPCloud(m2m) CRPCloud(m)
Load
11 (kN) CRPCloud(m2m)
2.71 CRPCloud(m)
1.42
26 11 2.71
2.16 1.42
1.50
40 26 2.42
2.16 1.69
1.50
54 40 1.43
2.42 0.78
1.69
Global RMS 2.23 1.39
54 1.43 0.78
Global RMS 2.23 1.39
Table 16. RMS (mm) in the CRP-DPA comparison beam#1 LC2 EOS M3.
Table 16. RMS (mm) in the CRP-DPA comparison beam#1 LC2 EOS M3.
Load (kN) CRPCloud(m2m) CRPCloud(m)
Load22(kN) CRPCloud(m2m)
1.23 CRPCloud(m)
0.35
22
48 1.23
1.08 0.89 0.35
Global
48RMS 1.16
1.08 0.68 0.89
Global RMS 1.16 0.68
Table 17. RMS (mm) in the CRP-DPA comparison beam#2 EOS M3.
Table 17. RMS (mm) in the CRP-DPA comparison beam#2 EOS M3.
Load (kN) CRPCloud(m2m) CRPCloud(m)
Load (kN) CRPCloud(m2m) CRPCloud(m)
19 2.10 1.69
39 19 2.10
1.76 1.69
1.59
68 39 1.76
4.32 1.59
1.00
68
Global RMS 4.32
3.14 1.00
1.33
Global RMS 3.14 1.33
Figure 13 shows the displacements for each load and for beam#1 second load condition and for
Figure 13 shows the displacements for each load and for beam#1 second load condition and for
beam#2.
beam#2.

0
Deflection (m)

0.05 Polynomial
Scanned edge 0 kN
Scanned edge 22 kN
Scanned edge 48 kN
0.1
0 0.5 1 1.5 2 2.5 3
Beam length (m)

0
Deflection (m)

Polynomial
0.05 Scanned edge 0 kN
Scanned edge 19 kN
Scanned edge 39 kN
Scanned edge 68 kN
0.1
0 0.5 1 1.5 2 2.5 3
Beam length (m)

Figure
Figure 13. 13.
CRP CRP polynomials
polynomials fitted
fitted along
along beamspan
beam spanfor
forbeam#1
beam#1LC2
LC2 (top)
(top) and beam#2
beam#2 (bottom).
(bottom).The
The
rounded
rounded markers
markers represent
represent a selection
a selection ofof
thethe meshpoints
mesh pointsused
usedin
inthe
thefitting
fitting process.
process.
Remote Sens. 2019, 11, 1940 15 of 20

Remote Sens. 2018, 10, x FOR PEER REVIEW 15 of 20

Figure 14 reports the comparison between the midspan displacements recorded by DPA, TLS,
Figure
CRP, and 14 the
DT for reports
threethe comparison
different between
experimental the(beam#1
tests midspan displacements
LC1, beam#1 LC2,recorded
beam#2) by DPA, TLS,
considering
CRP,
the and DT for them2m
two methodologies: three
anddifferent
m. experimental tests (beam#1 LC1, beam#1 LC2, beam#2)
considering the two methodologies: m2m and m.
80 80
DPA DPA
DT DT
60 TLSm2m 60 TLSm
Load (kN)

Load (kN)
CRPm2m CRPm
40 40

20 20

0 0
0 10 20 30 40 0 10 20 30 40
Displacement (mm) Displacement (mm)
(a) (b)
80 80
DPA DPA
DT DT
60 TLSm2m 60 TLSm
Load (kN)

Load (kN)
CRPm2m CRPm
40 40

20 20

0 0
0 10 20 30 40 0 10 20 30 40
Displacement (mm) Displacement (mm)
(c) (d)
80 80
DPA
DT
60 60 TLSm
Load (kN)

Load (kN)

CRPm
40 DPA 40
DT
TLSm2m
20 CRPm2m 20

0 0
0 10 20 30 40 0 10 20 30 40
Displacement (mm) Displacement (mm)
(e) (f)
Figure
Figure 14.14.
TLSTLSandand CRP
CRP midspan
midspan displacements
displacements comparison
comparison forfor beam#1
beam#1 and
and #2.#2.
(a)(a) beam#1
beam#1 LC1
LC1 m2m
m2m
comparison;
comparison; (b)(b) beam#1
beam#1 LC1
LC1 mm comparison;
comparison; (c)(c) beam#1
beam#1 LC2
LC2 m2m
m2m comparison;
comparison; (d)(d) beam#1
beam#1 LC2
LC2 mm
comparison;
comparison; (e)(e) beam#2
beam#2 m2m
m2m comparison;
comparison; (f)(f) beam#2
beam#2 m comparison.
m comparison.

4. 4.
Discussion
Discussion
The comparisons
The comparisonsbetween the beam
between the displacements obtained with
beam displacements different
obtained instruments
with different highlighted
instruments
the limitations and potential of the two adopted geomatic techniques: TLS
highlighted the limitations and potential of the two adopted geomatic techniques: and CRP. In and
TLS particular,
CRP. In
this paper reports on two different methodologies for extracting the metric information
particular, this paper reports on two different methodologies for extracting the metric information from the
point cloud.
from the point cloud.
The
Theresults
resultsobtained
obtainedwith
with TLS
TLSusing
usingthe two
the two m2m
m2m and
andmm methodologies
methodologies are
aresimilar
similartoto
those
those
obtained by Gordon [39] and Tsakiri [38] with their tests on RC and wood beams. Also, the
obtained by Gordon [39] and Tsakiri [38] with their tests on RC and wood beams. Also, the results results ofof
the CRP are in line with some of the tests performed by Remondino
the CRP are in line with some of the tests performed by Remondino [45]. [45].
In the m2m case, the modeling of the point cloud is deterministic, thus the mesh is influenced by
all the noise caused by the measurement and by the object’s own coarseness in the measurement
position, no matter what happens in the neighborhood. This behavior can explain the trend of the
Remote Sens. 2019, 11, 1940 16 of 20

In the m2m case, the modeling of the point cloud is deterministic, thus the mesh is influenced
by all the noise caused by the measurement and by the object’s own coarseness in the measurement
position, no matter what happens in the neighborhood. This behavior can explain the trend of the
displacements measured by the CRP m2m in Figure 14e. Indeed, looking at only this single dataset
would lead to an incorrect assessment of the structural behavior of the beam.
In order to have a reliable estimation of the beam displacements, CRP data should be obtained
exploiting all the available information, as happens in the polynomial modeling. Also, in order
to strengthen the reliability of the calculations, groups of points whose positions are fixed and not
influenced by the loads should be checked during all processes. In this study, the positions of 41 points
in a stable area were monitored in each measurement series, obtaining residuals of the same order of
the method accuracy.
Considering the TLS, the overall RMS obtained with the m2m comparison process is 0.97 mm for
the beam#1 LC1 (see Table 6), 0.96 mm for the beam#1 LC2 (see Table 7) and 0.69 mm for the beam#2
(see Table 8). Modeling the beam edge with a polynomial led to an overall RMS of 0.54 mm for the
beam#1 LC1 (see Table 6), 0.92 mm for the beam#1 LC2 (see Table 7), and 0.63 mm for the beam#2 (see
Table 8), with an increment of accuracy of about 20%.
Better results were obtained for the CRP: In the m2m comparison, the overall RMS was 2.23 mm
for the beam#1 LC1 (see Table 15), 1.16 mm for the beam#1 LC2 (see Table 16), and 3.14 for the beam#2
(see Table 17). While, if the beam edge is fitted with second-degree polynomials, the overall RMS is:
1.39 mm for the beam#1 LC1 (see Table 15), 0.68 mm for the beam#1 LC2 (see Table 16), and 1.33 mm
for the beam#2 (see Table 17), with an increase in the accuracy of about 45%.
In this work, all the modeling calculations, including routines to fit portions of point clouds with
respect to geometric 2D/3D shapes, were developed in the Matlab™ environment.
For the CRP test, two non-professional Canon digital cameras were used—the PowerShot S110
and the more performant EOS M3. The results obtained with the two methodologies (m2m and m)
were almost similar. The EOS M3 demonstrated a slightly better performance. This is definitely due to
the fact that a much larger point cloud density was obtained—using a lower number of images and the
same processing parameters—in the EOS M3 images compared to the PowerShot S110 images.

5. Conclusions
The goal of this research was to investigate the obtainable accuracy in the structural displacement
measurements with CRP based on the Structure from Motion algorithm, and to compare this with the
TLS technique. A case study on two RC beams tested under a four-points bending loading condition
was presented. In order to assess the amount of displacement, point clouds at different beam loading
states were acquired and compared. A complete description of the instrumentation used and the
experimental environment, along with a comprehensive report on the calculations and results obtained
was reported. Two kinds of point clouds comparison were investigated: Mesh to mesh (m2m) and
modeling (m), with simple geometric primitives as second-order polynomials.
The comparison between the m2m and the m approaches showed that the latter leads to significantly
better results for both TLS and CRP.
The results obtained with the TLS for both the m2m and m methodologies present RMS below
1 mm; while for CRP, the RMS is around a few millimeters for m2m and around 1 mm for m.
The CRP results were highly accurate for two main reasons: The distance between the instruments
and the beam was quite short (3–4 m, see Figure 6) with a very small GSD; the dimensional scale of
the photogrammetric images was obtained through laser tracker measurements. This last approach is
not typical when working in the field, but the problem can be overcome by introducing two or more
calibrated scale bars.
The obtained results allow for consideration of these two techniques, which are suitable to evaluate
the deformation of structures. Both tested techniques can be considered fast and effective, but the
difference in cost should be highlighted. On the one hand, the TLS cost is quite high, both in terms of
Remote Sens. 2019, 11, 1940 17 of 20

instrumentation and data processing software; while on the other hand, the CRP can be developed
with non-professional digital cameras at an affordable cost and low-cost commercial or open-source
software can be used for the image processing and point clouds management.
Further developments are expected considering both new analyses of the already obtained
experimental data and new field tests considering larger structures and longer distances between
instruments and measured objects. Indeed, the techniques adopted in this paper to reduce the
uncertainty of the coordinates of the photogrammetric point clouds can be used when it is necessary to
accurately know the dimensions of structures such as steel pylons. In addition, the image processing
techniques with subpixel capabilities, see [61–64], seem very promising for structural health monitoring.

Author Contributions: F.S. and F.M. developed the structural part while G.S. and G.V. performed the geomatic
survey. The geomatic and metrological analysis were developed by G.S. and G.V. All the authors wrote the paper.
Funding: This study was supported by the research grant for the project “Healthy Cities and Smart Territories”
(2016/17) funded by Fondazione di Sardegna and the Autonomous Region of Sardinia. Flavio Stochino
acknowledges the financial support of the Autonomous Region of Sardinia under grant P.O.R. SARDEGNA
2014–2020, CCI: 2014- IT05SFOP021, Project: Retrofitting, rehabilitation and requalification of the historical cultural
architectural 506 heritage (R3-PAS).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Mechelke, K.; Sternberg, H.; Keller, F. A monitoring system with TLS sensor. In Proceedings of the Joint
International Symposium on Deformation Monitoring, Nottingham, UK, 9–10 September 2013.
2. Meloni, P.; Mistretta, F.; Stochino, F.; Carcangiu, G. Thermal Path Reconstruction for Reinforced Concrete
Under Fire. Fire Technol. 2019, in press. [CrossRef]
3. Stochino, F.; Mistretta, F.; Meloni, P.; Carcangiu, G. Integrated Approach for Post-fire Reinforced Concrete
Structures Assessment. Period. Polytech. Civ. Eng. 2017, 61, 677–699. [CrossRef]
4. Vacca, G.; Mistretta, F.; Stochino, F.; Dessi, A. Terrestrial Laser Scanner for monitoring the deformations
and the damages of buildings. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. 2016, XLI-B5, 453–460.
[CrossRef]
5. Giresini, L.; Sassu, M. Tests Results and Simple Structural Analysis of the Main Lighthouse in the Harbor of
Livorno (Italy). Adv. Mater. Res. 2014, 834, 1299–1303. [CrossRef]
6. Stochino, F.; Fadda, M.L.; Mistretta, F. Assessment of RC bridges integrity by means of low-cost investigations.
Frat. Ed Integrita Strutt. (Fract. Struct. Integr.) 2018, 12, 216–225. [CrossRef]
7. Pucci, A.; Puppio, M.L.; Giresini, L.; Sousa, H.; Matos, J.; Sassu, M. Method for sustainable large-scale bridges
survey. In Towards a Resilient Built Environment Risk and Asset Management; International Association for
Bridge and Structural Engineering: Guimarães, Portugal, 2019; pp. 1034–1041.
8. Han, J.; Guo, J.; Jiang, Y. Monitoring tunnel profile by means of multi-epoch dispersed 3-D LIDAR point
clouds. Tunn. Undergr. Space Technol. 2013, 33, 186–192. [CrossRef]
9. Deidda, M.; Sanna, G. Integration of Geomatics Techniques for the Urban Cavity Survey. ISPRS Ann.
Photogramm. Remote Sens. Spatial Inf. Sci. 2013, II-5/W1, 91–96. [CrossRef]
10. Kasperski, J.; Delacourt, C.; Allemand, P.; Potherat, P.; Jaud, M.; Varrel, E. Application of a terrestrial laser
scanner (TLS) to the study of the Séchilienne Landslide (Isère, France). Remote Sens. 2010, 2, 2785–2802.
[CrossRef]
11. Xu, Z.; Wu, L.; Shen, Y.; Li, F.; Wang, Q.; Wang, R. Tridimensional reconstruction applied to cultural heritage
with the use of camera-equipped UAV and terrestrial laser scanner. Remote Sens. 2014, 6, 10413–10434.
[CrossRef]
12. Giresini, L.; Gioeli, A.; Sassu, M. Seismic Reinforcement of a RC Building with External Steel Frameworks:
The Case of the Primary School XXV April of Arcola (Italy). Adv. Mater. Res. 2014, 834, 697–700.
13. Sassu, M.; Andreini, M.; De Falco, A.; Giresini, L.; Puppio, M.L. Structural protection after landslide
phenomena: A case study in Northern Italy. In Civil Engineering and Urban Planning IV, Proceedings of the 4th
International Conference on Civil Engineering and Urban Planning, Beijing, China, 25–27 July 2015; Liu, Y.M.,
Fu, D., Tong, Z.X., Bao, Z.Q., Tang, B., Eds.; CRC Press: Boca Raton, FL, USA, 2016; pp. 241–246.
Remote Sens. 2019, 11, 1940 18 of 20

14. Giannattasio, C.; Grillo, S.M.; Vacca, G. Interdisciplinary study for knowledge and dating of the San Francesco
convent in Stampace, Cagliari—Italy (XIII-XXI century). ISPRS Ann. Photogramm. Remote Sens. Spatial Inf.
Sci. 2013, II-5/W1, 139–144. [CrossRef]
15. Stochino, F.; Fadda, M.L.; Mistretta, F. Low cost condition assessment method for existing RC bridges. Eng.
Fail. Anal. 2018, 86, 56–71. [CrossRef]
16. Lindenbergh, R.; Pietrzyk, P. Change detection and deformation analysis using static and mobile laser
scanning. Appl. Geomat. 2017, 7, 65–74. [CrossRef]
17. Mukupa, W.; Roberts, G.W.; Hancock, C.M.; Al-Manasir, K. A review of the use of terrestrial laser scanning
application for change detection and deformation monitoring of structures. Surv. Rev. 2017, 49, 99–116.
[CrossRef]
18. Vosselman, G.; Maas, H.G. Airborne and Terrestrial Laser Scanning; Whittles Publishing: Dunbeath, UK, 2010.
19. Lague, D.; Brodu, N.; Leroux, J. Accurate 3D comparison of complex topography with terrestrial laser
scanner: Application to the Rangitikei canyon (N-Z). ISPRS J. Photogramm. Remote Sens. 2013, 82, 10–16.
[CrossRef]
20. Rutzinger, M.; Rüf, B.; Höfle, B.; Vetter, M. Change detection of building footprints from airborne laser
scanning acquired in short time intervals. Int. Arch. Photogramm. Remote Sens. Spat Inf. Sci. 2010, 38, 475–480.
21. Xu, S.; Vosselman, G.; Oude Elberink, S. Detection and classifi- cation of changes in buildings from airborne
laser scanning data. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci. 2013, II-5/W2, 343–348. [CrossRef]
22. Pesci, A.; Teza, G.; Bonali, E.; Casula, G.; Boschi, E. A laser-scanning based method for fast estimation of
seismic-induced building deformations. ISPRS J. Photogramm. Remote Sens. 2013, 79, 185–198. [CrossRef]
23. Wang, T.; Jaw, J.; Chang, Y.; Jeng, F. Application and validation of profile-image method for measuring
deformation of tunnel wall. Tunn. Undergr. Space Technol. 2009, 24, 136–147. [CrossRef]
24. Wujanz, D.; Neitzel, F.; Hebel, H.P.; Linke, J.; Busch, W. Terrestrial radar and laser scanning for deformation
monitoring: First steps towards assisted radar scanning. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci.
2013, II(5/W2), 325–330. [CrossRef]
25. Hanssen, R.F. Radar Interferometry; Kluwer Academic Publishers: Dordrecht, The Netherlands, 2001.
26. Bremer, M.; Sass, O. Combining airborne and terrestrial laser scanning for quantifying erosion and deposition
by a debris flow event. Geomorphology 2012, 138, 49–60. [CrossRef]
27. Young, A.P.; Olsen, M.J.; Driscoll, N.; Flick, R.; Gutierrez, R.; Guza, R.; Johnstone, E.; Kuester, F. Comparison
of airborne and terrestrial lidar estimates of seacliff erosion in southern california. Photogramm. Eng. Remote
Sens. 2010, 76, 421–427. [CrossRef]
28. Olsen, M.; Kuester, F.; Chang, B.; Hutchinson, T. Terrestrial laser- scanning-based structural damage
assessment. J. Comput. Civ. Eng. 2010, 24, 264–272. [CrossRef]
29. Pesci, A.; Casula, G.; Boschi, E. Laser scanning the Garisenda and Asinelli towers in Bologna (Italy): Detailed
deformation patterns of two ancient leaning buildings. J. Cult. Herit. 2011, 12, 117–127. [CrossRef]
30. Riveiro, B.; Arias, P.; Armesto, J.; Rial, F.; Solla, M. Multidiscipliniar approach to historical arch bridges
documentation. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2008, XXXVII(B5), 247–252.
31. Grosse-Schwiep, M.; Piechel, J.; Luhmann, T. Measurement of rotor blade deformations of wind energy
converters with laser scanners. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci. 2013, II(5/W2), 97–102.
[CrossRef]
32. Scaioni, M.; Marsella, M.; Crosetto, M.; Tornatore, V.; Wang, J. Geodetic and Remote-Sensing Sensors for
Dam Deformation Monitoring. Sensors 2018, 18, 3682. [CrossRef] [PubMed]
33. Jaafar, H.A.; Meng, X.; Sowter, A. Terrestrial laser scanner error quantification for the purpose of monitoring.
Surv. Rev. 2017. [CrossRef]
34. Zhao, P.; Hu, Q.; Wang, S.; Ai, M.; Mao, Q. Panoramic Image and Three-Axis Laser Scanner Integrated
Approach for Indoor 3D Mapping. Remote Sens. 2018, 10, 1269. [CrossRef]
35. Park, H.S.; Lee, H.M.; Adeli, H.; Lee, I. A new approach for health monitoring of structures: Terrestrial laser
scanning. Comput. Aided Civ. Infrastruct. Eng. 2007, 22, 119–130. [CrossRef]
36. Girardeau-Montaut, D.; Roux, M.; Marc, R.; Thibault, G. Change detection on points cloud data acquired
with a ground laser scanner. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2005, 36, 30–35.
37. Xu, X.; Bureick, J.; Yang, H.; Neumann, I. TLS-based composite structure deformation analysis validated
with laser tracker. Compos. Struct. 2018, 202, 60–65. [CrossRef]
Remote Sens. 2019, 11, 1940 19 of 20

38. Tsakiri, M.; Lichti, D.; Pfeifer, N. Terrestrial laser scanning for deformation monitoring. In Proceedings of the
3rd IAG/12th FIG Symposium, Baden, Germany, 22–24 May 2006.
39. Gordon, G.; Lichti, D.D.; Franke, J.; Stewart, M.P. Measurement of structural deformation using terrestrial laser
scanners. In Proceedings of the 1st FIG International Symposium on Engineering Surveys for Construction
Works and Structural Engineering, Nottingham, UK, 28 June–1 July 2004.
40. Luhmann, T.; Robson, S.; Kyle, S.; Boehm, J. Close Range Photogrammetry: 3D Imaging Techniques; Walter De
Gruyter Inc.: Berlin, Germany, 2014.
41. Szeliski, R. Computer Vision: Algorithms and Applications; Springer: Berlin, Germany, 2010.
42. Mancini, F.; Dubbini, M.; Gattelli, M.; Stecchi, F.; Fabbri, S.; Gabbianelli, G. Using Unmanned Aerial Vehicles
(UAV) for High-Resolution Reconstruction of Topography: The Structure from Motion Approach on Coastal
Environments. Remote Sens. 2013, 5, 6880–6898. [CrossRef]
43. Mathews, A.; Jensen, J. Visualizing and quantifying vineyard canopy LAI using an unmanned aerial vehicle
(UAV) collected high density structure from motion point cloud. Remote Sens. 2013, 5, 2164–2183. [CrossRef]
44. Turner, D.; Lucieer, A.; Watson, C. An automated technique for generating georectified mosaics from
ultra-high resolution unmanned aerial vehicle (UAV) imagery, based on structure from motion (SfM) point
clouds. Remote Sens. 2012, 4, 1392–1410. [CrossRef]
45. Remondino, F.; Spera, M.G.; Nocerino, E.; Menna, F.; Nex, F. State of the art in high density image matching.
Photogramm. Rec. 2014, 29, 144–166. [CrossRef]
46. Swinfield, T.; Lindsell, J.A.; Williams, J.V.; Harrison, R.D.; Habibi, A.; Gemita, E.; Schönlieb, C.B.; Coomes, D.A.
Accurate Measurement of Tropical Forest Canopy Heights and Aboveground Carbon Using Structure from
Motion. Remote Sens. 2019, 11, 928. [CrossRef]
47. Jiang, Q.; Fang, S.; Peng, Y.; Gong, Y.; Zhu, R.; Wu, X.; Ma, Y.; Duan, B.; Liu, J. UAV-Based Biomass Estimation
for Rice-Combining Spectral, TIN-Based Structural and Meteorological Features. Remote Sens. 2019, 11, 890.
[CrossRef]
48. Ni, W.; Dong, J.; Sun, G.; Zhang, Z.; Pang, Y.; Tian, X.; Li, Z.; Chen, E. Synthesis of Leaf-on and Leaf-off
Unmanned Aerial Vehicle (UAV) Stereo Imagery for the Inventory of Aboveground Biomass of Deciduous
Forests. Remote Sens. 2019, 11, 889. [CrossRef]
49. Morgan, J.A.; Brogan, D.J.; Nelson, P.A. Application of Structure-from-Motion photogrammetry in laboratory
flumes. Geomorphology 2017, 276, 125–143. [CrossRef]
50. Grillo, S.M.; Pilia, E.; Vacca, G. Integrated study of the Beata Vergine Assunta dome with structure from
motion and diagnostic approaches. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2019, XLII-2/W11,
579–585. [CrossRef]
51. Kersten, T.P.; Lindstaedt, M. Automatic 3D object reconstruction from multiple images for architectural,
cultural heritage and archaeological applications using open-source software and web services. Photogramm.
Fernerkund. Geoinf. (PFG) 2012, 6, 727–740. [CrossRef]
52. Remondino, F.; Del Pizzo, S.; Kersten, T.P.; Troisi, S. Low-cost and open-source solutions for automated image
orientation a critical overview. In Progress in Cultural Heritage Preservation; Springer: Berlin/Heidelberg,
Germany, 2012; pp. 40–54.
53. Barazzetti, L.; Scaioni, M. Development and implementation of image-based algorithms for measurement of
deformations in material testing. Sensors 2010, 10, 7469–7495. [CrossRef] [PubMed]
54. Barazzetti, L.; Scaioni, M. Crack measurement: Development, testing and applications of an automatic
image-based algorithm. ISPRS J. Photogramm. Remote Sens. 2009, 64, 285–296. [CrossRef]
55. Deiana, G.L.; Dessi’, A.; Mistretta, F.; Sanna, G.; Stochino, F.; Urru, E.; Vacca, G.; Valdes, M. Terrestrial Laser
Scanner and Close Range Photogrammetry point clouds accuracy assessment for the structure deformations
monitoring. In Geospatial Information for a Smarter Life and Environmental Resilience, Proceedings of the FIG
Working Week; Fédération Internationale des Géomètres (FIG): Hanoi, Vietnam, 2019.
56. Agisoft Software. Available online: http://www.agisoft.com/ (accessed on 10 January 2019).
57. Photoscan Manual. Available online: http://www.agisoft.com/pdf/photoscan-pro_1_3_en.pdf (accessed on
10 January 2019).
58. Westoby, M.; Brasington, J.; Glasser, N.F.; Hambrey, M.J.; Reynolds, J.M. ‘Structure from-motion’
photogrammetry: A low-cost, effective tool for geoscience applications. Geomorphology 2012, 179, 300–314.
[CrossRef]
Remote Sens. 2019, 11, 1940 20 of 20

59. Fonstad, M.A.; Dietrich, J.T.; Courville, B.C.; Jensen, J.L.; Carbonneau, P.E. Topographic structure from
motion: A new development in photogrammetric measurement. Earth Surf. Process. Landf. 2013, 38, 421–430.
[CrossRef]
60. Fathi, H.; Dai, F.; Lourakis, M. Automated as-built 3D reconstruction of civil infrastructure using computer
vision: Achievements, opportunities, and challenges. Adv. Eng. Inform. 2015, 29, 149–161. [CrossRef]
61. Chen, J.G.; Wadhwa, N.; Cha, Y.J.; Durand, F.; Freeman, W.T.; Buyukozturk, O. Modal identification of simple
structures with high-speed video using motion magnification. J. Sound Vib. 2015, 345, 58–71. [CrossRef]
62. Mas, D.; Espinosa, J.; Roig, A.B.; Ferrer, B.; Perez, J. Image based subpixel techniques for movement and
vibration tracking. In Proceedings of the 11th European conference on non-destructive testing (ECNDT
2014), Prague, Czech Republic, 6 October 2014.
63. Zhang, D.; Guo, J.; Lei, X.; Zhu, C. A high-speed vision-based sensor for dynamic vibration analysis using
fast motion extraction algorithms. Sensors 2016, 16, 572. [CrossRef]
64. Javh, J.; Slavič, J.; Boltežar, M. The subpixel resolution of optical-flow-based modal analysis. Mech. Syst.
Signal Process. 2017, 88, 89–99. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).