Anda di halaman 1dari 13

sensors

Article
Development and Verification of a Novel
Robot-Integrated Fringe Projection 3D Scanning
System for Large-Scale Metrology
Hui Du 1,2 , Xiaobo Chen 1,2 , Juntong Xi 1,2,3, *, Chengyi Yu 1,2 and Bao Zhao 1,2
1 School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China;
sdduhui@hotmail.com (H.D.); xiaoboc@sjtu.edu.cn (X.C.); jg5bvictor@sjtu.edu.cn (C.Y.);
zhaobao1988@sjtu.edu.cn (B.Z.)
2 Shanghai Key Laboratory of Advanced Manufacturing Environment, Shanghai 200030, China
3 State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University,
Shanghai 200240, China
* Correspondence: jtxi@sjtu.edu.cn; Tel.: +86-21-3420-5693

Received: 5 November 2017; Accepted: 7 December 2017; Published: 12 December 2017

Abstract: Large-scale surfaces are prevalent in advanced manufacturing industries, and 3D


profilometry of these surfaces plays a pivotal role for quality control. This paper proposes a
novel and flexible large-scale 3D scanning system assembled by combining a robot, a binocular
structured light scanner and a laser tracker. The measurement principle and system construction
of the integrated system are introduced. A mathematical model is established for the global data
fusion. Subsequently, a robust method is introduced for the establishment of the end coordinate
system. As for hand-eye calibration, the calibration ball is observed by the scanner and the laser
tracker simultaneously. With this data, the hand-eye relationship is solved, and then an algorithm is
built to get the transformation matrix between the end coordinate system and the world coordinate
system. A validation experiment is designed to verify the proposed algorithms. Firstly, a hand-eye
calibration experiment is implemented and the computation of the transformation matrix is done.
Then a car body rear is measured 22 times in order to verify the global data fusion algorithm. The 3D
shape of the rear is reconstructed successfully. To evaluate the precision of the proposed method,
a metric tool is built and the results are presented.

Keywords: large-scale metrology; robot integrated system; structured light profilometry; hand-eye
calibration; global data fusion

1. Introduction
Large-scale thin wall and surface components are widespread in modern high-end manufacturing
industries, especially in the automotive, shipbuilding, astronautical and aeronautical industry. In these
fields, the surface forming quality usually equals the manufacturing quality of the corresponding
component to some extent, and finally affects the assembly quality to a great degree. Bad performance
in quality control means more pauses, modifications and even failures in the manufacturing process.
This will undoubtedly lengthen the production cycle and the costs will increase for this reason.
Therefore, to meet the precision requirements of manufacturing plants and reduce time/manpower
costs, an automatic, flexible and accurate large-scale 3D measurement method is of great importance
or even indispensable in some application scenarios.
Various methods have been developed in the field of large-scale 3D shape measurement and
a lot of instruments have been introduced for this purpose [1–4]. Among off-the-shelf products,
the Leica T-Scan has a good performance in large range metrology for its high speed and hand-held

Sensors 2017, 17, 2886; doi:10.3390/s17122886 www.mdpi.com/journal/sensors


Sensors 2017, 17, 2886 2 of 13

property [5]. However, it is not so suitable for on-site automatic inspection which is increasingly
important in intelligent manufacturing. Traditionally, the coordinate measuring machine (CMM) has
been extensively used in 3D shape measurement. For different products of different sizes, different
CMMs are correspondingly developed. The measurement range spans from less than 1 m to several
meters. With the development of visual sensor, computer vision technology and computation ability,
more and more kinds of vision measurement equipment are integrated with CMMs, which are usually
used in contact measurement for the dies, molds and so on [6–8]. However, the biggest drawback of
this kind of method is that only limited types and numbers of products can be sampled and brought
to the CMM for inspection. This means that even for relatively important products, it is difficult
for us to obtain the quality data of all the components, which may lead to possible quality failure.
If the CMM is integrated with the production line, it usually will not perform well in precision and
robustness. Recently, with the improvement in precision manufacturing, robot kinematics and control
engineering, robotic technologies have gone through huge developments. All these factors make robots
increasingly economical and practical in manufacturing industry. Therefore, more and more visual
sensors are integrated into robots to form a more flexible measurement system [9,10]. Furthermore, in
some researches, a turntable surrounding the robot is introduced to extend the metrology range of the
robot system [11,12]. Because it combines robot’s flexibility and accuracy of visual profilometry, it is
promising for online inspection of large-scale parts. Another 3D shape measurement technology widely
used in large volume profilometry is laser scanning [13–15]. The advantage of this kind of method
is that it is easier to implement and more economical. Under the assumption of good calibration,
the accuracy can also be assured. However, the weakness of this method is clear. In principle, laser
scanning can only acquire data along one line or several lines for each measurement. To enhance the
data quantity at every planned measurement position, a movement mechanism should be integrated
with the scanner [16]. Normally a linear rail or a rotary mechanism is necessary to achieve this function.
However, the introduction of the movement mechanism brings movement errors into the system.
To compensate the errors, the movement mechanism should be calibrated, which is also a challenging
task. Compared to the laser scanning method, structured light profilometry [17–21] can acquire the
data on a surface for each measurement. Without the movement mechanism, the accuracy of every
measurement depends only on the calibration of the visual sensors. If only the calibration is designed
and implemented carefully, the accuracy can be assured [22,23]. Additionally, this method acquires
much bigger data than laser scanning, which, compared to the line scanning method, produces more
accurate metrology results. Meanwhile, thanks to the area scanning property, it has a better time
performance. For all these advantages, the structured light scanning method will be promising if it is
used in large-scale metrology. Some work has been made in this direction. Paoli et al. [24] mounted
a binocular structured light scanner at the end of an anthropomorphic robot arm, and mounts the
robot with two linear guides (horizontal and vertical). By building a series of coordinate systems, the
measured data is unified into a fixed coordinate system defined by a total station. This approach works
well in the measurement of a large yacht hull. However, as stated in Paoli’s paper, the position of the
total station must be set carefully to ensure that all the optical reflectors can be observed simultaneously.
To some extent, this limits the flexibility of the system.
Overall, compared to other technologies, structured light scanning is an accurate and efficient
method for 3D metrology. The integration of structured light scanner and robot will dramatically
enlarge the measurement volume. By appropriate hand-eye calibration and coordinate transformation,
a software independent algorithm can be built, which makes large-scale data fusion a less challenging
task. Until now, the study about this kind of system is limited. Therefore, more work should be done
to improve the performance of this type of method.
In this paper, a novel integrated robotic scanning system is proposed for flexible and robust
large-scale measurement. The system is composed of a binocular structured light scanner, a robot
with six degrees of freedom (DOF) and a laser tracker. The structured light scanner is used to get
the surface data in specific measurement positions. An optical target corner cube reflector (CCR) is
Sensors 2017, 17, 2886 3 of 13

fixed on the base of the scanner. The end coordinate system (ECS) is built by rotation of the scanner.
The laser tracker is used to get the center of target ball and finish the data fusion. As for the hand-eye
calibration, different from the traditional method, by observing the target ball using the scanner
Sensors 2017, 17, 2886 3 of 13
and
the end coordinate system, the transformation matrix is computed. After obtaining the transformation
between end coordinate
calibration, system
different from the and world method,
traditional coordinate system (WCS),
by observing all ball
the target the using
data isthe
combined into the
scanner and
samethe end coordinate
coordinate system. Insystem, the the
this way, transformation
data fusion ismatrix
finishedis automatically.
computed. After obtaining the
transformation
The rest of thebetween
paper end coordinateas
is organized system and world
follows: Sectioncoordinate systemthe
2 introduces (WCS), all the
overall data is
measurement
combined into the same coordinate system. In this way, the data fusion is finished
principle and system construction. The building of the end coordinate system and hand-eye calibrationautomatically.
The rest of the paper is organized as follows: Section 2 introduces the overall measurement
algorithm is also stated in this part. Section 3 introduces the results of hand-eye calibration and global
principle and system construction. The building of the end coordinate system and hand-eye
data fusion. A quantitative evaluation is presented in this part. The paper will finish in Section 4 with
calibration algorithm is also stated in this part. Section 3 introduces the results of hand-eye calibration
a short
andconclusion.
global data fusion. A quantitative evaluation is presented in this part. The paper will finish in
Section 4 with a short conclusion.
2. The Proposed Approach
2. The Proposed Approach
2.1. Measurement Model and System Construction
2.1. Measurement
The integrated Model
roboticandscanning
System Construction
system incorporates an industrial robot with six degrees of
freedom (DOF), a laser tracker,
The integrated a digital system
robotic scanning fringe projecting
incorporates(DFP) scanner and
an industrial robota with
CCR sixwhich is mounted
degrees of
freedom
on the scanner.(DOF),
Thea DFP
laser tracker,
scannera digital
is fixedfringe projecting
at the (DFP) scanner
end effector and a CCR
of the robot. When which
theissystem
mounted works,
on thelocates
the robot scanner. The
the DFP scanner
scanner to theisplanned
fixed at the end effector
discrete of theand
positions, robot.
theWhen the system
scanner acquires works,
the 3Dthepoint
cloudrobot locates
of that the scanner region.
corresponding to the planned discrete positions, and the scanner acquires the 3D point
cloud of that corresponding region.
Instead of choosing an off-the-shelf product, the scanner is a binocular DFP equipment developed
Instead of choosing an off-the-shelf product, the scanner is a binocular DFP equipment
according to the implementation circumstances (e.g., work distance, illumination and reflectivity
developed according to the implementation circumstances (e.g., work distance, illumination and
of the surface).ofFor
reflectivity its characteristics
the surface). of high resolution
For its characteristics and low
of high resolution and sensitivity to to
low sensitivity thetheambient
ambient light,
the three-frequency and three-step phase shifting scheme is adopted to encode
light, the three-frequency and three-step phase shifting scheme is adopted to encode the fringe the fringe patterns.
To access the data
patterns. of a region,
To access the datanine patterns
of a region, arepatterns
nine sequentially projected onto
are sequentially the surface
projected onto theby the projector.
surface by
Thenthe
they are acquired
projector. Then theybyare
two cameras
acquired by from different
two cameras directions.
from After phase
different directions. Afterdecoding and image
phase decoding
and image
registration, theregistration, the final
final 3D point data3D point data is
is acquired. Theacquired.
working The workingof
principle principle
the DFP of scanner
the DFP scanner
is illustrated
is illustrated
in Figure 1. in Figure 1.
Captured image

CameraL CameraR
Projector

Figure1.1.Measurement
Figure principleofofthe
Measurement principle the scanner.
scanner.

LikeLike
any any other vision metrology system, the cameras should be previously calibrated. To enhance
other vision metrology system, the cameras should be previously calibrated. To enhance
the accuracy of calibration and measurement, a more accurate calibration method is applied [25]. This
the accuracy of calibration and measurement, a more accurate calibration method is applied [25].
approach acquires the calibration points in a form of grid point array, and gets the calibration results
This by
approach acquires the calibration points in a form of grid point array, and gets the calibration
Zhang’s algorithm [26]. With this approach, high accuracy of calibration and measurement can
results by Zhang’s
be assured. algorithm [26]. With this approach, high accuracy of calibration and measurement
can be assured.
The coordinate system of the integrated system comprises measurement coordinate system
(MCS), ECS and WCS, which is shown in Figure 2. The ECS is defined by the rotation of the scanner.
The WCS is fixed with the laser tracker.
Sensors 2017, 17, 2886 4 of 13

The coordinate system of the integrated system comprises measurement coordinate system (MCS),
ECS and WCS, which is shown in Figure 2. The ECS is defined by the rotation of the scanner. The WCS
Sensors 2017, 17, 2886 4 of 13
is fixed with the laser tracker.
TE-W
Robot CCR

ECS
WCS
TM-E

MCS

Structured light Laser tracker


scanner

Figure 2. Definition of the coordinate system.


Figure 2. Definition of the coordinate system.

P a point
BeingBeing in the
P a point robot
in the workspace,
robot workspace,the
the mapping relationship
mapping relationship between
between PW inPWCS
coordinate
coordinate W in WCS
and
and PM inPMCS
M in MCS is expressed
is expressed as as follows:
follows:
PW  TE W TM E PM (1)
PW = TE−W TM− E PM (1)
TM−E is the transformation relationship between MCS and ECS, TE−W denotes the transformation
TM −E isbetween
matrix ECS and WCS.
the transformation The acquired
relationship data of MCS
between this integrated
and ECS,system
TE −W isdenotes
alignedtheandtransformation
assessed
in the world frame which is defined by the laser tracker.
matrix between ECS and WCS. The acquired data of this integrated system is aligned and assessed in
By combining the area scanning structured light equipment, the laser tracker and the robot, this
the world frame which is defined by the laser tracker.
system reaches a high-level equilibrium in flexibility, velocity and accuracy. With the DFP scanner,
By combining the area scanning structured light equipment, the laser tracker and the robot,
3D shape information of the object will be acquired at one robot position. Through off-line
this system reachesthea whole
programming, high-level
surfaceequilibrium
of the workpiece in flexibility, velocitycan
or specified features and accuracy.InWith
be measured. the DFP
this case,
scanner,
the 3D
robotshape information
is only used to carryofthe the objectand
scanner, willallbethe
acquired
acquiredat one
data is robot
unifiedposition. Through to
to WCS. Compared off-line
programming, the whole
other methods, surface
this scheme of the
avoids theworkpiece or specified
error accumulation features
of multiple can betransformation
coordinate measured. Inand this case,
robotis main
the robot only body
used calibration.
to carry the Therefore,
scanner,high andaccuracy could be data
all the acquired expected. The adoption
is unified to WCS.ofCompared
area
scanning ensures high efficiency and resolution of the system, which is
to other methods, this scheme avoids the error accumulation of multiple coordinate transformation crucial for subsequent data
analysis. For a specified position, the scanning can be finished within 3 s, including the fringe
and robot main body calibration. Therefore, high accuracy could be expected. The adoption of area
projection time. With this integrated system, the complete measurement could be executed in a short
scanning ensures high efficiency and resolution of the system, which is crucial for subsequent data
period and the accuracy could be maintained.
analysis. For a specified
Mounting position,
the scanner andthe
CCRscanning can be finished
on the industrial within
robot, and 3 s, including
putting the fringe
the laser tracker API T3projection
in
time. With
front of the six degree of freedom (DOF) Fanuc robot M-710iC, the integrated 3D scanning system is and
this integrated system, the complete measurement could be executed in a short period
the accuracy could
constructed as be maintained.
illustrated in Figure 3a. The construction of the structured light scanner is shown in
Figure 3b. the scanner and CCR on the industrial robot, and putting the laser tracker API T3 in
Mounting
front of theAssix
shown
degreein Figure 3a, to construct
of freedom (DOF) Fanucthe ECS,robot
a CCRM-710iC,
is set on the scanner.
the When3D
integrated the scanning
system works,system is
the scanner acquires the point cloud first, and then rotates to another two positions.
constructed as illustrated in Figure 3a. The construction of the structured light scanner is shown in Using these three
points, the ECS is built. Obviously, this method reduces the constraints of the relative position
Figure 3b.
between the laser tracker and the reflectors. Actually, only one CCR is used in this method and the
As shown in Figure 3a, to construct the ECS, a CCR is set on the scanner. When the system works,
ECS is constructed by the rotation of the scanner. Therefore, if only the laser tracker is put at an
the scanner acquires
appropriate the point
position from the cloud first,the
scanner, and then
ECS canrotates to another
be smoothly twobypositions.
acquired three timesUsing these three
of rotation.
points,Compared
the ECS istobuilt. Obviously, this method reduces the constraints of the
other methods, this approach avoids most of the occlusion and is relatively more relative position between
the laser tracker
flexible and the reflectors. Actually, only one CCR is used in this method and the ECS is
and robust.
constructed by the rotation of the scanner. Therefore, if only the laser tracker is put at an appropriate
position from the scanner, the ECS can be smoothly acquired by three times of rotation. Compared to
other methods, this approach avoids most of the occlusion and is relatively more flexible and robust.
Sensors 2017, 17, 2886 5 of 13
Sensors 2017, 17, 2886 5 of 13

Sensors 2017, 17, 2886 5 of 13

Scanner

Scanner
CCR

CCR
Robot

Robot
Laser tracker

Laser tracker

(a)
(a)

(b)
(b) system; (b) Binocular structured light scanner.
Figure 3. (a) Construction of the integrated 3D scanning
Figure 3. (a) Construction of the integrated 3D scanning system; (b) Binocular structured light scanner.
Figure 3. (a) Construction of the integrated 3D scanning system; (b) Binocular structured light scanner.
2.2. End Coordinate System Construction
2.2. End Coordinate System Construction
2.2. End Coordinate
As the System
beginning stepConstruction
of building the global data fusion algorithm, the ECS should be previously
As the beginning
constructed. step of
The robot’s J6 building thescanner
axis and the global aredata fusion
used algorithm,
to implement the
this ECSThe
work. should
CCR be previously
is put on
As the beginning step of building the global data fusion algorithm, the ECS should be previously
the basement
constructed. The of the scanner.
robot’s J6 axis When
and thescanner
the system areworks,
usedthetoscanner is positioned
implement this to a The
work. planned
CCR point
is put on
constructed. The robot’s J6 axis and the scanner are used to implement this work. The CCR is put on
and its position
the basement of the isscanner.
acquired When
by the laser
the tracker. works,
system Then fringes
the are projected
scanner is to acquire to
positioned theapoint cloudpoint
planned
the basement of the scanner. When the system works, the scanner is positioned to a planned point
of position
and its the corresponding
is acquired surface
bybyarea.
the After that, J6Thenaxis rotates
fringesfor another two times, and thetheposition
and its position is acquired thelaser
lasertracker.
tracker. Then fringes are
are projected
projected to acquire
to acquire the point point
cloudcloud
is recorded by the laser tracker. Finally, after rotation for three times, three points (P1, P2, P3) are
of theofcorresponding
the corresponding surface area.
surface area.After
Afterthat,
that, J6
J6 axis rotatesfor
axis rotates foranother
another twotwo times,
times, andand the position
the position
recorded. The first point P1 can be taken as the origin point. Connecting P1 and P2 into a straight line,
is recorded
is recorded by the
by the laserlaser tracker.Finally,
tracker. Finally,after
after rotation
rotation for
forthree
threetimes,
times, three points
three (P1,(P1,
points P2, P3)
P2,areP3) are
X-axis is in the same direction. Z-axis is built by vector cross product. Y-axis can be obtained by the
recorded.
recorded. The The
first first
pointpoint
P1 P1 can
can be be taken
taken as
as the
the origin
origin point.
point. Connecting
Connecting P1 and
P1 P2
and into
P2 a straight
into a line, line,
straight
same method. In this way, the ECS is constructed. This process is explained in Figure 4.
X-axisX-axis
is in isthe
in same
the same direction.
direction. Z-axisisisbuilt
Z-axis built by
by vector
vector cross
crossproduct.
product. Y-axis cancan
Y-axis be obtained
be obtainedby the by the
same method. In this way, the ECS is constructed. This process is explained in Figure 4.
same method. In this way, the ECS is constructed. This process is explained in Figure 4.
z
z
P1
P1 y
y
P2
xP2 P3

x P3
Figure 4. Construction of ECS.

Figure4.4. Construction
Figure Construction ofofECS.
ECS.
Sensors 2017, 17, 2886 6 of 13
Sensors 2017, 17, 2886 6 of 13

2.3.
2.3. Hand-Eye
Hand-Eye Calibration
Calibration
In
In classic
classichand-eye
hand-eyecalibration
calibration algorithm,
algorithm, to acquire
to acquirethe hand-eye
the hand-eye transformation
transformation matrix, the robot
matrix, the
should
robot should take the eye to several different positions and observe the same calibration rig. The robot
take the eye to several different positions and observe the same calibration rig. The robot
kinematics
kinematics parameters
parameters are are used
used toto solve
solve the
the transformation
transformation matrix.
matrix. Different
Different fromfrom the
the traditional
traditional
method, in the proposed method, the robot is only used as an orienting
method, in the proposed method, the robot is only used as an orienting device. It is unnecessary device. It is unnecessary for for
us
to apply robot’s kinematics parameters and the kinematics error can
us to apply robot’s kinematics parameters and the kinematics error can be bypassed. This benefitsbe bypassed. This benefits the
improvement
the improvement of calibration accuracy.
of calibration accuracy.
In
In the proposed method, a CCR
the proposed method, a CCR is is used
used asas the
the calibration
calibration target
target ball.
ball. In
In the
the calibration
calibration process,
process,
the ball is measured by the structured light scanner and laser tracker
the ball is measured by the structured light scanner and laser tracker simultaneously simultaneously (Figure(Figure
5). Firstly,
5).
to acquire the center of target ball, the scanner is used to get the point cloud data.
Firstly, to acquire the center of target ball, the scanner is used to get the point cloud data. This data is This data is used
to obtain the ball as Xas i . For
𝑖 ease of use, it should be
used to obtain thecenter in MCS,
ball center and this
in MCS, and center can be
this center candenoted
be denoted M 𝑋𝑀 . For ease of use, it should
saved
be savedin the form
in the of homogeneous
form of homogeneous coordinate.
coordinate.At theAtsame time,time,
the same the CCR ball isball
the CCR measured by thebylaser
is measured the
tracker. Actually, based on the principle of the laser tracker, the ball center can
laser tracker. Actually, based on the principle of the laser tracker, the ball center can be acquired in be acquired in this way.
It can
this be denoted
way. It can beas Ci . To transform
denoted Ci into ECS,
as Ci. To transform the ECS,
Ci into ECS isthefirst
ECSbuilt as stated
is first built asinstated
Sectionin 2.2.
SectionTaking
2.2.
P1 as the
Taking P1origin
as thepoint,
originthe coordinates
point, can be can
the coordinates acquired by projecting
be acquired P1 Ci onto
vectorvector
by projecting P1Ci the
ontothree ECS
the three
axes. This new
ECS axes. Thiscoordinate of Ci canofbeCdenoted
new coordinate i can beas XEi , which
denoted as is𝑋𝐸𝑖also transformed
, which is also to the homogeneous
transformed to the
coordinate form. By putting the CCR at several different positions
homogeneous coordinate form. By putting the CCR at several different positions in the in the scanner vision field,
scanner two
vision
homogeneous coordinatecoordinate
field, two homogeneous vector groups vectoraregroups
constructed, which are shown
are constructed, which are as follows:
shown as follows:

M XM XiMMi · · · 
h i
1 1 2X M2 3 3
X MX =  X MX M XX
MM · · · X (2)
(2)
h i
11 2 2 3 3 ii
XEX E = XX X X · · · XEE 
X · · · (3)
(3)
 EE XE E XE E

To solve the transformation matrix between the scanner and ECS, an equation is built as follows:
To solve the transformation matrix between the scanner and ECS, an equation is built as follows:
X M = TE− M XE (4)
X M  TE  M X E (4)

where TTEE−M
where −Misisthe
thehand-eye
hand-eyetransformation
transformationmatrix.
matrix.ItItcan
canbe bewritten
writteninto
intothe
thefollowing
followingform:
form:

RR T33×1 
" #
333 T
M  

11 
E =
TE−TM (5)
(5)
00

133 444×4

(R T)
MCS ECS

Xi M XiE

C1 C2 Ci

Figure 5.
Figure 5. Hand-eye
Hand-eye calibration.
calibration.

In this matrix, R means the rotation matrix, and T the translation vector. According to the
In this matrix, R means the rotation matrix, and T the translation vector. According to the property
property of rotation matrix, it exists the following constraint:
of rotation matrix, it exists the following constraint:
RTT R  I (6)
R R=I (6)
In this way, the computation of the transformation matrix can be transformed into an
optimization problem with a constraint:


min X M  TE  M X E
2

 F (7)

 s.t RT R  I 33
Sensors 2017, 17, 2886 7 of 13

In this way, the computation of the transformation matrix can be transformed into an optimization
problem with a constraint: (
mink X M − TE− M XE k2 F
(7)
s.t R T R = I3×3

By eliminating the translation term, this optimization problem can be converted to an


orthogonal-force-consistency problem, which can be solved by single value decomposition.

2.4. Global Data Fusion Model


For the structured light profilometry, the data on a surface area is acquired in every scanning.
Therefore, to get the entire data of a large-scale component, the measurement should be implemented
for a lot of times according to the size. In this process, the scanner is carried to different positions
by robot. The surface point cloud data at every position can be obtained. To combine all the data,
the position and pose of the end-effector should be tracked.
For every measurement position, an {ECS}i is built by tracking the CCR ball. Let TEi −W be the
transformation between {ECS}i and WCS. TEi −W can be written into the following form:
 
nx ox px tx
 ny oy py ty 
TEi −W =  (8)
 
nz oz pz tz

 
0 0 0 1

In this matrix, (nx ny nz )T , (ox oy oz )T , (px py pz )T are the corresponding unit vectors of {ECS}i
coordinate axes in WCS. (tx ty tz )T is the position of the origin point of {ECS}i in WCS. Until now,
both the hand-eye transformation matrix TE-M and the transformation matrix between {ECS}i and WCS
(TEi −W ) have been obtained. To combine all the data, the following equation can be used:

i i
TM −W = TE− M TE−W (9)

Here TMi
−W is the transformation matrix between MCS and WCS. With this equation, all the
acquired data can be unified into WCS, and the data fusion can be finished automatically.

3. Results
To verify the effect of proposed methodologies, several experiments are designed and
implemented. Through hand-eye calibration experiment, the transformation matrix between MCS and
ECS is computed. Based on this relationship, the global data fusion experiment is executed and the 3D
shape of a car body rear is acquired. To the end of quantitative assessment, a metric tool is constructed
and the evaluation results are demonstrated.

3.1. Hand-Eye Calibration


The hand-eye calibration algorithm has been introduced in Section 2.2. According to the algorithm,
the experiment is designed, which is shown in Figure 6. In the calibration process, the scanner and
robot should be kept still. The target ball is put on 15 different positions in the scanner vision field
range. For every position, the ball is measured by the scanner and the laser tracker simultaneously.
After this, the ball is set on the scanner. After three times’ rotation, the ECS is constructed. The data is
shown in Table 1.
Sensors 2017, 17, 2886 8 of 13
Sensors 2017, 17, 2886 8 of 13

CCR

Figure6.6.Hand-eye
Figure Hand-eye calibration
calibration experiment.
experiment.

Table 1. Hand-eye calibration data.


Table 1. Hand-eye calibration data.
WCS ECS MCS
X WCSY Z X Y
ECS Z X Y MCS Z
No.1 −603.234 −570.824 −623.064 −654.862 −154.646 −104.806 289.155 88.323 605.415
No.2 X−527.529 −541.327
Y Z
−623.513 X
−667.577 Y
−167.702 Z
−25.628 X 9.639 Y 622.489 Z
285.935
No.1 No.3 −440.814 −505.516 −621.711
−603.234 −570.824 −623.064 −679.309
−654.862 −183.062
−154.646 −104.806 289.155−81.623
66.197 280.02 639.968605.415
88.323
No.2 No.4 −456.599−541.327
−527.529 −455.086 −623.513
−621.325 −−649.743
667.577 −226.843
− 167.702 67.418
− 25.628 227.691
285.935−83.1199.639
640.253622.489
No.3 No.5 −540.046−505.516
−440.814 −475.961 −621.711
−622.876 −−631.899
679.309 −223.167
− 183.062 −16.662
66.197 220.646
280.02 0.517 624.339639.968
−81.623
No.4 No.6 −634.259−455.086
−456.599 −513.428 −621.325
−622.370 −−616.476
649.743 −206.324
− 226.843 −115.448
67.418 225.462
227.69198.891 602.866640.253
−83.119
No.5 No.7 −657.153 −469.599 −621.906
−540.046 −475.961 −622.876 −−587.494
631.899 −245.701
− 223.167 −122.851 176.813
−16.662 220.646 106.193 601.124624.339
0.517
No.6 No.8 −581.189−513.428
−634.259 −435.511 −622.370
−622.558 −−598.298
616.476 −262.613
− 206.324 −42.041
−115.448169.253
225.46225.72398.891
618.982602.866
No.7 No.9 −476.216−469.599
−657.153 −395.233 −621.906
−620.951 −−614.414
587.494 −278.986
− 245.701 68.033
− 122.851165.075
176.81349.782 551.496601.124
106.193
No.8 No.10 −494.938 −344.801 −620.797
−581.189 −435.511 −622.558 −−584.017
598.298 −323.345
− 262.613 66.487 111.551
−42.041 169.253 −82.539 640.278618.982
25.723
No.9 No.11 −577.894 −369.441 −621.975
−476.216 −395.233 −620.951 −−567.816
614.414 −316.284
− 278.986 −18.234
68.033 107.993
165.075 1.811 623.791551.496
49.782
No.10 No.12 −671.393−344.801
−494.938 −671.393 −620.797
−621.606 −−549.986
584.017 −304.248
− 323.345 −114.556
66.487 107.749
111.55197.758 603.093640.278
−82.539
No.11 No.13 −690.162−369.441
−577.894 −353.111 −621.975
−621.409 −−520.717
567.816 −346.523
− 316.284 −116.918
−18.234 56.707
107.99399.8611.811602.826623.791
No.12 No.14 −618.261 −325.513 −621.534
−671.393 −671.393 −621.606 −−532.745
549.986 −358.414
− 304.248 −41.782 53.710
−114.556 107.749 25.136 618.746603.093
97.758
No.13 No.15 −513.567 −284.161 −620.443
−690.162 −353.111 −621.409 −−548.676
520.717 −375.996
− 346.523 68.258 48.626
−116.918 56.707 −84.303 640.652602.826
99.861
No.14 −618.261 −325.513 −621.534 −532.745 −358.414 −41.782 53.710 25.136 618.746
No.15 With this data,
−513.567 and by using
−284.161 the algorithm
−620.443 proposed
−548.676 in Section
−375.996 2.2, TE−M
68.258 is solved−finally,
48.626 84.303 which
640.652
is shown as follows:

With this data, and by using the  0.5634


algorithm0.0051 0.8262
proposed 2.2, TE−M is solved finally, which
in Section8.2035
is shown as follows:  0.8262 0.0065 0.5633 53.9117 
TE  M     (10)
−0.5634
0.0025 10.0051
.0000 −0.0079
0.8262 11.7849 
8.2035
  0.8262 
  0.0000 .0000 −00.5633
00.0065 .0000 .0000  
−153.9117
TE− M =  (10)

 0.0025 −1.0000 −0.0079 −11.7849 

3.2. Global Data Fusion 0.0000 0.0000 0.0000 1.0000

3.2. GlobalAfter
Datathe calibration of the hand-eye relationship, a car body rear with a size of 1400 mm × 500 mm
Fusion
× 400 mm was measured to verify the proposed scheme. The experiment system is illustrated in Figure 3.
After the calibration
According of the hand-eye
to the path planning results, therelationship, a carbybody
scanner is carried rear to
the robot with 1400 mm ×
a size ofpositions.
22 different
500 mm × 400 mm was measured to verify the proposed scheme. The experiment system is illustrated
in Figure 3. According to the path planning results, the scanner is carried by the robot to 22 different
positions. For every position, corresponding surface data is acquired by the structured light scanner.
Figure 7 shows the point cloud data in a form of triangular meshes representation.
SensorsSensors
2017, 17, 2886
2017, 17, 2886 9 of 139 of 13
Sensors 2017, 17, 2886 9 of 13
For every position, corresponding surface data is acquired by the structured light scanner. Figure 7
Simultaneously,
shows
For everythe position,
point cloud{ECS}datai is
correspondingin constructed
a form by is
of triangular
surface data tracking
meshes the
acquired CCR
representation.
by the ball. According
structured light scanner. to Figure
the method
7
shows
proposed Simultaneously,
inthe point cloud
Section i
2.3, T{ECS}
data i isa form
in
—the constructed
of by tracking
triangular
transformation meshes
matrix the CCR ball.
representation.
between {ECS} According
and to
WCS—is the method
constructed.
E −W i
𝑖
proposed
The hand-eye in Section 2.3,
Simultaneously,
transformation 𝑇𝐸−𝑊
{ECS} —the
is transformation
constructed
i matrix matrix
by tracking
was presented between
by the CCR {ECS}
Formula ball.
(9) in and WCS—is
iAccording
Section to constructed.
3.1. the method the
Therefore,
𝑖
i The
TM−W —the hand-eye
proposed in transformation
Section 2.3,
transformation matrix 𝑇 𝐸−𝑊 matrix
—the was presented
transformation by
matrix Formula
between (9)
{ECS}in iSection
and
between MCS and WCS—can be built according to the algorithm 3.1.
WCS—is Therefore, the
constructed.
𝑖
𝑇
The
𝑀−𝑊 —the
hand-eye transformation
transformation matrix between
matrix was MCS
presentedand WCS—can
by Formula be built
(9) in according
Section to
3.1. the algorithm
Therefore, the
proposed𝑖
in Section 2.3. With this data, the point cloud data in each position can be transformed
𝑇𝑀−𝑊
proposed—the in transformation
Section 2.3. With this data,
matrix betweenthe point
MCScloud data in each
and WCS—can beposition
built accordingcan be transformed
to the algorithminto
into WCS. The global data fusion is implemented automatically. Figure 8 shows the multicolor
WCS.
proposed Thein global
Sectiondata fusion
2.3. With thisisdata,
implemented
the point cloud automatically.
data in eachFigure
position8 can shows the multicolor
be transformed into
representation
representationof point
WCS. The global
cloud
of point
data cloud
data atiseach
fusiondata
measurement
at each
implemented measurement position (Figure
position
automatically. (Figure
Figure
8a)
8 8a)
and
shows
thethe
and triangular
the meshes
triangular
multicolor
representation
meshes
representation (Figure
representation 8b) (Figure
of point of the holistic
cloud 8b) of
data carholistic
at the
each rear surface.
car rearItposition
measurement is illustrated
surface. It that8a)there
is illustrated
(Figure andthatexist
the overlapping
there exist
triangular
areasoverlapping
between
meshes adjacent
representation scans.
areas between These
(Figure 8b)overlapping
adjacent ofscans. Theseareas
the holistic car have
overlapping been
areas
rear surface. usedhave
It istoillustrated
evaluate
been used the fusion
to evaluate
that accuracy.
the
there exist
By a fusion
properaccuracy.
overlappingpathareas
planning,
Bybetween
a proper the percentage
path
adjacent planning, ofthethe
scans. These overlapping
percentage
overlapping thearea
of areas is set
overlapping
have beentoused10%istoto
area set40%, which
to 10%
evaluate to
the is
enough40%,
fusion which is
for accuracy. enough
precisionBy for precision
a proper pathIn
computation. computation.
planning, In the
the percentage
the fusional fusional
data, aboutof the data, about
20 overlapping
million points 20 million
areaare
is set points are
to 10% to
acquired, which
acquired,
40%, which
is redundant which
for is redundant
isaccuracy
enough for for accuracy
precision
evaluation. evaluation.
computation.
Therefore, In athe
by Therefore,
fusional data,
resampling by aabout
resampling
algorithm, 20 the algorithm,
million
number the
pointshas
are been
number
acquired,
reduced has
to about been
which reduced
is redundant
2 million. to about 2 million.evaluation. Therefore, by a resampling algorithm, the
for accuracy
number has been reduced to about 2 million.

Figure
Figure 7. Surface
7. Surface point
point clouds
clouds ofof2222times
timesmeasurement
measurement represented
representedinin
a form of triangular
a form meshes.
of triangular meshes.
Figure 7. Surface point clouds of 22 times measurement represented in a form of triangular meshes.
A simple visual inspection can be used to assess the alignment accuracy even without proper
A simple
metric
A tool.
visual inspection
Thevisual
simple stripped can
patterns
inspection onbe
can
used
the to assess
triangular
be used meshes
to assess
thealignment
alignment
the surface
accuracy
represent
accuracytheeveneven without
misalignment proper
errors
without proper
metric tool.tool.
between
metric The stripped
overlapping
The patterns
areas
stripped ononthe
of different
patterns triangular
thepoint meshes
cloudsmeshes
triangular surface
(Figuresurface represent
8b). represent thethe misalignment
misalignment errorserrors
between overlapping
Although
between the areas
overlappingvisual of
areas different
assessment
of different point
can clouds
afford
point (Figure
a qualitative
clouds 8b).
evaluation for the alignment precision,
(Figure 8b).
Although
the result the the
Although visual
cannot assessment
bevisual
considered can
canafford
as exhaustive.
assessment affordToa acquire
qualitative evaluation
the evaluation
qualitative quantitative for
thethe
result
for alignment
of the precision,
misalignment
alignment precision,
error,
the result the
the result proximity
cannot
cannot between
be considered
be considered the
asasoverlapping
exhaustive. To
exhaustive. areas (Figure
To acquire
acquire the 9) of different
thequantitative
quantitative point
result clouds
of the
result is computed.
misalignment
of the misalignment
error,error, the proximity
the proximity between
between thethe overlappingareas
overlapping areas (Figure
(Figure 9)
9)ofofdifferent
differentpoint
pointclouds is computed.
clouds is computed.

(a)
(a)

Figure 8. Cont.
Sensors 2017, 17, 2886 10 of 13
Sensors2017,
Sensors 17,2886
2017,17, 2886 10ofof13
10 13
Sensors 2017, 17, 2886 10 of 13

(b)
(b)
(b)
Figure 8. (a)
Figure Point
8. (a) cloud
Point clouddata
datafusion
fusion result; (b)triangular
result; (b) triangular meshes
meshes representation.
representation.
(a)Point
Figure8.8.(a)
Figure Pointcloud
clouddata
datafusion
fusionresult;
result;(b)
(b)triangular
triangularmeshes
meshesrepresentation.
representation.

Figure 9. Overlapping areas between all the aligned point clouds.


Figure
Figure 9. 9. Overlapping areas areas between
between all all the
the aligned
aligned pointpoint clouds.
clouds.
A metric tool has been
Figure developedareas
9. Overlapping to compute
between thealltranslation
the alignedand point rotation
clouds. error. Compared to
the perpendicular directions (x and y direction), the misalignment error along the optical scanner
A metric
A metrictool toolhashasbeen beendeveloped
developedtoto compute
compute thethe translation
translation andand rotation
rotation error.error. Compared
Compared to
to the
Aviewing
metric direction
tool has (z direction)
been developeddominatesto [24]. For the
compute mosttranslation
of the car rear and surface,
rotation the error.
curvature is low. to
Compared
the perpendicular
perpendicular directions
directions (x and(xy and y direction),
direction), the misalignment
the misalignment error along errorthe along
optical thescanner
optical viewing
scanner
the Therefore, the directions
perpendicular error along (x theand
z direction
y For is the most
direction), the significant
misalignment for alignment
error precision
along the isevaluation.
optical scanner
viewing
direction direction
(z direction) (z direction)
dominates dominates
[24]. [24].
most For
of themostcar of
rearthe car
surface,
The translation error is defined as a projection of the distance between the nearest points to the
rear thesurface,
curvature the curvature
low. is low.
Therefore,
viewing
Therefore, direction
the (z
error direction)
theofz along dominates
the plane.
z direction [24].
is the For most of the car rear surface, the curvature is low.
the error
normal along vector direction
the fitting is the Asmostshown in most
significantFigure significant
for10,alignment
given two forprecision
alignment
different precision
evaluation.
point clouds (PC1 evaluation.
and
Therefore,
The
The the error
translation along
error the
is z direction
defined is the
as aa projectionmost
projection significant
of the
the distancefor alignment
between precision
the Then
nearest evaluation.
points to to the
the
PC2),translation
for each point error set, is
thedefined as
mean distance (dm) between of alldistance
the pointsbetween
are acquired. the nearestthe points
distance
The
normal
normal translation
vector
vector
is used of the
of the
to get error
fitting
fitting
a radius is defined
plane.
plane.
which as
As a projection
shown in Figure
can be used to define a circle. of the
10, distance
given
10, given twobetween
different
two different
With these points, andthe nearest
point
point points
clouds
by aclouds
least-square (PC1to
(PC1 andthe
and
normal
PC2),plane
PC2), vector
for
for each
each of
fitting the
point fitting
estimation
point set, theplane.
set, the As
mean distance
computation,
mean shown
distance (dm)
the(dm) in Figure
normal between 10,all
vector
between given
all
(n1,the
the two
anddifferent
points
n2)points arebest
the
are point plane
acquired.
fitting
acquired. clouds
Then(π1,
Then the(PC1 π2) and
distance
distance
PC2), for
canto
is used
is used each
beget
to get point
computed
aa radiusset,
radius the
[27].which
which mean
Thencan can distance
the nearest
be used
be used (dm)
point between
pairs are
to define
to define all
searched
aa circle. the
circle. With points
and these
With the
theseare acquired.
distance
points,
points, between Then
and by
and bythesethe distance
points
aa least-square
least-square
isplane
usedareto get
acquired.a radius
In thiswhich
way, can
the be used
distance d to
fromdefine
C1 toa circle.
PC2 is With
defined these
as 𝑑 = ̅̅̅̅̅̅
|𝐶
points,
1 𝐶 2 |𝑐𝑜𝑠∠𝐸𝐶
and by1 𝐶
a 2least-square
.
plane fitting estimation computation, the normal vector (n1, n2) and the best fitting plane (π1,(π1,
fitting estimation computation, the normal vector (n1, n2) and the best fitting plane π2)
π2) can
plane fittingTheestimation
rotation error is defined as the
computation, the angle
normal value between
vector (n1, the unit
n2) andvectors
the n1 and
best n2. Byplane
fitting traversing(π1, π2)
cancomputed
be be computed [27].[27].
ThenThen the nearest
the nearest point point
pairspairs are searched
are searched and and the distance
the distance betweenbetweenthese these
pointspoints
are
can all the points[27]. in point cloud, the translation and rotation error computation is ultimately
̅̅̅̅̅̅ finished. As
arebe computed Then thedistance
nearest pointC1pairs are searched and
as dthe as distance
between these points
acquired.
acquired. In this In this
way,way, the
the distance d from d from toC1 PC2to PC2
is is defined
defined = 𝑑C1=C2|𝐶 cos𝐶
1 2 ∠ |𝑐𝑜𝑠∠𝐸𝐶
EC C . 𝐶
1 2 .
stated inIn [24], theway,
accuracy of the least squareC1 fitting algorithm significantly ̅̅̅̅̅̅
𝑑 = depends
|𝐶 𝐶2and on1the2 radius
|𝑐𝑜𝑠∠𝐸𝐶 . r,
are acquired.
The
The rotation
rotation thiserror
error is
is the
defined
defineddistance as
as thed angle
the from
angle valueto PC2
value betweenis defined
between the
the unit
unitas vectors
vectors 1n1
n1 and n2.
n2. 1 𝐶2traversing
By
By traversing
which can only be estimated by empirical analyses [28]. In the presented case, the value is defined as
Thepoints
all the rotation error iscloud,
defined theas the angleand value between thecomputation
unit vectors n1 and n2. Byfinished.
traversing
all r = 6 dm. in in point
pointcloud, the translation
translation and rotation
rotation error
error computation is ultimately
is ultimately finished. As
all
As the
stated points
stated inthe
in [24], point
in [24], the cloud, of
accuracy
accuracythe translation
the
of least
the andfitting
square
least square rotation error computation
fittingalgorithm
algorithm is depends
significantly
significantly ultimately
depends ononfinished.
the Asr,
the radius
stated
which in
which can [24],
can onlythe
only beaccuracy of
be estimated
estimated bythe least square
by empirical fitting
empirical analyses algorithm
analyses [28].
[28]. In the
In significantly
the presented depends
presented case,
case, the on
the value the
value is defined r,
radius
is defined as
as
n2
which can only be estimated by π2 PC2 analyses [28]. In the presented case, the value is defined as
empirical
rr == 6 dm.
r = 6 dm. C2 E
n1
nd2
π2 PC2
C1
n2
π PCπ21
2
PC1 C2 E
r Cn21 E
d
n1 C1 d
π1 C1
Figureπ10. Definition of the distance and angle error.
1
PC1
PC1 r
r

Figure 10. Definition of the distance and angle error.


Figure
Figure10.
10.Definition
Definitionofofthe
thedistance
distanceand
andangle
angleerror.
error.
Sensors 2017, 17, 2886 11 of 13
Sensors 2017, 17, 2886 11 of 13

By using the metric tool, the translation and rotation errors are computed, and Figure 11 shows
the results.
results. In
In this
thisFigure,
Figure,the
thehorizontal
horizontalaxis
axisrepresents
representsthethe error
error value
value and and
thethe vertical
vertical axisaxis denotes
denotes the
the percentage
percentage of the
of the corresponding
corresponding error.
error. It isItillustrated
is illustrated in Figure
in Figure 11a11a that,
that, forfor most
most ofof the
the points
points (witha
(with
a percentageofof88.53%),
percentage 88.53%),the
thedistance
distanceisisless
lessthan
than0.60.6mm.
mm.IfIfthe
thethreshold
threshold value
value isis set
set to
to 1 mm,
mm, almost
all the
thepoints
points(97.76%)
(97.76%)arearecomprised.
comprised.A similar
A similar situation occurs
situation for the
occurs forrotation error error
the rotation (Figure 11b). Most
(Figure 11b).
of
Mosttheoferror valuevalue
the error is less thanthan
is less 10 10
degrees
degrees (97.21%). ToTo
(97.21%). demonstrate
demonstratethe theerrors
errorsmore
moreclearly,
clearly, the
maximum value (max), minimum value (min), mean value (µ (µ)) and standard deviation value (σ) are
also summarized in Table 2. With these quantitative statistic results, the quality of the data fusion can
be assessed objectively.
objectively.

(a) (b)
Figure
Figure 11.
11. (a)
(a) Distribution
Distribution of
of the
the translation
translation error;
error; (b)
(b) distribution
distribution of
of the
the angle error.
angle error.

Table 2. Statistic of the distance and angle error.


Table 2. Statistic of the distance and angle error.
max min 𝛍 𝛔
max min µ σ
d (mm) 1.5081 0 0.2965 0.2465
d (mm)
A (deg.) 1.5081
20.0841 0
0.0029 0.2965 2.6185
2.8333 0.2465
A (deg.) 20.0841 0.0029 2.8333 2.6185

4. Conclusions
4. Conclusions
This paper presents an integrated system for large-scale component profilometry. In this system,
This paper
a structured presents
light scanneranisintegrated systemsurface
built to acquire for large-scale component
point cloud data atprofilometry.
each position. InThe
this robot
system,is
a structured
only used aslight scanner is
an orienting built in
device to acquire surfaceBy
large volume. point cloud data
establishing theattransformation
each position. The robot is
relationship
only usedmeasurement
between as an orienting device in
coordinate large (MCS)
system volume. andByworld
establishing
coordinatethe system
transformation
(WCS), all relationship
the data is
combined into WCS which is defined by laser tracker. For this system, the constructionall
between measurement coordinate system (MCS) and world coordinate system (WCS), of the
the data
end
is combined into WCS which is defined by laser tracker. For this system,
coordinate system (ECS) plays a pivotal role. Here the CCR is mounted on the base of the scanner. the construction of the
end coordinate system (ECS) plays a pivotal role. Here the CCR is
After three times’ rotation, the ECS is constructed. Additionally, different from classic hand-eye mounted on the base of the
scanner. After three times’ rotation, the ECS is constructed.
calibration method, in this scheme, the hand-eye transformation matrix is computed by a Additionally, different from classic
hand-eye calibration
synchronized observationmethod, in this
of the scheme,
scanner andthe hand-eye
laser tracker.transformation
This approachmatrix makes is the
computed
hand-eye by
a synchronized observation of the scanner and laser tracker. This approach
calibration independent from robot kinematics parameters, which makes the calibration more robust makes the hand-eye
calibration
and easier to independent
be implemented. from robot kinematics
An algorithm parameters,
is also built to which makes
solve the the calibration
transformation more
matrix robust
between
and easier to be implemented. An algorithm is also built to solve the
ECS and WCS. In this way, all the data can be automatically combined to the unified coordinate transformation matrix between
ECS andTo
system. WCS. In this
verify the way,
effectallofthe
thedata can be automatically
proposed combined toexperiments
method, corresponding the unified coordinate
are designed system.
and
To verify the effect of the proposed method, corresponding experiments
conducted. With this data, the transformation relationship between MCS and WCS is computed. are designed and conducted.
With
Finally,thisalldata, the transformation
the data is combined into relationship between MCS
the same coordinate andand
system, WCS theisshape
computed. Finally,
of a car body all
rearthe
is
data is combined into the same coordinate system, and the shape of
reconstructed successfully. To evaluate the precision of the proposed method, a metric tool is a car body rear is reconstructed
successfully.
developed andTothe
evaluate
accuracy thedata
precision of the proposed
is presented. method,
The translation a metric
error is lesstool is 0.6
than developed
mm for andmosttheof
accuracy data is presented. The translation error is less than 0.6 mm
the points (88.53%). A mean/maximum value of 0.2965/1.5081 mm is detected in the work volume. for most of the points (88.53%).
A
Themean/maximum
standard deviation valueisof0.2465
0.2965/1.5081
mm. Formm is detected
rotation error,inthethemean
workand volume. The standard
maximum deviation
value are 2.8333
is 0.2465 mm. For rotation error, the mean and maximum value are
and 20.0841 degrees respectively. The standard deviation of the rotation error is 2.6185 degrees. 2.8333 and 20.0841 degrees
respectively.
The mean Thevalue
standard deviationdeviation
and standard of the rotation error is 2.6185
demonstrate that thedegrees.
integrated system exhibits good
accuracy which is comparable to accuracy of the existing system [16,24]. It is believed that the proposed
scheme is of relatively high-efficiency and easy to be implemented. It is quite suitable for the
measurement of large-scale components, such as car bodies, ship plates and astronautical/aeronautical
Sensors 2017, 17, 2886 12 of 13

The mean value and standard deviation demonstrate that the integrated system exhibits good
accuracy which is comparable to accuracy of the existing system [16,24]. It is believed that the
proposed scheme is of relatively high-efficiency and easy to be implemented. It is quite suitable for the
measurement of large-scale components, such as car bodies, ship plates and astronautical/aeronautical
large-scale thin wall components. Future work will focus on more intelligent path planning algorithm
and the improvement of measuring accuracy.

Acknowledgments: This work was supported by the National Basic Research Program of China (973 Program,
No. 2014CB046604); National Science and Technology Major Projects of—Numerical Control Machine Tool
and Based Manufacturing Equipment with High Range (04 Special Program, No. 2014ZX04015021); National
Science and Technology Major Projects-Key and Common Technology in Ship Intelligent Manufacturing
(No. 17GFB-ZB02-194); National Natural Science Foundation of China (No. 51575354); Interdisciplinary Program
of Shanghai Jiao Tong University (No. YG2014MS04, No. YG2015MS09). The authors would like to express their
sincere appreciation to them. Comments from the reviewers and the editors are very much appreciated.
Author Contributions: Hui Du designed the principle of this novel system, built the experiment system
environment, finished the verification experiment/data processing and wrote the paper. Xiaobo Chen contributed
to the access of point cloud, including the design of the fringe projection scanner and camera calibration.
Juntong Xi provided the necessary experiment environment (e.g., robot, laser tracker and so on) and the project
support. Chengyi Yu helps to design the principle. Bao Zhao helps to build the metric tool.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Chen, F.; Brown, G.M. Overview of 3-D shape measurement using optical methods. Opt. Eng. 2000, 39, 10–22.
2. Mendikute, A.; Yagüefabra, J.A.; Zatarain, M.; Bertelsen, Á.; Leizea, I. Self-calibrated in-process
photogrammetry for large raw part measurement and alignment before machining. Sensors 2017, 17, 2066.
[CrossRef] [PubMed]
3. Sun, B.; Zhu, J.; Yang, L.; Yang, S.; Guo, Y. Sensor for in-motion continuous 3D shape measurement based on
dual line-scan cameras. Sensors 2016, 16, 1949. [CrossRef] [PubMed]
4. Jin, Z.; Yu, C.; Li, J.; Ke, Y. Configuration analysis of the ers points in large-volume metrology system. Sensors
2015, 15, 24397–24408. [CrossRef] [PubMed]
5. Hexagon T-Scan. Available online: http://www.hexagonmi.com/products/3d-laser-scanners/leica-tscan-5
(accessed on 7 December 2017).
6. Kosarevsky, S. Practical way to measure large-scale 2D parts using repositioning on coordinate-measuring
machines. Measurement 2010, 43, 837–841. [CrossRef]
7. Feng, C.X.J.; Saal, A.L.; Salsbury, J.G.; Ness, A.R.; Lin, G.C.S. Design and analysis of experiments in cmm
measurement uncertainty study. Precis. Eng. 2007, 31, 94–101. [CrossRef]
8. Saito, K.; Miyoshi, T.; Yoshikawa, H. Noncontact 3-D digitizing and machining system for free-form surfaces.
CIRP Ann. Manuf. Technol. 1991, 40, 483–486. [CrossRef]
9. Gong, C.; Yuan, J.; Ni, J. Nongeometric error identification and compensation for robotic system by inverse
calibration. Int. J. Mach. Tools Manuf. 2000, 40, 2119–2137. [CrossRef]
10. Ye, S.; Wang, Y.; Ren, Y.; Li, D. Robot Calibration Using Iteration and Differential Kinematics. J. Phys. Conf. Ser.
2006, 48, 1–6. [CrossRef]
11. Li, J.; Guo, Y.; Zhu, J.; Lin, X.; Xin, Y.; Duan, K.; Tang, Q. Large depth-of-view portable three-dimensional
laser scanner and its segmental calibration for robot vision. Opt. Lasers Eng. 2007, 45, 1077–1087. [CrossRef]
12. Larsson, S.; Kjellander, J.A.P. An Industrial Robot and a Laser Scanner as a Flexible Solution towards an
Automatic System for Reverse Engineering of Unknown Objects. In Proceedings of the ASME Biennial
Conference on Engineering Systems Design and Analysis, Manchester, UK, 19–22 July 2004; pp. 341–350.
13. Yu, C.; Chen, X.; Xi, J. Modeling and calibration of a novel one-mirror galvanometric laser scanner. Sensors
2017, 17, 164. [CrossRef] [PubMed]
14. Li, J.; Chen, M.; Jin, X.; Chen, Y.; Dai, Z.; Ou, Z.; Tang, Q. Calibration of a multiple axes 3-D laser scanning
system consisting of robot, portable laser scanner and turntable. Opt.-Int. J. Light Electron Opt. 2011,
122, 324–329. [CrossRef]
Sensors 2017, 17, 2886 13 of 13

15. Stenz, U.; Hartmann, J.; Paffenholz, J.A.; Neumann, I. A framework based on reference data with
superordinate accuracy for the quality analysis of terrestrial laser scanning-based multi-sensor-systems.
Sensors 2017, 17, 1886. [CrossRef] [PubMed]
16. Yin, S.; Ren, Y.; Guo, Y.; Zhu, J.; Yang, S.; Ye, S. Development and calibration of an integrated 3D scanning
system for high-accuracy large-scale metrology. Measurement 2014, 54, 65–76. [CrossRef]
17. Kumar, U.P.; Somasundaram, U.; Kothiyal, M.P.; Mohan, N.K. Single frame digital fringe projection
profilometry for 3-d surface shape measurement. Opt.-Int. J. Light Electron Opt. 2013, 124, 166–169. [CrossRef]
18. Bräuerburchardt, C.; Breitbarth, A.; Kühmstedt, P.; Notni, G. High-speed three-dimensional measurements
with a fringe projection-based optical sensor. Opt. Eng. 2014, 53, 112213. [CrossRef]
19. Gao, B.Z.; Wang, M.; Peng, X.; Liu, X.; Yin, Y. Fringe projection 3D microscopy with the general imaging
model. Opt. Express 2015, 23, 6846–6857.
20. Zhang, C.; Zhao, H.; Gu, F.; Ma, Y. Phase unwrapping algorithm based on multi-frequency fringe projection
and fringe background for fringe projection profilometry. Meas. Sci. Technol. 2015, 26, 045203. [CrossRef]
21. Zhang, C.; Zhao, H.; Zhang, L.; Wang, X. Full-field phase error detection and compensation method for
digital phase-shifting fringe projection profilometry. Meas. Sci. Technol. 2015, 26, 035201. [CrossRef]
22. Chen, X.; Xi, J.T.; Jiang, T.; Jin, Y. Research and development of an accurate 3D shape measurement system
based on fringe projection: Model analysis and performance evaluation. Precis. Eng. 2008, 32, 215–221.
23. Chen, X.; Xi, J.; Jin, Y. Accuracy improvement for 3D shape measurement system based on gray-code and
phase-shift structured light projection. In Proceedings of the International Symposium on Multispectral
Image Processing and Pattern Recognition, Wuhan, China, 15 November 2007.
24. Paoli, A.; Razionale, A.V. Large yacht hull measurement by integrating optical scanning with mechanical
tracking-based methodologies. Robot. Comput.-Integr. Manuf. 2012, 28, 592–601. [CrossRef]
25. Chen, X.; Xi, J.; Jin, Y.; Sun, J. Accurate calibration for a camera–projector measurement system based on
structured light projection. Opt. Lasers Eng. 2009, 47, 310–319. [CrossRef]
26. Zhang, Z. A flexible new technique for camera calibration. IEEE Trans. Pattern Anal. Mach. Intell. 2000,
22, 1330–1334. [CrossRef]
27. Rusu, R.B. Semantic 3D object maps for everyday manipulation in human living environments. KI Künstliche Intell.
2010, 24, 345–348. [CrossRef]
28. Mitra, N.J.; Nguyen, A. Estimating surface normals in noisy point cloud data. Int. J. Comput. Geom. Appl.
2004, 14, 0400147. [CrossRef]

© 2017 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/).