Professional Documents
Culture Documents
Basics of Terrestrial Laser Scanning
Basics of Terrestrial Laser Scanning
Basics of Terrestrial Laser Scanning
This chapter introduces the basic concepts of the automatic 3D surveying by means
of laser scanning techniques, a methodology developed less than 20 years ago (see,
e.g., Boehler et al. 2002) thanks to the technological improvement of the topograph-
ical instrumentations, in particular in the movement and in the electronical compo-
nents. The main topic here explained concerns the geometrical measurement princi-
ple of a 3D scene by means of different automatic scans, so performing different 3D
point clouds, and the analytical steps to refer such scans to the one chosen coordi-
nate system. Laser scanning technology described in this chapter considers terrestrial
static systems, anyway in the last part, some hints are reported about scans acquired
by moving systems, as those mounted on aerial vehicles or on portable devices.
7.1 Background
First of all, just to introduce this technique of surveying, let us specify that the
similar definitions Terrestrial Laser Scanner and Terrestrial Laser Scanning, both
corresponding to the TLS acronym, have a different meaning: the scanner is the
instrument allowing the surveying by scanning, although roughly speaking the con-
cepts are analogous.
A first classification dealing with TLS technique is the belonging of these sys-
tems to the active sensors, namely instrumentations emitting radiations in a certain
direction, obviously toward the investigated object, and detecting the radiation that
is reflected or backscattered from it. Passive sensors, as the CCD/CMOS cameras
acquiring images, detect instead (generally) natural radiation reflected by the ob-
ject being observed: of course, the reflected sunlight is the most common radiation
measured by passive sensors.
First prototypes of TLS instrumentations have been applied for industrial and
civil engineering investigations, while nowadays they are largely used also for the
surveying of architectural, archeological and cultural heritage sites, namely working
Fig. 7.1 Left: a laser scanner surveying of an archeological area with the X S, Y S, Z S (SOCS)
reference frame centered on the scanner. Right: the obtained 3D point cloud
Consider such laser beam in a generic spatial direction: the coordinates of the hit
point are given by the following equation:
XS
⎛ ⎞ ⎛ ⎞
sin ϕ sin ϑ
r S = ⎝ Y S ⎠ = d ⎝sin ϕ cos ϑ⎠ (7.1.1)
ZS cos ϕ
where:
• d is the distance from the TLS instrument center to the hit point, computed as
explained before;
• ϕ is the “vertical” angle of the ray direction, known since imposed by the rotation
of the mirror around the“horizontal” rotating axis;
• ϑ is the “horizontal” angle of the ray direction, also this known since imposed by
the rotation of the mirror around the“vertical” Z S -axis.
From the mathematical point of view, Eq. (7.1.1) expresses the (bi-univocal) trans-
formation from spherical coordinates d, ϕ, ϑ (nearly the same measures acquired by
a topographic total station), to Cartesian coordinates X S, Y S, Z S .
Coming back to the TLS surveying, the scanning effect is achieved quickly re-
flecting (nowadays up to one million times per second!) the emitted laser beam in
different directions by changing the mirror deflection angle of small increments. For
each point, the value of d is measured, while those of ϕ and ϑ are imposed: by con-
tinuously applying Eq. (7.1.1), the corresponding X S, Y S, Z S coordinates are then
automatically computed in quasi-real-time.
High accuracy in recording the angular values is truly important, since it settles
on the precision of the obtained X S, Y S, Z S coordinate point: in general, the error in
the 3D point positioning is restricted to a few millimeters.
90 7 Basics of Terrestrial Laser Scanning
As already mentioned, acquiring TLS data from different points of scanning, i.e., dif-
ferent SOCSs, originates the problem, known as registration of the scans, to merge
such independent clouds of points. Sometimes, this process is called alignment.
Considering Fig. 7.2, the problem is that, applying Eq. (7.1.1) for the same point
P, the coordinates X S1, Y S1, Z S1 with respect to SOCS1 are clearly different from
7.2 Registration and Geo-referencing of Laser Scanning Point Clouds 91
those X S2, Y S2, Z S2 with respect to SOCS2 . Obviously, this fact is much more boring
dealing with any other further ith scan and relating SOCSi . In other words, the point
scans/clouds have to be suitably referred to a unique common reference frame, called
Project Coordinate System (PCS). In general, the origin of this frame is in the center
of the surveyed area and with the Z P -axis exactly vertical. Sometimes, a specific
vertical scan position, the more central one, can be chosen as PCS.
Referring again to Fig. 7.2 for sake of simplicity, considering SOCS1 as the final
absolute reference, i.e., the PCS, and SOCS2 as the initial local reference, i.e., any
SOCSi , the analytical problem is described by the following relationship:
⎞ ⎛ P⎞
XP
⎛ ⎛ Si ⎞
X Si X
r P = ⎝ Y P ⎠ = ⎝ Y SiP ⎠ + R (, , K ) ⎝ Y Si ⎠ (7.2.1)
P
ZP Z Si Z Si
where:
• X P, Y P, Z P are the coordinates of point P with respect to PCS;
P
• X Si , Y SiP , Z Si
P
are the unknown SOCSi origin coordinates with respect to PCS;
• R is the 3×3 matrix of the unknown rotation from SOCSi to PCS, defined by three
rotation angles , , K around the three principal directions;
• X Si, Y Si, Z Si are the coordinates of point P with respect to SOCSi .
Relationship (7.2.1) states that, after the registration transformation, the coordi-
nates with respect to SOCSi (finally) correspond to the coordinates with respect to
PCS, apart from the observation errors.
From the geometrical point of view, the registration problem is very simple, involv-
ing a similarity transformation (rotation and translation) but without scale variation
(namely a rigid roto-translation). As usual in geomatics, the six unknowns are com-
puted by solving a system with a redundant number of equations (7.2.1) on the same
unknowns and by performing a least squares estimation. From the computational
point of view, the registration requires the “double” 3D coordinates of at least two
points in the two reference frames PCS and SOCSi , for this reason called double
points or tie-points. Please note that this problem can be solved with the Extended
Orthogonal Procrustes Analysis (EOPA), extensively described in Sect. 2.4.
92 7 Basics of Terrestrial Laser Scanning
In principle, in the overlapped zone between two point clouds, millions of quasi-
common points are available, with the word “quasi” remembering us that the scanned
point is never exactly the same in the different scans. Hence, theoretically millions
of double points can be exploited to analytically solve the registration, but with
the titanic problem of the correct pairing of the “same” point among the scans. All
these troubles can be completely skipped by exploiting the reflecting targets as tie-
points, considering their barycentric SOCS coordinates, and automatically finding
the correct point pairing by choosing, among all the possible combinations, the one
producing the minimum error (sum of the squares of the residuals). Some details
about this point pairing can be found in Beinat et al. (2007). The accuracy of the
obtained registration can be easily evaluated by considering the residuals on the
tie-points.
Following the same approach for each couple of scans and/or considering the
common targets among more than two scans, three translation and three rotation
parameters per scan are therefore computed. By applying these transformations to
each point of each cloud, they “move” from their SOCS frame to a common PCS
frame: as a consequence, a unique cloud of millions of points is finally obtained.
A possible procedure to solve the registration problem is the well-known Itera-
tive Closest Point (ICP) algorithm (Chen and Medioni 1992; Besl and McKay 1992),
already described in Chap. 3. Just to recall the basic idea, the estimation of the six
transformation parameters is iteratively carried out by means of a least squares mini-
mization of a “metric error”, usually the distance, between the overlapping surfaces.
Hence, it is required to compute the surfaces generated from two or more point
clouds (in truth, randomly considering a subsample of such point clouds). In gen-
eral, wide rotations/translations are first computed by model (7.2.1) and then refined
rotations/translations are later estimated via the ICP algorithm. In the former case,
the residuals are minimized onto few tie-points, in the latter for all the overlapping
surfaces.
Once all the scans have been registered, another topographic problem arises: the
just obtained unique PCS point cloud should be referred to a Global Coordinate
System (GCS), e.g., a cartographic reference frame, thus requiring the so-called geo-
referencing of the laser scanning data.
The problem to transform X P, Y P, Z P coordinates into X G, Y G, Z G ones can be
analytically solved by means of a direct or indirect method of geo-referencing (see,
e.g., Schuhmacher and Böhm 2005). In the first case, in truth not much applied,
the TLS system is integrated with a Global Navigation Satellite System (GNSS)
receiver and/or topographic devices allowing to directly measure the transformation
parameters from SOCSi to GCS frame, hence bypassing the registration step.
Following instead the indirect method, some scanned point (e.g., the same reflect-
ing targets) constitutes a set of control points surveyed by means of a topographic
total station from the vertices of a topographic network defined into a geodetic datum.
Nowadays, the novelty in this approach is the fixing of the network geodetic datum
by GNSS measurements on some vertices.
7.2 Registration and Geo-referencing of Laser Scanning Point Clouds 93
From the analytical point of view, the indirect geo-referencing problem is de-
scribed by a relationship very similar to the previous (7.2.1):
⎞ ⎛ G⎞
XG
⎛ ⎛ P⎞
XP X
r G = ⎝ Y G ⎠ = ⎝ Y PG ⎠ + R () ⎝ Y P ⎠ (7.2.2)
ZG Z GP ZP
where:
• X G, Y G, Z G are the coordinates of point P with respect to GCS;
• X GP , Y PG , Z GP are the unknown PCS origin coordinates with respect to GCS;
• R () is the 3×3 matrix of the unknown rotation from PCS to GCS, defined by
only one rotation angle since Z P -axis and Z G -axis are both vertical.
Also the least squares solution of relationship (7.2.2) follows the same strategy
of the previous registration step (7.2.1), where now the double points are given from
the control points, with topographically measured X G, Y G, Z G coordinates and reg-
istered X P, Y P, Z P ones. From a geometrical point of view, the availability of points
of known position, realistically with a good accuracy and a uniform positioning,
makes it possible to locally constrain the various clouds on “true 3D positions”.
In this sense, by means of the registration step, tie-points are only joined together
while, through the geo-referencing step, such tie-points are constrained to coincide
with the topographically measured positions. As a result, the total point cloud is
locally deformed, avoiding the effects of errors propagation of the registration and,
consequently, improving the surveying accuracy.
Summarizing, scans registration and geo-referencing both involve the estimation
of translation and rotation parameters and, for such reason, can be simultaneously
solved. It is important to underline how these transformation problems, involving
nonlinear estimations and sets of coordinates of double points in two data, can be
very efficiently solved thanks to the algorithms from Procrustes Analysis reported
throughout this book.
As a last consideration, these TLS steps are analogous to the photogrammetric
problems (see Chap. 6) of the relative orientation between two images with the
construction of a stereoscopic model (registration) and of the absolute orientation
of the so-created stereoscopic model (geo-referencing).
Once the whole point cloud has been geo-referenced, the laser scanning data
processing can be considered accomplished: the subsequent steps of surface modeling
by mesh, surface optimization, and image texturing are similar to those applied to 3D
point clouds obtained from images via the Structure from Motion approach (Hartley
and Zisserman 2003).
94 7 Basics of Terrestrial Laser Scanning
Positioning (7.1.1), registration (7.2.1) and geo-referencing (7.2.2) models are based
on the assumption that the terrestrial laser scanner is alternatively static on various
scan positions and well-distributed tie-points are available.
More complex scanning procedures can be carried out by using scanners mounted
on moving vehicles such as aircrafts and helicopters, but nowadays also on Unmanned
Aerial Vehicles (UAVs), vans, as well as very recent “portable by man” devices. All
these kinds of systems belong to the Mobile Mapping Systems (MMS) category, a
multisensors technology introduced more than 20 years ago (ASPRS 1995; Schwarz
and El-Sheimy 1996) to survey roads, mainly exploiting, at that time, photogram-
metric imaging sensors. Nowadays, systems are instead basically equipped with a
laser scanner for measuring, while as navigation sensors a GNSS receiver and an In-
ertial Navigation System (INS) are employed. In few words, GNSS/INS navigation
sensors enable the estimation of the position and the attitude of the moving vehicle,
and then of the TLS, at every instant.
Considering Fig. 7.3, now the position r G of a generic point with respect to the
GCS (PCS is no more exploited) is given by the more complex equation below:
r G = rgnss/ins
G G
ins tls ins
+ Rins Rtls r + atls (7.3.1)
where:
G
• rgnss/ins is the position of the GNSS/INS sensors, sometimes embedded, with
respect to GCS, suitably computed;
G
• Rins is the 3 × 3 matrix of the rotation from the INS frame to the GCS frame, with
the instantaneous rotation angles measured by the three INS gyroscopes;
ins
• Rtls is the 3 × 3 matrix of the rotation from the TLS frame to the INS frame, with
fixed angles suitably estimated in the system calibration procedure;
• rtls is the position of the surveyed point with respect to the TLS frame (analogous
to r S of Eq. (7.1.1));
ins
• atls is the position of the TLS frame with respect to the INS frame, with fixed
components generally of few centimeters, nevertheless also this to be suitably
measured/estimated in the system calibration step.
The kinematic positioning of every scanned point by means of Eq. (7.3.1) can be
reinterpreted as the static positioning rtls by (7.1.1) from a moving TLS system whose
G G G ins
instantaneous position is rtls = rgnss/ins + Rins atls , while the instantaneous rotation
G G ins
is Rtls = Rins Rtls . In other words, the six roto-translation registration parameters (in
principle for each laser beam since TLS is moving!) are directly estimated by means
of GNSS/INS measurements and not indirectly by using tie-points as in (7.2.1).
The most common application of model (7.3.1) is in the Aerial Laser Scanning
(ALS) (see, e.g., Baltsavias 1999), widely exploited for the territorial surveying:
although rtls could be kilometric and the aircraft/helicopter speed tens of km/h, the
accuracy of the positioning r G is surprising, generally limiting the errors under 10 cm.
Concluding this chapter, some words are devoted to the last (for now!) frontier
of the TLS technology represented by the portable systems, developed very recently
and variously called as human, handheld, backpack, trolley systems. From the tech-
nological point of view, also these multisensor systems have one or more scanning
(measuring) sensors and one or more navigation sensors, although with generally
lower performances. To underline the terrestrial but moving features of this technol-
ogy, the term Indoor Mobile Mapping Systems encloses all such instrumentations
(Thomson et al. 2013; Nocerino et al. 2017; Lehtola et al. 2017; Tucci et al. 2018),
where the word indoor remember us that GNSS data could be not available for a part
or all the surveying area.
By applying model (7.3.1), with some systems already in real time, the posi-
tioning r G is clearly affected by the propagation of scanning and navigation errors.
Anyway, this “first-obtained” point cloud is exploited to re-compute the TLS posi-
tioning and orientation. The analytical process making possible such improvement
is very complex and comes from algorithms developed in robotics (Nüchter et al.
2007). Summarizing, the fundamental step is a suitable subdivision of the whole
point cloud into a lot of n “virtual single static scans”, each one considering a certain
TLS localization. The obtained clouds have geometrical discrepancies, due to errors:
nevertheless, these are minimized by analytical procedures generally based again on
ICP algorithm, i.e., by suitably reducing the distance among the surfaces of these n
clouds.
96 7 Basics of Terrestrial Laser Scanning
7.4 Summary
From the geometrical point of view, the TLS measurement principle and the con-
sequent generation of 3D point cloud are very easy to understand, since they sim-
ply involve a transformation of topographically observed spherical coordinates into
Cartesian coordinates referred to the TLS center.
In spite of this simplification, since the TLS system has to be moved into a second,
third and so on scan position, any new point cloud is referred to a moving and local
reference frame: hence, the registration of all the scans to a single chosen project
coordinate system is a fundamental step. More complete results are obtained if the
chosen coordinate system is a global/geographic reference frame and, for this aim,
coordinates acquired by EDM topographic total stations and GNSS receivers are
mandatory to solve the geo-referencing step. Involving rigid rotations and translations
on point sets, registration, and geo-referencing transformations can be conveniently
solved by means of Procrustes Analysis, as it will be shown in Part III of this book.
To conclude this chapter, some aspects of the new trends in laser scanning technol-
ogy were briefly described, just to underline the potentialities of the nowadays new
moving systems, exploiting also navigations sensors to solve the geo-referencing
problem.
References
ASPRS. 1995 mobile mapping symposium, in Proceedings of the First International Workshop on
Mobile Mapping Technology, Columbus, OH, USA (May 1995)
C. Balletti, F. Guerra, The survey of cultural heritage: a long story. Rend. Lincei 26(1), 115–125
(2015)
E.P. Baltsavias, Airborne laser scanning: basic relations and formulas. ISPRS J. Photogramm.
Remote Sens. 54(2), 199–214 (1999)
A. Beinat, F. Crosilla, D. Visintini, F. Sepic, Automatic non parametric procedures for terrestrial
laser point clouds processing. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 36(3), W49B
(2007)
P. Besl, N. McKay, A method for registration of 3-D shapes. IEEE Trans. Pattern Anal. Mach. Intell.
14(2), 239–256 (1992)
W. Boehler, G. Heinz, A. Marbs, The potential of non-contact close range laser scanners for cultural
heritage recording. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 34(5/C7), 430–436
(2002)
Y. Chen, G. Medioni, Object modeling by registration of multiple range images. Image Vis. Comput.
10(3), 145–155 (1992)
S. Gonizzi Barsanti, F. Remondino, D. Visintini, 3d surveying and modelling of archaeological
sites—some critical issues. ISPRS Photogramm. Remote Sens. Spat. Inf. Sci. 2(5/W1):145–150
(2013)
R.I. Hartley, A. Zisserman, Multiple View Geometry in Computer Vision, 2nd edn. (Cambridge
University Press, Cambridge, 2003)
V.V. Lehtola, H. Kaartinen, A. Nüchter, R. Kaijaluoto, A. Kukko, P. Litkey, E. Honkavaara,
T. Rosnell, M.T. Vaaja, J.-P. Virtanen, M. Kurkelaand, A. El-Issaoui, L. Zhu, A. Jaakkola, J.
References 97
Hyyppa, Comparison of the selected state-of-the-art 3d indoor scanning and point cloud genera-
tion methods. Remote Sens. 9(8), 796 (2017)
E. Nocerino, F. Menna, F. Remondino, I. Toschi, P. Rodríguez-Gonzálvez, Investigation of indoor
and outdoor performance of two portable mobile mapping systems, in Videometrics, Range Imag-
ing, and Applications XIV, vol. 1033 (International Society for Optics and Photonics, 2017), p.
103320I
A. Nüchter, K. Lingemann, J. Hertzberg, H. Surmann, 6d slam—3d mapping outdoor environments.
J. Field Robot. 24(8–9), 699–722 (2007)
N. Pfeifer, C. Briese, Laser scanning–principles and applications, in GeoSiberia 2007—
International Exhibition and Scientific Congress (Citeseer, 2007)
S. Schuhmacher, J. Böhm, Georeferencing of terrestrial laserscanner data for applications in ar-
chitectural modeling, in International Archives of Photogrammetry, Remote Sensing and Spatial
Information Sciences, Proceedings of the ISPRS Working Group V/4 Workshop 3D-ARCH 2005,
vol. 36 (2005)
K.P. Schwarz, N. El-Sheimy, Kinematic multi-sensor systems for close range digital imaging. Int.
Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 31, 774–785 (1996)
C. Thomson, G. Apostolopoulos, D. Backes, J. Boehm, Mobile laser scanning for indoor modelling.
ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci. 2(5/W2), 289–293 (2013)
G. Tucci, V. Bonora, Geomatics and management of at-risk cultural heritage. Rend. Lincei 26(1),
105–114 (2015)
G. Tucci, D. Visintini, V. Bonora, E.I. Parisi, Examination of indoor mobile mapping systems in a
diversified internal/external test field. Appl. Sci. 8(3), 401–430 (2018)
G. Vosselman, H.-G. Maas, Airborne and Terrestrial Laser Scanning (Whittles Publishing, 2010)