ArticlePDF Available

Mobile LiDAR System: New Possibilities for the Documentation and Dissemination of Large Cultural Heritage Sites

Authors:

Abstract and Figures

Mobile LiDAR System is an emerging technology that combines multiple sensors. Active sensors, together with Inertial and Global Navigation System, are synchronized on a mobile platform to produce an accurate and precise geospatial 3D point cloud. They allow obtaining a large amount of georeferenced 3D information in a fast and efficient way, which can be used in several applications such as the 3D recording and reconstruction of complex urban areas and/or landscapes. In this study the Mobile LiDAR System is applied in the field of Cultural Heritage aiming to evaluate its performance with the purpose to document, divulgate, or to develop an architectural analysis. This study was focused on the Medieval Wall of Avila (Spain) and, specifically, the performed accuracy tests were applied in the "Alcazar" gate (National Monument from 1884). The Mobile LiDAR System is then compared to the most commonly employed active sensors (Terrestrial Laser Scanner) for large Cultural Heritage sites in regard to time, accuracy and resolution of the point cloud. The discrepancies between both technologies are established comparing directly the 3D point clouds generated, highlighting the errors affecting the architectural structures. Consequently, and based on a detailed geometrical analysis, an optimization methodology is proposed, establishing a segmented and classified cluster for the structures. Furthermore, three main clusters are settled, according to the curvature: (i) planar or low curvature; (ii) cylindrical, mild transitions and medium curvature; and (iii) the abrupt transitions of high curvature. The obtained 3D point clouds in each cluster are analyzed and optimized, considering the reference spatial sampling, according to a confidence interval and the feature curvature. The presented results suggest that Mobile LiDAR System is an optimal approach, allowing a high-speed data acquisition and providing an adequate accuracy for large Cultural Heritage sites.
Content may be subject to copyright.
remote sensing
Article
Mobile LiDAR System: New Possibilities for the
Documentation and Dissemination of Large
Cultural Heritage Sites
Pablo Rodríguez-Gonzálvez 1, *, Belén Jiménez Fernández-Palacios 1,Ángel Luis Muñoz-Nieto 1,
Pedro Arias-Sanchez 2and Diego Gonzalez-Aguilera 1
1TIDOP Research Group, University of Salamanca, Polytechnic School of Avila. Hornos Caleros, 50,
05003 Avila, Spain; belenjfp@gmail.com (B.J.F.-P.); almuni@usal.es (Á.L.M.-N.); daguilera@usal.es (D.G.-A.)
2Department of Natural Resources & Environmental Engineering, University of Vigo,
School of Mining Engineering, Maxwell s/n, 36310 Vigo, Spain; parias@uvigo.es
*Correspondence: pablorgsf@usal.es; Tel.: +34-920-353-500
Academic Editors: Rosa Lasaponara, Nicola Masini and Prasad S. Thenkabail
Received: 10 November 2016; Accepted: 20 February 2017; Published: 23 February 2017
Abstract:
Mobile LiDAR System is an emerging technology that combines multiple sensors.
Active sensors, together with Inertial and Global Navigation System, are synchronized on a mobile
platform to produce an accurate and precise geospatial 3D point cloud. They allow obtaining a large
amount of georeferenced 3D information in a fast and efficient way, which can be used in several
applications such as the 3D recording and reconstruction of complex urban areas and/or landscapes.
In this study the Mobile LiDAR System is applied in the field of Cultural Heritage aiming to evaluate
its performance with the purpose to document, divulgate, or to develop an architectural analysis.
This study was focused on the Medieval Wall of Avila (Spain) and, specifically, the performed
accuracy tests were applied in the “Alcazar” gate (National Monument from 1884). The Mobile
LiDAR System is then compared to the most commonly employed active sensors (Terrestrial Laser
Scanner) for large Cultural Heritage sites in regard to time, accuracy and resolution of the point cloud.
The discrepancies between both technologies are established comparing directly the 3D point clouds
generated, highlighting the errors affecting the architectural structures. Consequently, and based on
a detailed geometrical analysis, an optimization methodology is proposed, establishing a segmented
and classified cluster for the structures. Furthermore, three main clusters are settled, according to
the curvature: (i) planar or low curvature; (ii) cylindrical, mild transitions and medium curvature;
and (iii) the abrupt transitions of high curvature. The obtained 3D point clouds in each cluster
are analyzed and optimized, considering the reference spatial sampling, according to a confidence
interval and the feature curvature. The presented results suggest that Mobile LiDAR System is an
optimal approach, allowing a high-speed data acquisition and providing an adequate accuracy for
large Cultural Heritage sites.
Keywords:
cultural heritage; mobile LiDAR system; point cloud; terrestrial laser scanner; accuracy
assessment; optimization
1. Introduction
Recent advances in Geomatics Science enable the use of a wide range of sensors to record,
catalogue and analyze Cultural Heritage (CH) sites: from RGB and multispectral cameras [
1
,
2
] and
Terrestrial Laser Scanner (TLS) [
3
], to Ground Penetrating Radar (GPR) [
4
] and Raman Spectroscopy [
5
].
Furthermore, airborne LiDAR (Laser Imaging Detection and Ranging) is used as a remote sensing
device for 3D surveying and modeling of extensive archaeological sites [
6
,
7
]. Nowadays, hybrid
Remote Sens. 2017,9, 189; doi:10.3390/rs9030189 www.mdpi.com/journal/remotesensing
Remote Sens. 2017,9, 189 2 of 17
sensors such as Mobile LiDAR Systems (MLS) [
8
,
9
] bring added value to record large and complex
sites. Although the most common solution based on MLS are boarded on vans and applied in the field
of civil engineering and construction [
10
], MLS has been useful in other fields, such as geology [
11
];
agriculture, biology or forestry [
12
14
]; and even in the inspection of high-voltage power lines [
15
].
The last developments allow the use of MLS boarded in a backpack [
16
] or even in autonomous
robots [
17
], offering a flexible solution in terms of accuracy, flexibility, point density and access to
indoor and non-transitable areas [
18
22
]. These specific solutions are still under development and
several approaches are being experienced and tested [23,24].
MLS process and management optimization has become a key point. Even though MLS reduces
extremely data collection phase, it still requires a lot of time and resources consumption to extract
meaningful information [
25
] and even more time for 3D modeling. At this point, and for large
and unstructured point cloud, some authors [
26
] showed that it is more efficient to carry out
the simplification of the point cloud before creating a mesh. In terms of workflow optimization,
it may be useful when the presence of particular characteristics (e.g., breaklines) is desirable [
27
].
MLS optimization, understood not only as decimation but a smart filtering, has proven to be relevant in
different fields, such as infrastructures management [
28
], reverse engineering [
29
], object detection [
30
],
object fitting [
31
] and visualization optimization through Web services by means of mobile devices [
32
].
It is a well-known fact that CH information systems such as Nubes Project [
33
,
34
] and Cultural
Heritage Information System Projects [
35
37
] become the basis for an effective management of CH.
However, they are also extremely important for its dissemination and decision making. These Systems
and Projects are usually based on 3D Web interfaces [
38
]. Thus, it is clear that, in order to guarantee
and improve the specific visualization requirements of 3D Web systems, the MLS complex geometric
datasets need to be optimized.
Based on the remarks above, this paper aims to analyze the suitability of MLS for CH
documentation and dissemination purposes from a two-fold approach. Firstly, an accuracy assessment
is performed. For this purpose, the acquired MLS data were compared with a ground truth defined
by a milimetric laser scanner point cloud. Secondly, the data suitability, in terms of 3D point cloud
optimization for the data dissemination, is evaluated.
This manuscript is organized as follows. After the Introduction, sensors and devices are described
in Section 2. In Section 3, the multi-sensor integration methodology and multi-data process workflow
proposed are detailed. In Section 4, the representative study case of the Medieval Wall of Avila (Spain)
is shown. Finally, in Sections 5and 6, discussion and conclusions are respectively discoursed.
2. Materials
The evaluated mobile LiDAR technology is a LYNX Mobile Mapper manufactured by Optech
(Figure 1) [
9
,
39
]. This system is composed of two LiDAR sensors, four RGB cameras and an Applanix
POS LV 520 IMU. The system is configured to take 500,000 points per second with a scan frequency
of 200 Hz. The maximum range of the sensors is 200 m, with a precision of 8 mm (one sigma)
and permission to obtain up to 4 echoes of the signal and the intensity reflected by the objects at a
1550 nm wavelength (Table 1). Along with the geometric measurement system, the MLS incorporates a
multisensory camera to acquire the radiometric texture of the scanned elements.
In this study, the MLS is mounted on a car, achieving speeds up to 40 km/h during the data capture.
Remote Sens. 2017,9, 189 3 of 17
Remote Sens. 2017, 9, 189 3 of 17
Figure 1. LYNX Mobile Mapper manufactured by Optech.
Table 1. Technical specifications of Optech LYNX Mobile Mapper.
MLS Technical Parameters
X, Y position 0.020 m
Z position 0.050 m
Roll and pitch 0.005°
True heading 0.015°
Measuring principle Time of Flight (ToF)
Maximum range 200 m
Range precision 8 mm (1σ)
Range accuracy ±10 mm (1σ)
Laser measurement rate 75–500 kHz
Measurement per laser pulse Up to 4 simultaneous
Scan frequency 80–200 Hz
Laser wavelength 1550 nm (near infrared)
Scanner field of view 360°
Operating temperature 10–40 °C
Angular resolution 0.001°
The use of MLS is not completely straightforward. It requires an initial data acquisition plan,
especially for the large cultural heritage sites. In this regard, any error in the INS will be directly
propagated to the point cloud, being especially complicated in scenarios with the presence of narrow
streets (e.g., urban CH sites), or dense vegetation, which could provide a loss in accuracy by the
multi-path and occlusion of the GNSS signal [40]. Besides, the boresight angles and level-arm
(related to the MLS assembly), which are kept constant throughout the whole scanning process,
require to be precisely determined to avoid any loss in the overall accuracy [41]. To this end, some
regular surfaces (e.g., road surface) should be measured prior to the data acquisition for its
calibration. The most critical aspect in data acquisition planning is the boresight, which could cause
that same scene acquisition in different trajectories to not overlap correctly. Additionally, the height
of the CH buildings could be a limitation to the system due to the vertical angle, so those trajectories
closer to the object should be avoided to reduce the occlusions in the final point cloud.
To assess the quality of the MLS, a Time-of-Flight (ToF) [42,43] TLS Trimble GX is employed,
found on direct time measurement. It is typically more precise than the phase-shift (PS) ones [43,44],
for instance, measurement amplitude-modulated continuous wave method (AMCW) based on
indirect time, with lower range but further point acquisition speed. The main technical specifications
are shown in Table 2. In a ToF System, the distance to the surface and the position of the points (X, Y,
Z) are calculated from the time of the flight of the pulse; a laser pulse is emitted towards the object
and, once it is reflected from the object surface, returns to the equipment. The distance to the object
can be determined by half of the round-trip range. Nowadays, both technologies are competing; PS
Figure 1. LYNX Mobile Mapper manufactured by Optech.
Table 1. Technical specifications of Optech LYNX Mobile Mapper.
MLS Technical Parameters
X,Yposition 0.020 m
Zposition 0.050 m
Roll and pitch 0.005
True heading 0.015
Measuring principle Time of Flight (ToF)
Maximum range 200 m
Range precision 8 mm (1σ)
Range accuracy ±10 mm (1σ)
Laser measurement rate 75–500 kHz
Measurement per laser pulse Up to 4 simultaneous
Scan frequency 80–200 Hz
Laser wavelength 1550 nm (near infrared)
Scanner field of view 360
Operating temperature 10–40 C
Angular resolution 0.001
The use of MLS is not completely straightforward. It requires an initial data acquisition plan,
especially for the large cultural heritage sites. In this regard, any error in the INS will be directly
propagated to the point cloud, being especially complicated in scenarios with the presence of narrow
streets (e.g., urban CH sites), or dense vegetation, which could provide a loss in accuracy by the
multi-path and occlusion of the GNSS signal [
40
]. Besides, the boresight angles and level-arm (related
to the MLS assembly), which are kept constant throughout the whole scanning process, require to be
precisely determined to avoid any loss in the overall accuracy [
41
]. To this end, some regular surfaces
(e.g., road surface) should be measured prior to the data acquisition for its calibration. The most critical
aspect in data acquisition planning is the boresight, which could cause that same scene acquisition in
different trajectories to not overlap correctly. Additionally, the height of the CH buildings could be a
limitation to the system due to the vertical angle, so those trajectories closer to the object should be
avoided to reduce the occlusions in the final point cloud.
To assess the quality of the MLS, a Time-of-Flight (ToF) [
42
,
43
] TLS Trimble GX is employed,
found on direct time measurement. It is typically more precise than the phase-shift (PS) ones [
43
,
44
],
for instance, measurement amplitude-modulated continuous wave method (AMCW) based on indirect
time, with lower range but further point acquisition speed. The main technical specifications are
shown in Table 2. In a ToF System, the distance to the surface and the position of the points (X, Y, Z)
are calculated from the time of the flight of the pulse; a laser pulse is emitted towards the object
and, once it is reflected from the object surface, returns to the equipment. The distance to the object
can be determined by half of the round-trip range. Nowadays, both technologies are competing;
Remote Sens. 2017,9, 189 4 of 17
PS technology increasing its range, while ToF technology improving the acquisition speed. Regarding
the TLS accuracy evaluation, please refer to [45,46].
Table 2. Technical specifications of the Trimble GX laser scanner.
TLS Technical Parameters
Measuring principle Time of Flight (ToF)
Laser wavelength 534 nm (visible-green)
Scanner field of view 360H×60V
Range precision 1.4 mm at 50 m
Measurement range 2–350 m
Spot size (beam diameter) 3 mm a 50 m
Scanning speed 5000 points/sec
In order to georeference the TLS point cloud in a global coordinate reference system (ETRS89) 50
GPS control points were acquired. The GPS points, well distributed through the entire area of interest,
were surveyed using two Leica 1200 GPS. This device is a dual frequency receiver which was used
with Real Time Kinematic (RTK) method. The a-priori precision of this measurement method is 1 cm
in horizontal plane and 2 cm in vertical axis.
3. Methods
In this section the two geomatics techniques employed, as well as the specific assessment
methodology and the developed algorithms for the optimization of point clouds, are described.
In Figure 2the global pipeline for the geometric accuracy and optimized assessment data aimed at 3D
web visualization and dissemination is shown.
Remote Sens. 2017, 9, 189 4 of 17
technology increasing its range, while ToF technology improving the acquisition speed. Regarding
the TLS accuracy evaluation, please refer to [45,46].
Table 2. Technical specifications of the Trimble GX laser scanner.
TLS Technical Parameters
Measuring principle Time of Flight (ToF)
Laser wavelength 534 nm (visible-green)
Scanner field of view 360° H × 60° V
Range precision 1.4 mm at 50 m
Measurement range 2–350 m
Spot size (beam diameter) 3 mm a 50 m
Scanning speed 5000 points/sec
In order to georeference the TLS point cloud in a global coordinate reference system (ETRS89)
50 GPS control points were acquired. The GPS points, well distributed through the entire area of
interest, were surveyed using two Leica 1200 GPS. This device is a dual frequency receiver which
was used with Real Time Kinematic (RTK) method. The a-priori precision of this measurement
method is 1 cm in horizontal plane and 2 cm in vertical axis.
3. Methods
In this section the two geomatics techniques employed, as well as the specific assessment
methodology and the developed algorithms for the optimization of point clouds, are described. In
Figure 2 the global pipeline for the geometric accuracy and optimized assessment data aimed at 3D
web visualization and dissemination is shown.
Figure 2. Pipeline for the assessment of Mobile LiDAR Systems (MLS) for Cultural Heritage
recording and modeling.
3D data acquired by active sensors (TLS and MLS) are processed following a conventional
workflow. In the case of TLS, the obtained unstructured point clouds (individual scans) are cleaned,
edited and aligned in a common local reference system. The alignment process can be automatically
carried out through Iterative Closest Point (ICP) points [47], by automatic recognition of objects [48]
or using artificial targets [49]. The 3D point cloud obtained with milimetric resolution can include
the intensity levels (I) and/or the color information (RGB). Finally, the aligned point cloud is
georeferenced in a global reference system by means of targets of known coordinates. Regarding the
MLS, three main types of sensors take part in data acquisition using three independent reference
systems. A multi-sensor calibration is mandatory to determine the translational and rotational
offsets among the different sensors and bring all the data into the same reference system [50]. As a
result, the unstructured point cloud is already aligned in a global reference system by means of the
navigation/positioning sensors with centimetric absolute precision (typically). The 3D point cloud
Figure 2.
Pipeline for the assessment of Mobile LiDAR Systems (MLS) for Cultural Heritage recording
and modeling.
3D data acquired by active sensors (TLS and MLS) are processed following a conventional
workflow. In the case of TLS, the obtained unstructured point clouds (individual scans) are cleaned,
edited and aligned in a common local reference system. The alignment process can be automatically
carried out through Iterative Closest Point (ICP) points [
47
], by automatic recognition of objects [
48
] or
using artificial targets [
49
]. The 3D point cloud obtained with milimetric resolution can include the
intensity levels (I) and/or the color information (RGB). Finally, the aligned point cloud is georeferenced
in a global reference system by means of targets of known coordinates. Regarding the MLS, three main
types of sensors take part in data acquisition using three independent reference systems. A multi-sensor
calibration is mandatory to determine the translational and rotational offsets among the different
sensors and bring all the data into the same reference system [
50
]. As a result, the unstructured
point cloud is already aligned in a global reference system by means of the navigation/positioning
sensors with centimetric absolute precision (typically). The 3D point cloud obtained uses to have also
Remote Sens. 2017,9, 189 5 of 17
a centimetric resolution, including at least the intensity levels of returned pulses or even the color
information (RGB).
3.1. Accuracy Assessment Protocol
Since the MLS involves several sensors (e.g., IMU, LiDAR, GNSS, etc.) to acquire a georeferenced
3D point cloud, the aim of this step is to analyze, in the CH field, the quality of the MLS point cloud
using a ground truth provided by a TLS. To dismiss the uncertainties associated to the global coordinate
geo-referencing, a registration phase with the Iterative Closest Point (ICP) technique is carried out,
prior to the MLS point cloud evaluation [
51
]. As a result, only those errors affecting the 3D point cloud
of the architectural elements are analyzed. Particularly, discrepancies are computed in consonance
with Multiscale Model to Model Cloud Comparison (M3C2) [
52
], which performs a direct comparison
of the 3D point clouds (TLS to MLS), avoiding the preliminary phase of meshing.
In order to prevent possible bias effects affecting the data processing, the normality assumption,
i.e., the hypothesis that errors follow a Gaussian distribution, are checked according to graphical
methods such as QQ-plot [
53
] well-suited for very large samples [
54
]. If the samples are not normally
distributed, either due to the presence of outliers or because of a different population hypothesis,
a robust model based on non-parametric estimation should be employed. In this case, the median
(m) and the median absolute deviation (MAD) are used as robust measures instead of the mean and
standard deviation, respectively. The MAD is defined as the median (m) of the absolute deviations
from the data’s median (mx):
MAD =m(|ximx|)(1)
where the x
i
values come from the discrepancies (Euclidean distance) between both point clouds
(TLS and MLS).
The computation of the discrepancies maps is carried out by Cloud Compare v2.7.0 software [
55
],
being a point cloud-to-point cloud based on the M3C2 plugin [
52
]. The robust statistical estimators
are computed by a custom script as well as the in-house statistical software (STAR (Statistics Tests for
Analyzing of Residuals)) [56].
3.2. Point Cloud Optimization
Despite recent technological breakthroughs, it is still common to face some limitations since
3D datasets obtained by active sensors (TLS and MLS) produce complex reality-based 3D point
clouds which are generally large. For extensive CH sites, the geometric component hinders the
access for analysis purposes through Web visualization, worsen when mobile devices (e.g., tablets or
smartphones) are used, due to the vast amount of information. Depending on the device used and the
size of the 3D models, a delay might occur that could affect the fluidity of the 3D scene. Moreover,
the 3D point cloud acts as mainframe for the extraction of semantic and/or vector information,
avoiding the meshing step. The common algorithms used to generate a mesh are time-consuming,
since they require human intervention to correct topological errors (e.g., non-manifold edges, loose
edges, overlapping faces, self-intersections, etc.) that could be generated, and the holes filling required
to rebuild those areas affected by errors and occlusions. Thus, different simplification and optimization
strategies would be desirable to be applied to the 3D point clouds, while keeping the relevant
information for the subsequent studies and/or operations in the global pipeline of CH conservation
and management.
The proposed optimization methodology (Figure 3) is based on an idealization process and an
adaptive sampling according to a confidence interval and local feature curvature. Thus, the more
significant are the points, the more 3D points are kept to represent those complex areas of the site.
On the contrary, for those parametric and simple geometrical CH elements such as planes, the point
cloud is considerably reduced in size.
Remote Sens. 2017,9, 189 6 of 17
The point cloud optimization strategy has been developed following three global phases:
(i) a preparation phase to setup the raw point cloud acquired from MLS; (ii) a clustering phase
based on a curvature segmentation; and (iii) a final weighted sampling phase.
Remote Sens. 2017, 9, 189 6 of 17
The point cloud optimization strategy has been developed following three global phases: (i) a
preparation phase to setup the raw point cloud acquired from MLS; (ii) a clustering phase based on a
curvature segmentation; and (iii) a final weighted sampling phase.
Figure 3. Pipeline for 3D point cloud optimization.
The preparation phase involves three sub steps in the following order:
An initial outliers filtering of the MLS point cloud based on the absolute deviation (Euclidean
distance) of the points from a fitted local plane in a spherical vicinity. The absolute threshold is
not set very strictly to avoid false positives. In the case a plane cannot be fitted in the
neighborhood, the point is also excluded.
Next, in order to ease the simplification, a local smoothing process by means of a moving least
squares [57] is applied. This operation involves a point displacement regarding the original
positions; however, the spherical search volume is chosen in relation to the precision obtained
from the accuracy assessment studies (see Section 3.1) to avoid a precision deterioration of the
final optimized point cloud.
Finally, a reduction of high point density areas, due to the MLS acquisition methodology, is
applied. For this task, a spatial sampling is carried out. The sampling value determines the final
results. Since there are several options for this sampling value, all of them being related to the
input point cloud precision, the proposed one is to setup the 95% confidence interval of the
error dispersion from the accuracy assessment studies. By this procedure, it is possible to
guarantee the optimized final results.
Once the point cloud is ready, the curvature clustering phase is carried out following the next
three sub steps:
Initially, the local curvature is computed in a wide spherical neighborhood to reduce noise
effects. The spherical radius for the curvature calculation is defined in relation to the main
geometrical elements of the CH site (i.e., a-priori length and height).
Next, the 3D point cloud is segmented according to the computed local curvature and the
a-priori approximate knowledge of the main geometrical primitives presented (e.g., planes and
radius of cylinders). These coarse values allow defining a discrete number of clusters according
to a similar curvature values. The final number of clusters is increased in one, since the highest
curvature values are related to non-parametric areas, as the break-lines, borders, corners,
abrupt areas, surface discontinuities or geometric and topological errors.
Finally, since the curvature computation and clustering could be affected by local errors,
especially in transition areas, a refinement based on connected component analysis is carried
out to reclassify them in the more suitable cluster. This analysis is based on an octree
representation of the point cloud, so a reference subdivision level has to be set. In order to find
connected components, the octree level has to be set slightly higher than the homogenized
spatial resolution. Since the spatial resolution was homogenized in the preparation phase, and an
octree level has to be fixed, it is possible to relate easily the component’s area by the number of
Figure 3. Pipeline for 3D point cloud optimization.
The preparation phase involves three sub steps in the following order:
An initial outliers filtering of the MLS point cloud based on the absolute deviation (Euclidean
distance) of the points from a fitted local plane in a spherical vicinity. The absolute threshold is not
set very strictly to avoid false positives. In the case a plane cannot be fitted in the neighborhood,
the point is also excluded.
Next, in order to ease the simplification, a local smoothing process by means of a moving least
squares [
57
] is applied. This operation involves a point displacement regarding the original
positions; however, the spherical search volume is chosen in relation to the precision obtained
from the accuracy assessment studies (see Section 3.1) to avoid a precision deterioration of the
final optimized point cloud.
Finally, a reduction of high point density areas, due to the MLS acquisition methodology, is applied.
For this task, a spatial sampling is carried out. The sampling value determines the final results.
Since there are several options for this sampling value, all of them being related to the input point
cloud precision, the proposed one is to setup the 95% confidence interval of the error dispersion
from the accuracy assessment studies. By this procedure, it is possible to guarantee the optimized
final results.
Once the point cloud is ready, the curvature clustering phase is carried out following the next
three sub steps:
Initially, the local curvature is computed in a wide spherical neighborhood to reduce noise effects.
The spherical radius for the curvature calculation is defined in relation to the main geometrical
elements of the CH site (i.e., a-priori length and height).
Next, the 3D point cloud is segmented according to the computed local curvature and the a-priori
approximate knowledge of the main geometrical primitives presented (e.g., planes and radius of
cylinders). These coarse values allow defining a discrete number of clusters according to a similar
curvature values. The final number of clusters is increased in one, since the highest curvature
values are related to non-parametric areas, as the break-lines, borders, corners, abrupt areas,
surface discontinuities or geometric and topological errors.
Finally, since the curvature computation and clustering could be affected by local errors, especially
in transition areas, a refinement based on connected component analysis is carried out to reclassify
them in the more suitable cluster. This analysis is based on an octree representation of the point
cloud, so a reference subdivision level has to be set. In order to find connected components,
the octree level has to be set slightly higher than the homogenized spatial resolution. Since the
Remote Sens. 2017,9, 189 7 of 17
spatial resolution was homogenized in the preparation phase, and an octree level has to be fixed,
it is possible to relate easily the component’s area by the number of points. For the main features
of the CH site, it is possible to define their minimum area. As result, the components inside a
cluster that do not verify this minimum area are reallocated to the neighbor cluster in crescent
curvature. No points are removed.
Lastly, a weighted sampling is applied to all the points inside a cluster based on the associated
feature (e.g., plane, cylinder, etc.) and based on the maximum and minimum curvature values
of the cluster. The highest curvature cluster is kept unmodified, since encloses all the conflictive
areas. The others cluster thresholds, low and medium curvature, are established according to its
geometric elements. For instance, the number of points of a wall could be drastically reduced if
points can be assimilated/idealized as a plane. In the cases of other elements as towers, the arc to
chord approximation was used. Thus, a final 3D optimized point cloud is obtained, retaining all the
geometrical relevant information for subsequent tasks, as vectorization, reverse engineering, finite
element (FEM) analysis and information management through HBIM or even for Web visualization.
4. Experimental Results
The case study was the Medieval Wall of Avila. Its construction is the most important example of
military architecture of the Spanish Romanesque style as well as an exceptional model of European
medieval architecture [
58
]. The construction of the city wall is perfectly adapted to the topography.
The wall was used not only to defend the town from possible invasions and protect the people from
possible pests or epidemics, but also to control the trade of the city with the outside. The southern
sectors are shorter as they are built upon a cliff that acts as a natural defense. The western and northern
sections grow in height, reaching the tallest and thickest points located in the east section. There are
nine gates giving access to the town, of which the most spectacular is “Puerta del Alcázar” (Gate of
the Fortress). In 1884, it was declared a National Monument and in 1985, the old city of Ávila and its
extramural churches were declared a World Heritage site by UNESCO. Some references state that the
building dates back to 1090. Other researchers have argued instead that the wall’s construction most
likely continued through into the 12th century. Its large size is a clear example of a challenging CH sites
for the recording and documentation processes. Despite its linear nature (Figure 4), the towers and
wall distribution, the gates, and neighbor city buildings add difficulties for acquisition and processing.
Since the case study presents a repetitive pattern (i.e., wall–tower–wall) along its whole extension,
only a significant part of its extension was evaluated. Thus, the obtained results could be extrapolated
to the whole medieval Wall. The main geometric information is shown in Table 3.
Table 3. Avila’s Medieval Wall main dimensions.
Parameter Value
Perimeter 2516 m
No. of towers 87
No. of Battlement elements (current/original) 2113/2379
No. of gates 9
Width of the wall Between 2.6 and 2.8 m
Average height of the wall 11.5 m
Average height of the towers 15 m
The 3D point cloud of the whole medieval wall coming from the TLS was carried out using the
process described in [
59
]. The spatial resolution achieved was 15 mm for an average scanning distance
of 20 m. The georeferencing of the TLS into a global coordinate system (ETRS89) was done by a GNSS
network of 50 control points, using two Leica 1200 GPS, distributed in the towers and the upper part
of the wall as well as along its base. The registration errors of the individual TLS scans have been
Remote Sens. 2017,9, 189 8 of 17
evaluated by a network of control points, analyzing its propagation in a close object. The discrepancies
reached up to 5 cm as stated in [59]. The TLS workload involved 159 h for the 2.5 km perimeter wall.
The 3D point cloud obtained (Figure 5) by means of MLS of the whole Medieval Wall of Avila
was carried out according to the process described in [
60
]. The spatial resolution acquired was 60 mm
for an average scanning distance of 25 m. The georeferencing of the LiDAR 3D point cloud into a
global coordinate system was done by the integrated IMU sensor (Applanix POS LV 520) on the LYNX
Mobile Mapper.
Remote Sens. 2017, 9, 189 8 of 17
The 3D point cloud obtained (Figure 5) by means of MLS of the whole Medieval Wall of Avila
was carried out according to the process described in [60]. The spatial resolution acquired was 60
mm for an average scanning distance of 25 m. The georeferencing of the LiDAR 3D point cloud into
a global coordinate system was done by the integrated IMU sensor (Applanix POS LV 520) on the
LYNX Mobile Mapper.
(a)
(b)
Figure 4. TLS 3D point cloud (a); and XY view (b) of the Wall of Avila.
(a)
(b)
Figure 5. MLS 3D point cloud (a); and XY view (b) of the “Puerta del Alcázar” (Gate of the Fortress)
of the Wall of Avila.
MLS and TLS workload, in terms of time consumption and accuracy, is compared in Table 4.
Table 4. Comparison between MLS and TLS field work and processed point cloud.
Trimble GX LYNX Mobile Mappe
r
Measuring principle Time of Flight (ToF) Time of Flight (ToF)
Range
350 m to 90% reflectivity
250 m to 10% reflectivity 200 m to 35% reflectivity
155 m to 18% reflectivity
Resolution 15 mm 60 mm
Scanning speed up to 5000 points per second up to 500 lines/sec
Scanned area (approximate) 30,000 m
2
250,000 m
2
No. of stations 98 1
No. of points 300,000,000 185,000,000
No. of images 215 420
Geodetic reference system-projection ETRS89 and UTM30 ETRS89 and UTM30
Acquisition time 150 h (laser) + 4 h (camera) + 5 h (GNSS) 1 h
Processing time 435 h 15 h
4.1. Point Cloud Accuracy Assessment
The accuracy assessment was focused in the named “Alcazar” door, one of the highest wall
entrances (Figure 6), and its vicinity, originally annex to the “Alcazar” fortress, where the defensive
system was reinforced with a barbican and a ditch.
Figure 4. TLS 3D point cloud (a); and XY view (b) of the Wall of Avila.
Remote Sens. 2017, 9, 189 8 of 17
The 3D point cloud obtained (Figure 5) by means of MLS of the whole Medieval Wall of Avila
was carried out according to the process described in [60]. The spatial resolution acquired was 60
mm for an average scanning distance of 25 m. The georeferencing of the LiDAR 3D point cloud into
a global coordinate system was done by the integrated IMU sensor (Applanix POS LV 520) on the
LYNX Mobile Mapper.
(a)
(b)
Figure 4. TLS 3D point cloud (a); and XY view (b) of the Wall of Avila.
(a)
(b)
Figure 5. MLS 3D point cloud (a); and XY view (b) of the “Puerta del Alcázar” (Gate of the Fortress)
of the Wall of Avila.
MLS and TLS workload, in terms of time consumption and accuracy, is compared in Table 4.
Table 4. Comparison between MLS and TLS field work and processed point cloud.
Trimble GX LYNX Mobile Mappe
r
Measuring principle Time of Flight (ToF) Time of Flight (ToF)
Range
350 m to 90% reflectivity
250 m to 10% reflectivity 200 m to 35% reflectivity
155 m to 18% reflectivity
Resolution 15 mm 60 mm
Scanning speed up to 5000 points per second up to 500 lines/sec
Scanned area (approximate) 30,000 m
2
250,000 m
2
No. of stations 98 1
No. of points 300,000,000 185,000,000
No. of images 215 420
Geodetic reference system-projection ETRS89 and UTM30 ETRS89 and UTM30
Acquisition time 150 h (laser) + 4 h (camera) + 5 h (GNSS) 1 h
Processing time 435 h 15 h
4.1. Point Cloud Accuracy Assessment
The accuracy assessment was focused in the named “Alcazar” door, one of the highest wall
entrances (Figure 6), and its vicinity, originally annex to the “Alcazar” fortress, where the defensive
system was reinforced with a barbican and a ditch.
Figure 5.
MLS 3D point cloud (
a
); and XY view (
b
) of the “Puerta del Alcázar” (Gate of the Fortress) of
the Wall of Avila.
MLS and TLS workload, in terms of time consumption and accuracy, is compared in Table 4.
Table 4. Comparison between MLS and TLS field work and processed point cloud.
Trimble GX LYNX Mobile Mapper
Measuring principle Time of Flight (ToF) Time of Flight (ToF)
Range
350 m to 90% reflectivity
250 m to 10% reflectivity
200 m to 35% reflectivity
155 m to 18% reflectivity
Resolution 15 mm 60 mm
Scanning speed up to 5000 points per second up to 500 lines/sec
Scanned area (approximate) 30,000 m2250,000 m2
No. of stations 98 1
No. of points 300,000,000 185,000,000
No. of images 215 420
Geodetic reference system-projection ETRS89 and UTM30 ETRS89 and UTM30
Acquisition time
150 h (laser) + 4 h (camera) + 5 h (GNSS)
1 h
Processing time 435 h 15 h
Remote Sens. 2017,9, 189 9 of 17
4.1. Point Cloud Accuracy Assessment
The accuracy assessment was focused in the named “Alcazar” door, one of the highest wall
entrances (Figure 6), and its vicinity, originally annex to the “Alcazar” fortress, where the defensive
system was reinforced with a barbican and a ditch.
Remote Sens. 2017, 9, 189 9 of 17
Figure 6. Segmented MLS point cloud of the analyzed zone (southeast walls), coded according to the
intensity values.
The georeferencing error of the MLS is dependent on three factors: the GNSS system, the
trajectory compensation and the calibration method of its sensors. The georeferencing error of the
TLS is associated with the two GPS receivers used. Thus, in order to evaluate the accuracy of a MLS
point cloud, a registration phase with ICP was carried out. Consequently, a comparison of both 3D
point clouds was done (TLS and MLS). After the removal of the non-overlap areas (e.g., parts
acquired by the MLS but not by the TLS), the discrepancies map was computed, as shown in Figure
7, separately for the east walls (Figure 7a) and the south walls (Figure 7c), which delimitated the
fortress. These discrepancies were computed using the point cloud-to-point cloud comparison,
obtained as the result of the signed error between both point clouds.
(a)
(b)
(c) (d)
Figure 7. Relative discrepancies maps between MLS and TLS in the “Alcazar” fortress (Left) and its
associated histogram (Right) for the: east walls (a,b); and south walls (c,d).
Figure 6.
Segmented MLS point cloud of the analyzed zone (southeast walls), coded according to the
intensity values.
The georeferencing error of the MLS is dependent on three factors: the GNSS system, the trajectory
compensation and the calibration method of its sensors. The georeferencing error of the TLS is
associated with the two GPS receivers used. Thus, in order to evaluate the accuracy of a MLS
point cloud, a registration phase with ICP was carried out. Consequently, a comparison of both
3D point clouds was done (TLS and MLS). After the removal of the non-overlap areas (e.g., parts
acquired by the MLS but not by the TLS), the discrepancies map was computed, as shown in Figure 7,
separately for the east walls (Figure 7a) and the south walls (Figure 7c), which delimitated the fortress.
These discrepancies were computed using the point cloud-to-point cloud comparison, obtained as the
result of the signed error between both point clouds.
Remote Sens. 2017, 9, 189 9 of 17
Figure 6. Segmented MLS point cloud of the analyzed zone (southeast walls), coded according to the
intensity values.
The georeferencing error of the MLS is dependent on three factors: the GNSS system, the
trajectory compensation and the calibration method of its sensors. The georeferencing error of the
TLS is associated with the two GPS receivers used. Thus, in order to evaluate the accuracy of a MLS
point cloud, a registration phase with ICP was carried out. Consequently, a comparison of both 3D
point clouds was done (TLS and MLS). After the removal of the non-overlap areas (e.g., parts
acquired by the MLS but not by the TLS), the discrepancies map was computed, as shown in Figure
7, separately for the east walls (Figure 7a) and the south walls (Figure 7c), which delimitated the
fortress. These discrepancies were computed using the point cloud-to-point cloud comparison,
obtained as the result of the signed error between both point clouds.
(a)
(b)
(c) (d)
Figure 7. Relative discrepancies maps between MLS and TLS in the “Alcazar” fortress (Left) and its
associated histogram (Right) for the: east walls (a,b); and south walls (c,d).
Figure 7.
Relative discrepancies maps between MLS and TLS in the “Alcazar” fortress (
Left
) and its
associated histogram (Right) for the: east walls (a,b); and south walls (c,d).
Remote Sens. 2017,9, 189 10 of 17
The obtained values were analyzed with a QQ-plot to confirm the non-normality of the sample
(Figure 8). This fact was hinted by the histogram shape (Figure 7b,d), but since it is directly
dependent on the bin size, it cannot be used as decision criterion. Since this sample does not follow
a normal distribution (Figure 8), it is not possible to infer the central tendency and dispersion of the
population according to Gaussian statistics parameters such as mean and standard deviation. For that
reason, the accuracy assessment was computed based on robust alternatives, using non-parametric
assumptions such as the median and the MAD value shown in Table 5.
Remote Sens. 2017, 9, 189 10 of 17
The obtained values were analyzed with a QQ-plot to confirm the non-normality of the sample
(Figure 8). This fact was hinted by the histogram shape (Figure 7b,d), but since it is directly
dependent on the bin size, it cannot be used as decision criterion. Since this sample does not follow a
normal distribution (Figure 8), it is not possible to infer the central tendency and dispersion of the
population according to Gaussian statistics parameters such as mean and standard deviation. For
that reason, the accuracy assessment was computed based on robust alternatives, using
non-parametric assumptions such as the median and the MAD value shown in Table 5.
Figure 8. QQ-plots of relative discrepancies between MLS and TLS in the “Alcazar” fortress for the:
east walls (left) and south walls (right).
Table 5. Robust statistical descriptors associated to relative discrepancies.
Statistics Value
East South Global
Mean 0.003 m 0.008 m 0.005 m
Standard deviation ±0.026 m ±0.017 m ±0.023 m
Median 0.003 m 0.007 m 0.005 m
MAD ±0.015 m ±0.006 m ±0.011 m
Quantile 25 0.011 m 0.000 m 0.006 m
Quantile 75 0.019 m 0.013 m 0.016 m
In this analysis, the median value is close to zero (2.8 mm in the east wall), as was expected after
the application of the ICP registration algorithm. However, a slightly higher value of 6.9 mm is
appreciated in the south wall, which points to some registration error, but it is inside the expected
range. Moreover, in Table 5, the dispersion value (MAD) is according to the expected error,
confirming the MLS a-priori reference values, especially in the south wall, which has a simple
geometry (Figure 7c). The analysis of the error sample by means of the non-parametric estimator for
the quantile, yields that the 95% of the error points are inside the [0.046 m; 0.047 m] interval. This
reference value is used in the optimization phase to carry out the point cloud simplification.
It is worth noting that, despite the symmetrical shape of the sample in the east wall (as shown in
the first and third quartile), there is a high difference (up to three times) between the standard
deviation and the MAD having as reference the south wall. The conclusions would seem wrong
without an adequate statistical parameter selection.
In conclusion, the spatial resolution achieved by the MLS and his distribution is directly related
to the sensor–object distance; therefore, there will be areas with higher point density than others
(Figure 9). Moreover, in Figure 9, some diagonal stripes are shown by cause of inhomogeneity in the
acquisition phase due to the oblique laser scanner heads arrangement in the vehicle (Figure 1). This
fact is not a critical issue, but reinforces the necessity of a 3D point cloud data optimization that
copes with those redundant areas.
Figure 8.
QQ-plots of relative discrepancies between MLS and TLS in the “Alcazar” fortress for the:
east walls (left) and south walls (right).
Table 5. Robust statistical descriptors associated to relative discrepancies.
Statistics Value
East South Global
Mean 0.003 m 0.008 m 0.005 m
Standard deviation ±0.026 m ±0.017 m ±0.023 m
Median 0.003 m 0.007 m 0.005 m
MAD ±0.015 m ±0.006 m ±0.011 m
Quantile 25 0.011 m 0.000 m 0.006 m
Quantile 75 0.019 m 0.013 m 0.016 m
In this analysis, the median value is close to zero (2.8 mm in the east wall), as was expected
after the application of the ICP registration algorithm. However, a slightly higher value of 6.9 mm
is appreciated in the south wall, which points to some registration error, but it is inside the expected
range. Moreover, in Table 5, the dispersion value (MAD) is according to the expected error, confirming
the MLS a-priori reference values, especially in the south wall, which has a simple geometry (Figure 7c).
The analysis of the error sample by means of the non-parametric estimator for the quantile, yields that
the 95% of the error points are inside the [
0.046 m; 0.047 m] interval. This reference value is used in
the optimization phase to carry out the point cloud simplification.
It is worth noting that, despite the symmetrical shape of the sample in the east wall (as shown in
the first and third quartile), there is a high difference (up to three times) between the standard deviation
and the MAD having as reference the south wall. The conclusions would seem wrong without an
adequate statistical parameter selection.
In conclusion, the spatial resolution achieved by the MLS and his distribution is directly related
to the sensor–object distance; therefore, there will be areas with higher point density than others
(Figure 9). Moreover, in Figure 9, some diagonal stripes are shown by cause of inhomogeneity in
the acquisition phase due to the oblique laser scanner heads arrangement in the vehicle (Figure 1).
This fact is not a critical issue, but reinforces the necessity of a 3D point cloud data optimization that
copes with those redundant areas.
Remote Sens. 2017,9, 189 11 of 17
Remote Sens. 2017, 9, 189 11 of 17
Figure 9. MLS point density distribution (points/m
2
) for the southeast walls.
According to an ideal equilateral triangles distribution for a circular neighborhood [61], the
average point density achieved is 58.8 mm, decreasing up to a spatial resolution of 135 mm for the
farthest zones (e.g., battements and upper parts of towers). In addition, the occlusion in these zones
underestimates the density computation since the point vicinity is incomplete.
4.2. Optimization Analysis
In the Medieval Wall of Avila, two main features are clearly recognized: the planar walls and
the cylindrical towers. On the one hand, the walls are considered to have zero curvature, but due to
the construction processes and the deterioration of the structures caused by the course of time, the
actual conservation state does not keep this property. Therefore, a slight curvature of 0.005 m
1
(200
m radius) can be assumed. Additionally, the individual stones of the wall also contribute to local
groups of higher curvature, so this threshold is extended to the annexed towers. On the other hand,
in the case of the towers, the diameters of the cylinders range from 4.5 to 10 m. Accordingly, the
upper limit for the curvature cluster is set as 0.4 m
1
. The rest of the elements (e.g., battlements,
arrow slits or natural rocks) are contained in a final cluster of higher local curvature.
The size of these main elements (walls and towers) suggests a wide neighborhood for the
computations to balance the curvature precision and noise effects. The minimum height of 10 m and
the typical wall width of 20 m favors this large neighborhood. By a selection of a spherical diameter
of 2 m for the curvature computation, implicitly, a security buffer of 1 m is kept between the
different clusters. Additionally, a minimum area of 1 m
2
per cluster is set as threshold for the
clustering refinement.
As the final parameter definition, the sampling interval for the homogenization is chosen as 5
cm, which, according to Section 4.1, is the nearest round value corresponding to the 95%
non-parametric confidence interval of the MLS point clouds. In the first step of the preparation phase,
0.4% of points were excluded as outliers, since they exceeded the point to local plane distance of 20
cm. As result, after the smoothing process and the spatial resolution homogenization, 521,090 points
were obtained, which implies a 47.4% reduction of the original input.
Next, the curvature analysis is carried out for a spherical neighborhood of 1 m of radius, which
is segmented according the three pre-established clusters: (i) planar or low curvature (<0.05 m
1
); (ii)
cylindrical, mild transitions and medium curvature (0.05–0.40 m
1
); and (iii) the abrupt transitions of
high curvature (>0.40 m
1
). The results are outlined in Figure 10.
Subsequently, these clusters are refined based on a connected components analysis. The octree
level is chosen in relation to a reference spatial sampling of 0.12 m, which corresponds to 160 points
per square meter. Table 6 shows the percentage of reallocated components per cluster and the final
classification. For the low curvature cluster, the main planar elements are coded by only 47
components, the rest being small components (about two thousand) located in the transition area of
the wall planes, and in the transition to the battlements. Despite the high number of affected
components, more than 97% of cluster components, they only represent a small fraction of the
classified 3D points in the cluster. Similarly, for the cylindrical, mild transitions and medium
Figure 9. MLS point density distribution (points/m2) for the southeast walls.
According to an ideal equilateral triangles distribution for a circular neighborhood [
61
],
the average point density achieved is 58.8 mm, decreasing up to a spatial resolution of 135 mm
for the farthest zones (e.g., battements and upper parts of towers). In addition, the occlusion in these
zones underestimates the density computation since the point vicinity is incomplete.
4.2. Optimization Analysis
In the Medieval Wall of Avila, two main features are clearly recognized: the planar walls and
the cylindrical towers. On the one hand, the walls are considered to have zero curvature, but due to
the construction processes and the deterioration of the structures caused by the course of time, the
actual conservation state does not keep this property. Therefore, a slight curvature of 0.005 m
1
(200 m
radius) can be assumed. Additionally, the individual stones of the wall also contribute to local groups
of higher curvature, so this threshold is extended to the annexed towers. On the other hand, in the
case of the towers, the diameters of the cylinders range from 4.5 to 10 m. Accordingly, the upper limit
for the curvature cluster is set as 0.4 m
1
. The rest of the elements (e.g., battlements, arrow slits or
natural rocks) are contained in a final cluster of higher local curvature.
The size of these main elements (walls and towers) suggests a wide neighborhood for the
computations to balance the curvature precision and noise effects. The minimum height of 10 m
and the typical wall width of 20 m favors this large neighborhood. By a selection of a spherical
diameter of 2 m for the curvature computation, implicitly, a security buffer of 1 m is kept between
the different clusters. Additionally, a minimum area of 1 m
2
per cluster is set as threshold for the
clustering refinement.
As the final parameter definition, the sampling interval for the homogenization is chosen as 5 cm,
which, according to Section 4.1, is the nearest round value corresponding to the 95% non-parametric
confidence interval of the MLS point clouds. In the first step of the preparation phase, 0.4% of points
were excluded as outliers, since they exceeded the point to local plane distance of 20 cm. As result,
after the smoothing process and the spatial resolution homogenization, 521,090 points were obtained,
which implies a 47.4% reduction of the original input.
Next, the curvature analysis is carried out for a spherical neighborhood of 1 m of radius, which
is segmented according the three pre-established clusters: (i) planar or low curvature (<0.05 m
1
);
(ii) cylindrical, mild transitions and medium curvature (0.05–0.40 m
1
); and (iii) the abrupt transitions
of high curvature (>0.40 m1). The results are outlined in Figure 10.
Subsequently, these clusters are refined based on a connected components analysis. The octree
level is chosen in relation to a reference spatial sampling of 0.12 m, which corresponds to 160 points
per square meter. Table 6shows the percentage of reallocated components per cluster and the final
classification. For the low curvature cluster, the main planar elements are coded by only 47 components,
the rest being small components (about two thousand) located in the transition area of the wall planes,
and in the transition to the battlements. Despite the high number of affected components, more than
97% of cluster components, they only represent a small fraction of the classified 3D points in the
Remote Sens. 2017,9, 189 12 of 17
cluster. Similarly, for the cylindrical, mild transitions and medium curvature cluster, the reallocated
components are located in the battlements and in some abrupt rocks distribution inside the curtain
walls. In the case of the battlements, this behavior is caused due to their small size in relation to the
curvature spherical vicinity used for the computation. However, due to the optimization protocol,
they are moved to a cluster with crescent curvature in order to keep a higher spatial density. In the
case of natural rocks aggrupation located in the wall, their curvature was lower than the threshold for
the high curvature cluster (>0.4 m
1
). As result of the minimum area threshold, they were reallocated
in the high curvature cluster since they were heterogeneously disposed in the curtain walls.
Remote Sens. 2017, 9, 189 12 of 17
curvature cluster, the reallocated components are located in the battlements and in some abrupt
rocks distribution inside the curtain walls. In the case of the battlements, this behavior is caused due
to their small size in relation to the curvature spherical vicinity used for the computation. However,
due to the optimization protocol, they are moved to a cluster with crescent curvature in order to
keep a higher spatial density. In the case of natural rocks aggrupation located in the wall, their
curvature was lower than the threshold for the high curvature cluster (>0.4 m1). As result of the
minimum area threshold, they were reallocated in the high curvature cluster since they were
heterogeneously disposed in the curtain walls.
Figure 10. Initial MLS set of clusters: (i) planar or low curvature in blue; (ii) cylindrical, mild
transitions and medium curvature in green; and (iii) the abrupt transitions of high curvature in red.
Table 6. Results of clustering refinement step (* the components are the result of the curvature
clustering phase according to the main geometric elements of the CH site).
Cluster
Curvature
Classification
Legend
Initial
Points Components * Reallocated
Components
Reallocated
Points
Final
Points
Low Blue 236,832 2022 1975 (97.7%) 13,716 (5.8%) 223,116
Medium Green 263,580 1610 1563 (97.1%) 18,576 (7.0%) 258,720
High Red 20,654 - - - 39,230
Finally, the weighted sampling is applied inside the clusters. For the low curvature, a resolution
of 1 m is chosen, since the smooth process and plane-to-curve idealization. In the intermediate
cluster, aimed for the towers, an arc to chord assimilation is applied. For a 50 cm chord, the
idealization error will cause a sagittal of 13 mm for the worst case, which is inside the acceptable
limits. Therefore, a 0.5 m weighted sampling is set. Lastly, the high curvature cluster is kept
unmodified. The final point cloud has 58,476 points (Figure 11), which implies a reduction of 88.8%
in relation to the spatially homogenized point cloud (521,090 points), and 94.1% in relation to the
raw point cloud.
Figure 11. Final MLS point cloud optimization and simplification: (i) planar or low curvature in blue;
(ii) cylindrical, mild transitions and medium curvature in green; and (iii) the abrupt transitions of
high curvature in red.
Figure 10.
Initial MLS set of clusters: (i) planar or low curvature in blue; (ii) cylindrical, mild transitions
and medium curvature in green; and (iii) the abrupt transitions of high curvature in red.
Table 6.
Results of clustering refinement step (* the components are the result of the curvature clustering
phase according to the main geometric elements of the CH site).
Cluster
Curvature
Classification
Legend Initial Points Components * Reallocated
Components
Reallocated
Points Final Points
Low Blue 236,832 2022 1975 (97.7%) 13,716 (5.8%) 223,116
Medium Green 263,580 1610 1563 (97.1%) 18,576 (7.0%) 258,720
High Red 20,654 - - - 39,230
Finally, the weighted sampling is applied inside the clusters. For the low curvature, a resolution
of 1 m is chosen, since the smooth process and plane-to-curve idealization. In the intermediate cluster,
aimed for the towers, an arc to chord assimilation is applied. For a 50 cm chord, the idealization error
will cause a sagittal of 13 mm for the worst case, which is inside the acceptable limits. Therefore,
a 0.5 m weighted sampling is set. Lastly, the high curvature cluster is kept unmodified. The final point
cloud has 58,476 points (Figure 11), which implies a reduction of 88.8% in relation to the spatially
homogenized point cloud (521,090 points), and 94.1% in relation to the raw point cloud.
Remote Sens. 2017, 9, 189 12 of 17
curvature cluster, the reallocated components are located in the battlements and in some abrupt
rocks distribution inside the curtain walls. In the case of the battlements, this behavior is caused due
to their small size in relation to the curvature spherical vicinity used for the computation. However,
due to the optimization protocol, they are moved to a cluster with crescent curvature in order to
keep a higher spatial density. In the case of natural rocks aggrupation located in the wall, their
curvature was lower than the threshold for the high curvature cluster (>0.4 m1). As result of the
minimum area threshold, they were reallocated in the high curvature cluster since they were
heterogeneously disposed in the curtain walls.
Figure 10. Initial MLS set of clusters: (i) planar or low curvature in blue; (ii) cylindrical, mild
transitions and medium curvature in green; and (iii) the abrupt transitions of high curvature in red.
Table 6. Results of clustering refinement step (* the components are the result of the curvature
clustering phase according to the main geometric elements of the CH site).
Cluster
Curvature
Classification
Legend
Initial
Points Components * Reallocated
Components
Reallocated
Points
Final
Points
Low Blue 236,832 2022 1975 (97.7%) 13,716 (5.8%) 223,116
Medium Green 263,580 1610 1563 (97.1%) 18,576 (7.0%) 258,720
High Red 20,654 - - - 39,230
Finally, the weighted sampling is applied inside the clusters. For the low curvature, a resolution
of 1 m is chosen, since the smooth process and plane-to-curve idealization. In the intermediate
cluster, aimed for the towers, an arc to chord assimilation is applied. For a 50 cm chord, the
idealization error will cause a sagittal of 13 mm for the worst case, which is inside the acceptable
limits. Therefore, a 0.5 m weighted sampling is set. Lastly, the high curvature cluster is kept
unmodified. The final point cloud has 58,476 points (Figure 11), which implies a reduction of 88.8%
in relation to the spatially homogenized point cloud (521,090 points), and 94.1% in relation to the
raw point cloud.
Figure 11. Final MLS point cloud optimization and simplification: (i) planar or low curvature in blue;
(ii) cylindrical, mild transitions and medium curvature in green; and (iii) the abrupt transitions of
high curvature in red.
Figure 11. Final MLS point cloud optimization and simplification: (i) planar or low curvature in blue;
(ii) cylindrical, mild transitions and medium curvature in green; and (iii) the abrupt transitions of high
curvature in red.
Remote Sens. 2017,9, 189 13 of 17
5. Discussion
The novel use of MLS applied to the 3D surveying of large Cultural Heritage sites is evaluated
and compared with the TLS, which is currently the most common technology used for this purpose.
MLS device combines different sensors on a mobile platform that allows obtaining an accurate
and georeferenced 3D point cloud. Active sensors together with Inertial and Global Navigation System
are synchronized in real time to collect and store data. The active sensor acquires detailed 3D points
from a local coordinate system, while the IMU and GNSS systems provide and correct 3D points under
a global reference frame. Moreover, some MLS can acquire RGB information captured by photographic
or video cameras. Although the MLS spatial resolution depends on the active sensor (laser scanner)
integrated on the mobile platform, usually the point clouds acquired by MLS cannot compete with
TLS in terms of high resolution. Nevertheless, for large CH sites, the use of MLS could be more
suitable, allowing the mapping of large and complex sites in an efficient way (e.g., historical cities
or landscapes).
Otherwise, MLS reduces the data acquisition time thanks to the use of a mobile platform,
while TLS needs multiple stations (sometimes even hundreds) to cover a large area. There are two
main advantages of MLS technology compared to other 3D recording systems: (i) MLS is less expensive
than airborne LiDAR systems boarded on airplanes or helicopters; and (ii) MLS can acquire accurate
3D data faster than TLS. On the contrary, its main disadvantage is that sometimes the transit with vans
or cars is not possible due to conservation reasons or limitations of space (e.g., narrow streets). In this
last case, it is possible to board the MLS on alternative platforms such as quads [
62
], backpack [
16
] or
even autonomous robots [17].
MLS also reduces the data processing time in comparison with TLS, especially in large and
complex CH sites. MLS directly produces a single point cloud, georeferenced with centimetric accuracy
due to the IMU and GNSS calibrated sensors, whereas the 3D data captured by TLS have to be aligned
in a relative or global reference system.
Finally, the visualization and management of large datasets still represent a bottleneck for both
techniques (MLS and TLS). The ongoing development of 3D recording sensors and data capture
technologies are advancing more rapidly than the management of large geometrical datasets. MLS and
TLS usually produce massive point clouds (millions of points) requiring storage spaces of many GB.
To advance this problem, 3D datasets (i.e., point clouds) need to be efficiently optimized and simplified
for architectural and structural analysis (e.g., FEM), documentation and management (e.g., HBIM)
or outreach purposes (e.g., Web Visualization). As additional benefits of the proposed point cloud
optimization, the optimized point cloud will be improved for subsequent 3D modeling tasks based
on meshing or reverse engineering procedures, improving the computational time and the resulting
geometric definition.
In this paper, our main contribution was based on the use of algorithms and methodologies for the
point cloud optimization, without any prior surface reconstruction. There are numerous algorithms to
reduce the 3D point cloud in automatic way, but our approach was focused on an adaptive sampling
that depends on the geometric features and also keeps the main structures without losing relevant
details. The optimal solution was to find a compromise between the resolution acquired, the accuracy
and the final size of the file, in order to obtain detailed 3D point clouds that could be stored, managed
and visualized with a fluent real-time interaction.
6. Conclusions
This paper deals with a novel use of Mobile LiDAR System for the 3D recording of large Cultural
Heritage sites. As part of the suitability assessment of this technology, an accuracy evaluation with a
Terrestrial Laser Scanner was carried out. The obtained results, checked in an appropriate study case,
allow us to confirm the suitability of Mobile LiDAR System techniques for 3D recording and modeling
of large Cultural Heritage sites.
Remote Sens. 2017,9, 189 14 of 17
On the one hand, we were focused on time consumption. Comparing both 3D recording
techniques, Mobile LiDAR System technology was a clear winner in terms of time used for data
acquisition and data processing. Data coming from our study case shows that Mobile LiDAR System
took only 1 h to sweep an area of 250,000 m
2
while the Terrestrial Laser Scanner took 159 h. This means
considerable time saving.
On the other hand, spatial resolution was our concern. Terrestrial Laser Scanner could achieve
a point cloud density of 15 mm (average of scanning distance: 20 m) while Mobile LiDAR System
density was 60 mm (average of scanning distance: 25 m). In order to evaluate data accuracy of the
Mobile LiDAR System, a comparison of both 3D point clouds was done, regarding the georeferencing
error. Considering that the represented area involved a few kilometers, the results show that 95%
of the MLS points acquired are inside of the tolerance range [
0.046 m; +0.047 m]. Even if Mobile
LiDAR System provides a less dense point cloud, it is conclusive that it has enough spatial resolution
and quality to provide the reconstruction of the most relevant architectural details in large Cultural
Heritage sites.
In any case, due to the huge amount of acquired data, the final point clouds should be necessarily
optimized for visualization and management purposes. A novel process for point cloud optimization
is introduced to facilitate its handling by scholars from various disciplines. The proposed optimization
methodology was based on a detail geometric analysis, allowing classifying the different structures
into three main clusters (low, medium and high curvature). Different optimization parameters were
used for each cluster according to their curvature, obtaining a reduced point cloud ca. 94.1% less in
relation to the entire raw point cloud and guaranteeing an optimal solution for multiple purposes.
The results confirm that Mobile LiDAR System also shows more efficiency in terms of operation
flexibility, acquisition and processing time, producing high quality and accurate data. However, shape
complexity and surrounding characteristics, such as the environment location or the access restrictions,
must be taken into account in further studies to define which is the best approach.
In the worst case, when the geometric complexity of the scene is high, the transit of vehicles is
difficult or they are restricted access to the archaeological sites, the novel indoor LiDAR technologies
(e.g., the backpack solution, mobile robots or the handheld LiDAR mapping system) could be used as
an alternative outdoor solution.
To conclude, it would be interesting to extend our approach integrating different outdoors and
indoors mobile techniques for large and complex Cultural Heritage sites.
Acknowledgments:
The authors wish to thanks Insitu Engineering S.L. for providing the LiDAR data. The first
author would also like to thank the University of Salamanca, for the financial support given through human
resources grants (Special Program for Post-Doctoral Contracts). The authors would like to thank the editors and
anonymous reviewers for their valuable feedback. This research has been partially supported by CHT2—Cultural
Heritage through Time—funded by JPI CH Joint Call and supported by the Ministerio de Economia y
Competitividad, Ref. PCIN-2015-071.
Author Contributions:
All of the authors conceived and designed the study. Pablo Rodríguez-Gonzálvez,
Diego Gonzalez-Aguilera and Ángel Luis Muñoz-Nieto acquired the TLS data and processed it. Pedro Arias-Sanchez
acquired the MLS data and processed it. Pablo Rodríguez-Gonzálvez and Belén Jiménez Fernández-Palacios
implemented the methodology and analysed the results. Pablo Rodríguez-Gonzálvez, Belén Jiménez Fernández-Palacios
and Diego Gonzalez-Aguilera wrote the manuscript and all authors read and approved the final version.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Guidi, G.; Micoli, L.L.; Gonizzi, S.; Brennan, M.; Frischer, B. Image-based 3D capture of cultural heritage
artifacts an experimental study about 3D data quality. In Proceedings of the 2015 Digital Heritage, Granada,
Spain, 28 September–2 October 2015; pp. 321–324.
2.
Del Pozo, S.; Herrero-Pascual, J.; Felipe-García, B.; Hernández-López, D.; Rodríguez-Gonzálvez, P.;
González-Aguilera, D. Multispectral radiometric analysis of façades to detect pathologies from active
and passive remote sensing. Remote Sens. 2016,8, 80. [CrossRef]
Remote Sens. 2017,9, 189 15 of 17
3.
Borrmann, D.; Heß, R.; Houshiar, H.; Eck, D.; Schilling, K.; Nüchter, A. Robotic mapping of cultural heritage
sites. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2015,XL-5/W4, 9–16. [CrossRef]
4.
Puente, I.; Solla, M.; González-Jorge, H.; Arias, P. NDT documentation and evaluation of the Roman Bridge
of Lugo using GPR and mobile and static LiDAR. J. Perform. Constr. Facil. 2013,29, 06014004. [CrossRef]
5.
Casadio, F.; Daher, C.; Bellot-Gurlet, L. Raman spectroscopy of cultural heritage materials: Overview of
applications and new frontiers in instrumentation, sampling modalities, and data processing. Top. Curr. Chem.
2016,374, 62. [CrossRef] [PubMed]
6.
Johnson, K.M.; Ouimet, W.B. Rediscovering the lost archaeological landscape of southern New England
using airborne light detection and ranging (LiDAR). J. Archaeol. Sci. 2014,43, 9–20. [CrossRef]
7.
Von Schwerin, J.; Richards-Rissetto, H.; Remondino, F.; Spera, M.G.; Auer, M.; Billen, N.; Loos, L.; Stelson, L.;
Reindel, M. Airborne LiDAR acquisition, post-processing and accuracy-checking for a 3D WebGIS of Copan,
Honduras. J. Archaeol. Sci. Rep. 2016,5, 85–104. [CrossRef]
8.
Hyyppä, J.; Jaakkola, A.; Chen, Y.; Kukko, A. Unconventional LIDAR mapping from air, terrestrial and
mobile. In Photogrammetric Week, 2013; Wichmann/VDE: Belin/Offenbach, Germany, 2013; pp. 205–214.
9.
Puente, I.; González-Jorge, H.; Martínez-Sáchez, J.; Arias, P. Review of mobile mapping and surveying
systems. Measurement 2013,46, 2127–2145. [CrossRef]
10.
Williams, K.; Olsen, M.; Roe, G.; Glennie, C. Synthesis of transportation applications of mobile LIDAR.
Remote Sens. 2013,5, 4652. [CrossRef]
11.
Lim, S.; Thatcher, C.A.; Brock, J.C.; Kimbrow, D.R.; Danielson, J.J.; Reynolds, B.J. Accuracy assessment of a
mobile terrestrial lidar survey at Padre Island National Seashore. Int. J. Remote Sens.
2013
,34, 6355–6366.
[CrossRef]
12.
Moskal, L.M.; Zheng, G. Retrieving forest inventory variables with terrestrial laser scanning (TLS) in urban
heterogeneous forest. Remote Sens. 2012,4, 1. [CrossRef]
13.
González-Ferreiro, E.; Miranda, D.; Barreiro-Fernandez, L.; Bujan, S.; Garcia-Gutierrez, J.; Dieguez-Aranda, U.
Modelling stand biomass fractions in Galician Eucalyptus globulus plantations by use of different LiDAR
pulse densities. For. Syst. 2013,22, 510–525. [CrossRef]
14.
Bauwens, S.; Bartholomeus, H.; Calders, K.; Lejeune, P. Forest inventory with terrestrial LiDAR:
A comparison of static and hand-held mobile laser scanning. Forests 2016,7, 127. [CrossRef]
15.
Zhang, S.; Wang, C.; Yang, Z.; Chen, Y.; Li, J. Automatic railway power line extraction using mobile laser
scanning data. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2016,XLI-B5, 615–619. [CrossRef]
16.
Lauterbach, H.; Borrmann, D.; Heß, R.; Eck, D.; Schilling, K.; Nüchter, A. Evaluation of a backpack-mounted
3D mobile scanning system. Remote Sens. 2015,7, 13753–13781. [CrossRef]
17.
Ziparo, V.A.; Zaratti, M.; Grisetti, G.; Bonanni, T.M.; Serafin, J.; Cicco, M.D.; Proesmans, M.; Gool, L.V.;
Vysotska, O.; Bogoslavskyi, I.; et al. Exploration and mapping of catacombs with mobile robots.
In Proceedings of the 2013 IEEE International Symposium on Safety, Security, and Rescue Robotics, Linköping,
Sweden, 21–26 October 2013; pp. 1–2.
18.
Bosse, M.; Zlot, R.; Flick, P. Zebedee: Design of a spring-mounted 3-D range sensor with application to
mobile mapping. IEEE Trans. Robot. 2012,28, 1104–1119. [CrossRef]
19.
Zlot, R.; Bosse, M.; Greenop, K.; Jarzab, Z.; Juckes, E.; Roberts, J. Efficiently capturing large, complex cultural
heritage sites with a handheld mobile 3D laser mapping system. J. Cult. Herit.
2014
,15, 670–678. [CrossRef]
20.
Sirmacek, B.; Shen, Y.; Lindenbergh, R.; Zlatanova, S.; Diakite, A. Comparison of Zeb1 and Leica C10 indoor
laser scanning point clouds. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci.
2016
,III-1, 143–149. [CrossRef]
21.
Calisi, D.; Giannone, F.; Ventura, C.; Salonia, P.; Cottefoglie, F.; Ziparo, V.A. ARIS—A Robotic Approach
to Digitization of Indoor and Underground Cultural Heritage Sites. In Proceedings of the International
Congress on Digital Heritage, Granada, Spain, 28 September–2 October 2015.
22.
Calisi, D.; Giannone, F.; Ventura, C.; Salonia, P.; Cottefoglie, F.; Ziparo, V.A. Digitizing Indoor and
Underground Cultural Heritage Sites with Robots. Sci. Res. Inf. Technol. 2016,6, 23–30.
23.
Rondeau, M.-C. New Technologies for city asset mapping: Setting the standard in city digitalisation.
GeoInformatics 2015,18, 20–21.
24.
Rönnholm, P.; Liang, X.; Kukko, A.; Jaakkola, A.; Hyyppä, J. Quality analysis and correction of mobile
backpack laser scanning data. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci.
2016
,III-1, 41–47.
[CrossRef]
Remote Sens. 2017,9, 189 16 of 17
25.
Sairam, N.; Nagarajan, S.; Ornitz, S. Development of mobile mapping system for 3D road asset inventory.
Sensors 2016,16, 367. [CrossRef] [PubMed]
26.
Pauly, M.; Gross, M.; Kobbelt, L.P. Efficient simplification of point-sampled surfaces. In Proceedings of the
Conference on Visualization, Boston, MA, USA, 27 October–1 November 2002; IEEE Computer Society:
Boston, MA, USA, 2002; pp. 163–170.
27.
Song, H.; Feng, H.-Y. A point cloud simplification algorithm for mechanical part inspection. In Proceedings
of the Seventh International Conference on Information Technology for Balanced Automation Systems in
Manufacturing and Services, Niagara Falls, ON, Canada, 4–6 September 2006; Shen, W., Ed.; Springer:
Boston, MA, USA, 2006; pp. 461–468.
28.
Charbonnier, P.; Tarel, J.-P.; Goulette, F. On the Diagnostic of Road Pathway Visibility; Transport Research Arena
Europe: Bruxelles, France, 2010.
29.
Yang, B.; Dong, Z. A shape-based segmentation method for mobile laser scanning point clouds. ISPRS J.
Photogramm. Remote Sens. 2013,81, 19–30. [CrossRef]
30.
Sun, H.; Wang, C.; El-Sheimyb, N.; traffics Wang, P. Surrounding detection for terrestrial mobile mapping
data quality evaluation. In Proceedings of the 6th International Symposium on Mobile Mapping Technology,
São Paulo, Brazil, 21–24 July 2009.
31.
Oude Elberink, S.; Khoshelham, K.; Arastounia, M.; Diaz Benito, D. Rail track detection and modelling
in mobile laser scanner data. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci.
2013
,II-5/W2, 223–228.
[CrossRef]
32.
Torres-Martínez, J.; Seddaiu, M.; Rodríguez-Gonzálvez, P.; Hernández-López, D.; González-Aguilera, D.
A multi-data source and multi-sensor approach for the 3D reconstruction and web visualization of a complex
archaelogical site: The case study of “Tolmo De Minateda”. Remote Sens. 2016,8, 550. [CrossRef]
33.
Jiménez Fernández-Palacios, B.; Stefani, C.; Lombardo, J.; De Luca, L.; Remondino, F. Web visualization of
complex reality-based 3D models with NUBES. In Proceedings of the IEEE Conference Digital Heritage 2013,
Marseille, France, 28 October–1 November 2013; Volume 1, pp. 701–704.
34. Nubes Project. Available online: http://www.map.cnrs.fr/nubes/ (accessed on 25 December 2016).
35.
Cultural Heritage Information System Project. Available online: http://lrv.ugr.es/chis/en/ (accessed on
25 December 2016).
36.
Torres, J.C.; López, L.; Romo, C.; Arroyo, G.; Cano, P.; Lamolda, F.; Villafranca, M.M. Using a cultural heritage
information system for the documentation of the restoration process. In Proceedings of the 2013 Digital
Heritage, Marseille, France, 28 October–1 November 2013.
37.
Von Schwerin, J.; Richards-Rissetto, H.; Agugiaro, G.; Remondino, F.; Girardi, G. QueryArch3D: A 3D WebGIS
System linking 3D visualizations to archaeological data. In Proceedings of the Computer Applications and
Quantitative Methods in Archaeology, Southampton, UK, 26–30 March 2012.
38.
Prandi, F.; Devigili, F.; Soave, M.; Di Staso, U.; De Amicis, R. 3D web visualization of huge cityGML models.
Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2015,XL-3/W3, 601–605. [CrossRef]
39.
Conforti, D.; Zampa, F. Lynx mobile mapper for surveying city centers and highways. Int. Arch. Photogramm.
Remote Sens. Spat. Inf. Sci. 2011,XXXVIII-5/W16, 219–222. [CrossRef]
40.
Hofmann, S.; Brenner, C. Accuracy assessment of mobile mapping point clouds using the existing
environment as terrestrial reference. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci.
2016
,XLI-B1,
601–608. [CrossRef]
41.
Chan, T.O. Feature-Based Boresight Self-Calibration of a Mobile Mapping System. Master’s Thesis,
University of Calgary, Calgary, AB, Canada, 2012.
42.
Grussenmeyer, P.; Landes, T.; Doneus, M.; Lerma, J. Basics of range-based modelling techniques in cultural
heritage 3D recording. In 3D Recording, Documentation and Management of Cultural Heritage; Whittles
Publishing: Dunbeath, UK, 2016.
43.
Schulz, T. Calibration of a Terrestrial Laser Scanner for Engineering Geodesy. Ph.D. Thesis, Technical
University of Berlin, Berlin, Germany, 2007.
44.
Remondino, F. Heritage recording and 3D modeling with photogrammetry and 3D scanning. Remote Sens.
2011,3, 1104–1138. [CrossRef]
45.
Gonzalez-Jorge, H.; Rodriguez-Gonzalvez, P.; Gonzalez-Aguilera, D.; Varela-Gonzalez, M. Metrological
comparison of terrestrial laser scanning systems Riegl LMS Z390i and Trimble GX. Opt. Eng.
2011
,50,
116201–116209.
Remote Sens. 2017,9, 189 17 of 17
46.
González-Aguilera, D.; Rodríguez-Gonzálvez, P.; Armesto, J.; Arias, P. Trimble GX200 and Riegl LMS-Z390i
sensor self-calibration. Opt. Express 2011,19, 2676–2693. [CrossRef] [PubMed]
47.
Besl, P.J.; McKay, N.D. A method for registration of 3-d shapes. IEEE Trans. Pattern Anal. Mach. Intell.
1992
,
14, 239–256. [CrossRef]
48.
Tombari, F.; Salti, S.; Di Stefano, L. Performance evaluation of 3D keypoint detectors. Int. J. Comput. Vis.
2013,102, 198–220. [CrossRef]
49.
Akca, D. Full automatic registration of laser scanner point clouds. In Proceedings of the Optical 3-D
Measurement Techniques VI, Zurich, Switzerland, 22–25 September 2003; pp. 330–337.
50.
Paparoditis, N.; Papelard, J.-P.; Cannelle, B.; Devaux, A.; Soheilian, B.; David, N.; Houzay, E. Stereopolis II:
A multi-purpose and multi-sensor 3D mobile mapping system for street visualisation and 3D metrology.
Rev. Fr. Photogramm. Télédétec. 2012,200, 69–79.
51.
Toschi, I.; Rodríguez-Gonzálvez, P.; Remondino, F.; Minto, S.; Orlandini, S.; Fuller, A. Accuracy evaluation
of a mobile mapping system with advanced statistical methods. Int. Arch. Photogramm. Remote Sens. Spat.
Inf. Sci. 2015,XL-5/W4, 245–253. [CrossRef]
52.
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, 10–26.
[CrossRef]
53.
Höhle, J. The assessment of the absolute planimetric accuracy of airborne lasers canning. Int. Arch. Photogramm.
Remote Sens. Spat. Inf. Sci. 2011,XXXVIII-5/W12, 145–150.
54.
Rodríguez-Gonzálvez, P.; Garcia-Gago, J.; Gomez-Lahoz, J.; González-Aguilera, D. Confronting passive and
active sensors with non-gaussian statistics. Sensors 2014,14, 13759–13777. [CrossRef] [PubMed]
55. Cloud Compare v.2.7.0. Available online: www.danielgm.net/cc/ (accessed on 7 October 2016).
56.
Rodríguez-Gonzálvez, P.; González-Aguilera, D.; Hernández-López, D.; González-Jorge, H. Accuracy
assessment of airborne laser scanner dataset by means of parametric and non-parametric statistical methods.
IET Sci. Meas. Technol. 2015,9, 505–513. [CrossRef]
57.
Levin, D. The approximation power of moving least-squares. Math. Comput. Am. Math. Soc.
1998
,67,
1517–1531. [CrossRef]
58. Serna, M. Walls: Speaking of Towers, Battlements, Gates. Diario de Ávila, 13 October 2002.
59.
Mancera-Taboada, J.; Rodríguez-Gonzálvez, P.; González-Aguilera, D.; Muñoz-Nieto, Á.; Gómez-Lahoz, J.;
Herrero-Pascual, J.; Picón-Cabrera, I. On the use of laser scanner and photogrammetry for the global
digitization of the medieval walls of Avila. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci.
2010
,XXXVIII,
169–174.
60.
Haala, N.; Peter, M.; Cefalu, A.; Kremer, J. Mobile lidar mapping for urban data capture. In Proceedings of
the 14th International Conference on Virtual Systems and Multimedia, Limassol, Cyprus, 20–25 October 2008;
pp. 95–100.
61.
Rodríguez-Gonzálvez, P.; Nocerino, E.; Menna, F.; Minto, S.; Remondino, F. 3D surveying & modeling of
underground passages in WWI fortifications. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci.
2015
,
XL-5/W4, 17–24.
62.
Kurkin, A.; Pelinovsky, E.; Tyugin, D.; Kurkina, O.; Belyakov, V.; Makarov, V.; Zeziulin, D. Coastal remote
sensing using unmanned ground vehicles. Int. J. Environ. Sci. 2016,1, 183–189.
©
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/).
... Light Detection and Ranging (LiDAR) provides big data in a relatively short time. Hence, it is a prevalent technique in measuring the Earth's surface or structural objects, vegetation cover, forests, mining damages, cultural heritage sites, urban environments, and buildings [1][2][3][4][5][6][7][8][9]. Depending on the problem, or measurement object, one can apply different LiDAR platforms: airborne laser scanning (ALS), terrestrial laser scanning (TLS), or mobile laser scanning (MLS) [10]. ...
Article
Full-text available
Terrestrial laser scanning (TLS) is a modern measurement technique that provides a point cloud in a relatively short time. TLS data are usually processed using different methods in order to obtain the final result (infrastructure or terrain models). Msplit estimation is a modern method successfully applied for such a purpose. This paper addresses the possible application of the method in processing TLS data from two different epochs to model a vertical displacement of terrain resulting, for example, from landslides or mining damages. Msplit estimation can be performed in two variants (the squared or absolute method) and two scenarios (two point clouds or one combined point cloud). One should understand that point clouds usually contain outliers of different origins. Therefore, this paper considers the contamination of TLS data by positive or/and negative outliers. The results based on simulated data prove that absolute Msplit estimation provides better results and overperforms conventional estimation methods (least-squares or robust M-estimation). In practice, the processing of point clouds separately seems to be a better option. This paper proved that Msplit estimation is a compelling alternative to conventional methods, as it can be applied to process TLS data disturbed by outliers of different types.
... Galeazzi (2016) [48] used 3D laser scanning and photogrammetry techniques to produce archaeological records of the microscopic strata of caves under extreme environment and light conditions. Rodriguez-Gonzalvez et al. (2017) [49] synchronized LiDAR sensors with the inertial and global navigation systems that are installed on mobile platforms to generate point clouds with absolute geographic coordinates. It significantly improved the data acquisition efficiency of large cultural heritage sites and reduced the problems caused by multi-site cloud splicing. ...
Article
Full-text available
Three-dimensional point cloud has been widely used in the cultural heritage field in the last two decades, gaining attention from both academic and industry communities. A large number of scientific papers have been published concerning this topic, which covers a wide range of journals, countries, and disciplines. There has been no comprehensive and systematic survey of recent literature performed in a scientometric way based on the complex network analysis methods. In this work, we extracted the terms (i.e., noun phrases included in the title, abstract and keywords), the documents, the countries that the research institutions are located in, and the categories that the literature belongs to from the Web of Science database to compose a term co-occurrence network, document co-citation network, collaborative country network and category co-occurrence network using CiteSpace software. Through visualizing and analyzing those networks, we identified the research hotspots, landmark literature, national collaboration, interdisciplinary patterns as well as the emerging trends through assessing the central nodes and the nodes with strong citation bursts. This work not only provides a structured view on state-of-art literature, but also reveals the future trends of employing 3D point cloud data for cultural heritage, aiding researchers carry out further research in this area.
... While the process of creating textured polygon models using photogrammetry dates to the 1980s, the employment of photogrammetry as a tool for the capturing and rendering of artworks is a more recent development, with its origins postdating the turn of the twenty-first century [5]. Additionally, the technical processes involved in using LiDAR equipment to scan architectural structures, edit individual scans, and stitch together laser-captured forms were established only recently [6][7][8][9][10][11]. Since 2005, pipelines have been established to create accurate point cloud models of important sites such as the basilica of S. Francesco in Assisi, the Medieval wall of Avila, the Roman Forum in Pompeii, and multiple structures produced during the early modern period in Florence, to name but a few [12][13][14][15]. More recently, specialists have combined photogrammetric models with point clouds for research purposes, layering the former over the latter in order to place textured surfaces on top of spatially accurate models produced with LiDAR equipment [16][17][18][19]. ...
Article
Full-text available
The Digital Humanities project Florence As It Was ( http://florenceasitwas.wlu.edu ) seeks to reconstruct the architectural and decorative appearance of late Medieval and early Modern buildings by combining 3D point cloud models of buildings (i.e. extant structures such as chapels, churches, etc.) with 3D rendered models of artworks that were installed inside them during the fourteenth and fifteenth centuries. This paper documents a novel bifurcated workflow that allows the construction of such integrated 3D models as well as an example case study of a church in Florence, Italy called Orsanmichele ( https://3d.wlu.edu/v21/pages/orsanmichele2.html ). The key steps worked out in the optimized workflow include: (1) art historical research to identify the original artworks in each building, (2) the use of LiDAR scanners to obtain 3D data (along with associated color information) of the interiors and exteriors of buildings, (3) the use of high resolution photogrammetry of works of art (i.e. paintings and sculptures) which have been removed from those buildings and are now in public collections, (4) the generation of point clouds from the 3D data of the buildings and works of art, (5) the editing and cojoining of a textured polygon model of artworks with a reduced size (using novel algorithms) point cloud model of the buildings in an open-source software tool called Potree so that artworks can be embedded in their original architectural settings, and (6) the annotation of these models with scholarly art historical texts that present viewers immediate access to information, archival evidence, and historical descriptions of these spaces. The integrated point cloud and textured models of buildings and artworks, respectively, plus annotations are then published with Potree . This process has resulted in the development of highly accurate virtual reconstructions of key monuments from the Florentine Middle Ages and Renaissance (like the fourteenth century building of Orsanmichele and the multiple paintings that were once inside it) as they originally appeared. The goal of this project is to create virtual models for scholars and students to explore research questions while providing key information that may assist in generating new projects. Such models represent a significant tool to allow improved teaching of art and architectural history. Furthermore, since the assigned location of some of the historic artworks within these sites are not always firmly known, the virtual model allows users to experiment with potential arrangements of objects in and on the buildings they may have originally decorated.
... The drone LiDAR sector is growing rapidly and especially over the past few years. In only a short period of time, manufacturers of the best aircraft LiDAR sensors have engineered LiDAR sensors for small drones [9]. LiDAR Drones will have a huge impact in all life sectors see ...
Article
Full-text available
Unmanned Aerial Vehicle (UAV) is nowadays a valuable source of data for inspection, surveillance, mapping and 3D modeling issues. UAVs can be considered as a low cost alternative to the classical manned aerial photogrammetry, because they provide users with high spatial resolution images up to a centimeter limit. Following a typical photogrammetric workflow, 3D results like Digital Surface or Terrain Models (DTM/DSM), contours, textured 3D models, vector information, etc. can be produced, even on large areas. The paper reports the state of the art of UAV for Geomatics applications, giving an overview of different UAV platforms, applications and showing also the latest developments of UAV programs. The research presents the sequence of operations should be follow on the images taken with these aircraft for maximum use in remote sensing and air survey. Recent photographs were taken using the DJI Phantom 4 UAV (which was awarded to the Nineveh Department of Antiquities and Heritage by the Italian Mission) for the archaeological Nineveh region in Mosul city and use as a case study.
... Even if L-Vox can correct for the effects of occlusion, constraints that are related to canopy structure and scanning configuration remain. One solution that is becoming more useful for overcoming these constraints is the use of mobile LiDAR units (e.g., (Bauwens et al., 2016;Liang et al., 2018;Rodríguez-Gonzálvez et al., 2017), where the ability to walk freely under the canopy and around the trees is not only efficient in reducing occlusion, but also in reducing data acquisition time. Yet, the point cloud resulting from these mobile units does not have similar specifications to the TLS: beam divergence is greater, scanning density is lower, and the ability to probe the upper canopy is poorer. ...
Article
Plant area density (PAD in m²·m⁻³) defines the total one-sided total plant surface area within a given volume. It is a key variable in characterizing exchange processes between the atmosphere and land surface. Terrestrial laser scanning (TLS) provides unprecedented detail of the 3D structure of forest canopies. Yet, signal occlusion and uneven sampling density of the TLS point clouds limit our capacity to characterize the 3D distribution of canopy components. Recent studies have made use of statistical estimators of PAD that are applied to TLS point clouds subdivided into three-dimensional (3D) cubes, or voxels. Computation of such metrics under actual field conditions with point clouds containing several millions of returns is challenging. Moreover, rigorous assessment of the estimated PAD and effects of occlusions in forests remain unclear due to laborious, time-consuming, and inaccurate field measurements. In the present study, we present L-Vox, a software that computes PAD per voxel for TLS scans acquired in forest environments, which is based upon recent development of unbiased estimators derived from maximum likelihood. Two applications are presented. First, the software is evaluated for virtual forest plots, which are detailed 3D models of individual trees with corresponding simulated TLS scans, for which reference data are known. Second, L-Vox is applied to actual scans that were acquired in hardwood and coniferous plots in New Brunswick and Newfoundland, Canada. Both test cases were used to investigate the effects of occlusion and the uneven sampling in estimating PAD. The test cases were also used to assess the influence of voxel size and the number of scans per plot on PAD estimates. Our results showed strong correlations between the estimated PAD profile from L-Vox and simulated PAD for virtual forest plots, with a mean R² = 0.98 and a mean coefficient of variation (CV) = 15.6%. We demonstrated that comparing multi-scan to single scan TLS acquisitions in real forest plots substantially reduced signal occlusion, resulting in an increase up to 50% in PAD values. Effects of voxel size on PAD estimates greatly depended upon the relative size of foliar and woody elements, with an optimal size around 10 cm in coniferous plots. L-Vox proved to be an efficient and accurate tool for computing 3D distributions of PAD from TLS measurements in natural forest environments.
... Mobile LiDAR Systems were evaluated by [72] for the analysis of cultural heritage sites, based on a two-fold approach. A clustering phase consisted of computing the local curvature, defining the number of clusters, according to similar curvature values, and conducting a component analysis to reduce errors. ...
Article
Full-text available
As innovative technologies emerge, extensive research has been undertaken to develop new structural health monitoring procedures. The current methods, involving on-site visual inspections, have proven to be costly, time-consuming, labor-intensive, and highly subjective for assessing the safety and integrity of civil infrastructures. Mobile and stationary LiDAR (Light Detection and Ranging) devices have significant potential for damage detection, as the scans provide detailed geometric information about the structures being evaluated. This paper reviews the recent developments for LiDAR-based structural health monitoring, in particular, for detecting cracks, deformation, defects, or changes to structures over time. In this regard, mobile laser scanning (MLS) and terrestrial laser scanning (TLS), specific to structural health monitoring, were reviewed for a wide range of civil infrastructure systems, including bridges, roads and pavements, tunnels and arch structures, post-disaster reconnaissance, historical and heritage structures, roofs, and retaining walls. Finally, the existing limitations and future research directions of LiDAR technology for structural health monitoring are discussed in detail.
... Since 2010, the use of new satellite resources in heritage has diversified Luo et al. 2019b). The LiDAR flights enable the gathering of information from the earth's surface with a high spatial resolution in areas of dense vegetation (Lieskovský et al. 2018;Rodríguez-Gonzálvez et al. 2017;Trier et al. 2021); and synthetic aperture radars (SAR radar) work at very high wavelengths and capture data in almost any climatic and environmental condition (Chen et al. 2017;Iadanza et al. 2013;Lopez et al. 2020;Tapete & Cigna 2017b;Themistocleous & Danezis, 2020a). The active sensors carried by LiDAR and the SAR radar go through the clouds and capture information from the earth's surface in situations where a passive satellite is not capable, such as a rain event, the presence of smoke associated with a fire or volcanic activity. ...
Article
Full-text available
The nature of rammed earth fortifications and the environmental conditions where they are located determine the pathologies that these structures suffer in the presence of humidity sources and strong winds. The objective of this project is to revise the main mechanisms of deterioration of rammed earth fortifications and evaluate the use of remote detection as a tool to register environmental threats that affect their preservation. The selected images and satellite results offer information about precipitation, ground humidity, temperature, wind intensity and direction and the presence of particles in the wind. The use of statistical analysis methodologies for large volumes of satellite images makes it possible to acquire daily, monthly and yearly maximums, averages and minimums of these variables. The application of satellite resources GPM, SMAP, MODIS, Merra-2 and the statistical analysis of large volumes of images for preventive conservation in Andalusia has become useful to monitor the main threats that affect rammed earth fortifications on a global level: humidity, wind and temperature.
... Since 2010, the use of new satellite resources in heritage has diversified Luo et al. 2019b). The LiDAR flights enable the gathering of information from the earth's surface with a high spatial resolution in areas of dense vegetation (Lieskovský et al. 2018;Rodríguez-Gonzálvez et al. 2017;Trier et al. 2021); and synthetic aperture radars (SAR radar) work at very high wavelengths and capture data in almost any climatic and environmental condition (Chen et al. 2017;Iadanza et al. 2013;Lopez et al. 2020;Tapete & Cigna 2017b;Themistocleous & Danezis, 2020a). The active sensors carried by LiDAR and the SAR radar go through the clouds and capture information from the earth's surface in situations where a passive satellite is not capable, such as a rain event, the presence of smoke associated with a fire or volcanic activity. ...
Article
Ge-conservación es una publicación periódica del GEIIC, cuyo objetivo es contribuir al desarrollo científico, a la difusión y al intercambio de los conocimientos en materia de conservación y restauración del Patrimonio Cultural.
Article
Full-text available
Augmented reality is a mature technology that uses the real world as a substrate and extends it by overlaying computer-generated information. It has been applied to several domains. In particular, the technology was proven to be useful for the management and preservation of Cultural Heritage. This study provides an overview of the last decade of the use of augmented reality in cultural heritage through a detailed review of the scientific papers in the field. We analyzed the applications published on Scopus and Clarivate Web of Science databases over a period of 9 years (2012–2021). Bibliometric data consisted of 1201 documents, and their analysis was performed using various tools, including ScientoPy, VOS Viewer, and Microsoft Excel. The results revealed eight trending topics of applying augmented reality technology to cultural heritage: 3D reconstruction of cultural artifacts, digital heritage, virtual museums, user experience, education, tourism, intangible cultural heritage, and gamification. Each topic is discussed in detail in the article sections, providing insight into existing applications and research trends for each application field.
Article
Full-text available
This research proposes a detailed analysis of the potential of MMS (Mobile Mapping Systems), supported by SLAM (Simultaneous Localisation And Mapping) algorithms, performed on a multiscale test field in order to make a concrete contribution to the morphological study of cities. These systems, developed with the aim of acquiring a large number of points in a short time, are able to map the surrounding area and automatically localise themselves in real time in relation to a determined reference system. The analysed area, located in Venice, was divided into three different test fields characterised by typical elements potentially comparable to those of other urban realities. The data were acquired using the LiBackPack C50, Kaarta Stencil and Heron Lite systems and compared quantitatively and qualitatively with data obtained from more traditional surveying techniques. Specifically, the data obtained from TLS (Terrestrial Laser Scanning) surveys, supported by topographic measurements, were the most accurate basis on which to evaluate the accuracy and completeness of the three different MMS devices. The standard deviation values were initially analysed in the final 3D global models using the C2C (Cloud to Cloud) and C2M (Cloud to Mesh) distance calculation methods. Subsequently, the geometric differences were investigated through the extraction of horizontal profiles, and two more specific 2D analyses were carried out: the first inspecting the residual parameters calculated after the Helmert transformation from two sets of control points obtained from the profiles, followed by a local strain analysis. The study of the local deformation parameters allowed us to validate the results obtained and to identify the real limits of these survey instruments. The aim was to make a concrete contribution to the formalisation of an operative protocol for the morphological study of the city, exploiting the potential of these technologies to overcome the differences in scale and the gap between outdoor and indoor spaces.
Conference Paper
Full-text available
Visibility distance on the road pathway plays a significant role in road safety and in particular, has a clear impact on the choice of speed limits. Visibility distance is thus of importance for road engineers and authorities. While visibility distance criteria are routinely taken into account in road design, only a few systems exist for estimating it on existing road networks. Most existing systems comprise a target vehicle followed at a constant distance by an observer vehicle, which only allows to check if a given, fixed visibility distance is available. We propose two new approaches that allow estimating the maximum available visibility distance, involving only one vehicle and based on different sensor technologies, namely binocular stereovision and 3D range sensing (LIDAR). The first approach is based on the processing of two views taken by digital cameras onboard the diagnostic vehicle. The main stages of the process are: road segmentation, edge registration between the two views, road profile 3D reconstruction and finally, maximal road visibility distance estimation. The second approach involves the use of a Terrestrial LIDAR Mobile Mapping System. The triangulated 3D model of the road and its surroundings provided by the system is used to simulate targets at different distances, which allows estimating the maximum geometric visibility distance along the pathway. These approaches were developed in the context of the SARI-VIZIR PREDIT project. Both approaches are described, evaluated and compared. Their pros and cons with respect to vehicle following systems are also discussed.
Article
Full-text available
In acquisition and processing of Digital Elevation Models (DEMs) derived from airborne laser scanning (ALS) blunders and systematic errors occur. An assessment of the geometric quality of DEMs is necessary during production and before using the final DEM for an application. The vertical as well as the horizontal (planimetric) accuracy have to be assessed. Commonly agreed accuracy measures and procedures are necessary. The work of existing procedures and standards is analyzed. A new method to derive the absolute planimetric accuracy of ALS point clouds reliably and accurately is described. It is based on derivation of roof planes which are intersected to lines and points from the laser foot prints. The coordinates of generated points are then compared with points determined by aerial photogrammetry. From the differences accuracy measures are derived. Tests regarding the type of error distribution are carried out in order to apply either standard accuracy measures or robust accuracy measures. The results of a practical test are used to assess the planimetric accuracy by means of various accuracy measures. Suggestions for a new standard for the assessment of the absolute planimetric accuracy of airborne laserscanning are made.
Article
Full-text available
We present a comparison of point cloud generation and quality of data acquired by Zebedee (Zeb1) and Leica C10 devices which are used in the same building interior. Both sensor devices come with different practical and technical advantages. As it could be expected, these advantages come with some drawbacks. Therefore, depending on the requirements of the project, it is important to have a vision about what to expect from different sensors. In this paper, we provide a detailed analysis of the point clouds of the same room interior acquired from Zeb1 and Leica C10 sensors. First, it is visually assessed how different features appear in both the Zeb1 and Leica C10 point clouds. Next, a quantitative analysis is given by comparing local point density, local noise level and stability of local normals. Finally, a simple 3D room plan is extracted from both the Zeb1 and the Leica C10 point clouds and the lengths of constructed line segments connecting corners of the room are compared. The results show that Zeb1 is far superior in ease of data acquisition. No heavy handling, hardly no measurement planning and no point cloud registration is required from the operator. The resulting point cloud has a quality in the order of centimeters, which is fine for generating a 3D interior model of a building. Our results also clearly show that fine details of for example ornaments are invisible in the Zeb1 data. If point clouds with a quality in the order of millimeters are required, still a high-end laser scanner like the Leica C10 is required, in combination with a more sophisticated, time-consuming and elaborative data acquisition and processing approach.
Article
Full-text available
Backpack laser scanning systems have emerged recently enabling fast data collection and flexibility to make measurements also in areas that cannot be reached with, for example, vehicle-based laser scanners. Backpack laser scanning systems have been developed both for indoor and outdoor use. We have developed a quality analysis process in which the quality of backpack laser scanning data is evaluated in the forest environment. The reference data was collected with an unmanned aerial vehicle (UAV) laser scanning system. The workflow included noise filtering, division of data into smaller patches, ground point extraction, ground data decimation, and ICP registration. As a result, we managed to observe the misalignments of backpack laser scanning data for 97 patches each including data from circa 10 seconds period of time. This evaluation revealed initial average misalignments of 0.227 m, 0.073 and -0.083 in the easting, northing and elevation directions, respectively. Furthermore, backpack data was corrected according to the ICP registration results. Our correction algorithm utilized the time-based linear transformation of backpack laser scanning point clouds. After the correction of data, the ICP registration was run again. This revealed remaining misalignments between the corrected backpack laser scanning data and the original UAV data. We found average misalignments of 0.084, 0.020 and -0.005 meters in the easting, northing and elevation directions, respectively.
Article
Full-text available
Research on power line extraction technology using mobile laser point clouds has important practical significance on railway power lines patrol work. In this paper, we presents a new method for automatic extracting railway power line from MLS (Mobile Laser Scanning) data. Firstly, according to the spatial structure characteristics of power-line and trajectory, the significant data is segmented piecewise. Then, use the self-adaptive space region growing method to extract power lines parallel with rails. Finally use PCA (Principal Components Analysis) combine with information entropy theory method to judge a section of the power line whether is junction or not and which type of junction it belongs to. The least squares fitting algorithm is introduced to model the power line. An evaluation of the proposed method over a complicated railway point clouds acquired by a RIEGL VMX450 MLS system shows that the proposed method is promising.
Article
Full-text available
Rooted in the long tradition of Raman spectroscopy of cultural heritage materials, in this work we provide a personal perspective on recent applications and new frontiers in sampling modalities, data processing, and instrumentation.
Article
Mobile mapping data is widely used in various applications, what makes it especially important for data users to get a statistically verified quality statement on the geometric accuracy of the acquired point clouds or its processed products. The accuracy of point clouds can be divided into an absolute and a relative quality, where the absolute quality describes the position of the point cloud in a world coordinate system such as WGS84 or UTM, whereas the relative accuracy describes the accuracy within the point cloud itself. Furthermore, the quality of processed products such as segmented features depends on the global accuracy of the point cloud but mainly on the quality of the processing steps. Several data sources with different characteristics and quality can be thought of as potential reference data, such as cadastral maps, orthophoto, artificial control objects or terrestrial surveys using a total station. In this work a test field in a selected residential area was acquired as reference data in a terrestrial survey using a total station. In order to reach high accuracy the stationing of the total station was based on a newly made geodetic network with a local accuracy of less than 3 mm. The global position of the network was determined using a long time GNSS survey reaching an accuracy of 8 mm. Based on this geodetic network a 3D test field with facades and street profiles was measured with a total station, each point with a two-dimensional position and altitude. In addition, the surface of poles of street lights, traffic signs and trees was acquired using the scanning mode of the total station. Comparing this reference data to the acquired mobile mapping point clouds of several measurement campaigns a detailed quality statement on the accuracy of the point cloud data is made. Additionally, the advantages and disadvantages of the described reference data source concerning availability, cost, accuracy and applicability are discussed.