Content uploaded by Elisa Mariarosaria Farella
Author content
All content in this area was uploaded by Elisa Mariarosaria Farella on Jul 12, 2017
Content may be subject to copyright.
3D MAPPING OF UNDERGROUND ENVIRONMENTS WITH A HAND-HELD LASER
SCANNER
RILIEVO TRIDIMENSIONALE DI AMBIENTI IPOGEI CON UN SISTEMA A
SCANSIONE LASER PORTATILE
Elisa Mariarosaria Farella1, 2
1 Dept. of Architecture, Federico II University, Naples, Italy
2 3D Optical Metrology (3DOM) unit, Bruno Kessler Foundation (FBK), Trento, Italy - http://3dom.fbk.eu, elifarella@fbk.eu
KEY WORDS: Hand-held mobile mapping system, LiDAR, underground environment, 3D surveying, representation, visualization
PAROLE CHIAVE : Sistema mobile mapping portatile, LiDAR, ambiente sotterraneo, rilievo 3D, rappresentazione,
visualizzazione
ABSTRACT:
The development of several instruments and techniques for reality-based 3D survey provides for new effective and affordable
solutions for mapping underground environments. Terrestrial laser scanning (TLS) techniques demonstrated to be suitable for
recording complex surfaces in high resolution even in low ambient lightning conditions. TLS approaches allow to obtain millions of
3D points and very detailed representations of complex environments, but these normally required a very high number of stations.
This paper presents the investigation and deployment of a hand-held laser scanning system, the GeoSlam Zeb1, for the fast 3D
digitization of underground tunnels. This active hand-held device was employed in two different typologies of underground
structures: the Grotta di Seiano (Fig.1 a-b), a 800 m long monumental passage used as entrance of a roman villa in Posillipo
(Naples), and some military fortifications (Fig.1 c-d) built during the First World War (WWI) on the hills around Trento. In the first
case study, owing to the length of the gallery and the lack of well-defined geometric features on its wall, errors in the alignment were
expected. Consequently, the final alignment of the numerous acquired scans was verified. In the second part, the research is focused
on suitable procedures for the final three-dimensional representation and visualization of complex underground passages, i.e. the
military tunnels. Using an automatic classification procedure on the point-clouds, vegetation was removed and, through a manual
segmentation approach, the rooms were classified according to their specific functions. In the paper, the results are critical presented
and discussed.
RIASSUNTO:
Lo sviluppo di strumentazioni e tecniche di rilievo 3D reality-based offre nuove efficaci ed accessibili soluzioni per la modellazione
3D di ambienti sotterranei. Le tecniche di laser scanning terrestre (TLS) hanno dimostrato di essere ideali per rilevare superfici
complesse ad alta risoluzione geometrica anche con scarse condizioni di illuminazione degli ambienti. Le soluzioni TLS statiche
permettono di ottenere milioni di punti tridimensionali e rappresentazioni molto dettagliate di ambienti complessi, ma normalmente
richiedono un numero elevato di stazioni. Questo articolo presenta lo studio e l’utilizzo di un laser scanner portatile, lo Zeb1 della
Geoslam, per la digitalizzazione dinamica di ambienti ipogei. Questa strumentazione attiva è stata impiegata in due diverse strutture
sotterranee: la Grotta di Seiano (Fig.1 a-b), un lungo tunnel monumentale utilizzato come ingresso di una villa romana a Posillipo
(Napoli), e alcune fortificazioni militari (Fig.1 c-d) costruite durante la Prima Guerra Mondiale sulle colline intorno a Trento. Nel
primo caso, a causa della lunghezza della galleria e della mancanza di pareti con caratteristiche geometriche ben definite, erano attesi
errori durante l’allineamento delle scansioni, che hanno richiesto ulteriori verifiche. Nella seconda parte, la ricerca si è concentrata
sulle migliori procedure per la rappresentazione tridimensionale finale e la visualizzazione di complessi camminamenti ipogei, come
i tunnel militari. Dopo l’utilizzo di una procedura di classificazione automatica delle nuvole di punti per il filtraggio della
vegetazione, gli ambienti sono stati classificati considerando le loro specifiche funzioni attraverso una tecnica di segmentazione
manuale. L’articolo presenta in maniera critica la tecnologia di rilievo, la sua caratterizzazione e i risultati ottenuti.
a) b) c) d)
Figure 1. a-b) Grotta di Seiano – Archaeological Site of Pausilypon (Naples); c-d) WWI underground fortifications around Trento –
Monte Celva.
1. INTRODUCTION
1
Mapping underground passages, such as tunnels and caves, has
2
always required the development of particular procedures.
3
Indeed, such structures are often characterized by particularly
4
complex surfaces, hardly accessible parts and low ambient
5
lightning conditions.
6
In the past, several specific measuring instruments (special
7
compasses, measuring tapes, plumbing tools, etc.) were
8
developed for acquiring data in natural or artificial underground
9
environments. Their representation has also required the
10
introduction of ad-hoc symbols (Mattes, 2015). More recent
11
tacheometric methods, based on mining compass with
12
inclinometer, theodolites and total station, increased the level of
13
accuracy of documentation, although very time-consuming
14
approaches. Many issues are still open in this research:
15
How to record complex surfaces and huge tunnels with high
16
level of details in a reasonable time?
17
How to share and access large 3D datasets of such complex
18
environments?
19
How to appropriately represent underground structures in
20
these particular environmental conditions?
21
Nowadays the development of reality-based 3D surveying
22
instruments and methods provides an important support in this
23
field. In recent years, geomatics techniques have been diffusely
24
adopted especially for heritage documentation (Galeazzi et al.,
25
2014; Nocerino et al., 2014a; Remondino and Campana, 2014;
26
Remondino, 2011).
27
Nevertheless, in underground environments, some approaches
28
are more suitable than others. While image-based techniques
29
(Remondino and El-Hakim, 2006) are strongly limited by low
30
ambient lightning conditions and small passages, whereas range
31
sensors such as terrestrial laser scanners (TLS) allows high
32
resolution geometric surveys also in subterranean spaces. TLS
33
have been frequently used for three-dimensional acquisition of
34
man-made and natural tunnels (Beraldin et al., 2011; Caputo et
35
al., 2011; Roncat et al., 2011; Laurent, 2014; Nocerino et al.,
36
2014b; Wang et al., 2014; Gallay et al., 2015; McFarlane et al.,
37
2015; Rodríguez-Gonzálvez et al., 2015) for various reasons:
38
reasonable instrument weight and transportability, capability of
39
acquiring millions of points even on complex surfaces,
40
possibility of working in different light conditions, etc.
41
However, a great number of TLS stations and many working
42
days may be required for large environments, consequently
43
producing huge amount of data often difficult to be managed
44
and processed.
45
This paper presents an approach for fast 3D digitization of
46
underground passages. After a laboratory characterization and
47
investigation, two case studies, featuring different shapes,
48
dimensions and constructive elements, are critically discussed,
49
mainly focusing on the issues related to their 3D surveying and
50
final representation. The employed surveying instrument is the
51
hand-held laser scanner system GeoSlam Zeb1
52
(http://geoslam.com/). The device, suitable for both indoor and
53
outdoor applications (Zlot et al., 2013), was already employed
54
for mapping underground caves and mines (Zlot et al., 2014).
55
The device, which does not acquire neither colour nor intensity
56
information, was selected for its portability, ease of use (the
57
data acquisition is performed simply by natural or artificial
58
walking through the environment) and possibility to operate
59
without GNSS signal.
60
61
62
2. THE ZEB1 HAND-HELD ACTIVE DEVICE
63
The GeoSlam Zeb1 (Fig. 2) is a hand-held active device
64
equipped with a 2D infrared laser scanner profilometer and an
65
inertial measurement unit (IMU) mounted on a spring. The
66
UTM-30LX laser scanner emits pulses at a high frequency that
67
reflect off surfaces and return to the sensor where signals are
68
converted into range measurements based on the time of flight
69
principle. IMU measurements of angular velocities and linear
70
acceleration, combined with laser data, allow to estimate the
71
device’s trajectory. A three-axial magnetometer records
72
magnetic interferences common in underground environments.
73
Laser scanner and IMU are connected to a micro-
74
computer/battery unit which fits in a backpack. This very low-
75
weight instrument acquires up to 43,000 measurement points
76
per second, within a field of view of 270° and with a maximum
77
range of 30 m (15 m outdoor). The device has a range precision
78
up to 3 cm, conditioned by the distance, the incidence angle and
79
the surface reflectivity. The scanning field of view is increased
80
by the swinging mechanism due to a spring that allows to
81
generate three-dimensional profiles of the environment roughly
82
scanned every second. The instrument head can oscillate (or
83
nod) in the front-back/walking direction or side by side (i.e.
84
orthogonally to the advancing direction).
85
86
Figure 2. GeoSlam Zeb1 hand-held system.
87
3D data are acquired simply walking through the environments
88
and keeping the device in one hand. Every dataset has to be
89
acquired in an average range suggested of 20-30 minutes.
90
Once followed the desired path for data acquisition, the device
91
has to be placed on the ground for some seconds, so as the IMU
92
can indicate the micro-computer to stop the acquisition and to
93
terminate the logging process.
94
In order to merge all the acquired profiles (by estimating 3D
95
scanner positions and orientations), the device uses a
96
simultaneous localization and mapping (SLAM) algorithm. This
97
solution requires to observe the same features several times. The
98
Zeb1 device acquires the local scene roughly once per second.
99
Local views of the scenes, obtained through the swinging of the
100
instrument, contain position and normal direction of every
101
element recorded. By matching pairs of surface elements
102
acquired in different times, the trajectory is estimated through
103
the relation between surface geometries.
104
3D point clouds and followed trajectories are provided in
105
standard point cloud file formats, i.e. laz and ply.
106
107
108
3. PRELIMINARY INVESTIGATION
109
Before running the field campaign, the Zeb1 scanner was
110
investigated in a challenging indoor scenario with a twofold
111
aim: (i) to understand the potentialities and limitations of the
112
sensor and (ii) to identify the best acquisition procedure. The
113
selected environment, a horizontal corridor (X,Y plane) with
114
walls along the vertical direction (Z plane), was characterized
115
by smooth walls with poor geometric features and few elements
116
along its main direction (Fig. 3).
117
118
119
Figure 3. The corridor used for evaluating sensor limitations
120
and best acquisition’s procedure.
121
122
Four different acquisition strategies were tested:
123
1) “round trip” (i.e. the data collection starts and finishes in
124
the same place) by nodding the scanner front-back w.r.t.
125
the walking direction;
126
2) “one way” (i.e. the data collection starts at the beginning
127
of the corridor and finishes at the end) by nodding the
128
scanner front-back w.r.t. the walking direction;
129
3) “round trip” by nodding the scanner side by side, i.e.
130
orthogonally w.r.t. the walking direction;
131
4) “one way” by nodding the scanner side by side, i.e.
132
orthogonally w.r.t. the walking direction.
133
The corridor length was measured with a Leica distance-meter,
134
obtaining a reference length of 52.77 m with cm accuracy.
135
Table 1 shows the length and height variation measured on the
136
point clouds obtained through the four acquisition protocols.
137
138
Acquisition protocol
Corridor length
∆Z variation
1)
52.51 m
< 0.02 m
2)
52.79 m
≈ 0.60 m
3)
52.74 m
< 0.02 m
4)
52.78 m
≈ 0.06 m
Table 1. Corridor length and height variation obtained through
139
four different acquisition strategies.
140
141
The results achieved by nodding the scanner in front-back w.r.t.
142
the walking direction produced the worst results. Indeed
143
protocol 1) provided a significantly shorter corridor length. The
144
point cloud obtained with protocol 2) showed significant
145
bending in the vertical direction, due to SLAM divergence. The
146
results obtaining with protocol 3) (round trip and side by side
147
scanner nodding) resulted within the declared sensor accuracy
148
of 3 cm, in agreement with the scanning procedure suggested by
149
the vendor for featureless corridors.
150
In the two case studies hereafter presented a hybrid approach,
151
comprising front-back nodding in cooperative structures (i.e.
152
with evident geometric features) and side by side oscillation for
153
smooth surfaces, was adopted.
154
155
156
4. SURVEY OF THE GROTTA DI SEIANO, NAPLES
157
The so-called Grotta di Seiano (Soprintendenza Archeologica
158
di Napoli e Caserta, 1999) is a monumental tunnel leading to an
159
ancient maritime Roman villa named “Villa di Pausilypon”
160
which contained also two large theatres. The tunnel is almost
161
800 m long and it was excavated through the soft volcanic tuff
162
of the hill of Posillipo. It is not clear today if it was realized
163
during the earlier phase of construction of this huge residence
164
or during its transformation into an imperial villa. After
165
centuries of abandon and several collapses, in the 19th century,
166
the passage was reactivated, through the construction of many
167
masonry strengthening arches still today visible (Fig. 4a). The
168
tunnel, used during the Second World War as air-raid shelter,
169
was reopened only in the last years to the public access (as the
170
entire archaeological site). The tunnel is characterized by an
171
elongated shape, with alternation of sections showing geometric
172
elements (strengthening arches) and parts with flat and
173
featureless walls. The tunnel is naturally illuminated, beside at
174
the entrances, by three intermediate ventilation and lightning
175
openings. The other stretches are poorly illuminated with
176
artificial lamps.
177
The entire archaeological site (tunnel, theatres and villa) was
178
surveyed with multiple techniques and Virtual Reality (VR)
179
applications were developed with the aim of promoting and
180
sharing via web the virtual reconstruction of the site (Farella et
181
al., 2016). The 3D surveying of the tunnel, with its peculiar
182
geometry, would have required a huge number of TLS stations;
183
moreover, low ambient lightning and time constraint excluded a
184
complete photogrammetric survey. Consequently, the hand-
185
handle Zeb1 was considered a good surveying alternative and,
186
in order to assess the reliability and accuracy of the acquisition
187
result in a so critical environment, a topographic survey was
188
also realized.
189
190
4.1 Data acquisition
191
The Zeb1 allowed to acquire data in only one day of
192
acquisitions, covering the entire 800 m underground passage
193
and the area outside the two entrances. Considering a limit of
194
acquisition suggested of 20 minutes per scan and the possible
195
SLAM divergences, eight different dataset were acquired,
196
walking about 150 m for each acquisition.
197
Every section was scanned according to the “round trip”
198
acquisition protocol (Section 3), within the recommended
199
scanning time and maintaining a mean speed of 0.9 Km/h. The
200
sections featuring the strengthening arches were acquired with
201
the front-back nodding procedure, whereas in the parts
202
characterized by smooth and featureless walls the side by side
203
nodding technique was employed.
204
Moreover some white wooden circular targets of 30 cm
205
diameter (Fig. 4b) were designed and placed in several locations
206
inside the tunnel in the overlapping area between consecutive
207
scans (about 40 m). At least five targets were planned to be
208
visible in each scanned section. The target centres were
209
topographically surveyed with a total station (Section 4.2) to
210
verify the quality and reliability of scanning results.
211
a) b)
Figure 4. Tunnel’s section with the supporting arches (a);
wooden circular targets used for the topographic survey (b).
4.2 The topographic network
212
A TOPCN GPT 7001i total station (Table 2) was employed to
213
survey the tunnel. Constrained by the passage geometry, a
214
combination of triangulation, trilateration and open traverse was
215
used. 3D coordinates of 25 circular targets were also obtained
216
using the adjustment of the open-source software GAMA
217
(Čepek, 2002), whose average coordinate precision in space
218
from least squares adjustment was σxyz <6 mm.
219
Topcon GPT7001i
Range measurement accuracy (non-prism)
±5 mm
Range (non-prism)
1.5 to 250 m
Angle measurement accuracy (non-prism)
1”
Tilt correction
Dual axis
Compensating range
±4”
Table 2. Main technical specifications of the total stations used
220
for the topographic survey of the Grotta.
221
222
4. 3 Data processing and evaluation of 3D results
223
The 8 acquisitions (raw scans) were firstly processed by
224
GeoSlam using their SLAM process in the Cloud. Then the
225
derived 3D point clouds were further processed in
226
CloudCompare to align and merge them. After a first manual
227
transformation for a rough alignment between consecutive
228
scans, a finer registration with a traditional ICP method was
229
performed. Considering the previous registered point cloud as
230
reference, this operation was repeated for every adjacent
231
dataset. The maximum RMSE in the registration of consecutive
232
scans was 0.14 m. The final 3D point cloud merged with this
233
method was about 24 mil points (Fig. 5-6-7).
234
The aligned Zeb1 point cloud was compared with topographic
235
surveying data. Firstly every target visible in the point cloud
236
was isolated and, considering the noise present in the Zeb1 data,
237
the precise coordinates of their centres (measured also with the
238
total station) were estimated in PolyWorks through best fitting
239
procedures (Fig. 8).
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
Figure 5. Particular of the Grotta di Seano: section with masonry arches.
255
256
257
258
Figure 6. Central part of the grotta showing the alternation of smooth and featureless walls (1) and geometric elements (2).
259
260
261
262
263
264
Figure 7. Top and side views of the whole Grotta di Seiano showing the aligned point clouds in different colours.
265
266
267
Figure 8. Selection of targets in the Zeb1 point cloud (above).
268
Best fitting of circular planes and extraction of centres (below).
269
270
Using as reference the topographic data, a rigid similarity
271
transformation was performed using topographic and laser
272
scanner coordinates of the targets. The final RMSE of the
273
alignment resulted of 9.44 m. The probable reasons of this value
274
are: (i) an error in identifying the centres of the targets (due to
275
the low-res and noisy Zeb1 point clouds) and (ii) a block
276
deformation of the acquired scans. For these reasons, the same
277
procedure was repeated employing only the coordinates of
278
targets visible in each scan and verifying the achieved RMSE.
279
This procedure allowed to highlight the point clouds with
280
higher alignment error (Table 3 – central column).
281
282
DATASET
RMSE (m) of single
complete scan
RMSE (m) of
segmented scans
1
3.266
0.072
2
0.607
0.637
3
0.042
0.050
4
5.824
0.082
5
2.027
0.109
6
0.041
0.089
7
0.023
0.034
8
0.862
0.051
Table 3. RMSE of the similarity transformation between the
283
topographic points and the single Zeb1 acquisitions (central
284
column) and for each segmented point cloud (last column).
285
286
The registration results were further investigated as big errors
287
were still present for those point clouds containing long walls,
288
with no geometrical elements (no strengthening masonry arches
289
– datasets 1, 4, 5). The registration was then repeated following
290
a new procedure: each single scan was segmented in
291
correspondence of the circular targets and only the segments
292
showing a low transformation error with respect to the
293
topographic coordinates were retained. With this procedure
294
much better RMSE were obtained (Table 3 – last column). The
295
final mean RMSE of the complete 3D point cloud registered
296
with this procedure was then 0.13 m.
297
This final point cloud will be used for traditional two–
298
dimensional drawings (plans, sections, details, etc.) used in the
299
archaeological investigation for highlighting different roman
300
constructive techniques adopted for this construction.
301
302
303
5. SURVEY OF WWI FORTIFICATIONS IN MONTE
304
CELVA, TRENTO
305
Before the First World Word (WWI) outbreak, numerous
306
Austro-Hungarian fortifications (tunnels, trenches, forts, etc.)
307
were built on plateaus, hills and mountain tops around the city
308
of Trento for protecting and monitoring the territory (Nocerino
309
et al., 2014). Indeed the Trentino – Alto Adige region was part
310
of the Austro-Hungarian Empire until 1918: it represented the
311
hot southern border with the Italian kingdom and, consequently,
312
it was disseminated of many military fortifications. The shape
313
and dimension of the built military fortifications were generally
314
planned a-priori but many structures (e.g. tunnels and trenches)
315
were normally decided directly on the field. The fortifications
316
were built in concrete, often hand-carved in the stone and
317
organized with main bodies, casemates and several connecting
318
galleries.
319
Some of these military structures are today still partly visible in
320
the Trento’s region, like on Monte Celva. The 996 m height
321
mountain represented a strategic place for the defence of Trento
322
and belonged to the so-called Fortress of Trento (Marzi and
323
Borsato, 2000; http://trentocittafortezza.fbk.eu). Some of the
324
underground military constructions present in Monte Celva,
325
(galleries, batteries in caves, shelves for ammunitions, rifle
326
emplacements, etc.), along with an outdoor (not underground)
327
defensive system (trench) connected to the artillery batteries in
328
the cave (Fig. 9), were surveyed with the Zeb1 hand-held
329
system. Monte Celva presents a huge and complex underground
330
network, with some parts difficult to be reached and others only
331
partially cleared from rubbles due to structural collapses.
332
a) b)
333
334
c)
335
336
Figure 9. The excavated trench connected to the tunnels and
337
artillery batteries in caves (a, b) and a rifle emplacement (c).
338
339
Therefore an active hand-held surveying device was the most
340
appropriate instrument. In addition to the surveying issues, the
341
(3D) representation of such complex underground systems
342
poses not trivial problems. Therefore, the last part of this case
343
study focuses on the identification of suitable procedures and
344
methods for the final representation and visualization of the
345
digitized tunnels.
346
5.1 Data acquisition
347
Five different areas were surveyed with the Zeb1,
348
corresponding to several fortifications that occupy the low and
349
middle part of Monte Celva. Each military structure was
350
surveyed in about 30 minutes, covering an average path of 200
351
m. Every fortification was acquired following the “round trip”
352
surveying approach (Section 3) with a mean speed of 0.8 Km/h
353
and alternating the front-back and side by side nodding
354
procedure. The entrances of the military structures were also
355
digitized, thus collecting many elements of the external natural
356
environment like trees, vegetation, rocks, etc.
357
After the raw data processing, the Zeb1 point clouds of each
358
area were further processed for data cleaning and classification
359
(Section 5.2), global alignment and final representation (Section
360
5.3).
361
362
5.2 Point cloud classification
363
The exterior and surrounding parts of tunnels and trenches,
364
although fundamental to co-register the Zeb1 data with the
365
LiDAR landscape model, need to be removed for better
366
understanding and representation. Instead of manually cleaning
367
the large and complex point clouds, an automated procedure
368
was run. The Canupo classification algorithm (Brodu and
369
Lague, 2012) implemented in CloudCompare was thus used to
370
separate natural and man-made structures (Fig. 10). The Canupo
371
plug-in allows to create own classes as well as to use existing
372
classifiers for segmenting point clouds into subsets (e.g.
373
vegetation, ground). This supervised method is based on 3D
374
geometrical properties of the point cloud across multiple scales
375
and, employing a probabilistic approach, the points with high
376
uncertainty can be removed from the wrong class. For creating a
377
new class, a sample of points representing each class have to be
378
manually identified. For the WWI structures and underground
379
passages, both new and available classifiers were tested:
380
LongRange Classifier: for brush and trees scanned at
381
intermediate resolution (down to 5-10 cm point spacing);
382
RangiCliff Classifier: for brush and trees scanned at high
383
resolution (down to 1-2 cm point spacing).
384
The RangiCliff Classifier provided the best classification results
385
and was then adopted for all the acquired point clouds. The
386
separation between man-made structures and natural elements
387
was useful to better represent and map the surveyed military
388
structures.
389
390
391
392
393
394
395
Figure 10. Point cloud classification with Canupo plug-in. RangiCliff classifier (red: man-made military structures; grey: vegetation
396
samples).
397
398
399
5.2 The 3D representation of WWI fortifications
400
The segmented point clouds were aligned among them and also
401
with the LiDAR-based terrain model of the area. Then they
402
403
404
405
were manually segmented in CloudCompare, highlighting the
406
functions of the different spaces and their extensions and curved
407
shape inside the mountain (Fig. 11-18).
408
Figure 11. WWI fortifications in Monte Celva.
Figure 12. Four registered point clouds: second battery or “100 steps stairway” (1), third battery (2), trench (3) and fourth battery (4).
Figure 13. Second battery in cave, the so-called “100 step
stairways”: entrance (1), riflemen emplacements (2) connecting
well with the upper trench (3).
Figure 14. Third battery in cave: connecting underground gallery (1), artillery
battery in cave (2), barracks (3), connection with trench (4).
Figure 15. The trench structure: gun emplacement (1),
casemate (2), foxhole defensive position (3), connection with
third battery (4).
Figure 16. Fourth battery in cave: gun emplacements (1), connecting
underground gallery (2), underground casemate (3).
409
410
Figure 17. First battery in the cave in the lower part of Monte Celva (1: entrance of the structure; 2: lower casemates; 3, 5: connecting
galleries; 4: riflemen emplacements; 6: guard post, 7: artillery battery in cave, 8: ammunition depot).
Figure 18. Views of some WWI underground passages in Monte Celva integrated into the 3D terrain model of the area.
411
412
413
6. LESSON LEARNT AND CONCLUSIONS
414
The paper described the investigation and use of a hand-held
415
laser scanner system to survey and map several underground
416
and complex heritage structures. The device (GeoSlam Zeb1) is
417
able to map in short time large and complex structures, although
418
on-site and post-processing procedures must be implemented to
419
verify its accuracy and reliability, especially for long and
420
featureless structures. Moreover, when textural information is
421
important, an additional acquisition (for example, through a
422
photogrammetric survey) is necessary. In the Grotta di Seiano
423
survey, the comparison between the Zeb1 3D data and classic
424
surveying showed a maximum RMS error of 10 m. This was
425
mainly due to a block deformation of the scans acquired,
426
especially in the segments with poor morphological features.
427
From the lab investigations and field experiences, the nodding
428
speed and direction, along with the walking speed were the
429
most critical factors while using the Zeb1 system. In this work it
430
was verified the importance of keeping a constant speed of
431
walking and a stable oscillation of the sensor to guarantee better
432
results. Moreover, an overlapping area between consecutive
433
data acquisition at least of 20%-30% is advisable. In case of
434
structures with alternation of featureless parts and rich
435
geometric elements, a hybrid acquisition approach (side by side
436
and front-back nodding) can provide most reliable results.
437
Moreover, a “round trip” approach (turning back in every scans
438
to the starting point) is essential to reduce errors and
439
deformations. The new version of the sensor offers an
440
automatically oscillating head in order to reduce user-dependent
441
results and facilitate the acquisition, avoiding undesirable and
442
erroneous motion or divergences in the acquired point clouds.
443
Investigations with this new head are planned.
444
ACKNOWLEDGMENTS
445
446
The authors are thankful to the Soprintentenza per i Beni
447
Archeologici di Napoli and Mesa srl (www.mesasrl.it) for their
448
valuable support with the GeoSlam Zeb1 sensor.
449
450
451
REFERENCES
452
453
Beraldin, J.-A., Picard, M., Bandiera, A., Valzano, V., Negro,
454
N., 2011: Best Practices for the 3D Documentation of the
455
Grotta dei Cervi of Porto Badisco, Italy. Proc. SPIE, Vol. 7864,
456
pp. 78640J-78640J-15.
457
458
Brodu N., Lague D., 2012. 3D Terrestrial LiDAR data
459
classification of complex natural scenes using a multi-scale
460
dimensionality criterion: applications in geomorphology In: The
461
International Archives of the Photogrammetry, Remote Sensing
462
and Spatial Information Sciences, Vol. 68, pp.121-134.
463
464
Caputo P., Farella E.M., Mauriello F., 2011. La Crypta Romana
465
di Cuma. Spazi e Culture del Mediterraneo. PRIN 2009-2011.
466
La scuola di Pitagora editrice.
467
468
Čepek, A., 2002. The GNU Gama project-adjustment of
469
geodetic networks. Acta Polytechnica, Vol. 42(3).
470
471
CloudCompare, 2015. EDF R&D, Telecom ParisTech.
472
http://www.cloudcompare.org/
473
474
Farella E.M., F. Menna, E. Nocerino, D. Morabito, F.
475
Remondino, M. Campi, 2016. Knowledge and valorization of
476
historical sites through 3D documentation and modeling. In:
477
The International Archives of the Photogrammetry, Remote
478
Sensing and Spatial Information Sciences, Vol. XLI, in press.
479
480
Galeazzi F, Moyes, H., Aldenderfer, M 2014. 3D
481
Documentation in Archaeology: Recording Las Cuevas Site,
482
Chiquibul Reserve, Belize. In: Archaeology in the Digital Era
483
Volume II, Amsterdam University Press, Amsterdam, pp. 350-
484
362.
485
Gallay M., Kaňuk J., Hofierka J., Hochmuth Z. & Meneely J.,
486
2015 – Mapping and geomorphometric analysis of 3-D cave
487
surfaces: a case study of the Domica Cave, Slovakia. In:
488
Zwoliński Z., Jasiewicz J., Mitasova H. & Hengl T. (Eds.),
489
Geomorphometry 2105, pp. 69-73 (available at:
490
http://geomorphometry.org/2015)
491
492
Laurent J., 2014. Use of 3D scanning technology for automated
493
inspection of tunnels. Proc. World 309 Tunnel Congress 2014.
494
495
Marzi, C., Borsato, T., 2000: Trento città fortezza. Le opere
496
militari in muratura ed in caverna. Persico ed., 361 p.
497
498
Mattes, J., 2015. Underground fieldwork – A cultural and social
499
history of cave cartography and surveying instruments in the
500
19th and at the beginning of the 20th century. In: International
501
Journal of Speleology, Vol. 44, pp. 251-266.
502
503
McFarlane D.A., Roberts W., Buchroithner M., Van Rentergem
504
G., Lundberg J. & Hautz S., 2015. Terrestrial LiDAR-based
505
automated counting of swiftlet nests in the caves of Gomantong,
506
Sabah, Borneo. In: International Journal of Speleology, Vol.
507
44(2), pp. 191-195.
508
509
Nocerino E., Fiorillo F., Minto S., Menna F., Remondino F.,
510
2014. A non-conventional procedure for the 3D modeling of
511
WWI forts. In: The International Archives of the
512
Photogrammetry, Remote Sensing and Spatial Information
513
Sciences , Vol. XL-5, pp. 457-464.
514
Remondino F., Campana S., 2014. 3D Recording and
515
Modelling in Archaeology and Cultural Heritage. Theories and
516
best practices. Archaeopress.
517
518
Remondino F., 2011. Heritage Recording and 3D Modelling
519
with Photogrammetry and Laser Scanning. In: Remote Sensing,
520
Vol. 3(6), pp. 1104-1138.
521
Remondino, F., El-Hakim, S., 2006. Image-based 3D
522
modelling: a review. The Photogrammetric Record, Vol.
523
21(115), pp. 269-291.
524
525
Rodríguez-Gonzálvez, P., Nocerino, E., Menna, F., Minto, S.,
526
Remondino, F., 2015. 3D surveying & modeling of
527
underground passages in WWI fortifications. The International
528
Archives of Photogrammetry, Remote Sensing and Spatial
529
Information Sciences, Vol. 40(5), pp. 17-24.
530
531
Roncat A., Dublyansky Y., Spotl, C., Dorninger P., 2011 –
532
Full-3D surveying of caves: A case study of Marchenhohle
533
(Austria). Proc. IAMG2011 Conference, pp. 1393-1403.
534
535
Soprintendenza Archeologica di Napoli e Caserta, La grotta di
536
Seiano e il Parco Archeologico del Pausilypon, Electa Napoli,
537
1999.
538
539
Wang W., Weisen Z., Huang L., Vimarlund V., Wang Z., 2014.
540
Applications of terrestrial laser scanning for 281 tunnels: A
541
review. In: Journal of Traffic and Transportation Engineering
542
Vol. 1, pp. 325-337.
543
544
Zlot, R., Bosse, M., 2014. Three-dimensional mobile mapping
545
of caves. In: Journal of Cave and Karst Studies, Vol. 76(3), pp.
546
191–206.
547
548
Zlot, R., Bosse, M., Greenop, K., Jarzab, Z., Juckes, E., Robert,
549
J., 2013. Efficiently capturing of large, complex Cultural
550
Heritage sites with a handheld 3D mobile laser mapping system.
551
In: Journal of Cultural Heritage, Vol. 15(6), pp. 670-678.
552
553
554