International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198

Contents lists available at ScienceDirect

International Journal of
Rock Mechanics & Mining Sciences
journal homepage: www.elsevier.com/locate/ijrmms

Semi-automatic extraction of rock mass structural data from high resolution

LIDAR point clouds
Giovanni Gigli a,n, Nicola Casagli b,1
a
Department of Earth Sciences, University of Florence, Largo Enrico Fermi 2, Arcetri, 50125, Florence, Italy
b
Department of Earth Sciences, University of Florence, Via Giorgio La Pira 4, 50121, Florence, Italy

a r t i c l e in f o abstract

Article history: In this paper a Matlab tool called DiAna (Discontinuity Analysis), for the 2D and 3D geo-structural analysis
Received 8 October 2009 of rock mass discontinuities on high resolution laser scanning data is presented.
Received in revised form The proposed approach is able to semi-automatically retrieve some relevant rock mass parameters,
26 June 2010
namely orientation, number of sets, spacing/frequency (and derived RQD), persistence, block size and
Accepted 27 November 2010
scale dependent roughness, by analyzing high resolution point clouds acquired from terrestrial or aerial
Available online 5 January 2011
laser scanners.
Keywords: In addition, with a specific DiAna option called filterveg, we are able to remove vegetation or other
Geomechanical survey disturbing objects from the point cloud, which is one of the main problems in LIDAR data processing.
Automatic
Some examples of the proposed method have demonstrated its ability to investigate rock masses
Matlab
characterized by irregular block shapes, and suggest applications in the field of engineering geology and
Laser scanning
Point cloud emergency management, when it is often advisable to minimize survey time in dangerous environments
DiAna and, in the same time, it is necessary to gather all the required information as fast as possible.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction Nowadays, several techniques are available for retrieving high

resolution 3D representations of land surface, such as digital
When we face engineering geology problems in rock, it is photogrammetry [3,4], laser scanning (terrestrial and aerial)
fundamental to reconstruct the 3D geometry and the structural [5,6] and SAR interferometry [7].
setting of the rock masses, sometimes in inaccessible areas. An In addition, the increased computational performance of per-
accurate description of the geometrical and mechanical properties sonal computers allows us to process large amounts of data in a
of the material is specifically required, as the overall mechanical relatively short time.
behavior of a rock mass depends on its structure, on the char- The advantage of employing remote and high resolution
acteristics of discontinuities and on the properties of intact rock. surveying techniques for geomechanical purposes is based on
Traditional geomechanical surveys are performed in situ, either the capability of performing both large scale [8,9] and small scale
in one dimension (scanline method) or two dimensions (window [10,11] analyses and to rapidly obtain information on inaccessible
method), and require direct access to the rock face for the collection rock exposures.
of the relevant parameters [1]. Sometimes, the features of interest can be very large [12], and
ISRM [2] selected the following ten parameters for the quanti- they could actually remain unnoticed if only a small scale field
tative description of discontinuities in rock masses: orientation, survey is performed. On the other side, the observation of small
spacing, persistence, roughness, wall strength, aperture, filling, details, such as discontinuity planes or traces and surface rough-
seepage, number of sets, and block size. ness, is a key element for the geomechanical characterization of the
For practical and safety reasons, traditional geomechanical rock mass.
surveys are often carried out on limited sectors of the rock mass, In order to perform correct analyses from a statistical point of
and usually they do not provide data for a complete reconstruction view, we need, therefore, to investigate a portion of the rock face as
of the full variability of a rock mass. wide as possible.
The capability of capturing small details depends primarily on
the resolution and on the accuracy of the survey method.
n
The main product of a long range laser scanning survey [13] is a high
Corresponding author. Fax: + 39 055 2055317.
E-mail addresses: giovanni.gigli@unifi.it (G. Gigli),
resolution point cloud, obtained by measuring with high accuracy
nicola.casagli@unifi.it (N. Casagli). (millimetric or centimetric) the distance of a mesh of points on the
1
Fax: +39 055 2756296. object, following a regular pattern with polar coordinates. The high

1365-1609/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijrmms.2010.11.009
188 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198

acquisition rate (up to hundreds of thousands of points/s) allows to The creation of a surface from a point cloud leads to a
immediately obtain the detailed 3D shape of the object. simplification and a first segmentation of the original data in
Laser scanning data can be processed by true coloring point elementary polygons (usually triangles), whose spatial location
clouds from high resolution optical digital images, or by triangulat- and orientation can be calculated. It is therefore possible to group
ing points in order to create Digital Surface Models (DSM). neighbor polygons with similar orientation, and to individuate the
One of the main tasks when we have to interpret the acquired extent of planar features within the investigated object [17,18,23].
data is the vegetation removal [14]. However, the capability of reconstructing discontinuity surfaces
Two main different levels of automation can be conceived, to depends primarily on the quality of the triangulation process and
extract the most relevant rock mass geomechanical characteristics, its resolution. Small features can be neglected, and complex shapes
hidden in the point cloud. could lead to incorrect and distorted polygonal surfaces.
On the other side, working on raw point cloud data gives the
 Manual: by inspecting the point cloud or the derived surface, advantage of keeping all the initial information, and the disadvan-
fitting local planes, taking measurements, drawing polylines of tage of a larger amount of input data to be processed, with a
interest, etc. [15]. This procedure has, however, a non-systema- consequently longer computational time.
tic character, is time-consuming and tends to neglect the This approach has been adopted by Jaboyedoff et al. [20] for the
smallest features. It is a subjective or biased analysis, as only assessment of the local orientation of ground surface, and for
those discontinuities, which appear to be important are inves- vegetation detection and removal, by means of eigenvalues.
tigated. The success of this approach depends on the quality of Roncella et al. [16], Voyat et al. [19] and Ferrero et al. [22]
digital data and on the skill and experience of the geologist. applied a procedure based on the Random Sample Consensus
 Automatic/semi-automatic: by selecting a specific algorithm for (RANSAC) algorithm [37] for the segmentation of point clouds into
the segmentation of the original data in clusters of points subsets, each made of points belonging to the same discontinuity
belonging to the same discontinuity. This can be defined as surface.
an objective or random analysis, since all detectable disconti- Slob et al. [21] proposed a method for the direct segmentation of
nuities within the surveyed area are sampled. Since raw data point clouds, by verifying if most of the points around a seed point
can contain up to tens of millions of points, the adopted lie close to the same flat plane. Should this happen, the operation is
algorithm should be optimized to make computational time iterated by selecting other neighboring points and verifying their
acceptable. In the author’s opinion, it is important to use both position with respect to the plane found before, in order to
approaches, because they can complement each other. individuate the extent of the discontinuity surface.

In this paper we present a Matlab tool called DiAna (Disconti-

nuity Analysis), for the 2D and 3D semi-automatic extraction of
rock mass structural information from high resolution point clouds 3. Proposed method
obtained from a terrestrial laser scanner.
In particular, six of the ten ISRM [2] parameters can be evaluated In this paper, for the 3D semi-automatic analysis of large point
(orientation, number of sets, spacing/frequency, persistence, block clouds finalized to the extraction of the geomechanical character-
size and scale dependent roughness) and a specific option for istics of rock masses, we propose a new approach based on the
vegetation removal (filterveg) is implemented. definition of least squares fitting planes on clusters of points
After a review of the state of the art for indirect rock face extracted by moving a sampling cube on the point cloud. By
characterization, both 2D and 3D versions of the proposed tool are selecting the cube dimension and a standard deviation threshold,
described, and a field application is presented to validate the semi- the adopted method has demonstrated its validity to investigate
automatically obtained results with traditional field survey data. even rock masses characterized by very irregular block shapes.
The advantage of using this approach lies in its capability to
investigate all the geomechanical parameters that do not require
2. Methods for indirect rock face characterization direct access to the rock mass, thus, making this a more complete
analysis in connection with the existing methods described in the
During the last years many authors have been working on the previous section. The output ISRM [2] parameters are: orientation,
extraction of 3D rock mass properties from remotely acquired high number of sets, spacing/frequency (and derived RQD), persistence,
resolution data, mainly digital photogrammetry and LIDAR [16–25]. block size and scale dependent roughness.
These efforts led to the development of specific tools or soft- With the aim of keeping all the original information, the input
ware, written either for personal research or commercial purposes: data of the proposed algorithm are unorganized point clouds,
i.e. Vulcan [26]; Jointmetrix3D [27]; Surpac [28]; Sirovision [29]; produced by merging data acquired from different positions or
3DM Analyst [30]; Split-FX [31]; 3DGeomec [32]; Coltop3D [20]. even diverse techniques, in order to avoid shadow areas and to
Since laser scanning is a more recent technique than digital obtain a regular distribution of points. The heterogeneity of point
photogrammetry, the software currently available for the latter clouds consequently is one of the main limitations of the employ-
application is more advanced [33] and there are many commercial ment of ground-based laser scanning in rock slope characteriza-
packages that include geological and structural mapping facilities tion, as outlined by Sturzenegger et al. [38].
[34–36]. Natural and man-made slopes show, however, very different
With the aim of semi-automatically extracting the geometrical morphology, which results in different processing conceptions.
characteristics of a rock mass from Lidar data, two basic approaches Often, especially in case of man-made excavations, the rock
can be pursued, by working on point clouds or derived surfaces. slope face is planar. In these circumstances, only discontinuity
First of all, since our primary object is the estimation of traces are detectable from the point cloud, provided the available
discontinuity plane orientations and the comparison with field data are associated with a digital image and their resolution is high
measurements, the raw data have to be georeferenced on a global enough. Discontinuity traces can be represented by 2D polylines on
reference system, in order to obtain true orientations of the the slope fitting plane, and a 2D quantitative analysis can be
surfaces of interest with respect to the horizontal (dip) and North performed, based on their length and their pitch angle on the
(dip direction). reference plane, the orientation of which is known.
 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198 189

On the other side, rock faces with rugged shape can be investigated
by inspecting the discontinuity surfaces exposed on the slope. Such
3D approach requires the extraction of clusters of points belonging to
the same discontinuity plane from the point cloud; subsequently,
a spatial analysis for the quantitative description of discontinuities
within the rock mass has to be performed.
Both cases have been dealt with the proposed method, by
applying two different Matlab [39] tools, called DiAna3D and
DiAna2D and fully described in the following sections.
Authors are presently working at the integration of the 2D and
3D methods in a single tool, after which their intent is to distribute
it for free.

3.1. DiAna3D

The three dimensional approach is based on the selection of a

sub-set of the point cloud (red points in Fig. 1) contained within a Fig. 2. Example of box analysis with a 2 m reference cube. A sub-set of the point
searching cube (red cube in Fig. 1). cloud contained within each cube is recursively selected for the definition of the best
The best fitting plane (blue grid in Fig. 1) according to the least fitting plane.

squares method is then found for the cubic selection, by applying

the singular value decomposition matlab function (svd) [40].
where a and b are the plane dip/direction and dip, respectively;
The plane is defined by its direction cosines (l, m, n) and has a
l, m and n are the direction cosines of the plane; Q is a constant,
general equation of the type
which assumes the following values: Q¼01 if l 40 and m 40;
ax þby þ cz þ d ¼ 0 ð1Þ Q¼3601 if l40 and mo0; Q¼1801 if l o0 and mo0 or if l o0 and
The standard deviation (s) of the selected points concerning the m40.
best fitting plane is then computed and a threshold value (st) is The adopted procedure allows both to isolate those points
chosen by the user, based on the searching cube dimension, the belonging to a planar surface and to make local face orientation
resolution of the point cloud and the large scale roughness measures, just like moving with a compass and clinometer on the
characteristics of the rock slope. rock face, taking measurements on discontinuity planes.
If s o st a cluster is formed and the selected points are extracted Different searching cube dimensions can be selected; the
from the point cloud. A minimum number of points within the smaller is the cube dimension, the smaller is the feature that can
selection cube can be chosen, to avoid the formation of small and be observed on the point cloud (Fig. 3).
unrepresentative clusters. All cluster orientations are then plotted on the stereographic
In order to cover the whole point cloud, the searching cube is projection. Fig. 4 shows the stereoplot of the clusters extracted with
moved along the geographic axes according to a regular pattern a reference cube dimension of 0.1 m (Fig. 3d), with a total of
(box analysis; Fig. 2), or can be centered at each single point of the 18,757 poles.
point cloud (discrete analysis). The main discontinuity sets can therefore be extracted from the
Once a valid cluster has been identified, the associated plane point cloud (Fig. 5), as all the clusters belonging to the same set are
orientation is found by applying the following equations [1]: assigned a common ID. In order to directly control the set
m recognition process, and to have an immediate validation with
a ¼ arctan þQ ð2Þ possible field data, we chose to perform this operation manually;
l
for this reason the whole 3D approach can be defined as semi-
  automatic.
n
b ¼ arctan pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð3Þ The next step is aimed at merging all the clusters belonging to
l2 þm2
the same discontinuity plane. This is done by comparing the
orientation of all neighboring clusters with the set they belong
to: if two clusters with similar orientation (same set) have been
extracted from adjoining cubes, they are supposed to belong to the
same discontinuity surface.
Based on this assumption a small enough reference cube
dimension has to be chosen to avoid the method to group together
two closely spaced parallel discontinuities.
This operation can be iterated by considering all the clusters the
centroids of which fall within a fixed distance each other, to obtain
the maximum extent of single discontinuity faces.
In order to fill holes, isolated clusters with unassigned ID are
attributed to a discontinuity plane if they are surrounded by
clusters assigned to the same discontinuity surface.
Fig. 6 reports the results of the merging process for the clusters
extracted with a 0.1 m reference dimension. A total of 79 disconti-
nuity surfaces have been identified by merging 18,757 clusters
(Figs. 3d and 4).
Fig. 1. Cubic selection (red points) and best fitting plane (blue grid) on high
resolution point cloud. The dimension of the red selection cube is 1 m. (For
By applying this procedure it is possible to extract even minor
interpretation of the references to color in this figure legend, the reader is referred discontinuity planes. Fig. 6 demonstrates that decimetric features
to the web version of this article.) can be observed, if the searching cube dimension is small enough.
190 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198

Fig. 3. Influence of the searching cube dimension on the total number of valid points (blue points) for a standard deviation threshold of 0.025 m. (a) 1 m selection cube;
(b) 0.5 m selection cube; (c) 0.25 m selection cube and (d) 0.1 m selection cube. The areal extent of the point cloud is about 16  10 m. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

Fig. 4. Definition of the rock mass discontinuity sets on the stereographic projection. Contour lines have been obtained by plotting the orientation of valid clusters with a
reference dimension of 0.1 m.

When all the points belonging to a discontinuity plane have discontinuity persistence, which implies the areal extent of the
been extracted, the dip and dip direction of the best fitting plane are surface [2]. It is important to emphasize that this method can
computed, and the discontinuity bounding polygon is found (Fig. 7) produce an underestimation of the discontinuity persistence, as the
by applying a convex hull algorithm [41]. exposed surface is often a sub-set of the entire fracture.
Minimum and maximum discontinuity persistence can be, For each discontinuity a cylinder is drawn, with the axis
therefore, calculated for each set, based on the bounding polygon perpendicular to the plane passing from the centroid and the base
dimensions. This procedure fits well with the original definition of containing the surface points (Fig. 8).
 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198 191

By counting the number of discontinuities belonging to the possible to assess the mean, minimum and maximum joint spacing,
same set intersected by the cylinder (hatched polygons, Fig. 8) and and the associated frequency of each set. If a set is defined to be
measuring the maximum distance between them (Fig. 8d) it is infinitely persistent (i.e. bedding planes), its spacing is calculated
by considering all the discontinuities belonging to that set,
independently from their position.
Block dimensions are evaluated by using the correlation pro-
posed by Palmstrom [42,43]

Vb ¼ bx Jv3 ð4Þ

where Jv is the Volumetric Joint Count, defined by Palmstrom

[44] and b is the block shape factor, which may be estimated by
b ¼ 20 þ 7 a3 =a1 ð5Þ
where a1 and a3 are the shortest and longest dimensions of the
block, respectively.
Discontinuity roughness plays a very important role in the
mechanical behavior of the rock mass, and the shear strength
criteria for rock discontinuities include it among the basic para-
meters [45–47].
Fig. 5. Discontinuity set representation on the point cloud (Jn1: blue points; Jn2: red
The most practical method for estimating the roughness of a
points; BG: green points). (For interpretation of the references to color in this figure discontinuity plane is by visual comparison of sampled roughness
legend, the reader is referred to the web version of this article.) profiles with standard profiles published by Barton and Choubey [48].

Fig. 6. Discontinuity plane detection from cluster merging.

Fig. 7. Boundary polygons of identified discontinuities (Fig. 6).

192 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198

1.80
1.60
Plane1
1.40
Plane2
1.20

Theta ( °)
1.00
0.80
0.60
0.40
0.20
0.00
10 100 1000 10000 100000
n.points

Fig. 9. Effect of instrument accuracy on the surface roughness estimation. Number

of points within the selection vs. angular deviation between the original plane and
the best fitting plane after the introduction of point range error.

Fig. 8. Calculation of the true spacing and frequency of discontinuity sets.

Continuous polygons: set 1; dashed polygons: set 2.
attenuated if the angle between the line of sight and the normal
to the plane is high (Plane 2; Fig. 9).
Fig. 10 reports the stereographic projection of the roughness
Drawing roughness profiles of discontinuities directly from characteristics for different reference dimensions, calculated for
lidar data is a very difficult issue, due to limitations in instrument the black central discontinuity in Fig. 6. We can observe how the
accuracy. pole scattering decreases with increase in reference cube dimen-
Moreover, it has been observed that discontinuity roughness is sion (left). By drawing the contour lines of the poles for each
characterized by a very marked scale effect [49]. reference dimension (right), we are able to quantitatively assess
To overcome this problem a 3D approach can be pursued, as the surface roughness at different scales. For the smallest features
suggested by ISRM [2], by sampling the local surface orientation (0.1, 0.2 m) the 95th percentile lines have been drawn to compen-
with a compass and disc clinometers with different diameters. This sate the scattering due to the small number of points contained
procedure requires however direct accessibility to the disconti- within the selections.
nuity plane and is quite time consuming, as a minimum number of Regarding the remaining four ISRM parameters (aperture,
250 measures is suggested for the operation to be valid. seepage, wall strength and filling), their estimation requires direct
The proposed algorithm fits in well with these concepts and access to the rock face and cannot be evaluated from conventional
allows us to rapidly perform quantitative measures of the rough- high resolution point clouds. Only centimetric joint apertures could
ness of the main discontinuity surfaces at various scales. be measured, based on the resolution and accuracy of original data;
A searching cube with different dimensions (0.1, 0.2, 0.4, 1, 2 m however, this capability has not been yet implemented within the
and maximum surface persistence) is moved along the selected presented algorithm.
discontinuity. If the number of points within the cube exceeds a By observing intensity colored point clouds obtained from laser
prescribed threshold (to make sure the selection is centered on the scanning surveys, it could be also possible to qualitatively identify
surface), the best fitting plane dip and dip direction are obtained, the seepage conditions of discontinuities, as wet areas are usually
and associated points are extracted from the surface. By plotting associated with lower reflectivity than dry ones. However, the
the orientation values on a stereoplot, the discontinuity roughness reflectivity of a surface depends on many factors, such as the
angles at various scales can be measured. orientation related to the laser line of sight, the type of material and
It is worth stressing that the reliability of this procedure the distance from the scanner; thus, it is quite difficult to extract the
depends mainly on the accuracy of the point cloud data; if it is wetness contribution, and to incorporate this capability within an
too low, this could lead to an overestimation of surface roughness automatic procedure.
[50], especially for small scale analyses (0.1, 0.2 m), or low One of the main difficulties arising when we process laser
resolution point clouds. scanning data is the removal of noisy objects, such as vegetation,
To investigate the effect of instrument accuracy on the surface buildings, cables, piles, vehicles, pedestrians, etc.
roughness estimation two virtual scans have been performed on If the shape of the object to be removed is regular, this operation
hypothetical perfectly planar surfaces 0.1 m  0.1 m wide. These can be easily performed by manually selecting the associated
have been located at a distance of 20 m from the observation point; points and deleting them from the point cloud.
the angle between the line of sight and the normal to the planes 1 Vegetation is sometimes characterized by very irregular shapes;
and 2 were 01 and 751, respectively. Each surface has been if bushes have to be removed on steep rock slopes, the manual
discretized in 36, 121, 441, 1156, 10,201 regularly spaced points procedure could take a very long time, and could bring to
and their position has been expressed in polar coordinates. Point unsatisfactory results.
range error has been added to the true values by considering a A specific option called filterveg has been implemented within
normal distribution with a standard deviation of 0.005 m, accord- the 3D version of the presented tool, for the removal of objects
ing to the experiments of Kersten et al. [51] for a Riegl LMS Z-420i characterized by chaotic shapes. It consists of the iterative applica-
instrument. Finally best fitting plane orientations have been tion of the first step of the 3D procedure described in this section
calculated for all the corrected sets of points and the angles with the aim of extracting clusters of points lying on regular
between each plane and the originals have been determined. surfaces, such as rock exposures.
The results of this analysis (Fig. 9) show that the higher is the With this option, after selecting specific cube dimensions and
number of points within the selection, the lower is the angular standard deviation thresholds and moving the searching cube
deviation due to inaccurate measurements. This effect is along the point cloud, the segmentation is iterated by applying
 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198 193

REF. DIMENSION (m)

0.1
0.2
0.4
1
2
6.9

W E

Fig. 10. Roughness characterization of the black discontinuity surface of Fig. 6: stereographic projection of poles (left) and contour lines of the poles for each reference
dimension (right).

Fig. 11. DiAna application of the filterveg option for rock and vegetation segmentation.

different cube dimensions, with the result that even the smallest we can not extract them from discontinuity traces. This informa-
rock surfaces are extracted (Fig. 11). tion can be obtained either from a traditional geomechanical
survey, or from the 3D analysis described in the previous section,
performed on a rugged sector of the slope.
3.2. DiAna2D By measuring the length of the projected lines, discontinuity
persistence can be evaluated for each trace [54]. It is important to
When a planar rock face has to be investigated, a 2D approach remark that while measuring trace lengths will always produce a
can be pursued by automatically extracting, or manually tracking, biased estimate of persistence in favor of the longest traces, the 3D
the discontinuity traces, and projecting them on the best fitting approach is more objective, provided that a small enough reference
plane of the rock face [52,53]. cube dimension is selected.
With reference to the described 3D approach, we need to know Once the modal orientations of discontinuity sets are known, it is
in advance the orientations of the modal planes of each set, since possible to associate each trace with the original set, by considering its
194 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198

trend and plunge (which are related to the pitch of the line on the
reference plane). This approach is similar to the one adopted by
Kemeny et al. [55].
The intersection lines between the fitting plane of the rock face
and each discontinuity set plane are hence computed. If the angle
between one of these intersections and the discontinuity traces is
lower than a fixed threshold, the trace is assigned to the
correspondent set.
In order to evaluate the joint frequency and spacing, virtual
scanlines are drawn on the slope fitting plane, and apparent
frequency and RQD [56] are calculated for each of them. It is
possible to trace many scanlines, with different pitches on the rock
face plane, in order to enhance the directional characteristic of
these parameters.
Based on the scanline and discontinuity sets orientations, and
on the coordinates of intersection points between discontinuity
traces and the virtual scanlines, true spacing and associated
frequency and mean persistence are then assessed for each set.
Finally, as described for the 3D version, block dimensions are
evaluated using the correlation proposed by Palmstrom [42,43].
The 2D approach is suitable for analyzing rock masses with
sharply distinct discontinuity sets and it could lead to misleading
results when applied to rock masses with irregular block structure,
or if important random discontinuities occur, in fact, the 2D
association process of a trace to a discontinuity set is not
unambiguous, since the same polyline can be attributed to more
than one sets.
Although this is a weak point of the proposed approach, mainly
considering photogrammetric methods, the 2D approach has been
conceived as an integration to the 3D version.

Fig. 12. The abandoned quarry of Maiano, near Florence.

4. Field application

Both versions of the method have been applied for the geome-
chanical characterization of a man-made rock face near Florence
(Fig. 12).
The abandoned quarry of Maiano is about 50 m high, and was
one of the most important areas of extraction of a feldspathic
greywacke, called Pietra Serena, belonging to the Monte Modino
Sandstones [57]. This stone is one of the main materials employed
in Florentine architecture from the 12th century, especially for
decorative purposes [58].
The choice of this area is supported by the availability of
traditional geomechanical survey data, by the presence of both
planar and rugged surfaces in the lower and upper part of the
eastern sector of the quarry, respectively (Fig. 13), and by the
irregular block shapes within the rock mass, in order to apply
the proposed methodology to a non-ideal real case, for enhancing
its main advantages and disadvantages.
A detailed point cloud of the rock face has been obtained from a
laser scanning survey. A long range 3D laser scanner (RIEGL
LMSZ410-i) has been employed, based on the time of flight
detection technique. This device is capable of determining the
position of up to 12,000 points/s, with a maximum angular Fig. 13. True colored high resolution point cloud from laser scanning survey. The
resolution of 0.0081 and an accuracy of 710 mm, from a maximum blue and red rectangles are the areas of application of the 3D and 2D approaches,
distance of 800 m. It is worth stressing that the accuracy specifica- respectively. The extent of the image is approximately 25  40 meters.
tions given for devices, which are built in small series (like laser
scanners) may vary from instrument to instrument, and is often high resolution digital images have been taken. A true colored high
determined under controlled laboratory conditions [59]. resolution point cloud of the rock face is shown in Fig. 13.
In order to completely cover the rock face, two surveys from A standard geomechanical survey has been performed, by
different scan positions have been performed. The resulting point applying the scanline method at the base of the rock wall. All
clouds have been linked to a global reference system with the aid of the relevant parameters required to make a detailed description of
reference points, the coordinates of which were defined by using a the discontinuity system in the rock mass have been recorded on a
GPS device. A total of more than 3 million points and associated specific form.
 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198 195

Fig. 14. (a) Stereoplot of discontinuities from traditional geomechanical survey; (b) semi-automatically extracted clusters from the high resolution point cloud; (c) field data,
after Terzaghi correction; (d) numerical data, after inverse Terzaghi correction (see text).

Table 1 The orientation of all the sampled discontinuities is plotted on

Comparison of the main geometrical parameters of discontinuities obtained from the stereographic projection in Fig. 14a using the Dips software
traditional field surveys, DiAna3D and DiAna2D analyses. X: true spacing; L: [60]. The contour plot of Fig. 14c is the result of the Terzaghi
persistence; JRC: Joint Roughness Coefficient; Vb: Block volume.
correction [61] applied to field data; in the same plot discontinuity
Field survey DiAna3D DiAna2D set poles and cyclographic traces have been drawn.
No. of sets 3 3 / The resulting main geometrical properties of discontinuities are
reported in Table 1.
Jn1 dip/dip dir (1) 88/190 84/189 /
The 3D approach has been tested on the blue area of Fig. 13. In
X (m) 1.21 1.15 1.32
L (m) 1.4 0.91 0.98 order to extract as many features as possible from the point cloud, a
JRC 10 Scale dep. / small searching cube (side of 0.1 m) has been employed for the
Jn2 dip/dip dir (1) 89/113 87/98 / semi-automatic 3D analysis. The standard deviation threshold has
X (m) 0.79 0.9 0.78 been set to 0.01 m.
L (m) 0.75 1.33 1.22
A total of 16,261 valid clusters have been obtained, and the contour
JRC 11 Scale dep. /
BG dip/dip dir (1) 14/331 13/339 / lines of their orientations are projected on a stereoplot in Fig. 14b.
X (m) 0.76 0.69 0.77 While for traditional field surveys discontinuities sub-parallel to the
L (m) / / / scanline have lower probability to be sampled, the 3D analysis will
JRC 10 Scale dep. /
favor those clusters perpendicular to the scan direction. For these
Vb (m3) 0.79 0.78 0.86
reasons an inverse form of the Terzaghi correction, to compensate the
196 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198

bias introduced in favor of those features, which are perpendicular to The geometrical properties of discontinuities retrieved with
the line of sight of the laser scanner, has been applied (Fig. 14d) the 3D approach are reported in Table 1 and compared with
those derived with standard scanline surveys performed in
1 the field.
o¼ ð6Þ
9cos y9 Joint roughness has been defined for the largest discontinuity
surface of each set, according to the method described in the
where y is the angle between the scan direction and the normal to the previous section.
rock face. The diagram of Fig. 16 (left) reports the roughness angle vs. the
By inspecting Fig. 14c and d, we can notice a good agreement reference dimension, to enhance the scale effect. The stereoplot
between the set orientations measured in the field and the ones (Fig. 16, right) illustrates the pole scattering for the three surfaces,
extracted semi-automatically. There is, in fact, a difference of about considering a 0.1 m reference cube.
151 in dip direction of set 2 (Jn2). Regarding the 2D analysis (Fig. 17), discontinuity traces (yellow
Fig. 14b shows a strong influence on the cluster concentration by lines) and virtual scanlines (blue lines) have been drawn on the best
the mean slope direction of the artificial slope. This contribution can fitting plane of the quarry face in the lower sector of Fig. 13 (red
be partially filtered by applying the Terzaghi correction (Fig. 14d). area). The resulting geometrical properties of discontinuities are
After the extraction of the clusters (blue points in Fig. 15a), all reported in Table 1.
valid points have been merged to obtain discontinuity sets It is worth stressing that the 2D approach does not allow to
(Fig. 15b) and discontinuity surfaces (Fig. 15c). extract information on the number of sets and discontinuity
Fig. 15d represents the bounding polygons of the identified orientation and roughness.
discontinuities. The estimation of the true spacing (X in Table 1) with both
It is worth stressing that, despite the small extension of versions (2D and 3D) of the method is in good agreement with the
discontinuities, their high roughness, and the irregular block results from the field survey.
shapes of the rock mass, the obtained results are very promising. Some differences however occur in connection with the persis-
They cannot be compared to an ideal case, where a regularly tence (L rows in Table 1); this could be explained by the practical
blocked rock mass occurs (Fig. 6). By observing Fig. 15b we can difficulties in measuring the length of discontinuities in the field
notice, for example, the accurate detection of the bedding plane set (since joint terminations are often difficult or impossible to reach);
(red areas), which is characterized by very narrow (centimetric) as a result discontinuity persistence is often roughly estimated
and long (metric) discontinuity planes. rather than measured in the field.

Fig. 15. DiAna 3D analysis results: (a) valid clusters (blue points) with 0.1 m reference dimension; (b) discontinuity set detection (green: Jn1; blue: Jn2; red: BG);
(c) discontinuity plane detection; (d) representation of discontinuity boundary polygons.
 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198 197

Fig. 16. Roughness estimation for the widest surface of each discontinuity set. Left: reference cube dimensions vs roughness angle diagram; right: Stereoplot of poles for a
reference cube with 0.1 m dimension. Green: Jn1; blue: Jn2; red: BG.

Fig. 17. DiAna2D analysis. Discontinuity traces (yellow lines) and virtual scanlines (blue lines) on textured surface. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)

5. Conclusions The presented method allows us to investigate larger portions of

the rock mass related to field surveys, and all output parameters are
With the aim of extracting 2D and 3D structural information quantitatively measured.
from high resolution point clouds, a Matlab tool, called DiAna It is important to remark that a field analysis based on geologic
(Discontinuity Analysis) has been compiled. experience is always required in order to validate the semi-auto-
The 3D approach is based on the definition of least squares matically extracted data, especially if the rock mass is composed of
fitting planes on clusters of points selected by moving in the space a different structural domains. In this case the semi-automatic
searching cube with variable dimensions. If the associated standard procedure could give misleading results and is necessary to
deviation is below a defined threshold, the cluster is considered integrate the analysis with field data for a preliminary segmenta-
valid. By applying geometric criteria it is possible to join all the tion of the point cloud.
clusters lying on the same surface; in this way discontinuity planes Finally, the proposed algorithm allows us to separate points
can be reconstructed, and rock mass geometrical properties are belonging to a regular surface (such as rock face or ground) from
calculated. One of the main outcomes of the described procedure is irregular point geometries, typical of vegetation.
the definition of surface roughness at different scales. For low Both these versions have been applied to a real case, and the
scales this operation is however limited by the accuracy of the obtained properties have been compared with the results from a
point cloud. traditional geomechanical survey.
The 2D approach is suitable for planar rock faces with no relief, The results of this comparison are very promising, as all the
and is based on the analysis of geometrical properties of disconti- computed parameters are in good agreement with the field ones.
nuity traces.
Six of the ten parameters suggested by ISRM [2] for the
quantitative description of discontinuities (orientation, spacing, References
persistence, roughness, number of sets and block size) can be semi-
automatically calculated with the proposed method. The remain- [1] Priest SD. Discontinuity analysis for rock engineering. London: Chapman &
ing four parameters (aperture, seepage, wall strength and filling) Hall; 1993.
[2] ISRM. Suggested methods for the quantitative description of discontinuities in
cannot be assessed from conventional high resolution point clouds, rock masses. International Journal of Rock Mechanics and Mining Sciences &
as their estimation requires direct access to the rock face. Geomechanics Abstracts 1978;15:319–68.
198 G. Gigli, N. Casagli / International Journal of Rock Mechanics & Mining Sciences 48 (2011) 187–198

[3] Chandler J. Effective application of automated digital photogrammetry for [30] Adam Technology. 3DM Analyst, Mining Analysis Software, home page,
geomorphological research. Earth Surface Processes and Landforms 1999;24: /http://www.adamtech.com.au/3dm/Analyst.htmlS (accessed August 2009).
51–63. [31] Split Engineering. Split-FX ver 2.0, home page, /http://www.spliteng.com/
[4] Lane SN, James TD, Crowell MD. Application of digital photogrammetry to split-fx/S (accessed August 2009).
complex topography for geomorphological research. Photogrammetric Record [32] Gexcel. 3DGeomec, home page, /http://www.gexcel.it/en/prodotti/3dgeo
2000;16:793–821. mecS (accessed August 2009).
[5] Kraus K, Pfeifer N. Determination of terrain models in wooded areas with [33] Haneberg WC. 3-D rock mass characterization using terrestrial digital photo-
airborne laser scanner data. ISPRS Journal of Photogrammetry and Remote grammetry. AEG News 2006;49:12–5.
Sensing 1998;53:193–203. [34] Birch JS. Using 3DM analyst mine mapping suite for rock face characterization.
[6] Gigli G, Mugnai F, Leoni L, Casagli N. Analysis of deformations in historic urban In: Tonon F, Kottenstette J, editors. Laser and photogrammetric methods for
areas using terrestrial laser scanning. Natural Hazards and Earth System rock face characterization; 2006. p. 13–32.
Science 2009;9:1759–61. [35] Gaich A, Pötsch M, Schubert W. Basics, principles and applications of 3D
[7] Ferretti A, Prati C, Rocca F. Multibaseline InSAR D.E.M. reconstruction: the imaging systems with conventional and high-resolution cameras. In: Tonon F,
wavelet approach. IEEE Transactions on Geoscience and Remote Sensing Kottenstette J, editors. Laser and photogrammetric methods for rock face
1999;37:705–15. characterization; 2006. p. 33–48.
[8] Oppikofer T., Jaboyedoff M., Blikra L.H., Derron M.H. Characterization and [36] Haneberg WC. Using close range terrestrial digital photogrammetry for 3-D
monitoring of the Åknes rockslide using terrestrial laser scanning. In: rock slope modeling and discontinuity mapping in the United States. Bulletin
Proceedings of the 4th Canadian conference on geohazards: from causes to of Engineering Geology and the Environment 2008;67:457–69.
management. 2008. p. 211–18. [37] Fischler MA, Bolles RC. Random sample consensus: a paradigm for model
[9] Oppikofer T, Jaboyedoff M, Keusen HR. Collapse at the eastern Eiger flank in the fitting with applications to image analysis and automated cartography. Comm
Swiss Alps. Nature Geoscience 2008;1:531–5. of the ACM 1981;24:381–95.
[10] Abellán A, Vilaplana JM, Martı́nez J. Application of a long-range terrestrial laser [38] Sturzenegger M, Yan M, Stead D, Elmo D. Application and limitations of
scanner to a detailed rockfall study at Vall de Núria (Eastern Pyrenees, Spain). ground-based laser scanning in slope characterization. In: Eberhardt E, Stead
Engineering Geology 2006;88:136–48. D, Morrison T, editors. Proceedings of the 1st Canadian US rock mechanics
[11] Lombardi L, Casagli N, Gigli G, Nocentini M. Verifica delle condizioni di symposium. London: Taylor & Francis; 2007. p. 29–36.
sicurezza della S.P. Lodovica in seguito ai fenomeni di crollo nella cava di [39] MathWorks. MATLAB, the language of technical computing. Version 7.5
Sesto di Moriano (Lucca). Giornale di Geologia Applicata, AIGA 2006;3: Release 2007a. [CD-ROM]. Mathworks Inc.
249–56. [In Italian]. [40] Golub GH, Reinsch C. Singular value decomposition and least squares solutions.
[12] Jaboyedoff M, Couture R, Locat P. Structural analysis of Turtle Mountain Numerische Mathematik 1970;14:403–20.
(Alberta) using digital elevation model: toward a progressive failure. Geo- [41] Preparata FP, Hong SJ. Convex hulls of finite sets of points in two and three
morphology 2009;103:5–16. dimensions. Communications of the Association for Computing Machinery
[13] Fröhlich C, Mettenleiter M. Terrestrial laser scanning—new perspectives in 3D 1977;20:87–93.
surveying. In: Thies M, Koch B, Spiecker H, Weinacker H, editors. Laser-scanners [42] Palmstrom A. RMi—a rock mass characterization system for rock engineering
for forest and landscape assessment, 36. International Archives of Photogram- purposes. PhD thesis 1995, University of Oslo, Department of Geology, 400 pp.
metry, Remote Sensing and Spatial Information Sciences; 2004. p. 8/W2. [43] Palmstrom A. Measurement of correlations between block size and rock
[14] Vosselman G, Gorte BGH, Sithole G, Rabbani T. Recognizing structure in laser quality designation (RQD). Tunnelling and Underground Space Technology
scanner point clouds. International Archives of Photogrammetry, Remote 2005;20:362–77.
Sensing and Spatial Information Sciences 2004;46:33–8. [44] Palmstrom, A. The volumetric joint count—a useful and simple measure of
[15] Feng QH, Röshoff K. In-situ mapping and documentation of rock faces using a the degree of jointing. In: Proceedings of the 4th international congress IAEG,
full-coverage 3-D laser scanning technique. International Journal of Rock New Delhi. vol. 5. 1982. p. 221–8.
Mechanics and Mining Sciences 2004;41:139–44. [45] Patton FD. Multiple modes of shear failure in rock. In: Proceedings of the 1st
[16] Roncella R, Forlani G, Remondino F. Photogrammetry for geological applica- congress of the international society of rock mechanics, Lisbon. vol. 1. 1966.
tions: automatic retrieval of discontinuity orientation in rock slopes. In: p. 509–13.
Videometrics IX, electronic imaging, IS&T/SPIE, Proceedings of the 17th annual [46] Barton NR. Review of a new shear strength criterion for rock joints. Engineering
symposium. 2005. p. 17–27. Geology 1973;7:287–332.
[17] Slob S, Hack R, Van Knapen B, Turner K, Kemeny J. A method for automated [47] Barton NR. The shear strength of rock and rock joints. International Journal of Rock
discontinuity analysis of rock slopes with 3D laser scanning. Transportation Mechanics and Mining Sciences & Geomechanics Abstracts 1976;13:1–24.
Research Record 2005;1913:187–208. [48] Barton NR, Choubey V. The shear strength of rock joints in theory and practice.
[18] Turner AK, Kemeny J, Slob S, Hack R. Evaluation and management of unstable Rock Mechanics 1977;10:1–54.
rock slopes by 3-D laser scanning. IAEG 2006;404:1–11. [49] Barton NR, Bandis S. Effects of block size on the shear behavior of jointed rock.
[19] Voyat IH, Roncella R, Forlani G, Ferrero AM. Advanced techniques for geo Keynote lecture. In: Proceedings of the 23rd U.S. symposium on rock
structural surveys in modelling fractured rock masses: application to two mechanics. 1982. p. 739–60.
Alpine sites. In: Tonon F, Kottenstette J, editors. Laser and photogrammetric [50] Rahman Z, Slob S, Hack R. Deriving roughness characteristics of rock mass
methods for rock face characterization; 2006. p. 97–108. discontinuities from terrestrial laser scan data. Proceedings of the 10th IAEG
[20] Jaboyedoff M, Metzger R, Oppikofer T, Couture R, Derron MH, Locat J, Turmel D. Congress, ‘‘Engineering geology for tomorrow’s cities’’. Nottingham, United
New insight techniques to analyze rock-slope relief using DEM and 3D- Kingdom: Geological Society of London; 6-10 September 2006 paper 437, 12 pp.
imaging cloud points: COLTOP-3D software. In: Rock mechanics: meeting [51] Kersten T, Mechelke K, Lindstaedt M, Sternbernberg H. Geometric accuracy
society’s challenges and demands. Proceedings of the 1st Canada–U.S. rock investigations of latest terrestrial laser scanning systems. In: Proceedings of
mechanics symposium, Vancouver, Canada, May 27–31, 2007,Eberhardt, FIG Working Week, Stockholm, Sweden. 2008. On CD-ROM.
E,Stead, D,Morrison, T, editors. London: Taylor & Francis. 2007. p. 61–8. [52] Reid TR, Harrison JP. A semi-automated methodology for discontinuity trace
[21] Slob S, Hack R, Feng Q, Roshoff K, Turner AK. Fracture mapping using 3D laser detection in digital images of rock mass exposures. International Journal of
scanning techniques. In: Proceedings of the 11th congress of the international Rock Mechanics and Mining Sciences 2000;37:1073–89.
society for rock mechanics: the second half century of rock mechanics, Lisbon, [53] Lemy F, Hadjigeorgiou J. Discontinuity trace map construction using photo-
Portugal, vol. 1, 9–13 July, 2007. p. 299–302. graphs of rock exposures. International Journal of Rock Mechanics and Mining
[22] Ferrero AM, Forlani G, Roncella R, Voyat HI. Advanced geostructural survey Sciences 2003;40:903–17.
methods applied to rock mass characterization. Rock Mechanics and Rock [54] Priest SD, Hudson JA. Estimation of discontinuity spacing and trace length
Engineering 2009;42:631–65. using scanline surveys. International Journal of Rock Mechanics and Mining
[23] Lato M, Diederichs MS, Hutchinson DJ, Harrap R. Optimization of LiDAR Sciences & Geomechanics Abstracts 1981;18:183–97.
scanning and processing for automated structural evaluation of discontinuities [55] Kemeny J, Turner K, Norton B. In: Tonon F, Kottenstette J, editors. LIDAR for rock
in rockmasses. International Journal of Rock Mechanics & Mining Sciences mass characterization: hardware, software, accuracy and best-practices
2009;46:194–9. In: Laser and photogrammetric methods for rock face characterization;
[24] Sturzenegger M, Stead D. Quantifying discontinuity orientation and persis- 2006. p. 49–61.
tence on high mountain rock slopes and large landslides using terrestrial [56] Deere DU. Technical description of rock cores for engineering purposes.
remote sensing techniques. Natural Hazards and Earth System Science 2009;9: Felsmechanik und Ingenieurgeologie 1963;1:16–22.
267–87. [57] Abbate E, Bruni P. Modino-Cervarola. Torbiditi oligo-mioceniche ed evoluzione
[25] Sturzenegger M, Stead D. Close-range terrestrial digital photogrammetry and del margine nord appenninico. Memorie della Societa Geologica Italiana
terrestrial laser scanning for discontinuity characterization on rock cuts. 1987;39:19–33. [In Italian].
Engineering Geology 2009;106:163–82. [58] Bastogi M, Geologia Fratini F. litologia, cave e deterioramento delle pietre
[26] Maptek. Vulcan home page, /http://www.maptek.com/products/vulcan/ fiorentine. Memorie descrittive della Carte Geologica d’Italia 2004;66:
index.htmlS (accessed August 2009). 27–42. [In Italian].
[27] 3G Software and Measurement. Jointmetrix3D, home page, /http://www. [59] Bohler W, Bordas VM, Marbs A. Investigating laser scanner accuracy. Interna-
3gsm.at/eng/jminfo_eng.aspS (accessed August 2009). tional Archives of Photogrammetry and Remote Sensing 2003;34:696–701.
[28] Gemcom. Surpac, Geology and Mine Planning, home page, /http://www. [60] Rocscience. Dips, graphical and statistical analysis of orientation data, Ver. 5.0,
gemcomsoftware.com/products/surpac/S (accessed August 2009). User’s manual; 2003.
[29] CSIRO. Sirovision, home page, /http://www.sirovision.com/S (accessed May 2010). [61] Terzaghi RD. Source of error in joint surveys. Geotechnique 1965;15:287–304.