Displays 76 (2023) 102356

Contents lists available at ScienceDirect

Displays
journal homepage: www.elsevier.com/locate/displa

Discontinuity pattern detection and orientation measurement for tunnel

faces by using structure from motion photogrammetry☆
Intan Norsheira Yusoff a, Mohd Ashraf Mohamad Ismail a, *, Hayato Tobe b, Takako Miyoshi b,
Kensuke Date c, Yasuhiro Yokota c
a
School of Civil Engineering, Universiti Sains Malaysia (USM), Engineering Campus, Pulau Pinang, Malaysia
b
Rock Mechanics and Hydro-geology Group, Kajima Technical Research Institute, Kajima Corporation, Tokyo, Japan
c
Kajima Technical Research Institute Singapore, Kajima Corporation, Singapore

A R T I C L E I N F O A B S T R A C T

Keywords: Crack is a common condition that affects water-rich loess tunnels, and it negatively influences the reliability and
Structure from motion safety of the tunnel. Jointed rock masses are commonly found during underground excavation. Many under­
CloudCompare ground openings have failed during excavation and operation. The evaluation of tunnel faces by using a new
Agisoft Metashape
measurement system based on structure from motion (SfM) photogrammetry is proposed in this study. The main
Discontinuity pattern
objective is to determine the discontinuity pattern and orientation of tunnel faces as an input for tunnel face
Orientation
Image analysis stability evaluation. A set of overlapping images obtained from a tunnel face replica using 3D polystyrene and the
SfM photogrammetry approach is utilized to generate a 3D point cloud model. The discontinuity pattern and
orientation are determined via facet extraction of the KD-tree plugin in CloudCompare. The same set of over­
lapping images is analyzed in four different quality settings (low, medium, high, and ultrahigh) with different
sets of point cloud numbers that control the accuracy of the measurement of discontinuity. Results show that the
high-quality setting presents a consistent measurement of the discontinuity pattern and orientation in contrast to
the real 3D polystyrene tunnel face model. 2D pattern results from CloudCompare are validated using the fractal
analysis method in an image analysis software, where the image is converted into a binary-one. Two sets of
discontinuity are derived. The means of dip and dip direction from the cluster analysis are 82◦ /164◦ and 84◦ /
307◦ , respectively. The orientations are verified through manual compass measurement. The orientations for
manual compass measurement for Sets 1 and 2 are 86◦ /151◦ and 80◦ /303◦ , respectively. The discontinuities
obtained from the high-quality setting of the point cloud and manual measurement of the orientation by using a
geological compass highlight the similarity of the discontinuity plane in both discontinuity sets for the tunnel
face replica. The proposed method for tunnel face evaluation has good judgment for tunnel support and the
prevention of biased analyses by professionals.

1. Introduction because they control the occurrence of unstable blocks in the tunnel and
the loads imposed by blocks on the excavation support [3,6,7].There­
Discontinuity is a general term that denotes any separation in a rock fore, discontinuities must be characterized in detail to determine the
mass with zero or low tensile strength. It is the collective term for most engineering properties of rocks.
types of joints, weak bedding planes, weak schistosity planes, weakness Quantitative analysis of crack patterns is an important aspect of the
zones, and faults [1,2]. Discontinuities make rock masses different from study of the cracking behavior of soils and rocks. Preston and Tang
other engineering materials. Discontinuities make up an individual mentioned that the shape, size, roughness, connectivity, and branching
factor with a strong influence on the strength, deformability, and of crack patterns are associated with the historical stresses and strains of
permeability of rock masses [3–5]; they are also crucial for tunnel design soils and rocks and have implications for their future stability and

This paper was recommended for publication by Prof Guangtao Zhai.Peer review under responsibility of All the manuscript should have the following footnote:
☆

This paper has been recommended for acceptance by <editor_name>..

* Corresponding author.
E-mail addresses: [email protected], [email protected] (M.A. Mohamad Ismail).

https://doi.org/10.1016/j.displa.2022.102356
Received 9 May 2022; Received in revised form 12 October 2022; Accepted 7 December 2022
Available online 16 December 2022
0141-9382/© 2022 Elsevier B.V. All rights reserved.
I.N. Yusoff et al. Displays 76 (2023) 102356

functionality [8,9]. Traditional manual characterization of crack pat­ behavior of rock masses and has to be accurately identified. Remote
terns is associated with low accuracy, low efficiency, and artificial er­ sensing techniques, such as LiDAR and close-range photogrammetry,
rors. In certain cases like in the study of Lima and Dasog, the original allow for a highly realistic and accurate measurement of discontinuity
crack pattern can be disturbed by human activities and equipment, spacing in three dimensions for each discontinuity set. Dip and strike are
which usually results in large measurement errors [10,11]. normally measured with a compass/clinometer, but the process is time
The rock masses of various structures have been exposed to long- consuming, risky in some cases, and not always practical for fractures
term tectonic movement over the period of geological history, which high up on the outcrop. Dewez mentioned that the use of a 3D virtual
endows them complexity and discontinuity. The area with poor geological outcrop can minimize these problems [30].
geological characteristics in the rock mass structure of a tunnel sur­ The current study developed a new measurement system for tunnel
rounding rock is the weak part in the rock mass of the tunnel [12]. face evaluation that uses the open-source software CloudCompare to
Tunnel face excavation requires reliable tunnel face evaluations. In detect the plane of discontinuities and its orientation. The rock outcrop
conventional evaluation methods, a tunnel face is assigned a rating for members of a discontinuity set are usually not perfectly parallel due to
various parameters, including rock strength, weathering grade, and natural structure complexity. Therefore, FACETS, which is a dedicated
distribution of major cracks, on the basis of pre-established criteria. The plugin in CloudCompare, was used in this study to perform planar facet
rock mass classification approaches that are commonly used to evaluate extraction and dip and dip direction calculations (i.e., azimuth of the
tunnel faces are rock mass rating, rock mass index, rock mass quality, steepest decent). FACETS implements two algorithms to divide the
and geological strength index. The results of these approaches are initial space: a tree with k dimensions (referred to as k-d tree) and a fast-
inconsistent in some cases and likely to vary from person to person [13]. marching method. Lowe used an approximate algorithm called the best-
The two most commonly employed remote sensing techniques to bin-first (BBF) algorithm [31,32]. The algorithm is approximate in the
generate point clouds are light detection and ranging (LiDAR) and dig­ sense that it returns the closest neighbor with high probability. The BBF
ital photogrammetry. Abellán, Jaboyedoff, Oppikofer, and Viero argued method modifies the search ordering of the k-d tree algorithm such that
that LiDAR and digital photogrammetry are widely accepted tools to feature space bins are searched based on the order of their closest dis­
produce point clouds for discontinuity analyses [14–17]. tance from the query location.
Mapping rock mass discontinuities as an input for slope stability Aside from SfM photogrammetry, image analysis is also discussed in
analysis, remediation, and blast evaluation is an essential but difficult, this paper. According to Kim, digital mapping and image analysis
time-consuming, and potentially dangerous aspect of engineering techniques have been developed to help minimize human errors given
geological fieldwork. Even when mechanical aid is unnecessary, field­ that visual assessment is subjective [33]. In the current study, image
work beneath overhanging, weathered, and possibly loose blocks of rock analysis was used to verify the pattern of discontinuities on a replica 3D
exposes field personnel to potentially dangerous conditions. Remote tunnel face. Tobe developed a quantitative analytical method based on
sensing techniques have been developed to compensate for the disad­ image analysis [13]. The method divides a tunnel face into several areas,
vantages of manual mapping, and new ways of digital and contact-free and it aims to determine the principal crack directions and crack in­
measurements with the potential of automated data processing have tervals in a short time.
been investigated [18]. However, their analyses are restricted to 2D, and The present study aims to develop a new measurement system for
the results are heavily influenced by lighting and rock mass conditions. identifying the crack pattern and orientation of dip and dip direction
The 2D geometric relations of discontinuity traces can be extracted using from a tunnel face model by using SfM photogrammetry with two soft­
general-purpose, image-processing methods that consider changes in ware programs, namely, Agisoft Metashape and CloudCompare. The
pixel intensities [19–22]. Close-range terrestrial digital photogram­ pattern and orientation from this method are verified using image
metry has evolved into a useful remote characterization tool for civil analysis and manual measurement with a geological compass.
engineering practitioners, particularly in situations where manual
discontinuity measurement may be impractical or dangerous [23]. Heal 2. Methodology
used digital photogrammetric modeling and mapping of rock faces to
evaluate ground support systems for underground mines [24]. Zhang This section comprehensively discusses the methodology used in this
proposed a method for automated discontinuity mapping by identifying study. Fig. 1 shows a flowchart of the proposed methodology. The
2D discontinuity traces in image data and linking these features to their workflow has two parts: crack pattern detection and crack orientation
3D point coordinates in a point cloud acquired by terrestrial laser identification. Six main steps are involved, and each step is discussed in
scanning or photogrammetry [25]. In the present study, data collection detail.
of images captured by a camera undergoes the process of structure from
motion (SfM), which is used to produce a 3D point cloud for disconti­ 2.1. SfM
nuity analysis.
3D analysis is important to identify the orientation of discontinuities SfM works on the same principles as stereoscopic photogrammetry in
and prevent key block failure. A tunnel key block is a complete convex or that it can resolve a 3D structure from a series of overlapping, offset
concave block with a face on the tunnel surface, and the other faces are images. It differs from conventional photogrammetry because the ge­
formed by pre-existing joint planes. A block must tend to move toward ometry of the scene, camera positions, and orientation are automatically
the excavated tunnel space without penetrating any of the bounding solved without the need to specify a network of targets with known 3D
blocks for it to be considered a key block [26]. Block theory is positions a priori. Instead, these are solved simultaneously by using a
geometrically based on a set of analyses that determine where poten­ highly redundant, iterative bundle adjustment procedure based on a
tially removable blocks can exist in a geological material intersected by database of features automatically extracted from a set of multiple
variously oriented discontinuities in three dimensions [27]. All of the overlapping images [34,35].
underlying 3D combined key blocks resulting from dense joint systems In this study, 124 images of a 3D polystyrene tunnel face model were
that intersect tunnels cannot be determined without block theory. For captured using a digital camera. Fig. 2 shows the 3D polystyrene tunnel
instance, García-Luna as cited in S. Wang, employed classical photo­ face model with 8 coded targets (CTs) used in this study; the targets are
grammetry to obtain the discontinuity sets needed to analyze the sta­ sufficient to cover the tunnel face model. Discontinuities were created
bility of blocks by using block theory and limit equilibrium [3,28]. with different orientations on the replica 3D tunnel face model. Pa­
García-Luna as cited in Zhu also adopted classical photogrammetry to rameters, such as persistence, aperture, and spacing, were considered
identify discontinuity sets and conducted a discontinuous deformation when building the 3D tunnel face model. Table 1 presents the justifi­
analysis of block stability [3,29]. Discontinuity plays a key role in the cation for each parameter. CTs can be used as markers to define the local

2
I.N. Yusoff et al. Displays 76 (2023) 102356

Fig. 1. Flowchart of the methodology.

coordinate system and scale of the model or as true matches to improve

the photo alignment procedure. Agisoft Metashape includes the auto­
matic detection and matching of CTs on source photos, which allows
users to benefit from marker implementation while reducing the time
spent on manual marker placement. Moreover, automatic CT detection
and marker placement are more precise than manual marker placement.
CTs must take up a considerable number of pixels in the original photos
for them to be detected successfully. This situation leads to a natural
limitation of CT implementation. CTs are generally useful in close-range
imagery projects.
All images are then transferred to Agisoft Metashape to process the
point cloud. Fig. 3(a) shows a schematic of the camera positions used to
produce images for SfM. The first step in [36] is to detect CTs in tunnel
face images. Agisoft Metashape detects and matches CTs and adds cor­
responding markers to the reference panel. Second, the images of the
tunnel face are aligned. Agisoft Metashape determines the camera po­
sition and orientation for each photo (Fig. 3(b)) and builds a sparse point
cloud model, followed by a dense one. Agisoft Metashape allows a dense
point cloud model to be generated and visualized. On the basis of the
estimated camera positions, the program calculates the depth informa­
tion for each camera to be combined into a single dense point cloud.
This study differentiates the low-, medium-, high-, and ultrahigh-
Fig. 2. 3D polystyrene tunnel face model with coded targets. quality results of photogrammetry. Each quality setting undergoes the
same steps to produce a 3D point cloud. However, the alignment pa­
rameters during photo alignment have different accuracies. Low-
Table 1 accuracy source files are downscaled by a factor of 16. The medium
Justification for the 3D tunnel face model. setting causes image downscaling by a factor of 4 (two times by each
Parameter Length Justification side). The software works with the photos of the original size in the high-
Orientation 2 sets of Different orientations to create various planes in accuracy setting, and the ultrahigh-accuracy setting upscales the image
joints CloudCompare by a factor of 4. Interpretation of the quality parameters of a dense cloud
Persistence 3–10 cm Persistence and non-persistence discontinuities are is similar to that of the accuracy settings given in the alignment pa­
created to represent a real tunnel face
rameters. The only difference is that the ultrahigh-quality setting in this
Aperture 0.1–0.4 cm Wide and closed openings for discontinuity are created
for CloudCompare to distinguish the different
case refers to the processing of original photos.
discontinuity sizes Different quality settings produce different results of the 3D point
Spacing 0.8–1.5 cm Different spacing to represent the condition on a real cloud. Table 2 shows the detailed results of each quality setting. High-
tunnel face quality settings can be used to obtain detailed and accurate geometry,
but they require a long processing time. The processing time of the
ultrahigh-quality setting is 5 h. However, the processing time depends
on the performance of the computer. High performance increases the
processing speed.

3
I.N. Yusoff et al. Displays 76 (2023) 102356

Fig. 3. (a) SfM photogrammetric principle (source: Structure from Motion (SfM) — Theia Vision Library, 2016). (b) Camera position and orientation of image capture.

Table 2
3D point cloud of the tunnel face under different quality settings.

Setting: Medium
Setting: Low
Total 3D point cloud: 4,314,359
Total 3D point cloud: 846,474
Processing time: 1 h
Processing time: 0.5 h

Setting: High Setting: Ultrahigh

Total 3D point cloud: 23,313,836 Total 3D point cloud: 42,631,505
Processing time: 3 h Processing time: 5 h

2.2. Discontinuity extraction using CloudCompare algorithm. K-d tree is a space-partitioning data assembly approach for
establishing points in a k-dimensional space. Fig. 4 shows an example of
CloudCompare is an open-source software where a 3D point cloud space-partitioning trees. The k-d tree method recursively splits the 3D
(and triangular mesh) is processed. CloudCompare can detect planes, point cloud into small planar patches until the point fits the best-fitting
especially discontinuities, on a surface area. Therefore, the 3D point plane. The planar patches are then back-grouped into large facets
cloud from Agisoft Metashape is transferred to CloudCompare with the through a co-planarity measure. Again, the FM approach splits the 3D
aid of the FACETS plugin to extract planes. Two approaches can be used point cloud into little patches and regroups them. Consequently, all of
to extract discontinuities: k-dimensional tree (k-d tree) and fast the patches have the same size [38,39]. K-d tree was chosen in this study
marching (FM). Both methods implement the least-square fitting because the research aim is to differentiate the quality of all parameters.

4
I.N. Yusoff et al. Displays 76 (2023) 102356

Fig. 4. (a) Space-partitioning algorithm and (b) 2D tree modified from [42].

K-d tree is a generalization of the binary tree, which is commonly These parameters affect the outcome of facets after processing. Each
used for sorting and searching. This approach takes advantage of the parameter is explained below.
coordinate information available in vector spaces, which is then used to
split the search space [40]. K-d tree is one of the most widely used Max angle: maximum angle between neighbor patches (in degrees).
nearest neighbor algorithms. At each node, the data points are divided Max relative distance: maximum distance between the merged
into two groups. K-d tree is a binary tree technique that always ends in patches and the current facet center.
two nodes. The median is frequently used as a split criterion [41]. Fig. 4 Distance criterion: used to estimate if a facet is still sufficiently
shows the basic working idea of the k-d tree algorithm. “flat” after merging a new patch; for instance, “Max distance @ 99
Two parameters area adopted in the technique for creating a k-d tree. %” means that 99 % of points must be close to the value specified in
The assumption is that the red points represent point P. Algorithm 1 the field on the right (0.2 in the abovementioned example).
shows one set of points P and an integer value that reflects the depth of Min points per facet: points per facet that are smaller than this
the subtree root. value is discarded.
Algorithm 1: K-d tree algorithm [43] Max edge length: used to extract the facet (concave) contour (the
Require: Set of points (x, y) named P and the tree depth value
contour should be closer to the points when the value is smaller).
Ensure: Root of the k-d tree
1: function: build k − d tree(P, depth) Each quality setting used the same k-d tree parameters, except for
2: if (P contains only one point) then minimum points per facet. The reason is that each model has a different
3: return the root with that point
total point cloud. Therefore, the minimum points per facet also differ.
4: else if (depth is an even number) then
5: divide the values of P into two subsets by using the median of the x coordinates of Table 3 shows a summary of the k-d tree parameters for each setting.
the P set.It After the facets are extracted, they are categorized by orientation
generates two subsets: (dip/dip direction) into a single or family plane. By using the FACET
6: if (Pxi < median(x)) then plugin, the discontinuities are easily distinguished by looking at the
7:P1xi : = subsetleft
different color facets. Fig. 5 shows the different facet colors of different
8: else
9:P2xi : = subsetright
planes that represent different dip values and dip directions. Same-color
10: end if facets form one plane that represents discontinuity. Fig. 6 shows the
11: else processes of producing facets and classifying them according to the
12:divide the values of P into two subsets by using the median of the y coordinates of family plane.
the P set.It
generates two subsets:
13: end if 2.3. Verification of discontinuity pattern by using fractal analysis
14: depth = depth + 1
15: end function Previous studies have proposed a method for analyzing tunnel faces
by using image processing. This method, which allows for a relatively
fast quantification of crack distributions on tunnel faces, is rooted on the
The time used in the k-d tree approach in k groups is O(k) for the depth fractal analysis of cracks visible in photographs of tunnel faces [45].
of the tree, and each partition has a time length of O(log(n)) because P is However, fractal analysis causes difficulty for workers on site to achieve
a finite set. Thus, the total time used in the algorithm is O(log(n) +k)
[43,44]. Table 3
In this study, the k-d tree FACET plugin was used to extract the Summary of k-d tree parameters for each quality setting.
discontinuities present on the polystyrene tunnel face model. After k- Setting/Parameter Low Medium High Ultrahigh
d tree’s plane segmentation of each quality setting on the polystyrene Max angle 10◦ 10◦ 10◦ 10◦
tunnel face model, facets were produced and grouped according to their Max relative distance 1.0 1.0 1.0 1.0
dip/dip directions. Before plane extraction, several parameters, such as Distance criterion 99 % 99 % 99 % 99 %
maximum angle, maximum relative distance, distance criterion, mini­ Min points per facet 10 14 48 169
Max edge length 0.01 0.01 0.01 0.01
mum points per facet, and maximum edge length, need to be considered.

5
I.N. Yusoff et al. Displays 76 (2023) 102356

Fig. 5. Color of facets with dip and dip direction.

If we convert a photograph of a tunnel face into a binary image, with

white pixels representing areas with large variance in brightness and
black pixels representing other areas, then the cracks will emerge as
white lines (Fig. 7). These cracks were subjected to a numerical analysis
in this study to determine the principal directions and continuity.
Fig. 6. Process of CloudCompare to produce facets and discontinuity The connection rate was calculated after the continuity of vertical
orientation. running cracks was determined. Rotating a single image resulted in
different connection rates depending on the angle turned (Fig. 8). It
an intuitive understanding of crack distributions. represents cracks that have same continuity and intervals but different
Rock cracks can be classified into several categories, including joint,
cleavage, and shear plane, depending on the origin of the discontinuity.
For the purpose of tunnel face evaluation, all weak surfaces in rocks are
treated in the same way. This study made no distinctions based on how a
crack came to be. We regarded all weak surfaces simply as cracks. When
a rock is exposed by excavation, it is likely to shear along a crack, which
is a dynamically weak surface, due to stress release. After excavation,
the cracks form lines on the uneven surfaces of the tunnel face. Illumi­
nating a tunnel face can highlight cracks because raised areas reflect
light, whereas depressions do not.
By applying a method of directly quantifying the connectivity of
minerals in rocks on the basis of percolation theory [46], we developed a
method of detecting principal crack directions that involves rotating the
image of a tunnel face and comparing how the cracks are connected
[13]. Similar to fractal analysis, percolation theory is associated with the
study of complex systems. It considers fundamental principles to analyze
the connections between elements [47]. Fig. 7. Example of a binary image.

6
I.N. Yusoff et al. Displays 76 (2023) 102356

Fig. 8. Rotation angle of artificial crack images from [13]. Fig. 10. Example of a principal crack image.

direction. The connection rate will be the highest in the image with the 2.4. Identification of crack orientation using Dips
most vertical cracks (0◦ ). We can apply this condition to identify the
principal crack direction in an image of a tunnel face. If the connection The dip and dip direction information obtained in CloudCompare
rates are calculated while the image is rotated, then the connection rate was imported into Dips Rocscience to perform a cluster analysis of the
will be the highest in the image with an angle showing the greatest orientation. Dips is a program that allows the user to interact with
number of vertical cracks. When measuring intervals in a visual in­ geological data based on orientation. The program has a wide range of
spection, the engineering geologist initially determines the area being applications and is intended for beginners and experienced stereo­
inspected (viewing area) and then determines the intervals while graphic projection users who want to utilize complex tools in geological
counting the number of cracks that cross the direction perpendicular to data analysis. Dips allows users to examine and visualize structural data
major cracks. Tobe developed an algorithm that corresponds to these by using approaches similar to those used in manual stereonets. It is a
steps of visual inspection. Considering that major cracks tend to be program that analyzes aspects that are relevant to the engineering
parallel to the principal direction, they counted the number of cracks in analysis of rock structures.
the viewing area that crossed the principal direction at right angles [13]. In this study, the identification of crack orientations only used a set
They believed that doing so would enable a quantitative calculation of of high-quality point clouds because it had the most consistent mea­
crack intervals in a manner consistent with visual inspection. Fig. 9 surement results in CloudCompare. A total of 948 orientations from facet
verifies that the connection rate in the image rotated at 0◦ is the highest. extraction in CloudCompare were processed by the computer software
When determining crack intervals through visual inspection, the Dips. All orientations were processed into sets via cluster analysis in
inspector focuses on the narrowest intervals and areas where cracks Dips. Cluster analysis quickly determines data clusters, such as joint sets,
concentrate. Similarly, we used the smallest crack interval in the on a stereonet by using an algorithm based on fuzzy cluster analysis. The
viewing area as the representative value in the developed image analysis “Sets from Cluster Analysis” option requires the user to manually select
method [13]. We counted the number of cracks included in the viewing the approximate center of a data cluster; then, the fuzzy cluster algo­
area that crossed the principal direction at right angles considering that rithm automatically determines which poles belong to the nearest cor­
major cracks tend to be parallel to the principal direction. Fig. 10 shows responding joint set within a specified maximum radius. Dominant
the principal crack image of the tunnel face model. The red line refers to discontinuity sets are distinguished based on equal-area stereographic
the first component of the principal crack that has the most dominant projection. The mean results on the dip and dip direction of the orien­
crack direction. The yellow line represents the second component of the tation are discussed in Section 3.3.
principal crack that has the next most dominant crack direction.

Fig. 9. Connection rates of artificial cracks from [13].

7
I.N. Yusoff et al. Displays 76 (2023) 102356

3. Results and discussion point clouds, and vice versa. High-quality settings can be used to obtain
detailed and accurate geometry, but they require a long processing time.
This section discusses the findings in detail. The results of this study However, the processing time depends on the performance of the com­
are divided into three parts, namely, discontinuity patterns, disconti­ puter. These 3D point clouds affect plane detection on CloudCompare.
nuity orientations, and verifications for discontinuity. In general, the 3D model is more detailed when the total 3D point
cloud is larger. Table 3 shows different 3D polystyrene tunnel face
models with joints. At first glance, the models do not appear to differ
3.1. Discontinuity pattern detection considerably. However, detailed results can be obtained after the
extraction of planes in CloudCompare. Table 4 shows two major
Visual observations and compass measurements of discontinuity discontinuity sets: Set 1 (purple) and Set 2 (pink). The mean dip/dip
orientations on tunnel faces are often the main source of information, directions for Set 1 (green) and Set 2 (purple) are 82◦ /164◦ and 84◦ /
particularly in shotcrete-supported tunnels. However, visual observa­ 307◦ , respectively. Table 3 shows the different orientations for each
tions cannot quantify orientations or distances accurately; meanwhile, color facet. As mentioned in Section 3.1, the produced facets differ based
the quality of compass measurements depends on the individual’s on their total point clouds. The low-quality setting produced the mini­
ability, and errors could arise from the way in which the compass is mum number of facets and could not show discontinuities on the tunnel
positioned and how its measurements are read [48]. Therefore, this face. The facets produced in the medium-quality setting showed the
study focused on a nondestructive method to identify cracks. orientation of some discontinuities. However, the orientation only
3D point cloud results for outdoor close-range photogrammetry have exhibited the major discontinuities. Meanwhile, the high-quality setting
four quality settings. The difference among these quality settings de­ produced good results. The facets could show the major and minor
pends on the total 3D point cloud. Low-quality settings have few total 3D

Table 4
Pattern of discontinuity for different quality settings in facet extraction.
Quality setting 3D point cloud Total facets Facets representing discontinuities

Low

Medium

High

Ultrahigh

a
J1 dip/dip direction: 82◦ /164◦ .
b
J2 dip/dip direction: 84◦ /307◦ refer to Fig. 5 for the legend of each color facet.

8
I.N. Yusoff et al. Displays 76 (2023) 102356

discontinuities on the polystyrene tunnel face model. This setting can measurement and both sets from CloudCompare have low values (22.58,
enable engineering geologists to easily imagine the actual tunnel face on 29.92, and 23.89, respectively), which denote a tight cluster.
site. With regard to the ultrahigh-quality setting, the facets produced
were highly detailed. Some unimportant lines were also detected in this 4. Advantages, limitations, and validity of the study
setting. Therefore, this setting can allow engineering geologists to ac­
quire an overview image of the tunnel face without concern for the dip This section discusses some of the advantages and limitations of the
and dip directions. In our opinion, the high-quality setting is the best proposed method. The advantages are as follows:
because it can produce results on major and minor discontinuities.
1. The importance of 3D analysis in identifying the discontinuity of a
tunnel face and preventing key block failure from happening is
3.2. Verification of the discontinuity pattern
revealed.
2. The proposed method provides information on the dip and dip di­
Using image analysis software is another method to discover dis­
rection of each discontinuity plane.
continuities on tunnel faces. Analysis of 2D images requires the appli­
3. The proposed method can help engineers collect data in high-risk
cation of a well-defined method that allows the determination of
and dangerous areas.
whether a pixel belongs to the discontinuities. Image segmentation is a
4. Immediate results on discontinuity patterns are provided by the 2D
method that must be dependable and accurate. Segmentation is the
image analysis software.
process of transforming a grayscale image into a binary image by
identifying two sets of pixels in the image. Fig. 11 shows an image of the
Some of the limitations of the proposed method are as follows:
polystyrene tunnel face model converted into a binary image, where the
cracks appear as white lines and then form a principal crack. The pri­
1. The processing time depends on the performance of the computer
mary directions and continuity of these cracks were determined via
used in point cloud generation and facet extraction.
numerical analysis. The most dominant crack directions are denoted by
2. Good-quality images are required to process point clouds with Agi­
red and yellow lines. However, this method can only provide qualitative
soft Metashape.
results. It cannot show the dip/dip direction of the discontinuities. The
results of the discontinuity pattern from the 2D image analysis method
The results of the discontinuity pattern were validated using 2D
were highly accurate in contrast to the CloudCompare analysis results.
image analysis software, and the principal cracks exhibited the same
pattern as those in facet extraction in CloudCompare. The results of the
3.3. Identification of the discontinuity orientation discontinuity orientation were validated using a geological compass in
image collection.
Crack detection is the process of detecting cracks in structures by
using any processing technique. The cluster analysis with Dips revealed 5. Concluding comments
the concentration of the orientations of the discontinuities of the 3D
tunnel face model in a stereogram (Fig. 12). The orientations extracted A new measurement system for tunnel face evaluation that uses
digitally from software are depicted in pole vector mode. The contour of CloudCompare is presented in this paper. The tunnel face evaluations
the pole vectors is facing the south-east direction in the stereogram. The include discontinuity pattern and orientation analyses. Discontinuity
means of dip and dip direction for the CloudCompare orientations for orientation is important because it can determine the key block that can
Sets 1 and 2 are 82◦ /164◦ (green) and 84◦ /307◦ (purple), respectively. cause tunnel face to fall. According to Mohan and Poobal (2018) who
The means for manual orientations for Sets 1 and 2 are 86◦ /151◦ and conducted a critical review and analysis of crack detection via image
80◦ /303◦ , respectively. Table 5 presents the statistical data from Dips. processing, various image-based crack detection techniques based on
Two sets for CloudCompare and the manual stereogram overlap in image types, such as camera, infrared, ultrasonic, and laser images, have
Fig. 13. The Fisher coefficient (k) is automatically calculated for each been used in different studies. They concluded that many researchers
set. It determines the distribution of joint orientations in the stereogram. have used the camera-type image for analysis with effective segmenta­
A high value of k denotes a tight cluster. The results of the 3D tunnel face tion algorithms, such as the threshold technique.
model show two sets of joint orientation in both stereograms. Set 1 in the The current study analyzed four quality settings for pattern analysis
geological compass measurement has a high value (113.86), which de­ by using CloudCompare. The high-quality setting showed the highest
notes a dispersed cluster. This condition might have been caused by the resemblance to the 3D tunnel face model. This study also proved that the
minimal data of Set 1. However, Set 2 from the geological compass

Fig. 11. Detecting the discontinuity pattern of the tunnel face model.

9
I.N. Yusoff et al. Displays 76 (2023) 102356

Fig. 12. (a) Stereogram orientations extracted by CloudCompare and (b) stereogram orientations measured by a geological compass.

results when discontinuity orientations are identified using CloudCom­

Table 5 pare are more accurate than those of the 3D tunnel face model and
Dips analysis information of 3D tunnel face model orientations.
manual measurement. The dip direction showed differences of 4◦ and
Statistical data Geological compass CloudCompare 13◦ in Sets 1 and 2, respectively.
Set of discontinuity Set 1 Set 2 Set 1 Set 2 This “distant” measurement technique is highly informative and
Dip/dip direction 86 /151
◦ ◦
80 /303
◦ ◦
82 /164
◦ ◦
84◦ /307◦
convenient when used for tunnel face discontinuity evaluation. This
Total entries 14 16 769 630 method prevents biased geological investigations by professionals.
Fisher coefficient (k) 113.86 22.58 29.92 23.89 However, the proposed method is time consuming because it provides
detailed results on discontinuities by giving the dip and dip directions
for each plane.
The application of the method to real tunnel cases can be studied in
the future. Other fields, such as architecture, aside from underground

10
I.N. Yusoff et al. Displays 76 (2023) 102356

Fig. 13. Overlapping of CloudCompare and manual measurement stereograms.

tunneling can use this method. Terrestrial laser scanning also can be [8] J. Preston, et al., Integrated whole-rock trace element geochemistry and heavy
mineral chemistry studies: aids to the correlation of continental red-bed reservoirs
used in the future study to obtain more precise 3D point cloud.
in the Beryl Field, UK North Sea, Pet. Geosci. 4 (1) (1998) 7–16, https://doi.org/
10.1144/petgeo.4.1.7.
[9] P. Jia, C.A. Tang, Numerical study on failure mechanism of tunnel in jointed rock
Declaration of Competing Interest mass, Tunn. Undergr. Sp. Technol. 23 (5) (2008) 500–507, https://doi.org/
10.1016/j.tust.2007.09.001.
The authors declare that they have no known competing financial [10] L.A. Lima, M.E. Grismer, Soil crack morphology and soil salinity, Soil Sci. 153 (2)
(1992) 149–153, https://doi.org/10.1097/00010694-199202000-00009.
interests or personal relationships that could have appeared to influence [11] G.S. Dasog, G.B. Shashidhara, Dimension and volume of cracks in a Vertisol under
the work reported in this paper. different crop covers, Soil Sci. 156 (6) (1993) 424–428.
[12] J. Wang, C. Wang, Z. Han, Y. Jiao, J. Zou, Study of hidden structure detection for
tunnel surrounding rock with pulse reflection method, Measurement 159 (2020),
Data availability 107791, https://doi.org/10.1016/j.measurement.2020.107791.
[13] H. Tobe, Y. Miyajima, S. Shirasagi, T. Yamamoto, J. Kawabata, A rapid image
Data will be made available on request. analyzing method for determining crack distribution and interval on tunnel faces,
in: ISRM Int. Symp. - 10th Asian Rock Mech. Symp. ARMS 2018, 2018.
[14] A.J. Riquelme, A. Abellán, R. Tomás, M. Jaboyedoff, A new approach for semi-
Acknowledgment automatic rock mass joints recognition from 3D point clouds, Comput. Geosci. 68
(2014) 38–52, https://doi.org/10.1016/j.cageo.2014.03.014.
[15] A.J. Riquelme, A. Abellán, R. Tomás, Discontinuity spacing analysis in rock masses
This study was funded by an RUI grant from Universiti Sains using 3D point clouds, Eng. Geol. 195 (2015) 185–195, https://doi.org/10.1016/j.
Malaysia (USM) (account number 1001/PAWAM/8014090) with the enggeo.2015.06.009.
topic “The Evaluation of Early-High-Strength Steel Fiber Reinforced [16] T. Oppikofer, M. Jaboyedoff, L. Blikra, M.-H. Derron, R. Metzger, Characterization
and monitoring of the Åknes rockslide using terrestrial laser scanning, Nat.
Shotcrete and its Interaction with the Surrounding Rock Mass as the
Hazards Earth Syst. Sci. 9 (3) (2009) 1003–1019.
Primary Tunnel Support in Overstressing Rock.” The first author thanks [17] A. Viero, G. Teza, M. Massironi, M. Jaboyedoff, A. Galgaro, Laser scanning-based
the USM Fellowship Scheme for providing financial support in the form recognition of rotational movements on a deep seated gravitational instability: the
of a scholarship. The authors would also like to thank the anonymous Cinque Torri case (North-Eastern Italian Alps), Geomorphology 122 (1–2) (2010)
191–204.
reviewers for their helpful and encouraging suggestions on how to [18] A. Buyer, W. Schubert, Extraction of discontinuity orientations in point clouds,
improve this paper. 2016, doi: 10.1201/9781315388502-197.
[19] J.A. Franklin, N.H. Maerz, C.P. Bennett, Rock mass characterization using
photoanalysis, Int. J. Min. Geol. Eng. 6 (2) (1988) 97–112, https://doi.org/
References 10.1007/BF00880801.
[20] G. Crosta, Evaluating rock mass geometry from photographic images, Rock Mech.
[1] ISRM, Suggested methods for the quantitative description of discontinuities in rock Rock Eng. 30 (1) (1997) 35–58.
masses, Int. J. Rock Mech. Min. Sci. Geomech. (1978) 319–368. [21] T.R. Reid, J.P. Harrison, A semi-automated methodology for discontinuity trace
[2] D.G. Price, Determination of discontinuity friction by rock mass classification detection in digital images of rock mass exposures, Int. J. Rock Mech. Min. Sci. 37
Détermination de friction de discontinuiteé par un système de classification d’ un (7) (2000) 1073–1089, https://doi.org/10.1016/S1365-1609(00)00041-1.
massif rocheux Bepalung von Reibungswiderstand von Discontinuiteiten mit ein [22] F. Lemy, J. Hadjigeorgiou, Discontinuity trace map construction using photographs
Klassifikationssystem H, Rock Mech. September (1995) 23–27. of rock exposures, Int. J. Rock Mech. Min. Sci. 40 (6) (2003) 903–917, https://doi.
[3] R. García-Luna, S. Senent, R. Jurado-Piña, R. Jimenez, Structure from Motion org/10.1016/S1365-1609(03)00069-8.
photogrammetry to characterize underground rock masses: Experiences from two [23] W.C. Haneberg, Using close range terrestrial digital photogrammetry for 3-D rock
real tunnels, Tunn. Undergr. Sp. Technol. 83 (2019) 262–273, doi: 10.1016/j. slope modeling and discontinuity mapping in the United States, Bull. Eng. Geol.
tust.2018.09.026. Environ. 67 (4) (2008) 457–469, https://doi.org/10.1007/s10064-008-0157-y.
[4] R.E. Goodman, Methods of Geological Engineering in Discontinuous Rocks, West [24] D. Heal, M. Hudyma, Y. Potvin, Assessing the in-situ performance of ground
Group, 1976. support systems subjected to dynamic loading, in: Ground Support in Mining and
[5] J.A. Hudson, J.P. Harrison, Engineering Rock Mechanics: An Introduction to the Underground Construction, Proceedings of the Fifth International Symposium on
Principles, Elsevier, 2000. ground support, Perth, Australia, 2004, pp. 28–30.
[6] R.D. Hammett, E. Hoek, Design of large underground caverns for hydroelectric [25] P. Zhang, Q. Zhao, D.D. Tannant, T. Ji, H. Zhu, 3D mapping of discontinuity traces
projects with particular reference to structurally controlled failure mechanisms, using fusion of point cloud and image data, Bull. Eng. Geol. Environ. 78 (4) (2019)
Recent Dev. Geotech. Eng. Hydro Proj. by Kulhawy FH) (1981) 192–206. 2789–2801, https://doi.org/10.1007/s10064-018-1280-z.
[7] A.L. Karzulovic, The Use of Keyblock Theory in the Design of Linings and Supports
for Tunnels, University of California, Berkeley, 1988.

11
I.N. Yusoff et al. Displays 76 (2023) 102356

[26] G.-H Shi, R.E. Goodman, The key blocks of unrolled joint traces in developed maps [38] S.K. Nagendran, M.A.M. Ismail, W.Y. Tung, Photogrammetry approach on
of tunnel walls, Int. J. Numer. Anal. Methods Geomech. 13(2) (1989) 131–158, doi: geological plane extraction using cloudcompare FACET plugin and scanline survey,
10.1002/nag.1610130203. Bull. Geol. Soc. Malays. 68 (2019) 151–158, https://doi.org/10.7186/
[27] R.E. Goodman, Block theory and its application 3 (1995) 383–423. bgsm68201916.
[28] S. Wang, P. Ni, M. Guo, Spatial characterization of joint planes and stability [39] W.Y. Tung, S.K. Nagendran, M. Ashraf, M. Ismail, IOP Conference Series: Earth and
analysis of tunnel blocks, Tunn. Undergr. Sp. Technol. Inc., Trenchless Technol. Environmental Science 3D rock slope data acquisition by photogrammetry
Res. 38 (2013) 357–367, https://doi.org/10.1016/j.tust.2013.07.017. approach and extraction of geological planes using FACET plugin in
[29] H. Zhu, W. Wu, J. Chen, G. Ma, X. Liu, X. Zhuang, Integration of three dimensional CloudCompare, Vol. 169, 2018, pp. 12051, doi: 10.1088/1755-1315/169/1/
discontinuous deformation analysis (DDA) with binocular photogrammetry for 012051.
stability analysis of tunnels in blocky rockmass, Tunn. Undergr. Sp. Technol. 51 [40] A. Serrano, L. Micó, J. Oncina, Which fast nearest neighbour search algorithm to
(2016) 30–40, https://doi.org/10.1016/j.tust.2015.10.012. use? Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect.
[30] T.J.B. Dewez, D. Girardeau-Montaut, C. Allanic, J. Rohmer, Facets: a cloudcompare Notes Bioinformatics) 7887 (2013) 567–574, https://doi.org/10.1007/978-3-642-
plugin to extract geological planes from unstructured 3d point clouds, ISPRS - Int. 38628-2_67.
Arch. Photogramm. Remote Sens. Spat. Inf. Sci. XLI-B5 (2016) 799–804, https:// [41] H. Marius, Tree algorithms explained: Ball Tree Algorithm vs. KD Tree vs. Brute
doi.org/10.5194/isprs-archives-xli-b5-799-2016. Force, 2020, https://towardsdatascience.com/tree-algorithms-explained-ball-tree-
[31] D.G. Lowe, Distinctive image features from scale-invariant keypoints, Int. J. algorithm-vs-kd-tree-vs-brute-force-9746debcd940 (accessed Apr. 21, 2022).
Comput. Vis. 60 (2) (2004) 91–110, https://doi.org/10.1023/B: [42] K. Tran, How to search data with KDTree. Given a task to find data in a certain… |
VISI.0000029664.99615.94. by Khuyen Tran | Towards data science, May 10, 2020. https://towardsdatascienc
[32] J.S. Beis, D.G. Lowe, Shape indexing using approximate nearest-neighbour search e.com/how-to-search-data-with-kdtree-aad5c82ebd99 (accessed Sep. 20, 2021).
in high-dimensional spaces, in: Proceedings of IEEE Computer Society Conference [43] J. Anzola, J. Pascual, G. Tarazona, R. González, A clustering WSN routing protocol
on Computer Vision and Pattern Recognition, 1997, pp. 1000–1006. based on k-d tree algorithm, Sensors (Switzerland) 18 (9) (2018) 1–26, https://doi.
[33] D.H. Kim, A.S. Balasubramaniam, I. Gratchev, Application of photogrammetry and org/10.3390/s18092899.
image analysis for rock slope investigation, Geotech. Eng. (2018). [44] H.M. Kakde, Range Searching using Kd Tree, Report (2005) 1–12.
[34] M.J. Westoby, J. Brasington, N.F. Glasser, M.J. Hambrey, J.M. Reynolds, [45] Y. Udagawa, Development of geological feature evaluation technique by fractal
‘Structure-from-Motion’ photogrammetry: a low-cost, effective tool for geoscience analysis of a tunnel face image, J. Japan Soc. Eng. Geol. 48 (3) (2007) 116–125.
applications, Geomorphology 179 (2012) 300–314, https://doi.org/10.1016/j. [46] H. Tobe, M. Chigira, Causes of shallow landslides of weathered granitic rocks from
geomorph.2012.08.021. the view point of weathering styles and petrologic textures, Disaster Mitig. Debris
[35] K.N. Snavely, Scene Reconstruction and Visualization from Internet Photo Flows, Slope Fail. Landslides. Univers. Acad. Press, Tokyo, 2006, pp. 493–501.
Collections, University of Washington, 2008. [47] K. Takayasu, et al., Hepatocellular carcinoma: treatment with intraarterial iodized
[36] Agisoft Metashape version 1.5, Agisoft Metashape User Manual, Agisoft Metashape, oil with and without chemotherapeutic agents, Radiology 163 (2) (1987) 345–351.
no. September, p. 160, 2019, [Online], Available: https://www.agisoft.com/pdf/m [48] S. Slob, B. Van Knapen, R. Hack, K. Turner, J. Kemeny, Method for automated
etashape-pro_1_5_en.pdf. discontinuity analysis of rock slopes with three-dimensional laser scanning, Transp.
Res. Rec. 1913 (2005) 187–194, https://doi.org/10.3141/1913-18.

12