Engineering Geology 279 (2020) 105878

Contents lists available at ScienceDirect

Engineering Geology
journal homepage: www.elsevier.com/locate/enggeo

A practical photogrammetric workflow in the field for the construction of a

3D rock joint surface database
Zhao Lianheng a, b, Huang Dongliang a, Chen Jingyu a, Wang Xiang a, *, Luo Wei c, Zhu Zhiheng a, *,
Li Dejian d, Zuo Shi a, *
a
School of Civil Engineering, Central South University, Changsha, Hunan 410075, China
b
Key Laboratory of Heavy-haul Railway Engineering Structure, Ministry of Education, Central South University, Changsha, Hunan 410075, China
c
School of Civil Engineering and Architecture, East China Jiaotong University, Nanchang, Jiangxi 330013, China
d
Department of Geotechnical Engineering, School of Civil Engineering, Southwest Jiaotong University, Chengdu, Sichuan 610031, China

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

Keywords: Collecting morphological characteristics of rock joint surfaces and building a digital rock joint library with
Rock joint database abundant sample types and sufficient sample numbers is important for fractured rock mass analysis. To solve the
Photogrammetry “inflexible field operation” problem in existing contact type and non-contact type data collection procedures, this
Geometry
work presents a photogrammetric workflow in the field for the construction of a 3D rock joint surface database.
3D digital model
3D reconstruction
Innovations in some steps make it practical for field scenarios: image collection procedures in the field and post-
Field investigation processing steps of the SfM-MVS-derived dense point cloud. The performance is considered as good as that of
white light scanning, and sub-millimetre accuracy is achieved during batch processing. Two rock joint databases
containing 310 samples are established, and two application examples show very good potential of the proposed
photogrammetric workflow in the evaluation of rock joint roughness and rock joint anisotropy.

1. Introduction randomness and anisotropy in space (Liu et al., 2018; Yang et al., 2011),
these newly proposed rock joint roughness models always require
Fractures, joints, faults, and fissures are widely distributed in nature, extensive digital rock joint data for validation (Bahaaddini et al., 2016;
and their morphological characteristics have significant influences on Thirukumaran and Indraratna, 2016; Wu et al., 2018; Zheng and Qi,
the hydraulic (permeability) and mechanical (shear strength and dila­ 2016). Thus, developing an efficient rock joint data collection method
tion) properties of the rock mass. Since Barton and Choubey (1977) has become a research topic of considerable interest.
proposed the Barton-Bandis failure criterion for rock mass, studies on Existing methods to acquire rock joint models can be generally
the relationship between joint roughness and shear strength have classified into contact type and non-contact type. By means of in­
continued to attract many scholars’ attention. However, most previous struments, contact type data collection involves direct contact with the
studies focused mainly on 2D joint profiles and ignored the 3D features joint surface, and samples are collected through point-by-point detec­
of roughness, so that reflecting the anisotropy of roughness is difficult. tion (Barton and Choubey, 1977; Singh and Basu, 2016; Tatone, 2009;
To overcome this problem, Grasselli and Egger (2003) proposed a new Yong et al., 2018). Non-contact data collection uses modern means to
shear strength model that fully considers the 3D features of joint scan and obtain high-precision topographic features of rock joints. Ac­
morphology. Grasselli held that only a convex micro-element on the cording to different working principles, non-contact data collection can
joint surface facing the shear direction is likely to be a contact part, so be divided into (1) 3D laser scanning technology (Belem et al., 2007;
the dip angle and area of the micro-element are used for roughness Fardin et al., 2001; Kulatilake et al., 2006; X. G. Liu et al., 2017b;
calculation. Thereafter, Grasselli’s model has been inherited and MŁynarczuk, 2010), (2) structured light 3D scanning technology (Jiang
improved by many researchers, e.g., Tatone (2009), Tang et al. (2012), et al., 2016; Tang and Jiao, 2017), (3) digital photography technology
Xia et al. (2013), Yang et al. (2015), Q. Liu et al. (2017a), and Tian et al. (Bae et al., 2011; Chen et al., 2017; Paixão et al., 2018), etc. An
(2018). Since the morphological characteristics of rock joints have large important way to batch-produce rock joints with the same natural

* Corresponding authors.
E-mail addresses: wang.xiang@csu.edu.cn (W. Xiang), zzh8207@163.com (Z. Zhiheng), zuoshi@csu.edu.cn (Z. Shi).

https://doi.org/10.1016/j.enggeo.2020.105878
Received 9 April 2020; Received in revised form 13 October 2020; Accepted 23 October 2020
Available online 27 October 2020
0013-7952/© 2020 Elsevier B.V. All rights reserved.
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 1. Array scanning frame for rock joint surfaces: (a) The four points on the calibration board are used to establish the local coordinate system; (b) The camera is
moved according to the array scanning frame, and the image sequence is collected directly opposite the joint surface; (c) The square grid line on the camera screen is
enabled during shooting. The overlap between horizontal and vertical adjacent images in the sequence is greater than 60%.

surfaces is 3D scanning (Jiang et al., 2020). However, the device is rock joint photogrammetry. Therefore, building digital rock joint li­
costly; thus, building a digital rock joint library with abundant sample braries with abundant sample types and sufficient sample numbers is
types and sufficient sample numbers is too expensive. Structured light possible.
3D scanning devices are high-precision and cost-effective, but the colour The photogrammetric scanning of rock joint surfaces can be done
distortion is serious. through the following steps: (1) photographing process in the field
Cost-effective digital photography technology can restore texture (section 2.1); (2) dense point cloud reconstruction using a Strcture from
colours very well and has high scanning accuracy. However, the pro­ Motion - Multi-View stereo (SfM-MVS) workflow (section 2.2); (3) point
posed workflows in the above papers are all operated in the laboratory, cloud segmentation (section 2.3); (4) tilt correction of point cloud seg­
and a complex set of equipment is required to ensure sufficient image ments (section 2.4); (5) noise filtering of point cloud segments (section
overlap and strict lighting conditions (Oglesby et al., 2017; Sirkiä et al., 2.5); and (6) resampling of point cloud segments (section 2.6). The
2016; Wernecke and Marsch, 2015). Uotinen et al. (2019) even used a photogrammetric workflow for rock joint surfaces is shown in Fig. 4.
motorized slider track to control the photographing process to obtain
better results. However, these approaches are difficult to carry out 2.1. Photographing process in the field
effectively in the field. As has become clear in previous literature
(Carrivick et al., 2016; Furukawa and Ponce, 2009), traditional photo­ As has become clear through the previous literature (Carrivick et al.,
grammetric methods require precise knowledge of the 3D location and 2016; Furukawa and Ponce, 2009), SfM-MVS depends entirely on the
pose of the camera and the precise 3D location of a series of control input of images taken from many viewpoints. In cases where the object
points in the scene. More flexible photogrammetric methods have since of interest is planar (e.g., an exposed rock joint surface), adjacent images
emerged with softcopy photogrammetry now used widely by the geo­ are taken by moving along the feature under near-parallel viewing
science community. A set of images can be used to reconstruct 3D scene conditions, with small translation movements and large overlaps. There
geometry where the extrinsic and intrinsic calibration parameters are is a requirement for near-parallel stereopairs of images with approxi­
unknown. Therefore, the existing photogrammetric workflow can be mately 60% overlap.
further improved. In practice, the camera is moved according to an array scanning
As an improvement of the existing methods, this work presents a frame (Fig. 1b), and an image sequence directly opposite the joint sur­
photogrammetric workflow in the field for the construction of a 3D rock face is collected. In the scanning frame, the calibration board is placed at
joint surface database. This approach produces digital models of the centre of the joint surface. Then, a local coordinate system is
equivalent or higher quality than previous models, allowing advanced established by using the ground control points on the four corners of the
and automated rock joint roughness analyses from a 3D perspective. The calibration board (Fig. 1a), where Point 3 is the origin of the co­
proposed method can be used to process the joint surfaces into standard ordinates; Point 4 and Point 1 are in the positive directions of the x-axis
data sets automatically; and construct a digital rock joint library with and y-axis, respectively; and the z-axis is outward. Therefore, the cali­
abundant sample types and sufficient sample numbers. To illustrate the bration board is in the xOy plane.
capability of the proposed framework, two rock joint databases con­ In SfM-MVS work, low-quality images are often ineffective, and
taining 310 samples are established, and two applications of the data­ adjusting appropriate camera parameters is necessary to obtain a clearer
bases are implemented. image with a larger depth of field. Many quality control measures for
photographing have been recorded in the previous literature (Oglesby
2. Description of photogrammetric workflow et al., 2017; Sirkiä et al., 2016; Uotinen et al., 2019; Wernecke and
Marsch, 2015). In the proposed approach, the distance of the camera
Based on the well-established photogrammetric measurement tech­ from the surface is controlled at approximately 0.5 m, roughly the length
nology, a digital camera and a calibration board (Fig. 1a) are needed for of an adult’s forearm. The camera mode is set to manual, and makes full
the proposed photographing process in the field. The photogrammetric use of the depth of field under a fixed focal length to ensure that the
measurement of rock joints in the field can make full use of natural objects are acceptably sharp.
lighting conditions, so there is no need to build a photo studio, which A simple trick controls the shooting positions along the moving
greatly improves work efficiency. The aforementioned devices are small route. The 3 × 3 square grid lines on the camera screen should be
in size. Researchers working in the field can easily complete the task of enabled during shooting. The image is equally divided into three parts

2
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 2. SfM-MVS workflow: (a) Estimation of camera position and pose; (b) Sparse point cloud; (c) Dense point cloud: deleted segments containing the calibration
board and incomplete segments on the dense point cloud boundaries.

by grid lines either in the horizontal or vertical directions (Fig. 1c). Each vertices. In other words, the meshing algorithm destroys the integrity of
part occupies 1/3 of the screen. Under the condition that the distance of the original data. Therefore, the SfM-MVS-derived dense point cloud is
the camera from the surface is approximately 0.5 m, the camera is the input for post-processing.
moved by 1/3 of the screen after each shot. In this way, the overlap
between adjacent images is theoretically 2/3. Considering the instability 2.3. Dense point cloud segmentation
of hand-held photographing, the overlap can reach 60% (Zhu, 2017; Zhu
et al., 2016). Scanning line by line or scanning column by column is The SfM-MVS-derived dense point cloud is seamlessly segmented
acceptable until the entire surface is covered. into fragments containing 200,000–300,000 points with a projection
size of 100 mm × 100 mm (Fig. 2c), which can be beneficial for appli­
2.2. Dense point cloud reconstruction using SfM-MVS workflow cation to the rock joint surface database. Fragments on dense point cloud
boundaries are incomplete in size and can be automatically filtered out
The 3D scene geometry of rock joint surfaces is reconstructed using by setting a point number threshold. Due to the established local coor­
the SfM-MVS workflow (Carrivick et al., 2016; Furukawa and Ponce, dinate system, the point cloud segment containing the calibration board
2009). The image sequence solving work is completed on the computer can also be automatically filtered out through the coordinate index. In
to obtain a sparse point cloud (Fig. 2b). Estimation of the relative camera the proposed approach, 20–30 valid point cloud segments can be ob­
position and pose is performed automatically with the SfM technique tained in one scan. The file is exported to the Stanford Triangle Format
(Fig. 2a-2b), embedded in the recently launched (commercial) computer (*.ply) for storage and is further resampled into a final standard data set.
vision Agisoft PhotoScan software v. 1.4.3, available at http://www.ag
isoft.com. After this step, a dense point cloud can be reconstructed 2.4. Tilt correction of point cloud segments
(Fig. 2c) using MVS algorithms (Fig. 2b-2c). In the SfM-MVS workflow,
georeferencing and scaling of the point cloud are completed using When the image sequence is collected in the field, the dip angle of the
ground control points on the calibration board. calibration board cannot be precisely controlled to be consistent with
A triangular mesh can also be generated. However, according to the that of the joint surface. After the calibration board is scaled and aligned
research of Paixão et al. (2018), the adopted meshing algorithm has a to the horizontal position, the point cloud model of the joint surface is
great influence on the goodness-of-fit of the triangular mesh to the given tilted. The tilting of the point cloud should be corrected if the calculation
point cloud, which seems to either decrease or increase the number of of joint surface roughness is required. Before correcting the tilt of the

Fig. 3. Tilt correction of point cloud segments: (a) The spatial position relationship between point cloud segment and standard raster model; (b) Direct ICP alignment
without preliminary position correction leads to incorrect solution: point cloud segment and standard raster model are not aligned; (c) After the translation
correction, the ICP alignment result is correct: first, translation and second, rotation.

3
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 4. Photogrammetric workflow for rock joint surfaces.

model, a standard raster reference plane is constructed in the xOy plane, two point cloud models. Each point cloud segment has its own co­
which is the same for each point cloud segment. Then, an iterative ordinates and has a certain spatial distance from the reference point
closest point (ICP) algorithm is applied to minimize the separation dis­ cloud model (Fig. 3a). If the ICP alignment is performed directly without
tance between the point cloud segment and the raster reference plane preliminary alignment, the results tend to be wrong (Fig. 3b). Therefore,
(Besl and McKay, 1992). Point cloud segments are then realigned to the a translation matrix can be used to perform the initial alignment of the
same orientation and location in space; that is, the tilt is corrected. point cloud. The translation matrix can be expressed as [− min (x), − min
The ICP algorithm is sensitive to the initial relative positions of the (y)] T, where x and y are the coordinates of each point cloud segment.

4
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 5. Photogrammetric scanning of rock joint surfaces for the verification experiment: (a) The three points on each joint surface are used for precision comparison;
(b) The camera is moved according to the array scanning frame, and the image sequence is collected directly opposite the joint surface; (c) The dense point cloud is
seamlessly segmented into 12 point cloud segments, and the number of points in each point cloud segment is approximately 200,000-300,000.

Then, the ICP algorithm is performed in two steps: first, translation and controlling the two parameters ‘NumNeighbours’ K and ‘Threshold’ α
second, rotation (Fig. 3c). (Rusu et al., 2008). A point is considered an outlier if the mean distance
to its K-nearest neighbours is greater than the specified threshold μ + αδ.
2.5. Noise filtering of point cloud segments
2.6. Resampling of point cloud segments
When the image sequence is collected, the light field, the tempera­
ture, and the reflection of light from the grains in the joint surface Referring to Barton’s ten standard JRC profiles, the point cloud
introduce noise into the aforementioned workflow. By traversing the segments are resampled into the standard raster data set using the
points on the segmentation point cloud, most noise can be eliminated by nearest-neighbour algorithm, which is a uniform grid with a sampling

Fig. 6. Post-processing of point cloud segment A-1: (a) Tilt correction of point cloud segment; (b) Noise filtering of point cloud segment (K = 4, α = 3.0). Point cloud
segment after applying gross outlier removal (points remaining: 2,932,728 out of 2,957,981, i.e., 99.15%); (c) Interpolated standard raster model; (d) Interpolated
point cloud model (with texture).

5
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 7. White light scanning process.

interval of 0.5 mm for both the x-axis and the y-axis. The grid is encoded For validation, 12 joint surface samples are first reconstructed by
by a triangular unit with a projection size of 100 mm × 100 mm and a photogrammetry-based scanning and then placed on the turntable one
total of 201 × 201 = 40,401 nodes. by one for white light scanning. The model, as presented on the com­
puter screen (see Fig. 7), shows that the morphological features of the
3. Verification experiment joint surface obtained by scanning are distinguishable, but the colour
distortion is serious, which is consistent with the conclusion reported by
3.1. Photogrammetric scanning of 12 rock joint surfaces Jiang et al. (2016).

To facilitate the subsequent verification of scanning accuracy, a rock 3.3. Precision comparison with white light scanning without post-
with a joint surface structure is cut into 12 independent blocks of 100 processing
mm × 100 mm × 50 mm, as shown in Fig. 5b. On the surface of each
block are three marked points with an inner diameter of 4 mm and an The accuracy of the white light scanning is up to 0.1 mm, which
outer diameter of 6 mm. The three points are triangularly distributed meets the requirements of a 0.5 mm sampling interval. However, since
(Fig. 5a) and numbered clockwise (− 1, − 2 and − 3). The lengths of the white light scanning is restricted with respect to the size of objects and
line segments (L12, L23 and L31) are used for scanning accuracy analysis. cannot realize batch processing, improving its scanning efficiency is
The digital camera used in the photogrammetric scanning is a Canon 7D difficult, and it is not suitable for large-scale scanning in the field. In this
Mark II 20.20 MPixel (APS-C format sensor: 22.4 × 15.0 mm) single lens respect, the photogrammetry method has great advantages. As long as
reflex (SLR) digital camera equipped with a Canon EF-S 18–135 mm f/ the scanning has an accuracy comparable to that of a white light scan,
3.5–5.6 IS STM Lens and digital shutter. The calibration board contains the aforementioned workflow can be used to build a 3D rock joint
169 marker points spaced 12.6 mm, with a length of 151 mm and a database in batches.
width of 151 mm. After scanning by the two methods, 24 point cloud models of 12 joint
After photographing along the scanning route at different positions, surface samples are obtained. In Fig. 8, the photogrammetry-based
the SfM-MVS-derived dense point cloud of the joint surface is segmented models (dense point clouds) and white light scanning models are
into 16 fragments. The 12 valid rock joint samples used for demon­ compared using the free version of the reverse engineering software
stration are labelled thereafter, as shown in Fig. 5c. The aforementioned Geomagic Studio 12, available at www.geomagic.com. For each sample,
three post-processing steps for the A-1 sample are shown in Fig. 6. Note the surfaces are registered, and indicators to analyse the deviation be­
that since the photogrammetry method retains the original colour of the tween objects are then calculated. Registration here is the rigid trans­
rock joint surface, the texture of the model can also be resampled formation of the photogrammetry-based model, taking the white light
(Fig. 6d). scanning model as a reference, so that identical regions are made to
coincide.
3.2. White light scanning of 12 rock joint surfaces A deviation value is the shortest distance from one point to a com­
parable point on the other object. In Fig. 8, colour scales of deviation
White light scanning is a well-established scanning technology for values are presented next to the point clouds. Reddish colours denote
the 3D reconstruction of various industrial objects. In this study, the rock outward differences, while blueish colours denote inward differences.
joint models obtained from white light scanning are employed as the The colour of most areas is green, indicating that the deviations are
benchmarks to verify the validity of the proposed photogrammetry- small. Table 1 presents additional information regarding the variation
based joint reconstruction method. The adopted white light scanning between point clouds (maximum distance, average distance, and stan­
equipment, named CAT-1, is a high-precision desktop white light 3D dard deviation). Observing Table 1 indicates that more than 50% of the
scanner with a single scanning accuracy up to 0.1 mm and a maximum surface area shows differences smaller than 0.01 mm, and the average
scanning size of 240 × 240 × 240 mm3 (see Fig. 7). When the devices are standard deviation of the defferences is approximately 0.016 mm. The
installed, the camera is fixed 0.5 m from the turntable to ensure that the value of maximum distance has a large dispersion among the samples.
object appears within the range of the focal length. Before scanning, the The data fluctuate between 0.180 mm and 1.334 mm, and the average
calibration of the system is carried out through the calibration board to value is 0.494 mm. Note that the maximum distance is greatly affected
ensure the dimensional accuracy of the model. When scanning, the by the noise of the model and might be improved further.
turntable automatically rotates by ±30◦ at equal time intervals. Then, Next, the three marked points on each joint surface are considered.
the camera scans each side and displays the point cloud model on the The lengths of the three line segments (L12, L23, and L31) are calculated,
computer in real time. and the absolute length deviations (D12, D23, and D31) between the two

6
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 8. Point cloud deviations between white light scanning and photogrammetry.

7
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 9. Location and description of joint surfaces digitized in-situ: (a)-(b): Overall and closeup photographs showing the rock joint surfaces exposed in the rock cut at
Narvik; (c)-(d): Overall and closeup photographs showing fresh rock joint surfaces exposed by the landslide in Libo County; (e)-(f): Photographs showing the real
scenarios of photographing the rock joint surface in the field (left: Narvik; right: Libo).

scanning models are shown in the right column of Table 1. The data are 3.4. Analysis of precision loss in post-processing
stable at 0.1 mm and have high accuracy. Generally, the performance of
the photogrammetry-based method is considered as good as that of The post-processing of the joint surface model includes two main
white light scanning: sub-millimetre accuracy is achieved during batch steps: noise filtering and point cloud resampling. Following the method
processing. The accuracy reported in this study is within the limits re­ in section 3.3, a precision comparison of the point cloud model after
ported by Paixão et al. (2018). post-processing is carried out. After comparison with the white light

Fig. 10. Typical samples of rock joint surface from two sites: (a)-(c) Narvik samples; (d)-(f) Libo samples.

8
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 11. Illustration of the roughness evaluation method based on triangulated 3D joint surface: (a) Schematic diagram illustrating the geometric definition of
azimuth, dip and apparent dip in relation to the selected analysis direction; (b) Potential contact micro-elements in the shear direction 0◦ ; (c) Fit of Grasselli method
(Tatone, 2009).

scanning model, the three deviation values are presented in Table 2. The due to the identification and deletion of outliers in the point cloud
values within and outside the parentheses correspond to the data before model. In addition, the values of the average distance and standard
and after post-processing, respectively. deviation are increased by more than ten times. This result is mainly the
The maximum distance of most joint surfaces can be seen to have effect of the significant reduction in the number of points after post-
decreased after post-processing (except for B-2 and C-1), which is mainly processing. Dense point clouds decrease the average value of the data.

Fig. 12. Scatter plots between Yi and Yj (horizontal axis = row variable, vertical axis = column variable).

9
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 13. Scatter plots between Xi and Xj (horizontal axis = row variable, vertical axis = column variable).

Overall, the deviation values after post-processing can better reflect the cracks. Six typical samples of rock joint surfaces from two sites are
true level of the scanning accuracy. presented in Fig. 10.
An alternative scanning frame for rock joint and the corresponding
analysis of precision loss in post-processing are presented in Appendix.
4.2. Roughness evaluation of a 3D rock joint surface

4. 3D rock joint database and its applications (1) Roughness evaluation based on Grasselli parameters.
Based on the undulating morphology of the triangulated joint surface
4.1. Constructing a 3D rock joint database from two sites and its shear failure mechanism, Grasselli and Egger (2003) proposed
the parameter θ*max/(C + 1) to evaluate the 3D roughness of the rock
The aforementioned photogrammetry-based procedure was applied joint surface. The relevant formulas are as follows:
to construct 3D rock joint databases, and 310 valid samples from two
sites were collected. Due to geographic differences, the samples were tanθ* = − tanθcosβ (1)
classified into two databases and named after the local place names: the
( * )C
Libo database (205 samples) and the Narvik database (105 samples). θ*max − θ
Aθ * = A0 (2)
Libo County is located in Guizhou Province, Southwest China (Fig. 9d-9e θ*max
and Fig. 9f). A large-scale landslide geological disaster occurred in 2017,
and many fresh rock joints were exposed (Zhao et al., 2019). The sliding where θ is the true dip angle of triangulated micro-element; θ* is the
body is moderately weathered limestone, with a layer thickness of 0.8 m. corresponding apparent dip angle; β is the angle measured clockwise
Narvik is a northern Norwegian city. On the north bank of the Harrow between the selected analysis direction and the true dip direction of the
Grand Bridge, numerous rock joints (Fig. 9b-9c and Fig. 9e) are exposed triangulated micro-element (Fig. 11a); θ*max is the maximum apparent
on both sides of the E6 road, and the lithology is granite. After a long dip angle of the surface in the chosen analysis direction; Aθ* is the area
period of exposure, moss began to grow on the rock surface, as well as in ratio of the specific fraction in the shear direction to the total surface

10
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 14. Polar plots of the 3D Grasselli roughness values for the six typical rock joint surfaces with varying shearing directions: (a)-(c) Narvik samples; (d)-(f)
Libo samples.

(Fig. 11b), describing the part that is more steeply inclined than pro­ θ*max/(C + 1) is, the rougher the joint surface.
gressively higher angular thresholds [θ*] (such as 0, 5, 10, …, θ*max); Later, Cai (2018) proposed a similar parameter, the projection area
and A0 is the Aθ* value when θ* = 0. The Aθ* vs. θ* curve is plotted, percentage (PAP), to evaluate the 3D roughness of the rock joint surface,
(Fig. 11c) and C is a dimensionless fitting parameter, calculated via a which can be obtained using Eq. 3.
non-linear least-squares regression analysis; the larger the value of

Fig. 15. Box plots of the anisotropy ellipse parameters of θ*max/(C + 1) for the two rock joint surface databases.

11
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Fig. 16. An alternative array scanning frame for a rock joint.

∑n
1 Ai
′
(0◦ ≤ θs < 90◦ ) is calculated according to Eq. 7. The inclination angle αk
PAP = × 100% (3) is defined as the angle between the normal vector of the triangulated
Ah
micro-element and the vertical or z-axis. The surface relative roughness
where A’i is the projection area of the i-th potential contact micro- coefficient Rs is defined in Eq. 8 as the ratio between the actual area At,
element on the plane, which is perpendicular to the shear direction and the nominal area An. The nominal area is the projection of the rock
(Fig. 11b); Ah is the projection area of the whole joint surface on the joint surface on its regression mean plane (which is parallel to the
horizontal plane; and n is the total number of potential contact micro- horizontal plane xOy in this study).
elements in the shear direction.
1 ∑m
(2) Roughness evaluation based on fractal parameters. θs = (αk )i (7)
In Tatone’s work (Tatone, 2009), the roughness-length method was m i=1
applied to calculate the fractal parameters of a 3D surface. Log-log plots
of S(w) versus w are constructed using nine local window sizes typically Rs = At /An with 1 ≤ Rs (8)
between 2% and 10% of the size of the surface to be analysed. Then, S(w)
is calculated and plotted versus w to estimate fractal parameters A and D 4.3. Application to evaluation of rock joint roughness
(Eq. 4 and Eq. 5).
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ The two established joint surface databases are employed to analyse
1 ∑nw
1 ∑( )2
the roughness characteristics of rock joint surfaces. Of the seven 3D
S(w) = zj − z (4)
nw i=1 mi − 2 j∈wi roughness parameters in section 4.2, PAP and θ*max/(C + 1) are related
to the shear direction. Here, the shear direction is uniformly set to the
S(w) = Aw3− D
(5) 0◦ direction (Fig. 11b). The analysis of the anisotropy of roughness pa­
rameters is covered in the following sections. The remaining five pa­
where nw is the total number of local sampling windows of side length w; rameters consider the roughness of the joint surface from statistical
mi is the number of points in the i-th window; zj is the residual surface perspectives (Rs, θs and Z2s) and fractal perspectives (A and D). After
height of each measurement point in the i-th window; − z is the mean calculation, 1270 multivariate data points were obtained at the two
residual surface height in the i-th window; A is the amplitude parameter; sampling sites.
and D is the fractal dimension. In general, the roughness parameters of the rock joint surfaces at
(3) Roughness evaluation based on traditional statistical parameters. different locations show a certain degree of fluctuation, and the values
According to the work of Belem et al. (2000), three roughness pa­ calculated by different roughness parameters also vary, which creates
rameters Rs, θs and Z2s are defined. The surface root mean square new problems for the roughness level decision. However, correlations
parameter Z2s can be obtained using Eq. 6 and represents the root mean between two roughness parameters are likely to be less site-specific than
square of the slopes of the triangulated micro-elements that make up the the statistics for single roughness parameters. In this regard, the multi­
entire surface. variate model proposed by Phoon and Ching (2013) can be used to
analyse the aforementioned seven 3D roughness parameters. The cor­

⎧ ⎡ ( )2 ( )2 ⎤ ⎫1/2
x− 1 N y− 1
⎪
⎪
⎪ 1 N∑ ∑ zi+1,j+1 − zi,j+1 + zi+1,j − zi,j ⎪
⎪
⎪
⎪
⎪ ⎢ 2 ⎥⎪
⎪
⎨ ⎢ Δx i=1 j=1 2 ⎥⎪⎪
1 ⎢ ⎥⎬
Z2s = ( )⎢
⎢ ( )2 ( )2
⎥
⎥⎪ (6)
⎪
⎪(Nx − 1) Ny − 1 ⎢ Nx − 1 Ny − 1 ⎥⎪
⎪
⎪
⎪ ⎣ + 1 ∑ ∑ zi+1,j+1 − zi+1,j + zi,j+1 − zi,j ⎦⎪⎪
⎪
⎪
⎩ ⎪
⎭
Δy2 i=1 j=1 2

where Nx is the number of points along the x-axis; Ny is the number of relation between different parameters is comprehensively judged from
points along the y-axis; Δx and Δy are the sampling steps along the x- the perspective of multivariate analysis.
and y-axis; and zi,j = z(xi, yi). Phoon’s multivariate model builds a more general Johnson system,
The mean three-dimensional inclination angle of the entire surface θs including three distribution types: SU, SB, and SL. A percentile-based

12
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Table 1 Table 3
Quantitative data on point cloud deviations before post-processing. Distribution type and distribution parameters for the seven roughness
Joint Maximum Average Standard D12 D23 D31
parameters.
Distance Distance Deviation (mm) (mm) (mm) Random Roughness Distribution Distribution parameters
(mm) (mm) (mm) variable parameters type
aX bX aY bY
A-1 0.413 0.003 0.019 0.0 0.0 0.1
A-2 0.256 0.002 0.011 0.0 0.0 0.1 Y1 PAP SB 0.821 0.611 0.193 0.051
A-3 0.833 0.002 0.017 0.1 0.2 0.1 Y2 θ*max/(C + SB 0.955 0.687 23.102 5.245
A-4 0.327 0.003 0.016 0.2 0.2 0.4 1)
B-1 0.618 0.007 0.037 0.1 0.0 0.0 Y3 Rs SB 0.828 1.125 0.346 1.017
B-2 0.332 0.002 0.011 0.3 0.1 0.3 Y4 θs SB 0.961 0.418 26.969 8.222
B-3 1.334 0.003 0.025 0.0 0.0 0.1 Y5 Z2s SB 0.904 0.738 0.857 0.186
B-4 0.184 0.002 0.010 0.1 0.2 0.2 Y6 A SB 1.443 2.193 0.172 0.012
C-1 0.219 0.002 0.011 0.1 0.1 0.0 Y7 D SU 1.750 − 0.694 0.184 2.112
C-2 0.180 0.002 0.010 0.0 0.2 0.1
C-3 0.880 0.002 0.014 0.1 0.1 0.1
C-4 0.348 0.002 0.011 0.0 0.1 0.0 classified into the distribution SB.
Average 0.494 0.002 0.016 0.1 Fig. 13 presents the pairwise correlation structure underlying the
seven roughness parameters after they have been transformed into
standard normal random variables X1, X2, X3, X4, X5, X6 and X7. Then,
method rather than a moment-based method is applied to identify the
these pairwise correlations can be quantified using Pearson correlation
distribution type of each parameter. Once the distribution type has been
coefficients (results shown in Table 4). This step is easily carried out in
identified, the four distribution parameters (aX, bX, aY, and bY) can be
MATLAB using the Corrcoef function. The correlation coefficients
computed. The distribution type and distribution parameters for random
among the parameters PAP, θ*max/(C + 1), Rs, θs, and Z2s are all greater
variables Y1 = PAP, Y2 = θ*max/(C + 1), Y3 = Rs, Y4 = θs, Y5 = Z2s, Y6 =
than 0.950, indicating that the correlations are strong. The correlation
A and Y7 = D are summarized in Table 3. Transforming the roughness
coefficients between the fractal parameter A and the aforementioned
parameters into standard normal random variables X1, X2, X3, X4, X5, X6,
five parameters are between 0.750 and 0.800, indicating that the cor­
and X7 is easy using Eq. 9. The calculation of the parameters for each
relations are relatively weak. The absolute value of the correlation co­
type of distribution is different, and there are many calculation for­
efficients between the fractal parameter D and the other parameters is
mulas, which are recorded in detail in Phoon’s work (Phoon and Ching,
approximately 0.300, and most of them are negative values. These
2013).
values imply that there are no correlations. In fractal theory, D is the
⎧
⎪ Yi − bY fractal dimension, which is a key parameter. This result shows that
⎪ Yn =
⎪
⎪
⎪ aY fractal theory is very different from Grasselli parameters and traditional
⎪
⎪
⎪
⎪
⎨ Xi = bX + aX sinh− 1 (Yn ) Yi ∈ SU statistical parameters in describing the rock joint roughness. Roughness
( ) (9) evaluation of rock joint surfaces cannot be applied only unilaterally with
⎪
⎪
⎪
⎪ Xi = bX + aX In
Yn
Yi ∈ SB fractal parameters.
⎪
⎪ 1 − Yn
⎪
⎪
⎪
⎩
Xi = bX + aX In(Yn ) Yi ∈ SL 4.4. Application to investigate the anisotropy of rock joint roughness

Scatter plots between Yi and Yj are shown in Fig. 12. This figure il­
The two rock joint databases are further used to analyse the anisot­
lustrates qualitatively that the correlations between the parameters PAP,
ropy of surface roughness (Bao et al., 2020), that is, to calculate the
θ*max/(C + 1), Rs, θs, and Z2s are strong, and each shows a positive
parameters PAP and θ*max/(C + 1) from multiple shear directions. More
correlation in different distribution forms. From a statistical perspective,
specifically, adopting a counter-clockwise angular convention, in which
they are more like parameters of the same type. The positive correlations
the positive y-direction is considered 0◦ (Fig. 11), the values of PAP and
between the fractal parameter A and the aforementioned five parame­
θ*max/(C + 1) corresponding to each shear direction can be displayed on
ters are relatively weak, while another fractal indicator D shows states
a polar plot to visualize the anisotropy in surface roughness (Fig. 14).
that are almost unrelated to the other parameters. This contrast may
Fig. 14 illustrates that the roughness values in various directions
explain what is presented in Table 3, where the fractal parameter D is
approach the elliptical contour in polar coordinates. The red contour is
classified into the SU distribution and the remaining parameters are
contained within the black contour, indicating that the values of the
parameter θ*max/(C + 1) are slightly larger than those of the parameter
PAP. However, the two roughness parameters exhibit nearly the same
Table 2
anisotropic characteristics, which confirms the reliability of the calcu­
Quantitative data on point cloud deviations after post-processing.
lations. Further, the roughness of each rock joint surface can be objec­
Joint Maximum Distance Average Distance Standard Deviation tively evaluated based on the size of the contour. The order of roughness
(mm) (mm) (mm)
for the Narvik samples is (a) < (c) < (b), which corresponds to the
A-1 0.186 (0.413) 0.014 (0.003) 0.023 (0.019) morphological features of the rock joint surface shown in Fig. 10.
A-2 0.234 (0.256) 0.014 (0.002) 0.022 (0.011)
Sample (a) is the smoothest, while sample (b) is the roughest due to the
A-3 0.418 (0.833) 0.015 (0.002) 0.031 (0.017)
A-4 0.287 (0.327) 0.020 (0.003) 0.028 (0.016)
B-1 0.366 (0.618) 0.017 (0.007) 0.028 (0.037) Table 4
B-2 0.380 (0.332) 0.016 (0.002) 0.028 (0.011) Pearson correlation matrix for X1, X2, X3, X4, X5, X6, and X7.
B-3 1.239 (1.334) 0.021 (0.003) 0.068 (0.025)
B-4 0.154 (0.184) 0.013 (0.002) 0.019 (0.010) X1 X2 X3 X4 X5 X6 X7
C-1 0.291 (0.219) 0.016 (0.002) 0.026 (0.011) X1 1.000 0.978 0.970 0.966 0.961 0.779 − 0.226
C-2 0.139 (0.180) 0.013 (0.002) 0.018 (0.010) X2 1.000 0.977 0.977 0.968 0.774 − 0.237
C-3 0.577 (0.880) 0.015 (0.002) 0.025 (0.014) X3 1.000 0.991 0.994 0.776 − 0.274
C-4 0.331 (0.348) 0.013 (0.002) 0.021 (0.011) X4 1.000 0.975 0.779 − 0.256
Average 0.384 (0.494) 0.016 (0.002) 0.028 (0.016) X5 SYMM 1.000 0.765 − 0.278
X6 1.000 0.313
Note: The values within and outside the parentheses correspond to the data
X7 1.000
before and after post-processing, respectively.

13
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

obvious crack on the surface. The roughness of sample (c) is slightly It should be noted that this article mainly explores the batch scan­
lower than that of sample (b). The rock joint surface is darker in colour. ning workflow of a rock joint surface in the field and the construction of
A thin layer of moss growing on the surface of the joint seems to have the corresponding database. The comparison between different joint
reduced its roughness. The Libo samples come from the fresh rock sur­ roughness parameters is only the first application to demonstrate the
face exposed by the landslide, so there are fewer factors affecting the multi-functional use of the database. Therefore, the results of this
roughness. The order is (d) < (e) < (f), which also corresponds to the research do not indicate which parameter is closer to the true joint
topographic features of the sample surface. roughness value. An in-depth study of the rock joint roughness needs to
According to Belem’s research (Belem et al., 2000), the roughness be combined with a shear test.
values can be fitted by an ellipse whose major and minor axes are clearly The fusion of morphological and textural features of rock joints and
distinguishable. By referring to Fig. 14a, the half-large axis R, the half- the corresponding application research are the focus of further devel­
small axis r, the principal direction angle of anisotropy φ and the opment. Relevant research (Chatterjee, 2013; Lepistö, 2005; Momma
anisotropy ratio λa are calculated using Eq. 10. The results of six typical et al., 2006) shows that the texture of rock joints has strong direction­
samples show that the principal direction angle of the samples from ality and that the granularities and colours of various lithologies are
Narvik is approximately 33◦ , while the angle of the samples from Libo is usually different. The colour space of rocks is suitable for describing the
approximately 55◦ . These angles are obviously site-specific results. It is weathering degree of joints. Tools such as support vector machines and
also worth noting that the orders of the anisotropy coefficient are (c) < neural networks can be used to classify rock types and weathering de­
(b) < (a) and (f) < (e) < (d), which is almost the opposite to the law of grees by the feature parameters of colour vectors and texture.
roughness, and the crack on the surface of sample (b) does not increase The rock joint surface database in this article can also be extended
its anisotropy. This situation implies that the source of the difference in using 3D printing technology. Artificial rock joint samples with real
the anisotropy coefficient is complicated. morphological features can be reproduced and tested in the laboratory.
⎧ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ Problems such as difficulties with the field test, high transportation
⎪
⎪
⎪
⎪ R = (XR )2 + (YR )2 costs, and sample damage can be avoided. In addition, a reconstructed
⎨ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
joint model can be applied to discrete element numerical simulation to
r = (Xr )2 + (Yr )2 (10)
⎪
⎪
⎪ φ = atan(YR /XR ) analyse the failure mechanism of joints in the process of shearing and
⎪
⎩
λa = R/r establish the relationship between the meso-morphological structure
and meso-mechanical characteristics of a rock joint.
Furthermore, box plots of the anisotropy ellipse parameters of θ*max/
(C + 1) are presented in Fig. 15 for all the samples from the two rock 6. Conclusions
joint surface databases. In this way, the roughness and anisotropy of the
two rock joint databases can be comprehensively evaluated. On the The authors demonstrate a practical photogrammetric workflow for
whole, the roughness of the Narvik sample is significantly smaller than the scanning of 3D rock joint surfaces: (1) photographing process in the
that of the Libo sample, but the anisotropy coefficient shows the oppo­ field; (2) dense point cloud reconstruction using a SfM-MVS workflow;
site rule. The principal direction angle of anisotropy for the Narvik (3) point cloud segmentation; (4) tilt correction of point cloud segments;
sample is also slightly larger than that of the Libo sample. These con­ (5) noise filtering of point cloud segments; and (6) resampling of point
clusions are consistent with that for the six typical samples, indicating cloud segments. Innovations in some steps make it practical for field
that the samples in section 4.1 are representative. scenarios. Through analysis, the following main conclusions can be
drawn.
5. Discussion (1) The proposed workflow has high operability and images can be
collected by hand-held photographing, which is suitable for batch
The proposed photogrammetric scanning workflow in the field is a scanning of rock joint surfaces in the field;
practical method to construct a joint surface database with low cost and (2) The standard raster data set of the 3D rock joint surface is well
high efficiency. Innovations in some steps of the proposed workflow established. All samples have realistic colourful textures. The perfor­
make it practical for field scenarios: image collection procedures in the mance is considered as good as that of white light scanning, and sub-
field and post-processing steps of the SfM-MVS-derived dense point millimetre accuracy is achieved during batch processing;
cloud. It solves the “inflexible field operation” problem in existing (3) Two rock joint databases containing 310 samples are established,
contact type data collection procedures and other non-contact type data and two application examples show very good potential of the proposed
collection procedures. Through batch processing, scanning efficiency photogrammetric workflow in the evaluation of rock joint roughness
can be further improved and it has been verified to have sub-millimetre and rock joint anisotropy.
accuracy.
The proposed workflow also has good colour reproduction capabil­ Data availability
ities. In the scanning of joint surface texture, the colour distortion
collected by structured light scanning is severe. In addition, structured Datasets related to this article can be found at https://data.mendele
light scanning cannot reconstruct the objects with black and white y.com/datasets/zmw8zc766s/draft?a=c759b2db-7366-4991-9792-8
surfaces (Jiang et al., 2016), so scanning certain grain features on the 28af4a1979b.
joint surface is impossible, which may cause local holes in the joint
surface model. Therefore, the samples that can be scanned by the pro­ Disclosure statements
posed workflow are more complete.
To facilitate the description of the photogrammetry-based scanning The authors declare that they have no conflict of interest.
workflow, a standard raster model is defined, with a length of 100 mm
and a sampling interval of 0.5 mm. In practice, the sampling interval of Author statement
the model can be adjusted according to requirements. The scanning
process described above is universal. Even the size of the array scanning Zhao Lianheng and Huang Dongliang designed the improved
frame can be expanded when necessary. On the basis of the current photogrammetric workflow and the experimental validation. Chen Jin­
scanning frame with a one-ring neighbourhood, the radius of the gyu and Wang Xiang conducted two applications of the proposed rock
neighbourhood can be recursively enlarged to the K-neighbourhood, joint surface database. Li Dejian and Luo Wei carried out the imple­
which will double the scanning efficiency. mentation. Zhao Lianheng and Huang Dongliang wrote the manuscript

14
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Table 5 calibration board is higher. From this point of view, there should be
Quantitative deviations of mark points for each joint piece. different scanning accuracies between columns within the sheet scan­
Joint D12 (mm) D23 (mm) D31 (mm) ning frame. However, this trend is not seen in the data in Table 5. This
result may occur because that the cumulative error of the distance is
A-1 0.0 (0.0) 0.0 (0.0) 0.0 (0.1)
A-2 0.1 (0.0) 0.0 (0.0) 0.0 (0.1) small at this scanning scale and can be ignored; thus, the scanning re­
A-3 0.0 (0.1) 0.1 (0.2) 0.1 (0.1) sults do not show the difference. This outcome shows that properly
A-4 0.0 (0.2) 0.0 (0.2) 0.2 (0.4) controlling the scanning scale of the joint surface is necessary. In the
B-1 0.0 (0.1) 0.1 (0.0) 0.1 (0.0) case of large-scale scanning, the accuracy of the ring scanning frame
B-2 0.2 (0.3) 0.1 (0.1) 0.2 (0.3)
B-3 0.1 (0.0) 0.1 (0.0) 0.1 (0.1)
should be higher than that of the sheet scanning frame.
B-4 0.0 (0.1) 0.1 (0.2) 0.1 (0.2)
C-1 0.2 (0.1) 0.1 (0.1) 0.1 (0.0) References
C-2 0.2 (0.0) 0.4 (0.2) 0.1 (0.1)
C-3 0.3 (0.1) 0.2 (0.1) 0.2 (0.1) Bae, D., Kim, K., Koh, Y., Kim, J., 2011. Characterization of joint roughness in granite by
C-4 0.1 (0.0) 0.1 (0.1) 0.1 (0.1) applying the scan circle technique to images from a borehole televiewer. Rock Mech.
Average 0.1 (0.1) Rock. Eng. 44, 497–504. https://doi.org/10.1007/s00603-011-0134-9.
Bahaaddini, M., Hagan, P.C., Mitra, R., Khosravi, M.H., 2016. Experimental and
Note: The values within and outside the parentheses correspond to the data of numerical study of asperity degradation in the direct shear test. Eng. Geol. 204,
the former and the latter array scanning frame, respectively. 41–52. https://doi.org/10.1016/j.enggeo.2016.01.018.
Bao, H., Zhang, G., Lan, H., Yan, C., Xu, J., Xu, W., 2020. Geometrical heterogeneity of
the joint roughness coefficient revealed by 3D laser scanning. Eng. Geol. 265,
with input from all authors. Zuo Shi and Zhu Zhiheng conceived the 105415. https://doi.org/10.1016/j.enggeo.2019.105415.
study and were in charge of overall direction and planning. Barton, N., Choubey, V., 1977. The shear strength of rock joints in theory and practice.
Rock Mech. Felsmechanik M canique des Roches 10, 1–54. https://doi.org/10.1007/
BF01261801.
Declaration of Competing Interest Belem, T., Homand-Etienne, F., Souley, M., 2000. Quantitative Parameters for Rock Joint
Surface Roughness. Rock Mech. Rock. Eng. 33, 217–242. https://doi.org/10.1007/
s006030070001.
We declare that we have no financial and personal relationships with Belem, T., Souley, M., Homand, F., 2007. Modeling surface roughness degradation of
other people or organizations that can inappropriately influence our rock joint wall during monotonic and cyclic shearing. Acta Geotech. 2, 227–248.
work, there is no professional or other personal interest of any nature or https://doi.org/10.1007/s11440-007-0039-7.
Besl, P.J., McKay, N.D., 1992. A method for registration of 3-D shapes. IEEE Trans.
kind in any product, service and/or company that could be construed as Pattern Anal. Mach. Intell. 14, 239–256. https://doi.org/10.1109/34.1217917.
influencing the position presented in, or the review of, the manuscript Cai, Y., 2018. Methods for Estimating the Roughness and the Peak Shear Strength of Rock
entitled “A practical photogrammetric workflow in the field for the con­ Discontinuities. China University of Geosciences.
Carrivick, J.L., Smith, M.W., Quincey, D.J., 2016. Structure from Motion in the
struction of a 3D rock joint surface database”. Geosciences. WILEY Blackwell.
Chatterjee, S., 2013. Vision-based rock-type classification of limestone using multi-class
Acknowledgments support vector machine. Appl. Intell. 39, 14–27. https://doi.org/10.1007/s10489-
012-0391-7.
Chen, C., Zhang, L.M., Zhu, H., 2017. A photographic method for measuring soil
This study was financially supported by the National Natural Science deformations during internal erosion under triaxial stress conditions. Geotech. Test.
Foundation of China (Nos. 51878668; 51978666), the Guizhou Pro­ J. 41, 20170031. https://doi.org/10.1520/GTJ20170031.
Fardin, N., Stephansson, O., Jing, L., 2001. The scale dependence of rock joint surface
vincial Department of Science and Technology - Science and Technology roughness. Int. J. Rock Mech. Min. Sci. 38, 659–669. https://doi.org/10.1016/
Support Program Key Project (No. [2018]2815), the key engineering S1365-1609(01)00028-4.
science and technology projects of the Jiangxi provincial department of Furukawa, Y., Ponce, J., 2009. Accurate camera calibration from multi-view stereo and
bundle adjustment. Int. J. Comput. Vis. 84, 257–268. https://doi.org/10.1007/
transportation (Grant No. 2019C0011), the Fundamental Research
s11263-009-0232-2.
Funds for the Central Universities of Central South University (Nos. Grasselli, G., Egger, P., 2003. Constitutive law for the shear strength of rock joints based
2020zzts167, 2020zzts154, 2019zzts009). All financial supports were on three-dimensional surface parameters. Int. J. Rock Mech. Min. Sci. 40, 25–40.
greatly appreciated. And the authors are immensely grateful for the https://doi.org/10.1016/S1365-1609(02)00101-6.
Jiang, Q., Feng, X., Gong, Y., Song, L., Ran, S., Cui, J., 2016. Reverse modelling of natural
editor’s and reviewers’ constructive comments, which help to enhance rock joints using 3D scanning and 3D printing. Comput. Geotech. 73, 210–220.
the contents and the presentation of this article. https://doi.org/10.1016/j.compgeo.2015.11.020.
Jiang, Q., Yang, B., Yan, F., Liu, C., Shi, Y., Li, L., 2020. New method for characterizing
the shear damage of natural rock joint based on 3D engraving and 3D scanning. Int.
Appendix A. An alternative scanning frame for rock joint J. Geomech. 20, 1–15. https://doi.org/10.1061/(ASCE)GM.1943-5622.0001575.
Kulatilake, P.H.S.W., Balasingam, P., Park, J., Morgan, R., 2006. Natural rock joint
To satisfy different scanning conditions, another array scanning roughness quantification through fractal techniques. Geotech. Geol. Eng. 24,
1181–1202. https://doi.org/10.1007/s10706-005-1219-6.
frame is presented in Fig. 16. The original joint surfaces arranged in a Lepistö, L., 2005. Rock image classification using color features in Gabor space.
ring shape are rearranged into a sheet shape, and the calibration board is J. Electron. Imaging 14, 040503. https://doi.org/10.1117/1.2149872.
placed on the side of the object. The joint surfaces still use the original Liu, Q., Tian, Y., Liu, D., Jiang, Y., 2017a. Updates to JRC-JCS model for estimating the
peak shear strength of rock joints based on quantified surface description. Eng. Geol.
uniform numbering, which is convenient for comparing the accuracy 228, 282–300. https://doi.org/10.1016/j.enggeo.2017.08.020.
between the two scanning frames. Liu, X.G., Zhu, W.C., Yu, Q.L., Chen, S.J., Li, R.F., 2017b. Estimation of the joint
The absolute length deviations (D12, D23, and D31) of the three line roughness coefficient of rock joints by consideration of two-order asperity and its
application in double-joint shear tests. Eng. Geol. 220, 243–255. https://doi.org/
segments between the photogrammetry-based model and the white light 10.1016/j.enggeo.2017.02.012.
scanning model are shown in Table 5. The values within and outside the Liu, Q., Tian, Y., Ji, P., Ma, H., 2018. Experimental investigation of the peak shear
parentheses correspond to the data of the former and the latter array strength criterion based on three-dimensional surface description. Rock Mech. Rock.
Eng. 51, 1005–1025. https://doi.org/10.1007/s00603-017-1390-0.
scanning frames, respectively. The difference between the two data sets
MŁynarczuk, M., 2010. Description and classification of rock surfaces by means of laser
fluctuates in the range of 0.2 mm. The absolute value of the data is less profilometry and mathematical morphology. Int. J. Rock Mech. Min. Sci. 47,
than 0.4 mm. For the sampling interval of 0.5 mm, this accuracy is 138–149. https://doi.org/10.1016/j.ijrmms.2009.09.004.
sufficient. Therefore, the sheet scanning frame can be used as an alter­ Momma, E., Ono, T., Ishii, H., 2006. Rock classification by types and degrees of
weathering. In: 2006 SICE-ICASE International Joint Conference. IEEE, Busan,
native array scanning frame to meet the requirements of different pp. 149–152. https://doi.org/10.1109/SICE.2006.315390.
conditions. Oglesby, J., Hudyma, N., Brown, S., Bliss, A., Harris, A., 2017. Development and
In photogrammetry, the calibration board works well within only a assessment of a photogrammetry system for rock specimen surface characterization,
in: SoutheastCon 2017. IEEE, pp. 1–5. https://doi.org/10.1109/
certain spatial range. The scanning accuracy has a cumulative distance SECON.2017.7925384.
error in space. Therefore, the scanning accuracy of the area closer to the

15
Z. Lianheng et al. Engineering Geology 279 (2020) 105878

Paixão, A., Resende, R., Fortunato, E., 2018. Photogrammetry for digital reconstruction and Infrastructure Development- Proceedings of the 14th International Congress on
of railway ballast particles – a cost-efficient method. Constr. Build. Mater. 191, Rock Mechanics and Rock Engineering, ISRM 2019. Foz do Iguassu, Brazil,
963–976. https://doi.org/10.1016/j.conbuildmat.2018.10.048. pp. 461–468.
Phoon, K.-K., Ching, J., 2013. Multivariate Model for Soil Parameters Based on Johnson Wernecke, C., Marsch, K., 2015. Mapping rock surface roughness with photogrammetry.
Distributions, in: Foundation Engineering in the Face of Uncertainty. American ISRM Regional Symposium, EUROCK 2015, 1175–1180.
Society of Civil Engineers, Reston, VA, pp. 337–353. https://doi.org/10.1061/ Wu, Q., Xu, Y., Tang, H., Fang, K., Jiang, Y., Liu, C., Wang, L., Wang, X., Kang, J., 2018.
9780784412763.027. Investigation on the shear properties of discontinuities at the interface between
Rusu, R.B., Marton, Z.C., Blodow, N., Dolha, M., Beetz, M., 2008. Towards 3D Point cloud different rock types in the Badong formation. China. Eng. Geol. 245, 280–291.
based object maps for household environments. Rob. Auton. Syst. 56, 927–941. https://doi.org/10.1016/j.enggeo.2018.09.002.
https://doi.org/10.1016/j.robot.2008.08.005. Xia, C.-C., Tang, Z.-C., Xiao, W.-M., Song, Y.-L., 2013. New peak shear strength criterion
Singh, H.K., Basu, A., 2016. Shear behaviors of “real” natural un-matching joints of of rock joints based on quantified surface description. Rock Mech. Rock. Eng. 47,
granite with equivalent joint roughness coefficients. Eng. Geol. 211, 120–134. 387–400. https://doi.org/10.1007/s00603-013-0395-6.
https://doi.org/10.1016/j.enggeo.2016.07.004. Yang, Z.-Y., Taghichian, A., Huang, G.-D., 2011. On the applicability of self-affinity
Sirkiä, J., Kallio, P., Iakovlev, D., Uotinen, L., 2016. Photogrammetric calculation of JRC concept in scale of three-dimensional rock joints. Int. J. Rock Mech. Min. Sci. 48,
for rock slope support design. In: Proceedings of the 8th International Symposium on 1173–1187. https://doi.org/10.1016/j.ijrmms.2011.06.010.
Ground Support in Mining and Underground Construction, pp. 622–634. https://doi. Yang, J., Rong, G., Hou, D., Peng, J., Zhou, C., 2015. Experimental Study on Peak Shear
org/10.13140/RG.2.2.21173.06880. Strength Criterion for Rock Joints. Rock Mech. Rock. Eng. 49, 821–835. https://doi.
Tang, Z.C., Jiao, Y.Y., 2017. A self-consistent model with asperity interaction for the org/10.1007/s00603-015-0791-1.
mechanical behavior of rock joints under compressive loading. Int. J. Rock Mech. Yong, R., Fu, X., Huang, M., Liang, Q., Du, S.-G., 2018. A rapid field measurement
Min. Sci. 100, 23–27. https://doi.org/10.1016/j.ijrmms.2017.10.009. method for the determination of Joint Roughness Coefficient of large rock joint
Tang, Z., Xia, C., Song, Y., Fu, P., 2012. Analysis of peak dilation strength and model of surfaces. KSCE J. Civ. Eng. 22, 101–109. https://doi.org/10.1007/s12205-017-
peak dilation angle for artificial joint. Chinese J. Rock Mech. Eng. 31, 3038–3044. 0654-2.
Tatone, B.S., 2009. Quantitative Characterization of Natural Rock Discontinuity Zhao, L., Li, D., Tan, H., Cheng, X., Zuo, S., 2019. Characteristics of failure area and
Roughness in-Situ and in the Laboratory by. University of Toronto. failure mechanism of a bedding rockslide in Libo County, Guizhou, China. Landslides
Thirukumaran, S., Indraratna, B., 2016. A review of shear strength models for rock joints 16, 1367–1374. https://doi.org/10.1007/s10346-019-01188-6.
subjected to constant normal stiffness. J. Rock Mech. Geotech. Eng. 8, 405–414. Zheng, B., Qi, S., 2016. A new index to describe joint roughness coefficient (JRC) under
https://doi.org/10.1016/j.jrmge.2015.10.006. cyclic shear. Eng. Geol. 212, 72–85. https://doi.org/10.1016/j.enggeo.2016.07.017.
Tian, Y., Liu, Q., Liu, D., Kang, Y., Deng, P., He, F., 2018. Updates to Grasselli’s Peak Zhu, Z., 2017. Study on the Multi-View Geometry Based Method and its Application for
Shear Strength Model. Rock Mech. Rock. Eng. 51, 2115–2133. https://doi.org/ Creating Tunnel Layout Panoramic Image. Central South University.
10.1007/s00603-018-1469-2. Zhu, Z.-H., Fu, J.-Y., Yang, J.-S., Zhang, X.-M., 2016. Panoramic image Stitching for
Uotinen, L., Janiszewski, M., Baghbanan, A., Caballero, E., Oraskari, J., Munukka, H., Arbitrarily shaped Tunnel Lining Inspection. Comput. Civ. Infrastruct. Eng. 31,
Szydlowska, M., Rinne, M., 2019. Photogrammetry for recording rock surface 936–953. https://doi.org/10.1111/mice.12230.
geometry and fracture characterization. In: Rock Mechanics for Natural Resources

16