Geotechnical Engineering Journal of the SEAGS & AGSSEA Vol. 49 No.

2 June 2018 ISSN 0046-5828

Application of Photogrammetry and Image Analysis for Rock Slope Investigation

D-H. Kim1, A. S. Balasubramaniam2 and I. Gratchev3
1
School of Engineering, Griffith University, Gold Coast, Australia
2
School of Engineering, Griffith University, Gold Coast, Australia
3
School of Engineering, Griffith University, Gold Coast, Australia
E-mail: [email protected]

ABSTRACT: This study reviews the applications of close range photogrammetry (CRP) on modelling for rock slope stability analysis and
weathering investigations focussing on the applicability of CRP to obtaining roughness characteristics. Current photogrammetric techniques
have a potential to provide roughness profiles with dense measurement intervals. However, the quality of the roughness data is still
questionable and the level of accuracy has not been sufficiently investigated. To advance the boundaries of the availability of CRP, this paper
presents methodologies which can quantify the degree of accuracy for the obtained roughness data and to detect the data noise as an
evaluation tool. Firstly, this study suggests an error model which measures the level of accuracy based on an ideal lab condition. The level of
accuracy of rock joint roughness coefficient (JRC) obtained from CRP can be interpreted by using the developed error model. Secondly, this
paper presents an image analysis workflow via a MATLAB image filtering code for the estimation of recession areas focussing on the
variation of rock surface roughness. This post-process of CRP supports more reliable interpretation of photogrammetric roughness data. The
developed error model and the combined image analysis with 3D photogrammetric models could compensate the limitations of the uses of
both digital photographs and 3D surface models to obtain roughness characteristics and for quantifying weathering patterns.

KEYWORDS: Photogrammetry, Rock surface roughness, Image analysis, MATLAB, Weathering.

1. INTRODUCTION aspect of weathering because the products of weathering appear in

different ways due to their geological formation.
In rock slope investigation, the procedure of rock mass In order to identify geological characteristics, digital images
characterization is simple and relies on visual impression of the rock have been effectively interpreted by using various image analysis
structure. Considering geological composition and joint surface techniques. For example, McEwan et al. (2000) introduced an image
conditions in small scale, visual assessment plays an important role analysis technique to assess ground erosion. As another example,
for the determination of RMR and Q values in engineering purposes. Trauth (2010) studied individual grain sizes of deposits and
As visual assessment is subjective, digital mapping methods and geological settings by analysing digital images. Microscopic images
image analysis techniques have assisted the investigations to were also employed to investigate the degree of weathering with a
minimize human errors. Recently, terrestrial laser scanning (TLS) quantitative image analysis technique (Puente et al., 2006).
and close range photogrammetry (CRP) have been employed However, it is inadequate to analyse recession areas by image
effectively to create 3D surface models and to generate the data of analysis technique itself. Also, combining the data obtained from
discontinuities with great accuracy. The obtained high-resolution both the colours of 2D images and the changes of 3D surface model
images from these techniques can contain detailed geological is complicated. The limitations of the application of photographs to
composition of rock surfaces and the images are analysed using quantify degradation areas by weathering was also discussed by
various analysis tools. Ortiz et al. (2013). Nonetheless, digital images effectively
Close range photogrammetry (CRP) has been used in recent characterize the localized roughness on the surface of rocks using
years to create 3D slope models for stability analysis in rock the information on the adjacent pixels of images. If the features of
engineering (Ferrero et al., 2011; Firpo et al., 2011; Brideau & interest on a rock surface are simple and its colour is fairly uniform,
Stead, 2012). Recent high-end equipment in the field of digital the changes of undulation and roughness can be more precisely
photographing can create high resolution images and this technology detected by the change of regional brightness. Thus, the correlations
has encouraged the spread of CRP to obtain surface features such as between the brightness and the asperities obtained from 3D images
undulation and roughness which require higher degree of accuracy may compensate the limitations of both methodologies.
of 3D models. As a major research interest of the authors, the In this paper, the degree of accuracy of CRP to obtaining
feasibility of CRP to investigate rock surface roughness has been roughness data is widely discussed and the relevant errors are
actively studied focusing on the accuracy of 3D models within interpreted by using a photogrammetric JRC error model. Also, this
allowable distances (Kim et al., 2015b; Kim et al., 2016a). The idea study suggests a workflow of image analysis combining the images
from the previous studies developed a JRC (joint roughness of 3D models and the changes of roughness data focussing on the
coefficient) error model to interpret the errors of JRC values relationships between roughness variation and colour information.
obtained from site photogrammetry surveys. To enhance the reliability of photogrammetric roughness data, the
In rock engineering, roughness is a crucial factor in determining data of roughness heights are processed with relative brightness
both the conditions of discontinuities and the weatherability of rock integers obtained from its greyscale images and a MATLAB image
masses. Using the geological strength index (GSI) system, the analysis technique is employed to filter the images.
roughness of rock joints can be quantified by using roughness
parameters such as JRC and Js (joint smoothness factor) relating to 2. ROCK SURFACE INVESTIGATION USING CLOSE
the friction characteristics of rock joints. In the case of weathering RANGE PHOTOGRAMMETRY
investigation, rock surface roughness has been employed to
investigate the rate of weathering as the change of roughness can 2.1 Image resolutions in close range photogrammetry
indicate the recession of rock surface by weathering (Meierding,
1993; McCarrorll and Nesje, 1996; White et al., 1998; Pope et al., Photogrammetry is a science to obtaining precise three dimensional
2002; Sancho et al., 2003). However, it should be noted that the models using two or more images. Close range photogrammetry
roughness values are not the only possible parameter to explain the (CRP) can be defined by the camera distance from the object of

49
 Geotechnical Engineering Journal of the SEAGS & AGSSEA Vol. 49 No. 2 June 2018 ISSN 0046-5828

interest. CRP can be simply separated by the camera-to-object obtained from closer c-to-o distances have been used to construct
(c-to-o) distance of less than around 300 metres from the basic model geometries for creating numerical models for various
principles of aerial photogrammetry (Matthew, 2008). In order to analyses such as landslides and rock falls. CRP has produced
investigate rock surface roughness, the ranges of c-to-o distances are satisfactory results in the detection of joint sets and in the accuracy
much closer than the guideline as the targets of surveys are focussed of the corresponding data for the joint orientations.
on the detailed features of rock surfaces. It can be defined as an Figure 2 is an example of small scale CRP models. This model
appropriate range of c-to-o that produces high resolution 3D images was created for a natural slope located along a drive road in the
with small pixel sizes to such an extent as to generate roughness of Tamborine Mountain area, Gold Coast, Australia (Kim et al.,
rock surfaces. 2015a). The photogrammetry survey was to obtain measurements of
In field conditions, the resolution of 3D models depends on the the orientations of the joint sets of the slope and the shape and size
scale at which it is being considered combined with the field of rock blocks at the inaccessible slope. Two images were captured
conditions. In the case of large scale photogrammetry surveys (point at two camera positions using a Nikon D7000 camera with a 24 mm
interval > 10 mm), several examples of achieved point intervals with focal length lens with 33 metres of c-to-o distance. The model was
different photogrammetry setups are presented in Table 1. With created by using a photogrammetry program Sirovision (CSIRO,
regard to the determination of image resolution, the point intervals 2012). Based on the exposed joint sets of the 3D model, it was
of images are controlled by the following three parameters; camera estimated that the rock mass consisted of polygonal blocks with the
sensor size, lens focal length and camera-to-object distance. These average volume being in the range of 0.4 ~ 1.8 m3.
parameters should be mutually adjusted in photogrammetry setups
to obtain the intended resolutions of 3D images.

Table 1 Examples of camera setups and image resolutions

authors ○1 ○2 ○3 ○4
device/ no. c-to-o lens point
software effect. dist. FL spac.
pixels (m) (mm) (mm)
ADAM Nikon D2x 12e6 600 180 74
Tech /3DM Analyst
ADAM Nikon D2x 12e6 1200 400 66
Tech /3DM Analyst
Gaich et Nikon D70s 6e6 30 35 25
al., (2006) /ShapeMetriX3D ~70
Voyat et Nikon D100 6e6 - 18 50
al., (2006) /Virtuozo
Poropat, Nikon D300 12e6 100 100 60
(2008) /Sirovision
○1 : number of effective pixels, ○ 2 : camera-to-object distance, ○
3 :

focal length of lens, ○

4 : point cloud spacing Figure 1 Relationships between camera-to-object distances and
point intervals in fields and laboratory conditions; FL = 50 mm (a),
For the case of small scale photogrammetry survey (e.g. point FL = 85 mm (b)
interval < 10 mm), some experimental data of point intervals are
presented in Figure 1. The data in the figure were collected over six
site investigations and under a laboratory condition, and obtained by
using a Nikon D7000 digital camera (sensor resolution: 4,928 ×
3,624 pixels) with fixed focal length lenses (Nikon Nikkor, FL = 24,
50, 85 mm). Relevant test data are well presented in the authors’
previous study (Kim et al., 2016a). The data distributions show that
the point intervals increase linearly. In the field conditions, the point
intervals show rather scattered patterns (R2 = 0.71 ~ 0.84) compared
to those obtained under laboratory conditions (R2 = 0.96 ~ 0.99). It
is clearly demonstrated in the data distribution that a combination of
a distance (around 5 metre) with a focal length lens (FL = 50 mm)
was able to achieve a range of point intervals in high resolution
ranges (e.g. less than 1.0 mm).

2.2 Application of CRP in rock engineering

2.2.1 Photogrammetric modelling
In rock engineering, photogrammetry has been effectively used to
provide detailed information about the discontinuities of rock
masses for various geotechnical and hydrogeological issues
(Sturzenegger & Stead, 2009; Ferrero et al., 2011; Firpo et al., 2011;
Brideau et al., 2012). The scales of the objects in photogrammetry
surveys are various from a large open pit to individual rock blocks.
In a large scale survey, 3D models provide detailed structural Figure 2 Photogrammetric model for a failed rock slope (a);
information which can be integrated into broader structural systems stereonet projection of main joint sets (b); a 3DEC model at a
obtained through a GIS based geological map and a terrain digital sliding step (c) (Kim et al., 2015a)
elevation model. On slope stability issues, 3D surface models
50
 Geotechnical Engineering Journal of the SEAGS & AGSSEA Vol. 49 No. 2 June 2018 ISSN 0046-5828

As shown in Figure 2, structural mapping of joint sets was MAEJRC=∑ni=1|JRCo,i – JRCp,i| / N (1)
performed by a stereonet projection method using the obtained joint
data. The 3D model was also utilized to obtain JRC values to assess The error models have been verified by using sets of field data
a range of friction angles of the joint sets for stability analysis. (See Figure 3 (b) (Kim et al., 2016a). The results showed that the
However, the required extent of resolution of 3D images for the most influential of all the factors on the accuracy of roughness data
estimation of JRC values can be an issue because the accuracy of was the point interval of 3D models. The results also showed that
photogrammetric JRC values has been insufficiently established. the oblique angles of the optical axis to the pole can significantly
The camera setup in this case study created a 3D image with 2.9 mm affect the accuracy of roughness data. The errors of
of an average pixel size. This value may be included in a sufficient photogrammetric JRC values obtained from field surveys are plotted
range of density for structural mapping of joint sets. Considering the in Figure 3 (b). The point intervals of roughness profiles in the
typical size of profile gauge for measurement of joint roughness graph was less than 1 mm.
(1mm interval), the resolution of the 3D image may lead to a range
of errors in the estimation of JRCs.

2.2.2 Measurement of rock surface roughness

Based on the advantages of various visual information obtained
from images and the feasibility of photogrammetry for creating 3D
surface models, many researchers have employed photogrammetry
for characterizing surface roughness (Jessell et al., 1995; Cravero et
al., 2001; Lee & Ahn, 2004; Unal et al., 2004; Bistacchi et al., 2011;
Sturzenegger & Stead, 2009; Nilsson et al., 2012). The applications
of photogrammetry have been also found in wide ranges from
archaeological to scientific purposes. Due to the importance of the
use of roughness data, the feasibility of photogrammetry for the
roughness investigation has been steadily studied. Recent studies
just covered basic issues for the applicability of photogrammetry to
estimate JRC (joint roughness coefficient) values (Haneberg, 2007;
Baker et al., 2008; Poropat, 2008). These previous studies addressed
the issues of the extent of accuracy of CRP with the image scales.
Evaluation of the accuracy and precision of CRP is difficult
because of the uncertainty of photogrammetric data derived from
various factors. The influencing factors can be introduced by both
inaccuracies in camera system and bad planning of camera network
geometry. For example, lens distortion is one of the most significant
factors that can lead to systematic errors and needs to be taken into
account for any photogrammetry applications. Also, depth accuracy
is a factor which is related to planning of camera network. These
influencing factors are well identified and discussed in various
references on photogrammetry.
Considering the various influencing factors, statistical
approaches have been employed to investigate the accuracy of
photogrammetry. To measure the accuracy of continuous variables,
the root-mean-squared-error (RMSE) and the mean absolute error
(MAE) are commonly employed. RMSE has been widely used to
identify the accuracy of data due to its high correlation between the
predicted values and the observed values. In a photogrammetry Figure 3 Photogrammetric JRC error models (a), an example of
standard, the accuracy of geospatial data obtained from RMSE and MAE of JRC values obtained from field
photogrammetry has been classified using the RMSE of data photogrammetry survey (FL = 50 mm) (b), camera oblique angle
coordinates (ASPRS, 2014). As a natural measure of average error and normal vector of a section of interest (c) (Kim et al., 2016a)
magnitude, advantages of MAE have also been also reported
(Willmott & Matsuura, 2005; Chai & Draxler, 2014). Within the range of the dense point interval (e.g. less than 1 mm),
Kim et al. (2015b) showed a quadratic function between the the errors could be interpreted using the developed error models
obtained errors from photogrammetry and camera-to-object with close correlations. Figure 3 (b) also shows that the effect of the
distances based on a set of ideal laboratory tests. The proposed photograph angles on the accuracy of roughness data within the
quadratic functions which are correlations between the difference of range of high oblique angles (e.g. θ1 > 60˚ and θ2 < 30˚). The
JRC values and the normalized JRC values showed upward relevant angles are identified by using a simplified figure (Figure 3
parabolic curves with different widths according to the employed (c)). The experimental study showed that the accuracy of asperity
focal length of lens. The photogrammetric JRC values were heights and roughness parameter such as JRC values obtained from
distributed in both over estimation and under estimation ranges. field photogrammetry surveys can be verified by comparing the
Since the laboratory data were predominantly plotted in the range of values with the developed error model.
underestimation, the quadratic regression curves were partly lack of
correlations in the range of overestimation. Considering the balance 2.3 Investigation for weathering
of data, it is reasonable that the data can be interpreted by dividing
As an indicator of the degree of weathering, rock surface roughness
the data range into underestimation and overestimation categories.
has been employed to investigate weathering states (McCarroll &
The proposed error functions have been then improved as shown in
Nesje, 1996; White et al., 1998; Gómez-Pujol et al., 2006; Pinho et
Figure 3 (a). With the use of absolute values, the basic form of the
al., 2006). McCarroll & Nesje (1996) showed that roughness is a
MAE of JRC is simpler than the RMSE form as given in Eq. (1).
useful indicator to quantify the degree of weathering through a set of

51
 Geotechnical Engineering Journal of the SEAGS & AGSSEA Vol. 49 No. 2 June 2018 ISSN 0046-5828

measurement of roughness profiles using profile gauges. Combined the sandstone data. The JVR data indicate different measures in the
with modern survey techniques, weathering investigation has been data distributions according to their geological bases.
expanded to the use of 3D digital surface models. Using the
triangulate point clouds obtained from a laser scanning method,
Medapati et al. (2013) attempted to use the γ-value, which is the
angle from the vertical axis to the normal vector, for identifying the
rock surface roughness in different weathering conditions. As high
density photogrammetry 3D models allow for characterizing the
detailed features of rock surfaces, any variations of the surfaces
relating to erosion, roughness and damages of rock surfaces from
weathering and excavation can be detected by performing long-term
photogrammetry surveys.
If a rock is vulnerable to weathering, the variations of roughness
on the exposed rock surfaces can be observed at the intersections of
the exposed joints within the surface areas of slopes in a relatively
short period. Depending on the geological components of rocks, the
roughness alterations will produce different forms of weathering
products. Figure 4 shows an example of the features of weathering. Figure 5 Data distributions of JVR obtained using Tse & Cruden
In this case, the weathering products can be dominantly founded function for shale (a) and sandstone (b) sections
from the edges of joints. Profiles including the intersection of joints
will be changed and the loss of roughness between the exposed joint 3. IMAGE ANALYSIS
edges leads to a decrease of their roughness indices. To investigate
the trend of roughness variation, statistical methods can be usefully 3.1 Use of photogrammetric greyscale images
employed. Medapati et al. (2013) showed that roughness data can A digital image can be regarded as a group of discrete pixels, each
form specific distributions according to their weathering conditions. of which has various ranges of colour and brightness information.
Colour images with various geological features can be more
complex to be analysed than greyscale images.
In the case of a greyscale image, it carries only intensity of
brightness and can simplify the image structures. For example, each
pixel of an 8 bit greyscale image has a luminance value which can
be measured on a scale from black (0) to white (255). If the colour
formations and textures of rocks are simple, the differences of the
brightness between the compared stages of surveys can be regarded
as an important factor arising from the different reflection of light.
Using the brightness information of rock surface, the authors
reported an application of image analysis comparing between before
Figure 4 Changes of 3D model presenting loss of roughness by and after shearing stages in direct shear tests of rocks (Kim et al.,
weathering in the intersection of joints 2014). The case study showed the applicability of image analysis to
obtain the roughness variation form simple 2D greyscale images
An experimental case study was investigated by performing which are obtained in the same photographing conditions. However,
annual photogrammetry surveys (Kim et al., 2016). In this case it is worth mentioning that the differences of the extent of brightness
study, the roughness variation due to weathering was quantified by between the surveys in a variety of field conditions may produce
using the changes of JRC values. The study slope is composed of misleading results.
the alternate bedding structures of sandstone and shale. Through the In the case study, for the same study slope in Section 2.3, the 3D
annual photogrammetry surveys, JRC values were obtained from 3D images obtained from the annual photogrammetry surveys were
surface models for sandstone and shale sections at two yearly analysed by using their greyscale images. As a distinct advantage of
intervals. The JRC values were calculated by a function of 3D images, extended meshes which cover all the areas of interest in
‘Sirovision’ in four radial directions (0˚, 90˚, 180˚ and 270˚). The a rock slope in orthogonal directions can be obtained as shown in
JRC data were formed with scattered patterns for both shale and Figure 6 (a). The 3D images present the bright information of
sandstone zones. This case study introduced a JRC variation rate geological features as well as the coordinates. This enables an
(JVR) to quantify the roughness variations by weathering process as analysis of the integer values of pixels combined with the 3D
presented in Eq. (2). The results showed JVR can reflect the roughness data. As shown in Figure 6 (b), the changes of intensity
geological structural characteristics of the rock material as well as values of greyscale 3D images are interpreted with the changes of
its durability. asperity heights obtained from photogrammetric profiles as shown
in Figure 6 (c).
JVR (JRC variation rate) = ((JRCt2 – JRCt1) / JRCt1) / year (2) The results are used to estimate the recession areas during the
time interval. This methodology can compensate the use of
where, JRCt1 is the previously measured value in a particular year roughness indices for quantifying weathering considering the
and JRCt2 is a measured value at present. If the JVR values are geological textural properties of rocks.
negative, the JRC values have been decreased during the specific
periods. The estimated JVR data showed different shapes of 3.2 Procedure of image analysis
distribution curves depending on the structure and durability of rock MATLAB® (The Mathworks Inc.) has been used as the main code to
material. Figure 5 demonstrates that the JVR values are distributed rotate 3D point clouds and extract roughness profiles. In
forming skewed curves. In the shale zone, the surface of the selected photogrammetry analysis codes, their algorithms have been
area has considerably changed due to the exfoliations controlled by improved by developing better codes to produce an accurate
the textural properties of the shale bedding structures. The JVR simulation of objects and a user-friendly graphic interface (Madeira
values of the shale zone indicate higher average values than those of et al., 2010; CSIRO, 2012). Also, MATLAB® is a well-suited

52
 Geotechnical Engineering Journal of the SEAGS & AGSSEA Vol. 49 No. 2 June 2018 ISSN 0046-5828

Figure 8 (a). Figure 8(b) shows images of the sections which are of
interest.

Figure 6 The use of a greyscale image of 3D model to investigate

surface recession by weathering; surface model in pixels in 3D
space (a), bright integers (b) changes of profile (c), (Kim et al.,
2016b)

platform for analysing colour images using various image analysis

tools. Further, these tools provide effective functions to visualize 3D
images. This advantage enables researchers to approach the analyses
of rock surface features considering the colour data of pixels of Figure 8 Images of the sampling areas on rock surfaces of the study
images and their 3D coordinates. The procedure of the image slope; micro scale images of the collected rock (a), images of the
analysis for the quantification of weathering is classified by the sections of interest (b)
following three main categories as demonstrated in Figure 7. The
first phase is to collect roughness data by performing Photogrammetry surveys were carried out for the study slope
photogrammetry in the same site. Secondly, the original RGB within a two year time period from 2012 to 2014. The digital images
images are converted to greyscale with 8 bit pixel depth to simplify were created using a DSLR with 16.2 million effective pixels in an
the image data. Thirdly, the variations of brightness integers (I) image and a 24 mm focal length lens. The pixel sizes of the 3D
obtained from greyscale 3D images are correlated with the changes images varied with their camera setups as presented in Table 2.
of asperity heights. Using an image thresholding technique of Stereo photographs were taken at two camera positions and the
MATLAB®, the pixels of the greyscale images are filtered within annual photographs were taken at the same positions. In the geo-
the obtained guide ranges of (I) values from the profile analysis. referenced 3D images, two sections of both shale and sandstone
zones were selected for the image analysis. The dimensions of the
selected areas and the density of the 3D images are summarized in
Table 2.

Table 2 Details of the 3D images in the sampling areas

camera setup & image size shale sandstone
c-to-o distances (m) 17.0 17.0
image size width 332 ~ 667 427 ~ 523
(mm) height 678 ~ 1,123 427 ~ 513
Image scale (mm/pixel) 3.3 ~ 5.1 3.7 ~ 7.6
profiles (mm) 430 ~ 1,050 480 ~ 710
Point intervals (mm) 2.0 2.0

With visual comparisons of the 3D images of 2012 and 2014, it

was found that there were roughness alterations especially on the
periphery of exposed joints in both rock zones. The images show a
large alteration of colours on the surfaces accompanied by the
Figure 7 Workflow of roughness investigation combined with exfoliation of thin rock flakes in the shale sections. In sandstone
sections, however, dominant changes occurred around the edges of
image analysis
joints as shown in Figure 8 (b). Two sections which showed
noticeable changes on the exposed rock surfaces between 2012 and
3.3 Case study
2014 surveys were selected in both rock types. In the 3D images of
In this study, the proposed image analysis is investigated through a the selected sections, JRC values were estimated for the profiles in
case study. The geological condition of the study slope is composed four radial directions (steepest, 45˚, 90˚, 135˚) as shown in Figure 9.
of the alternate bedding structures of sandstone and shale in the The changes of JRC values are different over time according to the
Nerangleigh-Fernvale beds. The texture of sandstone is medium to rock types. For the given two years, the JRC values for the shale
coarse grained and joint sets are frequently found from the slope sections increased due to the exfoliation with flakes. In the
surface. The texture of shale is fine and has a laminated structure, sandstone regions, the JRC values diminished slightly according to
and the orientations of the beddings are steeply inclined. The texture its rounded edges. Also, the positions of loss of roughness due to
was observed using digital images taken by an 8 megapixel weathering were detected by comparing of the extracted profiles
microscope camera from collected rock samples, as shown in between the two years.
53
 Geotechnical Engineering Journal of the SEAGS & AGSSEA Vol. 49 No. 2 June 2018 ISSN 0046-5828

Figure 9 Alteration of photogrammetric JRC values in radial

directions over the two year interval in the shale section (a), and the
sandstone section (b)

The colour data of the pixels of the 3D images was converted to

greyscale with 8 bit pixel depth. Along the locations of the profiles
which were employed to estimate JRC values, the brightness integer
(I) values were extracted and the changes of (I) values were
compared with the changes of asperity heights of the profiles. The
variance of brightness integers increases with the increase of
roughness and that sandstone samples showed large values of bright
integers compared to shale due to its bright geological colour base.
The greyscale images of the shale sections show more inconsistent
variations of brightness than those of sandstone sections caused by
discoloration of weathered products. Because of the changeable
colours, it was difficult to define the correlation between the
changes of roughness and the changes of (I) values.
However, it can be said that overall trend of the variation of (I)
values responds to the roughness profile differences. The sandstone
sections showed more noticeable relationship between the recession
Figure 10 Changes of brightness integers (I) and the loss of
regions and the change of (I) values than the shale sections because
roughness of profiles in vertical directions; shale (a), sandstone (b)
of the simplicity of its texture. Also, the change of (I) value of
sandstone was strongly influenced by differences of luminance
between the photographing in 2012 and 2014. The difference of
brightness in surveys led to an extraneous increment of (I) values as
shown in Figure 10. Using the correlations between (I) values and
profile differences, the guide range of brightness integer can be
obtained from the same positions which indicate marked differences
of roughness along the profile. This procedure is well demonstrated
in the previous study (Kim et al., 2016b).
Using an image threshold technique of MATLAB, the pixel data
of the greyscale images were filtered within the obtained guide
ranges of (I) values. The obtained threshold images are presented in
Figure 11. The percentage of the filtered area to the total area is used
to quantify the products of weathering. In the case of sandstone,
because the range of brightness is relatively simple, the filtered
areas can be used as an indicator of the products of weathering. The
recession area in the sandstone sections was considerably increased
(8.6% to 25.4 %). By contrast, the recession areas are inconsistently
distributed on the surfaces of the shale sections. This is a limitation
of this technique which is related to the differences of luminance.
This result can be explained by the responses of chemical
weathering due to the lamination characteristics of shale. However,
Figure 11 Comparisons of filtered areas by MATLAB threshold
the image analysis technique shows that the brightness data in the
technique over the two year interval; shale section 2 (a), sandstone
pixels of 3D images can be used to study the progress of weathering
section 2 (b)
of rocks considering their geological conditions.
in another. That can be one of the reasons why photogrammetry for
4. DISCUSSION
roughness investigations has been insufficiently studied and used to
Photogrammetry has proven to be inexpensive and efficient in data. However, many examples of successful roughness simulations
creating geospatial data of large scale rock slopes. This technology using close range photogrammetry have been found in various areas
has successfully employed to create 3D models for numerical such as geology, archaeology and medical science. It can be said
analysis. With regard to roughness investigation, photogrammetry is that close range photogrammetry is a useful technique of providing
questionable due to various influencing factors. When dealing with deterministic design values in the field of rock slope engineering
the factors, uncertainties in one may be associated with uncertainties under the condition when the obtained data lie within reasonable
54
 Geotechnical Engineering Journal of the SEAGS & AGSSEA Vol. 49 No. 2 June 2018 ISSN 0046-5828

ranges. In this respect, it is important to mention that the feasibility with photogrammetric roughness data could compensate the
of photogrammetry depends on the required accuracy of the limitations of both the use of 2D roughness data and the use of the
produced models. integer values of 2D images. It can be concluded that the image
From an engineering point of view, the influence of the accuracy analysis using the brightness data in the pixels of 3D images can be
of 3D models on the corresponding roughness coefficients should be used to investigate the progress of weathering combined with the
evaluated when considering the influence of the values on the changes of roughness data on rock surfaces.
mechanical behaviour. This study demonstrates that the point
interval is an important factor and determines the level of accuracy 6. REFERENCES
of photogrammetric roughness data. It is obvious that accurate data
can be obtained when the images are taken at close distances with American Society for Photogrammetry & Remote Sensing (2014)
long focal lengths and thus the point intervals are short. As "ASPRS positional accuracy standards for digital geospatial
mentioned in Section 2.2.2, for the same measurement interval with data", PE & RS, 81(3), A1-A26.
manual measurement (e.g. less than 1.0 mm), field photogrammetric Baker, B. R., Gessner, K., Holden, E. J., Squelch, A. (2008)
roughness data could be interpreted using the developed error "Automatic detection of anisotropic features on rock
models by the authors with close correlations (Kim et al., 2015b). It surfaces", Geosphere, 4(2), pp.418-428.
is worth mentioning that this model is valid as an ideal standard for Bistacchi, A., Griffith, W. A., Smith, S. A. F., Di Toro, G., Jones,
the estimation of the amount of errors beyond the predicted error R., Nielsen, S. (2011) "Fault roughness at seismogenic depths
ranges. Also, the orientations of measurement surfaces in relation to from LIDAR and photogrammetric analysis", Pure and
the line of sight for cameras have considerable influence on the Applied Geophysics, 168, pp2345-2363.
accuracy of roughness data. Field data proposed a high-angle Brideau, M. A., Stead, D. (2012) "Evaluating kinematic controls on
oblique range to provide a good match between the error models and planar translational slope failure mechanisms using three-
photogrammetric JRC data distributions. dimensional distinct element modelling", Geotechnical and
The study presented in this paper also emphasizes the benefits of Geological Engineering, 30, pp991-1011.
the use of digital images, which possess colour data in the pixels of Chai, T., Draxler, R. R. (2014) "Root mean square error (RMSE) or
the images for the analysis of the variation of roughness. The use of mean absolute error (MAE)?-Arguments against avoiding
greyscale images is adequate to simplify the analysis when the RMSE in the literature, Geoscientific Model Development,
geological features of rocks are uncomplicated. The simplified 7(3), pp1247-1250.
greyscale images of 3D models are used to study the quantification Cravero, M., Iabichino, G., Ferrero, A. M. (2001) "Evaluation of
for rock weathering combined with the photogrammetric roughness joint roughness and dilatancy of schistosity joints",
data. As presented in Section 3.3, the case study shows close Proceedings of Eurock 2001, Espoo, pp217-222.
correlations between the change of image brightness and the change CSIRO (2012) "Field procedures for photogrammetric pit mapping".
of surface roughness. In order to analyse the brightness values in CSIRO Exploration & Mining
pixels of images, a threshold technique of MATLAB® image Ferrero, A. M., Migliazza, m., Roncella,R., Rabbi, E. (2011) "Rock
analysis tool provides to apply for filtering for the areas of interest. slopes risk assessment based on advanced geostructural
The filtered area can be used to indicate the vulnerability of survey techniques", Landslides, 8, pp221-231.
weathering. However, the intensity of brightness should be carefully Firpo, G., Salvini, R., Francioni, M., Ranjith, P. G. (2011) "Use of
interpreted by considering both luminance conditions for digital terrestrial photogrammetry in rocky slope stability
photographing and the geological characteristics of rocks. Lighting, analysis by distinct elements numerical methods",
camera angles, lens aperture values and the conditions of the International Journal of Rock Mechanics & Mining Sciences,
surfaces of interest at the moment of photographing can influence on 48, pp1045-1054.
the brightness integers of images. The limitations of this technique Gómez-Pujol, L., Fornos, J. J., and Swantesson, O. H. (2006) "Rock
are also related to the level of accuracy of photogrammetry models. surface millimeter-scale roughness and weathering of
supratidal Mallorcan carbonate coasts (Balearic Islands)".
5. CONCLUSION Earth Surface Processes and Landforms, 31, pp1792-1801.
Haneberg, W. C. (2007) "Directional roughness profiles from three
This study discusses the feasibility of close range photogrammetry dimensional photogrammetric or laser scanner point clouds",
from various points of view for any investigations into rock slopes Proceedings of the 1st Canada-US rock mechanics
focussing on the uses of rock surface roughness. As the most symposium, Vancouver, pp.101-106.
influential factor for the accuracy of 3D images, photogrammetry Jessell, M. W., Cox, S. J. D., Schwarze, P., Power, W. (1995) "The
setups and the corresponding point intervals were widely discussed anisotropy of surface roughness measured using a digital
through literature review and collected data from experiments and photogrammetric technique", Geological Society, London,
field surveys. In order to study roughness characteristics from rock Special publications, 92, pp27-37.
surfaces, CLP requires quite close camera-to-object distances. The Kim, D. H., Chung, M., Gratchev, I. (2014) "Assessment of rock
allowable distances can be established by considering camera setup joint roughness using image analysis of damaged area in
to obtain the point intervals less than 1 mm. direct shear tests". Proceedings of the 23rd Australasian
Field photogrammetry surveys produce scattered data and Conference on the Mechanics of Structures and Materials
plotted mainly overestimated JRC values. In order to interpret the (ACMSM23), Byron Bay, pp745-750.
level of accuracy of photogrammetric JRC values this study Kim, D. H., Gratchev, I., Balasubramaniam, A. S. (2015a) "Back
introduced an error model based on laboratory tests and field data. analysis of a natural jointed rock slope based on the
The error model which increases existing knowledge of photogrammetry method", Landslides, 12, pp.147-154.
photogrammetric errors is helpful as a guideline which can compare Kim, D. H., Gratchev, I., Balasubramaniam, A. S. (2015b)
with the obtained roughness data in specific photograph setups to "Improvement of photogrammetric JRC data distributions
estimate the accuracy of photogrammetric roughness data. based on parabolic error models". Int. J. Rock Mech. Min.
This study also discussed the use of image analysis using 3D Sci., 80, pp19-30.
photogrammetry models combined with roughness data for rock Kim, D. H., Poropat, G., Gratchev, I., Balasubramaniam, A. S.
weathering investigation. Photogrammetric 3D models provide two (2016a) "Assessment of the accuracy of close distance
dimensional roughness profiles and the 3D RGB images can be photogrammetric JRC data". Rock Mech. Rock Eng., 49,
converted to greyscale images and used as a source of the threshold pp4285-4301.
image analysis tool of MATLAB. The combined image analysis
55
 Geotechnical Engineering Journal of the SEAGS & AGSSEA Vol. 49 No. 2 June 2018 ISSN 0046-5828

Kim, D. H. Gratchev, I., Oh, E., and Balasubramaniam, A. S. Pinho, A., Rodrigues-Carvalho, J., Gomes, C., and Duarte, I. (2006)
(2016b) "Evaluation of the vulnerability of rock weathering "Overview of the evaluation of the state of rock weathering
based on monitoring using photogrammetry". Proceedings of by visual inspection", IAEG 2006 Engineering Geology for
the 5th International conference on geotechnical and Tomorrow’s Cities, The Geological Society of London, paper
geophysical site characterization, ISC5 2016, Gold Coast, 260.
Australia, pp1087-1092. Pope, G. A., Meierding, T. C., Paradise, T. R. (2002)
Lee, H. S., Ahn, K. W. (2004) "A prototype of digital "Geomorphology’s role in the study of weathering of cultural
photogrammetric algorithm for estimating roughness of rock stone", Geomorphology, 47, pp211-225.
surface, Geosciences Journal, 8(3), pp.333-341. Poropat, G. V. (2008) "Remote characterization of surface
Madeira, S., Goncalves, J., Bastos, L. (2010) "Photogrammetric roughness of rock discontinuities". Proceedings of the
mapping and measuring application using MATLAB". workshop of Laser and photogrammetric methods for rock
Computers & Geosciences, 36, pp699-706. face characterization, Colorado, pp63-75.
Matthew, N. A. (2008) Aerial and close-range photogrammetric Puente, M. E., Rodriguez-Jaramillo, M. C., Li, C. Y., Bashan, Y.
technology: providing resource documentation, interpretation, (2006) "Image analysis for quantification of bacterial rock
and preservation. Technical note 428. U.S. department of the weathering". Journal of Microbiological Methods, 64, pp275-
interior, Bureau of land management, National operation 286.
center, Denver, Colorado, p42 Sancho, C., Fort, R., Belmonte, A. (2003) "Weathering rates of
McCarroll, D., and Nesje, A. (1996) "Rock surface roughness as an historic sandstone structures in semiarid environment (Ebro
indicator of degree of rock surface weathering". Earth Surface basin, NE Spain)", CATENA, 53, pp53-64.
Processes and Landforms, 21, pp963-977. Sturzenegger, M., Stead, D. (2009) "Close-range terrestrial digital
McEwan, I. K., Sheen, T. M., Cunningham, G. J., Allen, A. R. photogrammetry and terrestrial laser scanning for
(2000) "Estimating the size composition of sediment surfaces discontinuity characterization on rock cuts", Engineering
through image analysis", ICE-Water and Maritime Geology, 106, pp163-182.
Engineering, 142(4), pp189-195. Trauth, M. H. (2010) "MATLAB® Recipes for earth sciences", 3 rd
Medapati, R. S., Kreidl, O. P., MacLaughlin, M., Hudyma, N., and edition, Verlag Berlin Heidelberg, Germany: Springer.
Harris, A. (2013) "Quantifying surface roughness of Unal, M., Yakar, M., Yildiz, F. (2004) "Discontinuity surface
weathered rock – examples from granite and limestone", roughness measurement techniques and the evaluation of
Proceedings of Geo-Congress 2013: Stability and digital photogrammetric method", Proceedings of the 20th
performance of slopes and embankments III, San Diego, international congress for photogrammetry and remote
California, pp120-128. sensing, ISPRS, pp1103-1108.
Meierding, T. C. (1993) "Inscription legibility method for estimating White, K., Bryant, R., and Drake, N. (1998) "Techniques for
rock weathering rates", Geomorphology, 6, pp273-286. measuring rock weathering: application to a dated fan
Nilsson, M.,Edelbro, C., Sharrock, G. (2012) "Small scale joint segment sequence in southern Tunisia", Earth Surface
surface roughness evaluation using digital photogrammetry", Processes and Landforms, 23, pp1031-1043.
Proceedings of the Eurock 2012, Stockholm. Willmott, C. J., Matsuura, K. (2005) "Advantages of the mean
Ortiz, P., Ortega, F. J., Vázquez, M. A., Martíin, J. M., Aparicio, P., absolute error (MAE) over the root mean square error
Ferruz, J., Caballero, F., Maza, I., Ollero, A. (2013) "The (RMSE) in assessing average model performance", Climate
diagnosis of the royal tobacco factory of Seville assisted by Research, 30(1), pp.79-82.
quad-rotor helicopters". Proceedings of the 1st conference on
Robotics Innovation for Cultural Heritage, Venice, Italy

56