MAPPING OF ROCK FRACTURES USING 3-D LASER SCANNING TECHNIQUE IN AN

AUTODESK 3-D STUDIO MAX ENVIRONMENT

A. K. L. Kwong and C. F. Lee
Department of Civil Engineering, The University of Hong Kong
Hong Kong, China (kwongakl@hkucc.hku.hk and leecf@hkucc.hku.hk)



ABSTRACT

During the process of investigation, design and upgrading of rock slopes, the orientation and spacing of structural
discontinuities or joints in a rock mass should be thoroughly mapped in the field so that any unstable blocks can be
identified and stabilization measures designed. The orientation of structural discontinuities is traditionally measured
manually using geological compass placed directly at the exposed surfaces. Scaffolding is usually erected to enable the
geologist to physically access the exposed rock surface. A large number of measurements are usually required in order
to obtain a statistical mean of the fracture orientation. With the ongoing advances of digital technology, 3-D laser
scanning technique can be used to replace direct physical access and large number of manual measurements can be
completely eliminated. A prototype system is described in this paper that combines the non-contact measurement
technologies of photogrammetric imaging and 3-D laser scanning to create dimensionally accurate and pictorially
correct 3 dimensional models and orthoimages of rock fractures. By taking photographs from at least two different
locations, lines of sight are mathematically intersected to produce the 3-D coordinates of the key reference points in a
rock face. Automatic 3-D laser scanning unit is then used to produce the 3-D coordinates of the entire rock surface. By
overlapping the images rectified from photogrammetry technique with the coordinates from 3-D laser scanning in an
Autodesk 3-D Studio Max environment, the coordinates of any objects in the photographs can be selected and their
orientation such as dip angle and dip direction calculated automatically. A case study is presented to compare the
orientations of fracture planes measured using geological compass and that from the prototype system developed.

KEYWORDS

Rock slopes, fracture spacing, dip angle, dip direction, photogrammetry, 3-D laser scanning, Autodesk 3d studio max.

INTRODUCTION

Hundreds of slopes in Hong Kong are being investigated, designed and upgraded every year to reduce the likelihood of
landslides and the potential loss of life and economic losses. During the process of investigation, design and upgrading
of rock slopes, the orientation and spacing of structural discontinuities or joints in a rock mass must be thoroughly
mapped in the field so that any unstable blocks can be identified and stabilization measures designed.

Conventional mapping of rock joint orientation and spacing requires scaffolding to access to the rock face so that a
hand-held geological compass can be placed directly at the rock joint surface to obtain the dip angle (maximum angle of
a plane relative to horizontal) and the dip direction (angle measured clockwise from the north to the line of the dip).
Because of scale effect, large amount of geological data has to be collected and plotted to come up with an
interpretation of statistically significant trends. The process is rather time-consuming, potentially hazard to the
geologists and workers, and limited to accessible areas only.

The purpose of this paper is to introduce the non-contact measurement techniques of Photogrammetry and 3-D Laser
Scanning and apply them to map the structural discontinuities so that the need to directly access the rock surface via
scaffolding is removed. By eliminating the need for scaffolding, the cost of site investigation and the potential of
workers falling off the slopes can be substantially reduced.

BACKGROUND OF PHOTOGRAMMETRY AND 3-D LASER SCANNING

Photogrammetry is a 3-Dimensional mapping technique that uses photographs as the fundamental medium for
measurement. By taking photographs from at least two different locations, lines of sight are mathematically intersected
to produce the 3-D coordinates of any points of interest. Control points and tie points are identified in the overlapping
photographs during the photogrammetry process in order that the orientation and location of the camera and the photos
are fixed relative to each other. Once the orientation and location of the photos are fixed relative to the camera and 3-D
coordinates of the tie points are determined, the 3-D coordinates of any points of interest between the tie points can be
interpolated. If the points of interest between the tie points have regular shape, for example a planar surface, then the
use of the photogrammetric process alone is adequate in forming a model that truly represents the coordinates and
photographic textures of an object in space.

The generic application of photogrammetry is in the creation of topographic maps at various scales from aerial
photography. The procedures for this operation are well established with commercial systems of various levels of
sophistication and automation readily available. Traditionally, topographic maps were produced in vector form the
features seen in the photographs represented by lines and symbols. The advent of digital techniques has allowed the
pictorial information to be included in the map. Such products are called orthimagemaps and are becoming a
standard topographical mapping product.

For object that has a regular shape, photogrammetry is a useful and accurate technique in converting objects within
pairs of 2-D photographs back to 3-D geometrical shape where measurements or coordinates can be taken directly from
the model. However, using photogrammetry technique to create dimensionally accurate model for fractures in a rock
slope surface is much less accurate because its surface profile is highly irregular between the tie points and identifying
common tie points in overlapping photos is not easy in rock fractures due to its plain texture appearance. A large
number of overlapping photographs are also needed to create a model for the entire slope surface.

The application of photogrammetry to geological mapping has mainly been confined to the use of aerial photography
and conventional photogrammetric mapping techniques to produce small-scale geological maps. A recently reported
application of ground-based photography (terrestrial photogrammetry) similar to that applied here is the work of
Robertson (1998) where he carried out the geological mapping of a dam abutment by applying photogrammetry
technique to historic photographs. It provided the identification of the geological features on copies of the photographs
and did not provide useful maps of the features. It is not reported what the accuracy achieved when comparing the joint
orientation obtained from photographs with that obtained from field mapping. In earlier research, Murlaz (1992)
reported the use of photogrammetric techniques in the geological mapping of the high walls of open-cast mines.
Orthophotographs were produced by registering photographs with information with map contents. No assessment of
the accuracy of the interpretation of the strata or geometric of the orthophotos was made.

Because photographs are a perspective projection, they cannot be used directly as an accurate map. The creation of
orthoimagemaps requires the effects of terrain relief to be removed by the application of a surface model through a
process called differential rectification. Surface models can be extracted from the overlapping photographs, but there is
the general requirement that the photographs are parallel with the coordinate system defining the plane of the surface to
be mapped. Most commercial photogrammetric systems have this requirement. In the mapping of slopes, the
coordinate system is defined in terms of local topographic coordinates XY forming the horizontal plane and Z being
the vertical direction therefore the slope surface is oblique to the coordinate system.

Terrestrial 3-D laser scanning is a relatively new technology that has found many applications in the engineering field.
A typical portable 3-D laser scanner is capable of recording positions of hundreds of thousands of points in 5 to 15
minutes with 2 mm accuracy in a 50m-target range. The minimum vertical and horizontal point-to-point measurement
spacing can be less than 4 mm at 50m-target range. With a field of view of 40 horizontally and vertically and a
maximum of 1000 rows and columns of points, this equates to an average point spacing of about 35 mm at a distance of
50 metres. Inside the scanner, two mirrors rapidly and systematically sweep narrow, pulsing laser beam over the chosen
target (or scene). A time-of-flight method is used to measure how long it takes for each laser pulse to hit a surface and
return to the scanner without the need of a reflector. Range measurements are generated for every laser pulse while
integrated optical encoders record mirror angles. The resulting positions commonly known as point cloud are
displayed graphically, as they accumulate in real-time, on the systems laptop.

The point cloud is a true 3-D representation of the surfaces seen by the scanner and can be coloured according to either
the intensity of the return laser pulse (which is influenced by the physical characteristics of the surface each pulse
strikes) or by the distance from the scanner to the surface. It can be viewed from any direction and accurate
measurements of distance, area and volume can be made between user-selected sets of points. Further processing of the
point cloud allows contours, profiles, 2 dimensional drawings and 3 dimensional CAD objects and models to be created.

Ono et al. (2000) tested a laser scanner for the mapping of mud slides in Japan and found that it was an efficient and
accurate method. Other work such as that of Lichti et al. (2002) have focused on the verification of the accuracy and
application of laser scanning; and Adams et al. (2001) through comparisons with photogrammetrically derived surface
models. In general they have found that laser scanners can produce the surface models to a higher level of accuracy and
density and in a shorter time than through photogrammetry.

Feng (2001) used a 3-D laser scanner (LARA) to quantify fracture geometry from exposed rock faces and compared
with that obtained from conventional reflectorless total station surveying equipment. It was found that good matches
were achieved between the two methods.

One shortcoming of the 3-D laser scanner is that although the points on the surface have 3 dimensional coordinates, the
integrated internal camera has very low resolution and is suitable only for defining the scan area. Its application to
rendering the model for accurate visual interpretation, such as that of geological strata, is limited. This can be
overcome by using high-resolution images from an independent camera, which could then be merged with the point
cloud.

The primary objective of this paper is to introduce a technique that will allow this to be done - integrate photographic
images from an independent camera using photogrammetric techniques with the true 3-D point coordinates obtained
from the laser scanning technique. This integration can be carried out through a 3-D AutoCAD environment. After the
photographic images and 3-D points are merged into an Autodesk 3-D Studio Max environment, the orientation of the
slope surface, dip angle and dip direction of specific fracture planes, spacing of joint planes can be clearly identified and
mapped.

PHOTOGRAMMETRIC PROCESS

Due to the geometric distortion associated with raw photographs, photographs have to be rectified before reliable
measurements can be obtained from it. The geometric distortion is caused by various systematic and nonsystematic
errors such as camera and sensor orientation, terrain relief, earth curvature, film and scanning distortion, and
measurement errors (Wolf, 2000).

To rectify image data, various geometric modeling methods such as finite element analysis (rubber sheeting) and
collinearity equations can be applied (Yang, 1997). The collinearity equations use the principle of triangulation to
determine 3-D positions, similar to the concept used by land surveying. It assumes collinearity relation exists among a
point in object space, its corresponding point on the projected plane, and the perspective center of the camera. A very
reliable modeling technique called bundle block adjustment (Leica Geosystems, 2003a) in the triangulation process,
uses measured image coordinates, control point information, line segments in object space and other observations to
determine both the camera orientation (i.e., camera position and aiming angle) and the 3-D object coordinates at the
same time. A block of images contained in a project is simultaneously processed in one solution. A statistical
technique known as least squares adjustment is used to minimize and distribute error for the entire block. The resulting
parameters are referred to as exterior orientation parameters. In order to estimate the exterior orientation parameters, a
minimum of three Ground Control Points (GCPs) is required for the entire block, regardless of how many images are
contained within the project (Leica Geosystems, 2003a).

Typical procedures involved in creating orthophotos are as follows:

1. Provide the geometric properties of the camera to the model (normally known as the Interior Orientation and
determined by calibration) such as defining the x and y pixel size of the digital camera Charge Coupled Device
(CCD), principal point x
0
, principal point y
0
, focal length of the camera, and lens distortion parameters. The
purpose of interior orientation is to define an image or photo-coordinate system within each image of the block,
in order to determine the origin and orientation of the image/photo-coordinate system for each image in the block.
2. Add images to the block.
3. Identify Ground Control Points (GCPs) in the images.
4. Perform Automatic Tie Point Generation. A tie point is the image/photo-coordinate position of an object
appearing on two or more images with overlapping areas. Automatic tie point generation makes use of digital
image matching techniques to automatically identify and measure the image positions of common points
appearing on two or more overlapping images.
5. Perform Block Aerial Triangulation. This is the process that uses the collinearity equations to mathematically
define the spatial relationships between the images contained within a block, the camera that obtained the images,
and the ground.

The systematic error associated with the block of images, camera positions, lens distortion and ground control point
coordinates etc. can be minimized using the method of least squares. This method is usually referred to as the standard
error, which is computed based on the summation of image coordinate residuals and ground coordinate residuals for
that particular iteration of processing. At each iteration of processing, the exterior orientation parameters of each
camera station and X, Y and Z coordinates of tie point are estimated. The newly estimated exterior orientation
parameters are then used along with the GCP and tie point coordinates to compute the new x and y image coordinate
values. The newly computed image coordinate values are then subtracted from the original image coordinate values.
The differences are referred to as the x and y image coordinate residuals. This process is iterated until the solution
converges below a pre-set threshold of say 0.001.

Once the process is finished, the position (x
0
, y
0
, z
0
) and orientation (Omega (), Phi ( ), Kappa ()) of each image in a
block, at the time of image capture, will be known. The six transformation coefficients relating the scale and rotation
differences between the file or pixel coordinate system of the image and the film or image space coordinate system are
also determined.

The X, Y and Z coordinates of tie points can be computed and converted to control points if necessary.

6. Create Orthoimages. If the effects of topographic relief displacement are large, an orthorectification process can
be used to rectify an image so that every point on an image would look as if an observer were looking straight
down, along a line of sight that is orthogonal to the earth. The orthorectification process takes the raw digital
imagery and takes each pixel of a Digital Terrain Model (DEM) and finds its equivalent position in the aerial
image. A brightness value is determined for this location based on resampling of the surrounding pixels. The
brightness value, elevation, and exterior orientation information are used to calculate the equivalent location in
the orthoimage file.

3-D LASER SCANNING PROCESS

The principles of using laser scanner to obtain the X, Y and Z coordinate of a point have been described. The typical
processes in 3-D laser scanning are as follows:

1. Set up control targets within the interest area required scanning.
2. Determine the range, sample spacing and number of points to be scanned.
3. Scan the area of interest and the control targets.
4. Carry out registration of the scan. Registration is the process of integrating more than one scanned areas into a
single coordinate system. The integration is based on a system of constraints, which are pairs of equivalent
objects that exist in two scanned areas. Results of the registration process will create an optimal overall
alignment transformation for each scanned object such that the constrained objects are aligned as closely as
possible in the resulting scanned area that has been merged together.
5. Create a mesh for the scanned object. A mesh is a series of triangles created using the points in a point cloud,
vertices, poly lines, or any combinations of the three as vertices.
6. Export the coordinates of the point or the mesh for further modeling and rendering.

INTEGRATING POINT CLOUD WITH ORTHOPHOTOGRAPHS IN AN AUTODESK 3-D STUDIO MAX
ENVIRONMENT

As stated earlier, laser scanning can produce a huge database of points with true 3-D local coordinates that can be
viewed but the surface textures of the object are poorly captured. By contrast, photogrammetry offers the real texture of
an object that can be visualized but the coordinates of the point are interpolated between the control points.

By combining the photo images from photogrammetry and accurate coordinates from 3-D laser scanning, the
photographs can be turned into dimensionally accurate and visually complete 3-D maps of the slope and hence the
structural discontinuities orientation and spacing can be manually mapped on the photographs, reducing the need and
extent of scaffolding.

For this mergence to become a reality, a platform of storage, manipulation and visualization of data acquired from both
photogrammetry and laser scanning is needed. Automated periphery functions such as the algorithms for computing
orientation (dip angle and dip direction) can be added into the system to enhance the value of the system.

The processes to be accomplished for achieving this scheme of mergence are summarized below with a flow chart
shown on Figure 1.

1. Carry out calibration for the camera.
2. Acquire suitable control points.
3. Capture images using the calibrated camera.
4. Determine interior orientation of the images.
5. Calculate exterior orientation of the images.
6. Capture point cloud using the laser scanner.
7. Carry out registration of different point clouds and import local coordinate systems.
8. Rotate the point clouds so that their coordinate system will align with that of the camera system.
9. Create a Triangulation Irregular Network (TIN, ESRI (2002)) from the rotated point cloud and export into 3-D
Studio Max as one layer. TIN is a data structure that consists of nodes that stores z-values connected by edges to
form contiguous, non-overlapping triangular facets.
10. Rectify the photos using the TIN created from point clouds.
11. Using the camera exterior orientation information, export the rectified image as another layer into the 3-D Studio
Max.
12. When the two layers are overlapped, point clouds and surface textures can be visualized and rotated together.
13. An automated algorithm can be written within the 3-D Studio Max program to select the fracture planes that are
of interest and perform automatic computation of orientation (dip angle and dip direction).
14. Repeat the automated algorithm to perform the dip angle and dip direction calculation of other fracture planes
that are of interest.

AUTOMATIC CALCULATION OF DIP ANGLE AND DIP DIRECTION OF FRACTURE PLANES

After the images and meshes are overlaid and attached together, an automated algorithm was written to select the
fracture planes that are of interest and perform automatic computation of orientation (dip angle and dip direction).

COMPUTATION OF DIP ANGLE AND DIP DIRECTION

In geological terms, Dip Angle (DA) is the maximum downward inclination of a structural discontinuity plane to the
horizontal, defined by (0

90

) as illustrated in Figure 2.

Dip Direction (DD) is the direction of the horizontal trace of a line to dip angle, measured clockwise from the North,
defined by (0

360

) as illustrated in Figure 2.

If a right-handed Cartesian system is adopted, then the positive Y-axis is directed to the North, the positive X-axis is
directed to the East and the positive Z-axis is directed upward for a X-Y plane that is horizontal.

The formula from analytical geometry can be used to compute the dip angle and dip direction of a fracture plane, which
is modeled as a best-fit plane intersecting the point clouds that are of interest.

Since the fracture plane is usually wavy, the orientation of the best-fit plane will be affected by the density and
distribution of the point cloud. Therefore, a least squares approach that uses multivariate linear regression method can
be used to define the best-fit plane representing the fracture plane.

The equation of the best-fit plane representing the fracture plane can simply be expressed as:

y b x b b z
2 1 0
+ + = (1)

Using the method of least squares, it can be assumed that the best-fit plane is produced when the minimal sum of the
deviations squared (least squares error) from a given set of data is found. In equation form, this is:

( ) |

+ + =
2
2 1 0
2
min
i i i
y b x b b z | (2)

For n number of points captured, the following matrix can be formed and the coefficients of the equation b
can be obtained by solving the matrix using either Gaussian Elimination or Crammers Theorem. In matrix algebra,
, , ,
2 1 0
b b
solution of the best-fit plane in space depends on the existence and uniqueness of the matrix and two cases must be
considered.

Photogrammetry Process Laser Scanning Process



Camera
Calibration



Determine Orientation
of Camera when
Object was captured
Rectification - Perform
Re-sampling of Pixels
Create Mesh or TIN
from Point Cloud
Based on Camera Orientation,
rotate point clouds so that the
two coordinate systems align
Capture
Terrestrial
Images
Provide Interior
Orientation of
Camera
Identify Control
Points and Tie
Points
Laser Scanning
of object
Laser Scanner
Registration

















Determine Exterior
Orientation of objects











3-D Visualization of
point cloud and texture
in 3-D Studio Max





Algorithm to select
fracture planes






Perform least square
calculations of best-fit
plane to obtain dip angle
an dip direction







Figure 1. Flowchart illustrating the procedures of merging photogrammetry with laser scanning

Figure 2. Determination of dip angle and dip direction of a fracture plane from point cloud

(
(
(
(
(
(
(

=
(
(
(

(
(
(
(
(
(
(

=
=
=
= = =
= = =
= =
n
i
i i
n
i
i i
n
i
i
n
i
i
n
i
i i
n
i
i
n
i
i i
n
i
i
n
i
i
n
i
i
n
i
i
z y
z x
z
b
b
b
y y x y
y x x x
y x n
1
1
1
2
1
0
1
2
1 1
1 1
2
1
1 1
(3)

where are coordinates of the captured points.
i i i
z y x , ,

If the determinant (D) of the matrix is equal to zero, then either the points captured are collinear (i.e., points rest on a
line and do not form a plane) or the fracture plane happens to be vertical to the X-Y plane and parallel to the Z-axis.
The former case can be avoided by making sure that the selected points are non collinear. For the latter case, the Dip
Angle is obviously equal to 90

and since the fracture plane is vertical, the equation of a plane can be degenerated into
an equation of a line, expressed simply as follows, again using the least squares method:

(
(
(
(

=
(

(
(
(
(

=
=
= =
=
n
i
i i
n
i
i
n
i
i i
n
i
i
n
i
i
y x
y
b
a
y x x
x n
1
1
1 1
1
(4)

This time, the solution of the best-fit line in the X-Y plane again depends on the existence and uniqueness of the matrix
and two cases need to be considered.

If the determinant (D) of the matrix is equal to zero, then there is no solution to Equation [4] and it implies that the line
is parallel to the Y-axis (North) and the Dip Direction is either 90

or 270

.

If the determinant (D) of the matrix is not equal to zero and b is greater than or equal to zero (means that slope is
positive in the X-Y plane), then the Dip Direction can be determined by:

(

|
|
.
|

\
|

|
|
.
|

\
|
=


0
1 0
0
1 0
180
tan 360
180
tan 180 b or b DD
(5)

If the determinant (D) of the matrix is not equal to zero and b is less than or equal to zero (means that slope is negative
in the X-Y plane), then the Dip Direction can be determined by:

(

|
|
.
|

\
|

|
|
.
|

\
|
=


0
1 0
0
1
180
tan 180
180
tan b or b DD
(6)

If the determinant (D) of the matrix describing the best-fit plane in Equation [3] is not equal to zero, then the fracture
plane is either inclined or parallel to the horizontal X-Y plane. The determination of Dip Angle and Dip Direction can
be obtained by comparing Equations [1] and [3].

If and are both equal to zero, then the fracture plane is horizontal and the Dip Angle is 90
1
b
2
b

and Dip Direction does

not exist. Otherwise, the fracture plane is inclined to the X-Y plane and determination of Dip Angle and Dip Direction
can be based on unit vector computation.

From analytical geometry point of view, Dip Angle is actually the angle between the fracture plane and the horizontal
plane and it can be determined by the inner product of the unit vector, ( ) 1 , 0 , 0 k , of the positive Z-axis and the normal
vector, , of the fracture plane (see Figure 3), as follows: ( 1 , ,
2 1
b b n )


(b
1
, b
2
, -1)

















Figure 3. Determination of Dip Angle based on Unit Vector

cos n k n k = (7)

or,

1
1
cos
2
2
2
1
1
+ +
=

b b

(8)

From analytical geometry point of view, Dip Direction is actually the angle counted clockwise from the positive Y-axis
to the horizontal component vector of the maximum inclination line or the horizontal component vector of the normal
vector of the fracture plane (see Figure 4).


(b
1
, b
2
, -1)















Figure 4. Determination of Dip Direction based on Unit Vector

Let be the unit vector of the positive Y-axis and ( 0 , 1 , 0 j ) ( ) 0 , ,
2 1
b b
h
n be the projection of the normal vector n onto
the horizontal plane. Similar to Equation [7], the Dip Direction can be calculated as follows:

cos
h h
n j n j = (9)

2
2
2
1
2 1
cos
b b
b
+
=

(10)

The normal vector of a plane may point upward relative to a horizontal plane or it may point downward relative to a
horizontal plane. Depending upon the pointing directions of the normal vector relative to the horizontal plane, the Dip
Direction calculated from Equation [10] may differ by 180

.

To follow the same geological definition, Dip Direction is determined by the upward normal vector of a fracture plane.
Therefore, the directional cosine value of the normal vector is checked to determine which quadrants the normal vector
is located in a Cartesian coordinate system and to identify whether the normal vector is upward or downward relative to
the horizontal plane. The directional cosine value of the normal vector is simply as follows:

2 2
2
2
1
2 2
2
2
1
2
2 2
2
2
1
1
1
1
cos
1
cos
1
cos
+ +

=
+ +
=
+ +
=
b b
b b
b
b b
b

(11)

where the three angles, () , () and () are the angles relative to the positive X-axis, Y-axis and Z-axis respectively.
For plane that does not pass through the origin, b is the distance from the origin to the plane. When b ,
0
0
0

2 2
2
2
1
1 + + b b should have negative - sign to satisfy all the points in Equation [1]. When
b
,
0
0
<
2 2
2
2
1
1 + + b b should have positive + sign.

Depending on whether the normal vector is upward or downward and which quadrant the plane rests relative to the
positive Y-axe (North), eight cases may be possible for determining the Dip Direction and is presented in Table 1.

Table 1. Calculation of Dip Direction based on direction of Normal Vector
Conditions Normal Vector Quadrant Dip Direction
0 cos 0 cos , 0 cos and
Upward North East
0 cos 0 cos , 0 cos < and
Upward South East 180

-
0 cos 0 cos , 0 cos < < and
Upward South West 180

+
0 cos 0 cos , 0 cos < and
Upward North West 360

-
0 cos 0 cos , 0 cos < and
Downward North East
0 cos 0 cos , 0 cos < < and
Downward South East 180

-
0 cos 0 cos , 0 cos < < < and
Downward South West 180

+
0 cos 0 cos , 0 cos < < and
Downward North West 360

-

CORRELATION COEFFICIENT

Correlation Coefficient, R
2
, also called multiple correlations or the coefficient of multiple determinations, is the percent
of the variance in the dependent explained uniquely or jointly by the independents. Simply saying, R
2
is a measure of
association between the variables. In this case, it is a measure of how well the equation obtained fits the plane. R
2

always lies between 0 and 1 and it represents the best fit when R
2
is equal to 1, where R
2
is given by:

( )( ) | |
( ) ( )

=
2
2
2
2


R
z z z z
z z z z
i i
i i
(12)

( ) ( y y b x x b b z
i i i
+ + =
2 1 0
)
(13)

where z y x and , are mean of respectively.
i i i
z y x and
,

ILLUSTRATIVE CASE: COMPARISON OF PLANES ORIENTATION CALCULATED FROM POINT
CLOUD WITH COMPASS

The algorithm described above for computation of Dip Angle and Dip Direction of a best-fit plane, after an
orthorectified photo has been correctly merged with the point cloud, has been programmed. Within the 3-D scene of the
3-D Studio Max program, any planes that are of interest can be rotated, enlarged and selected with a pointing device.
The orthorectified photo carrying realistic photographic textures greatly allows any planes that are of interest to be
correctly selected. After the points are selected, they can be exported to the sub-routine program for computation of
orientations.

An illustrative case is presented below, which describes a comparison of planes orientations calculated from
multivariate regression analysis with that measured using a conventional compass.

The site selected is a rock slope with multiple anchor blocks behind the building of the Civil Engineering Department of
The University of Hong Kong (see Figure 5). The orientations of the five different faces of the anchor blocks (see also
Table 2) are ideal for comparisons because they are planar and can be accurately measured with a conventional compass,
the readings of which can be used as the reference. Results of exposed rock planes below the block will be presented in
a separate paper.

A 3-D laser scanner Cyrax 2500 model (see insert on Figure 5) was used for scanning with point-to-point spacing of 4
mm, within a distance of less than 50 m. Scans from three different directions were carried out and they were registered
and tied into the local coordinate system of Hong Kong using 6 nos. of control points (see locations Figures 5 and 6),
the coordinates of which were obtained from Global Positioning Position (GPS) device that has an accuracy of 5 mm.
After registration was completed for the three different scans, regression analysis shows that the root mean square error
of all the control points was less than 5 mm which indicated that the three different scans were correctly merged
together.

Figure 6 shows the merged point cloud in far view and in close-up view. The far view of this point cloud is made up of
about 1,000,000 points, with each point carrying true local coordinates. The points are so closely spaced that it almost
gives the impression of a real object. A close-up view of the three blocks (point cloud and photos of Block B, Block C
and Block D) is also presented on Figure 6 and these blocks will be used for plane orientations computation. It is worth
mentioning that each scan only requires about 20 minutes and all the fieldwork was completed in less than 2 hours.



























Figure 5. Overall view of Rock Slope and Anchor Blocks at HKU

A photogrammetry program, Leica Photogrammetry Suite (Leica Geosystems, 2003a), was used to produce the
orthorectified photos. The camera used was Nikon D70 with a focal length of 35 mm and Charge Coupled Device
(CCD) size of 3060 pixel (horizontal) x 2030 vertical pixel (6 Megabytes). Three overlapping photos were used for
exterior orientation, and the control points used were the same as that used for registration of point cloud. After interior
orientation was defined, exterior orientation (X
o
, Y
o
, Z
o
, , and ) was computed based on the known control point
coordinates using the bundle block adjustment and additional tie points were generated. These tie points were
generated based on some image matching techniques (Leica Geosystems, 2003a). The standard error after bundle block
adjustment was less than 1 pixel, indicating that the spatial relationship of all the control points and tie points among the
overlapping photos was correctly set.

After the triangulation process was completed, the exterior orientation parameters were used for rotating the point cloud
to align with the photographic axis as described above. The next step was to create a TIN model (with elevation
contoured) using the point cloud. At the same time, one of the photos was orthorectified using the nearest neighbor
method in the re-sampling technique. A 2
nd
-order polynomial was used because the control points and tie points were
sufficiently accurate and closely spaced already.

After the orthorectified photo is correctly overlain on top of the TIN model as shown in the 3-D Studio Max scene on
Figure 7, the final step is simply selecting any plane of interests for orientation computation. A sub-routine program
using was written as an added-in module in 3-D Studio Max for selecting and computing the Dip Angle and Dip
Direction.

Table 2 summarizes the results and compares the Dip Angle and Dip Direction obtained from multivariate regression
analysis with that obtained from conventional compass. Examination of Table 2 indicates that for the Dip Angle, the
absolute difference between the two methods is less than 2% and the percentage difference between the two methods is
less than 3%. For the Dip Direction, the absolute difference between the two methods is less than 4% and the
percentage difference between the two methods is less than 6%. The overall average difference of all the orientations is
less than 1%. This small discrepancy amongst the two methods is beyond practical significance. Results of similar
measurements on exposed rock planes below the block will be presented in a separate paper.

SUMMARY AND CONCLUSIONS

A prototype system that combines the non-contact measurement technologies of photogrammetric imaging and 3-D
laser scanning to create dimensionally accurate and pictorially correct 3 dimensional models and orthoimages of a rock
slope is developed. These products allow safely mapping of the structural discontinuities orientation (Dip Angle and
Dip Direction) and spacing in an exposed rock slope surface, for which the need to erect scaffolding and access to the
rock face is reduced. Very high level of accuracy can be achieved based on the illustrative case described.

The implications of this successful prototype to the industry is far-reaching such as the following:

Scaffolding and access to slope surface will be greatly reduced in the investigation stage resulting in a
substantial savings in cost of any slope design program.
Reduce the risk of workers falling from the scaffolding.
Reduce the amount of bamboo used in the construction industry.
Generate a permanent record of the slope via photographs of rock joint surface that are pictorially correct in 3
dimensional space.
Allow design of slope upgrading works such as concrete buttresses, dowels or bolts directly drawn on the
rectified photographs for construction.
The creation of an accurate 3 dimensional models of the slope for additional investigations such as boulder
falls and natural landslide scar study.
The use of computer visualization methods leads to a much better understanding of the fracture pattern because
the slope features can be viewed from any virtual location, even from inside the slope.

The possible outcome of this research project may include establishing procedures for applying photogrammetry and 3-
D laser scanning techniques to other types of slope studies in Hong Kong. This may include identification and
measuring of boulder field in natural terrain hazard study and identification and measuring of corestone in soil slope
study. The findings from this project will also allow as-built drawings be produced which can couple with true-scaled
photographs for inclusion into the slope maintenance manual.

ACKNOWLEDGMENTS

The authors gratefully acknowledge the scanning carried out by the Hong Kong office of Leica Geosystems.

The financial support by the Research Grant No. HKU7117/03E provided by the Research Grants Council of the Hong
Kong Special Administrative Region, China is greatly appreciated.

NOTATION

The following symbols are used in this paper:

D = determinant
R
2
=

correlation coefficient
x
0
=

principal point of camera in direction x
y
0
=

principal point of camera in direction y
z
0
=

principal point of camera in direction z
X, Y, Z = ground coordinates
z y x , , = image coordinates system
z y x , , = point cloud coordinates system
= dip direction
= dip angle
, , = orientation of camera

REFERENCES

Adams, J. C., Smith, M. J. and. Bingley, R. M. (2001). Terrestrial laser scanning and close-range photogrammetry for
cliff face mapping. Proceedings of the First Annual Conference of the Remote Sensing and Photogrammetry
Society, 12-14 September 2001, 537-548.
ESRI. (2002). Using ArcGISTM 3D Analyst. 251p.
Feng, Q. (2001). Novel methods for 3-D semi-automatic mapping of fracture geometry at exposed rock faces. Ph.D.
thesis, Department of Civil and Environmental Engineering, Royal Institute of Technology, Stockholm, Sweden,
69p.
Leica Geosystems. (2003a). Leica Photogrammetry Suite OrthoBASE & OrthoBASE Pro Users Guide, 516p.
Leica Geosystems. (2003b). ERDAS Field Guide, Seventh Edition, 698p.
Lichti, D. D., Gordon, S. J. and Stewart, M. P. (2002). Ground-based laser scanners: operation, systems and
applications. Geomatica, 56 (1), 21-33.
Mularz, S. (1992). Remote sensing monitoring of open-cast mine. International Archives of Photogrammetry and
Remote Sensing, XXIX (B5), 311-317.
Ono, N., Tonoko, N. and Sato, K. (2000). A case study on the landslide by the 3-D laser mirror scanner. International
Archives of Photogrammetry and Remote Sensing, XXXIII (B5), 593-598.
PhotoModeler Pro. 4.0. (2001). Eos Systems Inc.
Robertson, G. (1998). Geological mapping of dam abutments utilizing historic images and digital feature matching.
International Society for Photogrammetry and Remote Sensing, Commission V, Hakodate, Japan.
Wolf, P. R. and Dewitt, B. A. (2003). Elements of Photogrammetry with Applications in GIS. 3rd ed., New York:
McGraw-Hill, Inc.
Yang, X. (1997). Georeferencing CAMS Data: Polynomial Rectification and Beyond. Ph. D. dissertation, University
of South Carolina, Columbia.
Block D
Block C
Block B Block A
Close-up View 2
Close-up View 1
Control Points for Registration of
Point Clouds and Triangulation
Far View of Point Cloud of Overall Rock Slope and Anchor Blocks

Block C
Block B
Close-up View 2
Block D
Close-up View 1










Block C
Block D
Block B







Close-up View of Photos corresponding to scanned Point Cloud
Figure 6. 3-D Scanned Point Cloud, Far view and Close-up view
14


Figure 7. Point Clouds Merged with Photos in 3-D Studio Max Environment
15
Table 2. Comparison of Dip Angle and Dip Direction calculated from Multivariate Regression Analysis with
measured from Hand Held Compass









Difference
(e)=(d)-(c)
Anchor
Block
(a)
Face
(b)
Multivariate Regression
Analysis based on Point Cloud
(c)
Hand Held Compass
(d)
(e)=[(d)-(c)]/(d)*100%
Dip Angle
*
Dip Direction
*
Dip Angle Dip Direction Dip Angle Dip Direction
1.0 0.2
B 1 73.0 358.8 74 359
1.4% 0.1%
-1.7 -1.0
2 53.7 0.0 52 359
-3.3% 0.3%
0.6 -1.4
3 73.4 340.4 74 339
0.8% -0.4%
-0.2 0.0
4 89.2 359.0 89 359
-0.2% 0.0%
1.3 0.3
5 72.7 18.7 74 19
1.8% 1.6%
1.2 0.0
C 1 69.8 0.0 71 0
1.7% 0.0%
-0.8 0.3
2 51.8 359.7 51 0
-1.6% 0.1%
0.7 -0.8
3 71.3 339.8 72 339
1.0% -0.2%
-1.0 0.0
4 88.0 0.0 87 0
-1.1% 0.0%
0.5 1.3
5 71.5 19.7 72 21
0.7% 6.2%
1.5 -1.5
D 1 72.5 355.5 74 354
2.0% -0.4%
-1.3 -0.9
2 51.3 354.9 50 354
-2.6% -0.3%
1.2 -4.3
3 72.8 344.3 74 340
1.6% -1.3%
-2.0 -2.5
4 88.0 356.5 86 354
-2.3% -0.7%
0.4 1.7
5 74.6 335.3 75 337
0.5% 0.5%
0.1 -0.6
Average
0.0% 0.4%
*
R
2
all greater than 0.95

2
2
2
5 1 3
1 3 5
5 1 3
4
4
4
Block B
Block D
Block C


16