You are on page 1of 8

Application of Computers and Operations Research in the Mineral Industry –

Dessureault, Ganguli, Kecojevic & Dwyer (eds)


© 2005 Taylor & Francis Group, London, ISBN 04 1537 449 9

Model-based camera vision evaluation of mining machine motion

G. Danko, J.S. Knowles & R. Tiwari


University of Nevada, Reno, NV, U.S.A.

ABSTRACT: Results of machine motion analysis for a front shovel are presented. Digging and loading trajec-
tories of the bucket of mine front shovels were recorded during regular mining operations at a surface mine. The
three-dimensional trajectory coordinates are evaluated using multi-camera stereo vision. A new, model-based
image analysis method is used for identifying the basic coordinate system and camera parameters, by relating
the known dimensions of the shovels to the positions of various marker points in a digital video image. The
results of the analysis are used to evaluate typical bucket trajectories for the definition of a suitable “software
kinematics” for the mining machine.

1 INTRODUCTION

A research project is being conducted to develop oper-


ator assistance for mining equipment that requires
multiple-joint coordination during cyclic loading and
digging operations. Front shovels, backhoes and
bucket loaders typically perform several hundred load-
ing cycles during a mining shift. In order to implement
a new control method for machine joint coordination
introduced by Danko (Danko 2001), it is necessary
to understand the nature of typical machine move-
ments during the repetitive loading operations. The Figure 1. Exemplary camera arrangement around a mining
basic concept of the new control architecture is that the shovel in operation.
machine’s hardware-defined kinematics is modified
(using robotics technology) into a software-defined
kinematics. The operator of the machine may then 3. What type of motion adjustment is needed from one
select a software-defined kinematics that is more load to the other?
suitable for a given mining task than the original, 4. What internal relationships may exist between
hardware-defined kinematics with respect to ease of motion parameters such as bucket trajectory posi-
operation and multiple-joint movement coordination. tion, bucket angle, and movement velocities?
It is necessary to evaluate machine motion dur-
ing typical mining operations in order to derive the The motion trajectory tracking system provided by
parameters of the software-defined machine kinemat- the Ariel Performance Analysis System (APAS) soft-
ics. Existing mining machines do not have position ware (Ariel Dynamics Inc 2004, Ariel 1980, 1991,
sensors for assisting in such task. The recent appli- Wilmington 1992, Wilson 1999) was found to be most
cations of GPS sensors on the cabin of large mine suitable for the task at hand. The system provides both
shovels do not provide any details of the bucket move- manual and automatic tracking of multiple markers, as
ment during digging and loading (Nieto & Kadri, well as evaluation of the three-dimensional (3-D) coor-
Johnson & Van Diggelen). An external motion mea- dinates of these markers, based on the simultaneous
surement method based on camera vision was needed records of two or more video cameras. An exemplary
to answer the following questions: arrangement of the cameras around a mining shovel
in operation is shown in Figure 1. In addition to a
1. What are the motion characteristics of the full motion analysis, the video images provide further
bucket during digging and loading under different information about the working conditions, material
conditions? type and fragmentation, muckpile shape, as well as
2. What is the similarity between consecutive cycles? bucket fill.

513

Copyright © 2005 Taylor & Francis Group plc, London, UK


A basic requirement for the APAS evaluation is to ZC
identify the 3-D coordinates of at least six initial points ZA x
P YC
that are used by the software for base coordinate sys-
tem definition, as well as camera calibration. This is XC
a difficult requirement to meet in an operating mine zC
in which access to the quickly-changing working area zA YA f x
is limited, and frequent re-arrangement of the camera xA yC y
set-up is needed due to the fast-paced digging and load- {A} {C}
xC
ing operations. Indeed, according to the field studies, AP Q
the typical recording time period for a loading opera- C
yA y
tion at a given working area was two hours, following Camera
a one-hour set-up preparation time with three cam- projection
XA xy-plane
eras. Within the three-hour time period, the loading
operation usually progressed enough that the equip-
ment moved out of the camera view field, requiring Figure 2. Transformation of camera coordinate system {C}
re-adjustment of the setup, and re-establishment of the to the base coordinate system {A}.
coordinates of the six initial points.
In order to overcome the difficulties in surveying {C} can be established using coordinate rotations, dis-
a minimum of six points according to APAS require- placements, and projections. [8] If C P and A P denote
ments, a new, model-based initialization concept was the coordinates of P in {C} and {A}, respectively, the
developed for the use with the APAS system. The basic relationship between them can be written as follows:
concept of the new method is that several marked
points on the mining equipment, exceeding the min-
imum requirement of six, are used as initial points.
The relationship between the marked points are known where Rz (a), Ry (e), and Rx (t) are rotational matrices
from the dimensions of the machine, therefore, the around the z, y, and x axes by camera azimuth angle
model of the machine can be used for calibrating a, elevation angle e, and tilt angle t, respectively. The
the unknown parameters of a single camera. Once A
P C displacement vector designates the origin of the
one camera is calibrated against the machine at its camera coordinate system {C} in base coordinate sys-
initial position, the coordinate system in which the tem {A}, shown in Figure 2. The point C P in {C}, from
machine is defined becomes the basic coordinate sys- Equation 1, can be expressed as:
tem of the task. The minimum of six machine points
with known coordinates become the initial points
needed for the APAS system, allowing tracking of the
marker point motion using multiple cameras. Since the
single-camera calibration already establishes a rela-
tionship between the 3-D machine coordinates and
the two-dimensional (2-D) camera image, the method
also shows promise as a stand-alone alternative to
APAS-based evaluation of 3-D motion trajectories.
The paper describes the formulation of the relation-
ship between the 3-D machine, and the 2-D camera
picture coordinates; the basic mathematical formula- The 2-D camera coordinates x and y of C P are projected
tion of the solution to the camera and the base point to an x − y camera plane, parallel to the YC − ZC plane
system calibration; the test of the method in the labora- at focal distance f from the origin of {C} along the XC
tory on a small excavator; and the application to field axis. The focal projection equations express simple
measurements at an operating mine. The results of the linear proportions, according to the similar triangles
3-D digging trajectories are used for the evaluation of shown in Figure 2, as follows:
machine motion.

2 MODEL BASED IMAGE ANALYSIS


The coordinates (x, y) of the 2-D camera projection
2.1 Camera model development
point Q corresponding to point P of the 3-D object
The relationship between the 3-D coordinates of a sin- are evaluated using Equations 2 and 3. An appro-
gle point P in the base coordinate system {A} and its 2- priate unit conversion scale factor m must also be
D coordinates in the camera picture coordinate system incorporated to convert from image pixels to standard

514

Copyright © 2005 Taylor & Francis Group plc, London, UK


units of length (e.g. meters). However, since the focal
Machine Marker Image Pixel
length itself behaves as a scale factor in Equation 3,
Coordinates Coordinates
m and f are incorporated into a single scale factor (or
P1 = [xA, yA, zA]1 QM,1 = [xM, yM]1
pseudo-focal-length) M = mf , reducing the total num- P2 = [xA, yA, zA]2 QM,2 = [xM, yM]2
ber of unknown camera parameters by one. In fact, ... ...
unless either m or f is known a priory, their individual Pn = [xA, yA, zA]n QM,n = [xM, yM]n
values are inaccessible using a parameter optimiza-
tion algorithm. However, the pseudo-parameter M can
be calculated as a fit parameter. This suggests the
following revision of Equation 3: Initial Guess
Parameters Camera Projection
a0, e0, t0, XC,0, Equations 2 & 4
YC,0, ZC,0, M0

The measured equivalent of point Q is QM = (xM , yM ), Calculated 2-D


where xM and yM are the planar pixel coordinates of Projections of Pi:
the point P as observed in a digital image. Modify Q1 = [x, y]1
parameters Q2 = [x, y]2
to obtain ...
2.2 Model-based image analysis/camera new guess Qn = [x, y]n
calibration algorithm
A camera calibration algorithm was developed based
on matching the calculated (x, y) coordinates from n
Equations 2 and 4 with the measured pixel coordi- E= 兺 ei where
i=1
nates (xM , yM ). Point matching is accomplished by
eliminating the difference e in Euclidian norm:
ei = (x⫺xM)2i + (y ⫺ yM)2i

The unknown parameters to be identified are the seven


camera parameters used in Equations 2 and 4: a, e, Best Fit
No E= Yes Parameters
t, X C , YC , ZC , and M. A flow chart of the camera
calibration algorithm is shown in Figure 3. Inputs minimum? a, e, t, XC,
to the algorithm are the P1 . . .Pn points, each with YC, ZC, M
three known machine marker coordinates, as well as
the corresponding 2-D pixel points QM ,1 . . .QM ,n eval-
Figure 3. Flow chart of model-based image analysis
uate from the image. The algorithm is in effect a algorithm.
seven-parameter optimization procedure in search of
the minimum of the error sum E = e1 + . . . + en for
the n points as an object function. The output of the
algorithm is the set of seven camera parameters. 3 CAMERA CALIBRATION ALGORITHM
A minimum of n = 4, non-coplanar machine points TESTS
are needed for successful parameter identification.
This can be verified by examining the relationship The model-based image analysis algorithm was ini-
between the known and unknown parameters. For three tially developed and tested using a pair of computer
input points, the outputs from Equations 2 and 4 are generated cubes of known size (Figure 4), whose spa-
the three (x – xM ) and (y – yM ) differences, i.e, a tial relationship to each other was known and fixed.
total 6 values. This is still deficient in number for Four corners of each cube were used as marker points,
identifying seven parameters by optimization. For four for a total of 8 non-coplanar marker points, exceeding
input points, the number of outputs is eight, slightly the minimum number of 4 required for camera param-
“over-determining” the solution. Since a minimization eter evaluation. Numerous tests were performed using
algorithm is used, any number of points larger than this object (blue in Figure 4), in order to evaluate the
four would provide a solution. In case of encountering effectiveness of the algorithm.
measurement errors in the QM pixel point coordinates, The computer generated cube object was “placed”
or inaccuracies in measurement of the machine point in a known position in the base coordinate system and
coordinates, P, it is beneficial to use as many points “observed” from a known camera location. “Obser-
as possible and/or available. vation” meant that a simulated camera image of the

515

Copyright © 2005 Taylor & Francis Group plc, London, UK


Figure 5. Bobcat® 225 Model (blue) plotted over machine
image (initial guess parameters).

Figure 4. Development of Model-Based Camera Calibra-


tion Algorithm using a Pair of Cubes. Inset: A close-up view
of one marker point after running the parameter fitting algo-
rithm shows that a perfect model (black) exactly fits the
“real” cubes (blue). A random position error (between −5%
and +5% for each marker point, emulating a marker posi-
tion measurement error) causes an imperfect fit (red), but
does not significantly impede the performance of the fitting
algorithm.

“real” cubes was created using arbitrary input cam-


Figure 6. Bobcat® 225 Model closely fits image after
era parameters. A set of “initial guess” parameters optimization using the camera and machine fitting algorithm.
was then generated by perturbing the known cam-
era parameters. Two computer models of the cube
4 METHOD APPLICATION TO A SMALL
object were created, an exact model and an approxi-
EXCAVATOR
mate model. The approximate model was generated by
adding a random error (ranging from −5% to +5%)
The model-based fitting algorithm was subsequently
to each coordinate of the marker points of the exact
expanded for use with hydraulic mining equipment
model (simulating modeling error due to imprecise
by allowing various internal machine parameters to
measurement of a real machine).
be calculated along with the camera parameters.
The model-based image analysis algorithm was
A 3-D computer model of a small Bobcat® 225
then used to calculate the camera parameters by min-
hydraulic excavator was created, which included the
imizing the error between: (1) the observed 2-D coor-
simplified shape of the machine with known dimen-
dinates of the 8 marker points in the simulated camera
sions, and several variable angles between adjacent
image of the “real” object; and (2) the expected loca-
machine links. These variable, unknown joint angles
tions of these marker points in a 2-D camera projection
(base, boom, arm and bucket rotation angles) were
of the model.
added to the set of fit parameters in the algorithm.
The process was repeated using both the exact and
It was determined that the theoretical minimum
approximate models. As shown in Figure 4, the result
number of marker points required for fitting both the
was a perfect fit in the case when an exact model was
camera and internal machine parameters can be calcu-
used and a very close fit when the approximate model
lated as follows. First, the undercarriage was chosen
was used.
as the “base link.” This base link has a fixed shape and
The test was repeated many times using various
it has no internal machine parameters associated with
combinations of “initial guess” parameter values with
it. Thus, a minimum of 4 marker points are required
good results in all but the most absurd cases (i.e. when
to fit the camera parameters with respect to this base
reasonable initial guess values were used, the correct
link. Each additional link has only one degree of free-
solution was always reached).
dom; therefore one additional marker point is needed

516

Copyright © 2005 Taylor & Francis Group plc, London, UK


data, output from the model-based fitting algorithm,
is used as input to the APAS system to facilitate 3-D
marker tracking using multiple cameras. In addition,
2-D (image coordinate) marker trajectory data output
from APAS can be fed into a multi-frame version of
the model based fitting algorithm, which calculates
the optimal machine parameters for sequential video
frames. The result is a series of machine parameters
which describe the position of each machine link in
each frame of the video. These machine parameters
are then used to calculate the 3-D trajectories of vari-
ous points on the machine, such as the tip of the bucket,
whose motion is to be studied.

Figure 7. APAS marker tracking test on Bobcat® 225


machine showing initial and final machine positions with 5 TRAJECTORY TRACKING OF MINE
marker trajectory overlay.
SHOVEL EXCAVATION
on each link for model identification. For practical
applications, in which a significant amount of model- 5.1 Motivation and methodology
ing error and/or camera image distortion is likely to In order to facilitate analysis of mining equipment
exist, the theoretical minimum number of points may motion using the new model-based image analysis
not be sufficient to produce good results. In addition, algorithm and the marker tracking capabilities of the
the marker points must be distributed over the machine APAS system, high-contrast markers (approx. 30 cm
links in such a manner that the position of each link black circles on white backgrounds) were painted on
relative to adjacent links is uniquely constrained. several pieces of equipment at an open-pit mine includ-
By adjusting the values of the camera and machine ing two Hitachi EX3500 front shovels (Figure 8).
parameters (starting from initial guess values provided The 3-D location of each marker was measured with
by the user), the algorithm varies the position of the respect to a reference coordinate system assigned to
model until it closely matches that of the real machine each machine link. The measured marker locations
as observed in the video image (Figures 5, 6). were used along with data obtained from the manu-
Assuming that a sufficient number of marker points facturer’s data sheet for the EX3500 machine, in order
are supplied, this process gives the initial position of to generate a computer model of each machine. Three
all machine links, as well as the camera viewpoint and video cameras were used to collect motion data during
camera perspective parameters. By feeding the result- normal equipment operation.
ing set of machine parameters back into the machine The general shape of the trajectory plots obtained
model, the initial 3-D coordinates of several marker from the videos, as well as the more clearly pronounced
points can be calculated. These points can then be used deviations from the general shape, have been visu-
as known reference points for the APAS analysis. In ally confirmed by watching the videos. The ability
this way, the position of all machine markers can be to visually verify the reasonableness of the data is
tracked relative to the initial position of the machine, one significant advantage of the use of video images
thus eliminating the need for external (surveyed) ref- over internal sensors for trajectory data acquisition.
erence markers. According to this technique, the initial Another advantage of video-based data collection
machine position is used for identifying the base coor- is that bucket motion due to bouncing and other
dinate system in which the subsequent machine motion machine base movement is clearly visible and eas-
is identified. ily distinguished from motion produced by hydraulic
The model-based fitting algorithm can be used for actuators on the machine links. It would be difficult,
both 2-D and 3-D motion analysis. Similarly, the APAS if not impossible, to measure the effects of elastic
system is capable of performing marker tracking in deformations using internal machine sensors.
2-D, using video data from a single camera, or in 3-D
using video from 2 or more cameras placed at different
positions with respect to the machine under analysis.
5.2 Bucket trajectory classification
Figure 7 shows the results of one 2-D marker tracking
test using the APAS marker tracking system with the Initial analysis of the bucket digging trajectories as
Bobcat® 225 excavator. observed in the videos for the Hitachi EX3500 front
Various combinations of 2-D and 3-D trajectory shovels, showed three major classes of bucket trajec-
analysis can be performed by using the two systems tory. These classifications are based on the bucket
together. As discussed above, 3-D marker location angle, and on whether the bucket is initially placed

517

Copyright © 2005 Taylor & Francis Group plc, London, UK


0 6

5
100

Vertical Displacement(m)
4

200
3

300 2

1
400
0
0 2 4 6 8
0 100 200 300 400 500 600 700 Horizontal Displacement (m)

(a) Type 1 excavation example 60

0
40

Bucket Angle (degrees)


100 20

0
200

-20
300 Type 1 -- 040811ds1_c1, # 1
Type 1 -- 040811ds1_c1, # 2
-40 Type 1 -- 040811ds2_c1, # 3

400
0 2 4 6 8
Horizontal Displacement (m)

0 100 200 300 400 500 600 700


Figure 9. Examples of Type 1 trajectory shapes (top) and
(b) Type 3 excavation example corresponding bucket angles (bottom).

Figure 8. Marker point trajectories plotted in camera coor-


dinates on initial video frame. 5.2.2 Type 2 trajectory
Bucket starts at the base of the pile and is initially
moved nearly horizontally as in Type 1. However, the
at the base of the pile, or midway up the pile. How- Type 2 trajectory (Figure 10) rapidly curves upward
ever, further analysis using more data, will be needed as the bucket is quickly filled. The main difference
for final confirmation of this categorization. Examples from Type 1 is that the bucket angle starts near 0,
of each type of trajectory, along with the associ- but increases dramatically as the maximum bucket
ated bucket angle plotted as a function of horizon- capacity is approached.
tal bucket displacement are shown in Figures 9–11.
Figure 12 defines the bucket angle which is plotted in 5.2.3 Type 3 trajectory
Figures 9–11. Table 1 summarizes the trajectory type Bucket starts midway up the pile (Figures 8b, 11), and
definitions. moves in an approximately linear trajectory, which is
nearly parallel to the slope face. The bucket is then
lifted and moved toward the haul truck as described in
5.2.1 Type 1 trajectory Type 1. Bucket angle begins near 0, and increases semi-
Bucket starts at the base of the pile (Figures 8a, 9), and linearly for most of the trajectory with a somewhat
is first pushed forward nearly horizontally (effectively sharper increase near the end.
smoothing the previously excavated area just in front
of the pile). The bucket edge trajectory then gradually
curves upward, and becomes approximately parallel 5.3 Application of trajectory classification to
to the slope face. Bucket is then lifted out of the pile robot-human control sharing
and (almost simultaneously) begins to move toward Research is ongoing to determine if any additional dis-
a nearby haul truck where it is to be emptied. Type 1 tinct trajectory types are observed. However, while it
Trajectories are characterized by an initial bucket angle is possible that additional types will be found, it is
which is significantly negative. Bucket angle increases believed that the majority of digging trajectories can be
approximately linearly, but remains negative for up to placed into a relatively small number of such classes.
50% of the trajectory’s arc length. Each class can be described in terms of a parameterized

518

Copyright © 2005 Taylor & Francis Group plc, London, UK


6 6

5 5

Vertical Displacement (m)


Vertical Displacement (m)

4 4

3 3

2
2

1
1

0
0 0 1 2 3 4
0 1 2 3 4 Horizontal Displacement (m)
Horizontal Displacement (m)

60 60

40

Bucket Angle (degrees)


40
Bucket Angle (degrees)

20
20

0
0 Type 3 - 040811ds1_c1, #1
Type 3 - 040811ds1_c1, #3
-20 Type 3 - 040811ds1_c1, #5
-20 Type 3 - 040811ds1_c1, #6
Type 3 - 040811ds1_c1, #7
Type 2 -- 040811ds1_c1, # 16 -40 Type 3 - 040811ds1_c1, #8
-40 Type 2 -- 040811ds2_c1, # 1
0 1 2 3 4 5
0 1 2 3 4 Horizontal Displacement (m)
Horizontal Displacement (m)
Figure 11. Examples of Type 3 trajectory shapes (top) and
Figure 10. Examples of Type 2 trajectory shapes (top) and corresponding bucket angles (bottom).
corresponding bucket angles (bottom).

general trajectory shape, and a set of possible devia-


tions from the general shape. Distinguishing between
the “general shape” of a trajectory, and the devia-
tions from that general shape, facilitates the integration
of human and robotic control. Computerized robotic
control is best suited for performance of the general θ
task, while the human operator is much more capable x
of evaluating the need for some types of deviations,
due to rapidly changing environmental factors such as Figure 12. Definition of bucket angle with respect to the
variations in diggability. horizontal as plotted in Figures 9–11.
Several kinds of deviations are observed:
1. Reactive Path Variations – Small, transient changes
in bucket direction due to resistance produced by Table 1. Summary of observed excavation trajectory types.
non-uniform density of material. These are pro-
Starting position
duced by reaction forces exerted on the bucket by
Initial bucket angle Base of pile Middle of pile
the material, and are facilitated by elastic defor-
mation of machine links, shaking or bouncing of Significantly negative Type 1
the machine’s undercarriage, and other similar phe- Near zero Type 2 Type 3
nomena. Reactive variations generally appear as
“noise” on the trajectory plot.
2. Operator Redirection – moderate to large changes
in bucket direction made by the operator. This type fill rate. Operator Redirection can take a variety of
of deviation may occur due to the presence of insur- forms, including sudden lifting or lowering of the
mountable obstacles, such as large boulders, hidden bucket, large, sudden changes in bucket angle, or
beneath the pile, or due to misjudgment of bucket even temporary reversal of bucket motion.

519

Copyright © 2005 Taylor & Francis Group plc, London, UK


3. Trajectory Scaling – proportional or non- and machine parameters. Parameter optimization is
proportional changes in overall trajectory dimen- based on the correlation between the known 3-D loca-
sions. Proportional scaling does not affect the aspect tions of machine points and the observed 2-D locations
ratio of the trajectory, but changes only its overall of these points in a given camera image.
size to match the size of the pile. Non-proportional The bucket movement of a front shovel was ana-
scaling allows for variations in the slope angle of lyzed using the model-based image analysis algorithm
the pile, by allowing the trajectory to be lengthened discussed above in order to evaluate the characteristics
in only one direction. of typical bucket trajectories to facilitate the defini-
tion of a suitable “software kinematics” for the mining
It is assumed that the last two categories of deviation
machine.
are best controlled by the operator, while the general
trajectory can be easily performed robotically through
the use of pre-programmed, parameterized trajectory ACKNOWLEDGMENTS
shapes.
In addition, it is assumed that there must be a well- US Department of Energy, Mining Industry of the
defined relationship between the direction of general Future. Bobcat Company, an Ingersoll-Rand business
trajectory motion and the bucket angle. Clearly, such Newmont Mining Corporation.
a relationship must exist to facilitate efficient loading
of the bucket, and indeed, the data suggest that there is
such a relationship. Both the existence of this relation- REFERENCES
ship and its importance for efficiency suggest that the
bucket angle might best be controlled by the robotic Ariel Dynamics Inc. 2004. “Ariel Performance Analysis Sys-
portion of the system, allowing the operator even more tem”. <http://www.arielnet.com/start/apas/default.html>
freedom to concentrate on controlling the deviations Ariel, G.B. 1980. “Human Movement Analysis”. Applied
from the general trajectory. Ergonomics 11: 61–62.
Ariel, G.B. 1991. “Functional capacity assessment applica-
Comparison of the bucket angles shown in Fig- tions”. Fifth Annual Scientific Sessions & Symposium,
ures 9-11, with the bucket motion direction is in American Acadamy of Disability Evaluating Physicians
progress. Optimization of this relationship is seen to November 14–16.
be important for reducing the digging force and related Craig, John J. 1989. Introduction to Robotics: Mechanics and
power consumption. The “software kinematics” of the Control. pp.19–60. Reading, MA. Addison-Wesley.
machine will be defined according to such optimiza- Danko, G. 2001. “Operator control architecture”. Proceed-
tion and used in the new machine control system. ings of the 4th Regional Symposium on computer appli-
cations in the minerals industries, Tampere, Finland,
September 3–5, 2001: 265–266.
6 CONCLUSIONS Johnson, Lyle & Van Diggelen, Frank. “Advantages of a com-
bined GPS + GLONASS precision sensor for machine
A new, model-based image analysis method is control applications in open pit mining”, Modular Mining
Systems Inc. and Magellan Corporation. Thales. accessed
described for identifying the basic coordinate sys-
June 25, 2004. <http://products.thalesnavigation.com/
tem and camera parameters of a stereo vision system. assets/techpapers/14_advantages_GPSGLONASS.pdf>
The basic coordinate system is linked to the known Nieto, Antonio & Dagedelen Kadri. “Development of dump
dimensions of the mining equipment being analyzed, edge and proximity warning system using GPS and wire-
eliminating the need for external surveyed reference less network communications to improve safety in open
points. The method has been tested and found to be pit mines”, Mining Engineering Department, Colorado
accurate by computer simulation as well as using a School of Mines, Golden, CO. <http://www.mines.edu/
small excavator in the laboratory. research/wmrc/Webpage/RP-6_gps_files/Publications/
Digging and loading trajectories of the bucket of Dump%20Edge%20and%20Vehicle%20Proximity.pdf>
Wilmington, R.P. 1992. “Methodology issues concerning
mine front shovels were successfully recorded dur-
the accuracy of kinematic data collection and analysis
ing regular mining operations at a surface mine. The using the Ariel Performance Analysis System”. NASA
three-dimensional trajectory coordinates were evalu- Contractor Report 185689.
ated from video data obtained using multiple cameras, Wilson D.J. 1999. “Accuracy of digitizing using automated
by simultaneous model-based optimization of camera and manual methods”. Physical Therapy 79: 558–564.

520

Copyright © 2005 Taylor & Francis Group plc, London, UK

You might also like