The applicability of a geomagnetic field based positioning technique with mobile phone to underground tunnels

Abstract

In underground mines and tunnels accurate and continuous positioning of moving machines and humans has been studied extensively. Geomagnetic field based positioning is an interesting possibility for this measurement need. Using a commercial smart phone and a new calculation service provided by a Finnish company, a feasibility and accuracy assertion tests were done in Finnish underground mine. First time a standard mobile phone, as far as we know, was used in underground tunnel environment to produce continuous position data and an accuracy assessment was performed. An L-shaped tunnel area, two tunnels crossing, was selected for the experiments. A total station was used for reference measurements. Accuracy of measurement in different static and dynamic situations was studied. The found RMSE of 7.32 meters should satisfy many needs. The use of standard smart phone for underground positioning worked fine, but more development work is needed it to be usable in everyday mining situations.

Full text

The Applicability of a Geomagnetic Field based Positioning
Technique with Mobile Phone to Underground Tunnels
T. Makkonena, R. Heikkiläa, A. Kaarankaa
aConstruction Technology Research Center, University of Oulu, Finland
E-mail: Tomi.Makkonen@oulu.fi, Rauno. Heikkila@oulu.fi, Annemari.Kaaranka@oulu.fi
Abstract
In underground mines and tunnels accurate and
continuous positioning of moving machines and
humans has been studied extensively. Geomagnetic
field based positioning is an interesting possibility
for this measurement need. Using a commercial
smart phone and a new calculation service provided
by a Finnish company, a feasibility and accuracy
assertion tests were done in Finnish underground
mine. First time a standard mobile phone, as far as
we know, was used in underground tunnel
environment to produce continuous position data
and an accuracy assessment was performed. An L-
shaped tunnel area, two tunnels crossing, was
selected for the experiments. A total station was used
for reference measurements. Accuracy of
measurement in different static and dynamic
situations was studied.
The found RMSE of 7.32 meters should satisfy
many needs. The use of standard smart phone for
underground positioning worked fine, but more
development work is needed it to be usable in
everyday mining situations.
Keywords -
Magnetic field; Underground positioning; Tunnel
1 Introduction
The need of continuous positioning in underground
mines is needed among other things for safety, logistics,
machine operations and machine automation [1], [2].
Position technologies are widely applied to above
ground level tasks mainly because GPS is so effortless
to use. There has not been commercial service which
could position underground effortlessly.
The best position method currently is obviously a
total station and they are widely used, however they
need to be set up and moved according to the
positioning needs. The narrow corridors add challenge
for setting up the device and also the requirement of
direct view complicates measurements.
Several positioning solutions have been proposed for
the underground mines, such as radio signal based
positioning [3], [4] and the use of RFID [5], [4]. CSIRO
has made lot of work with their Inertial Navigation
system to advance the field of automated continuous
mining equipment [6].
A geomagnetic field based positioning technique for
the underground mines was suggested and tested by
Haverinen and Kemppainen [7]. The magnetic field of
Earth is not constant, but has a number of anomalies
inside the buildings [8] as well as in mines when
magnetic minerals are present. Measuring these
anomalies and generating a magnetic field map, one can
use this data to successful positioning.
Haverinen and Kemppainen made a positioning test
underground with an accuracy of about 1.5 meters. They
used a magnetometer sensor array at approximate length
of 1 meter. [7] Their work with geomagnetic positioning
technique lead to company called IndoorAtlas which
has a business model to provide web based service for
position calculations using Monte-Carlo Localization
from recorded geomagnetic maps and tools in mobile
phone to do the mapping inside the buildings. In most
mobile phones there already is a three-axis
magnetometer costing less than one dollar making the
device a magnetometer [9].
Magnetic field positioning indoors has been studied
increasingly in recent years [10]–[15].There were not
prior test with geomagnetic positioning underground
using mobile phone thus the results are quite unique.
The measurements were done in the Outokumpu
Kemi mine, a chromium mine producing 2.7 million
tons of ore per year in northern Finland, about 500
meters below the surface.
2 Measurements and setup
Main tools for measurements were:
1. IndoorAtlas web service for creating and
maintaining maps
2. IndoorAtlas Mobile APK for creating magnetic
maps and visual testing on the site
The 31st International Symposium on Automation and Robotics in Construction and Mining (ISARC 2014)

3. KemiMineDataAquisition APK mobile software
for recording the location data
4. Android Nexus 4 phone
5. Kemi Mine wireless network
Because of the novel use of cell phone to get
position data undergrounds a group tests were
performed using setup seen in Figure 1. Performed tests
where: A. Visual inspection of usability B. Positioning
error on the central line of the tunnel, where the
mapping was made in 10 separate locations. C.
Positioning error on the central line of the tunnel, when
walking distance was increased by including every
second point. D. Positioning error on the side of the
tunnel in 10 separate locations. E. Initial tests with
vehicles for future use.
Figure 1. Schematic map about the experiment
2.1 Principle of magnetic field based
positioning
The magnetic field is usually presented as a constant
field as in Figure 2. Magnetic field has local variation as
well as time based variation. Figure 3 shows a more
localized map of the size of 5 km x 5 km, where
variations are clearly seen.
Finally on the mine tunnel level (Figure 4) we see
the same variations on the magnetic field. Using this
data as a map a position can be calculated. Fluctuations
on the magnetic field can be both natural or generated
by steel and reinforced concrete structures, electric
power systems, electric and electronic appliances [8]. In
the mines sufficient variations in magnetic fields to
enable magnetic field positioning is still a question mark
as the man made magnetic constructions and local
environmental variations are different around the
globebut it has been shown to work now at least in two
mines in Finland.
It has to be noted that the magnetic map is not
unique in given location, but algorithms are used to
derive position when object moves through the map.
More there is anomalies the easier the location is to
define [10]. Widely used algorithm in robotics to this
task is Monte Carlo localization.
Figure 2. Earth’s magnetic field. Source:
Wikimedia Commons. Image by Zurek [16]
Figure 3. Magnetic map [µT] of the Pyhäsämi
mining area in Finland. Image courtesy of J.
Haverinen and A. Kemppainen [7]
Figure 4. Magnetic field map [µT] of the tunnel.
Image courtesy of J. Haverinen and A.
Kemppainen [7]
POSTER PAPER

2.2 Data acquisition preparations
Pre data acquisition a map of the studied tunnel was
imported to the IndoorAtlas service. Three points from
the map were chosen as wide apart as possible (600
meters) and were mapped to correspond to known
locations. Points in the picture were chosen as carefully
as possible using image viewer with full zoom to get xy-
pixel coordinates. Then these points were referenced to
known reference coordinate points (EUREF89) using
the IndoorAtlas web service, see Figure5. The image
was automatically resized when moved to the cloud by
the web service and thus scaling with image size was
calculated to three reference picture points.
Two sources of error are present here. Firstly,
selecting the correct pixels contains an error and
secondly, reference points are transformed from mine
specific coordinate system to KKJ (an older Finnish
standard) and then again to EUREF89. Further the
IndoorAtlas uses WGS84-coordinate system and there is
an error element between the two of magnitude of 30
cm in the year 2009 [17].
Figure 5. A map of the tunnel used in the
measurements as seen in the IndoorAtlas web
service. Three reference points to fix map to
correct location was used.
A program, KemiMineDataMain, was created to
Android phone by modifying IndoorAtlas API Example
0.53, see Figure 6. Data saving, time stamp and
measurement running index were added. The
programming language JAVA was used with Eclipse
IDE. The program will list current time, number of
measurement, roundtrip time to the servers, Lat and Lon
in WGS84, X and Y in meters from the map’s upper
corner, I and J pixel coordinates again from the map’s
upper corner, heading in degrees, and server side
estimation of probability about the correctness of the
position.
On the site initial mapping was performed using
Android 4.4.2 LG Nexus 4 phone and software
IndoorAtlas Mobile 0.6.2.1382 APK. Figure 1. shows
the area mapped. The area was divided to two lines AB-
BC and magnetic field recording and validation was
made.
Two lines of reference points were measured.
Middle line was as close of the recorded and validated
line (Figure 1) as possible and the second reference line
was measured to the very side of the tunnel, of where
we assume the largest errors are due the measurement
and validation method.
Figure 6. KemiMineDataMain. Android data
collection software running on Nexus 4 phone.
2.2.1 Measurement A
Visual inspection of usability was performed using
IndoorAtlas Mobile APK where researchers and mine
worker walked around the mapped area and made their
observations (Figure 9). In this experiment a map of the
area was presented and a person’s location was moving
just like it does in everyday GPS-based localization
software.
2.2.2 Measurement B
Positioning error on the central line of the tunnel
was measured next. Ten reference points were measured
with Leica Viva R1000 robotic total station (Figure 8.)
created in an L-shape (Figure 1) with 4 points in the
cross tunnel and 6 in the main tunnel. These points were
measured in the Kemi mine coordinate system. A total
of 40 measurements were made, every point four times,
The 31st International Symposium on Automation and Robotics in Construction and Mining (ISARC 2014)

along the path 1 to 10 - 10 to 1 - 1 to 10 - 10 to 1. First
half of the data measurements were made using statiivi
so that the phone stays totally still for one minute. After
realizing the results do not change if the movement of
the phone is small, the rest 20 points were measured
keeping the phone still by hands only the time that was
need to take down the measurements.
2.2.3 Measurement C
Positioning error on the central line with increased
walking distance was performed using the same 10
reference points than in measurement B. The distance
was increased because we made an assumption that
system will perform better this way due its location
principles. A total of 20 measurements were made in
order (see Figure 1.) 10-8-6-4-2-1-3-5-7-9.
2.2.4 Measurement D
Positioning error on the side of the tunnel was
performed along the path 11 to 20, see Figure 1. The
points were measured as close to the tunnel wall as
reasonable. Procedure was same than what was used in
the Measurement B. A total of 10 reference points were
used with two measurement each.
Figure 7. Mapped area in the tunnel. In the centre
of the tunnel red line is recorded and black line is
currently validated.
2.2.5 Measurement E
Vehicle tests were performed as an initial
preparation for full tests in the future. A magnetic field
magnitude was measured at the distances of 15 m, 10 m,
5 m and 0 m away from the operational mining drill (in
Figure 8.) using mobile application on Iphone4, with ten
measurements on each point. Then a magnetic field was
measured using the same application when mine van car
drove from the 10 meters to 10 cm from the device and
also by visually checking the position data. Also xy-
coordinates were recorded 3 times when the mine van
car drove from about 5 meters to the close proximity of
magnetic field positioning device.
Figure 8. Total station and mine drill used on the
measurement E.
2.3 Coordinate transformations and error
analysis
Error calculations were performed in the Kemi mine
coordinate system (KMCS) and are presented in meters.
The procedure to calculate errors was same in
Measurements B,C and D.
For calculating coordinate transformation to KMCS
from WGS84 three known points were used as far apart
as possible on the mine map (Figure 5) and mapping
was done using six parameter affine transformation,
further, ETRS89 and WGS84 were treated as a same
coordinate system. Kemi Mine uses wide net of
reference points for total stations which are known in
KMCS and KKJ. These points were used to check the
calculated transformation with the standard
transformation KKJ to ETRS89.
Measurement reference points (Figure 1) are known
in KMCS, but measured positions with the phone are
acquired in WGS84. Again using assumption given
earlier and a six parameter affine transformation a
transformation chain WGS84-ETRS89-KMCS is ready.
POSTER PAPER

These values are then compared to the values measured
with a total station in the KMCS.
Root-mean-square-errors (RMSE) were calculated,
Equation (1) where  is predicted value and 
reference value.
 

 
(1)
Average error, also in this case know as mean bias error
(MBE) [18], is calculated using equation 2.
 

 
For precision evaluation standard deviation is
calculated.
(2)
Because mapping was done on the central line of the
tunnel, the results on the Measurement D will suffer a
bias as the measurements are projected to this mapped
line. There seems to be no difference in Measurement B
and C, thus combining these measurements for a larger
measurement group is used as a main data set to study
error.
Additionally errors for distance d from

x and

y
was calculated, defined in Equation (3).

(3)
For errors in vehicle test average magnitude was
used when measurements were made. The visual
inspections were performed as subjective opinions.
3 Results
Visual inspection of the usability of the geomagnetic
field based positioning technique with mobile phone
500 meters underground gave good results. The position
presented graphically in the screen of the smart phone
was deemed satisfactory for positioning a person
underground by all four testers. In Figure 9 a
positioning view is shown, where position of the phone
is shown inside a blue circle. Notice how predicted
position circle gets smaller, when convergence error is
estimated to get smaller.
Root-mean-square error was calculated for
Measurements B, C, D and combined B+C and also
total distance error d (Table 1). Other measurements
include: standard deviation (Table 2), Maximum errors
(Table 3) and MBE (Table 4)
Table 1. Calculated RMSE values
Measurement
x(m)
y(m)
d(m)
B
6.27
3.12
7.00
C
7.51
2.47
7.90
B+C
6.71
2.92
7.32
D
8.83
3.56
9.52
Table 2. Calcualted standard deviation
Measurement
x(m)
y(m)
d(m)
B
4.88
2.46
4.23
C
4.57
1.47
4.32
B+C
6.70
2.89
4.22
D
4.57
1.47
4.32
Table 3. Calculated maximum errors
Measurement
x(m)
y(m)
d(m)
B
19.98
7.20
20.62
C
19.15
4.84
19.48
B+C
19.97
7.20
20.62
D
16.45
5.01
17.20
Table 4. Calculated MBE
Measurement
x(m)
y(m)
d(m)
B
3.39
3.09
4.17
C
4.88
2.46
4.23
B+C
0.36
0.36
5.98
D
-7.56
3.25
8.48
The transformation error on the measurement area
between KMCS and WGS84 was on average 0.04 cm
and -2.58 cm for x and y directions, with standard
deviation 0.03 cm and 0.21 cm.
The averages for magnetic field magnitudes when
approaching drilling machine were: 54.2 µT (15 m),
49.6 µT (10 m), 53.7 µT (5 m, another vehicle close)
and 44.3 µT (0 m).
Mine van approaching gave average errors 1.73 m
and 1.06 m, for x and y directions.
4 Conclusion
The experiments and measurements for magnetic
field based positioning error using mobile phone
showed great promise. Talks with the Kemi mine staff
revealed that first encounter with this positioning
technique was pleasant. RMSE value of 7.32 m on the
central line of the tunnel was calculated and for
unmapped area (side of the tunnel) 9.52 m. The standard
deviation behaved nicely between 4.22 - 4.23 meters
between the different measurements. This accuracy is
The 31st International Symposium on Automation and Robotics in Construction and Mining (ISARC 2014)

enough for safety and logistic operations. However,
when we compare the accuracy to earlier work [7] we
see a quite large difference of almost 6 meters. The
obvious difference between the two is the measurement
device, in this measurement was done using standard
smart phone, in earlier work a specially designed sensor
array was used.
The max distance error is quite large and reason for
that is unknown, it can be a glitch or Monte-Carlo
Localization was performing badly due magnetic field
being too monotone or a convergence error is possible
too. Also one possible source for it is a break in a data
connection.
The vehicle experiments showed also great promise
to the future work. There seemed to be no problems
when vehicle was near, although we do have to
remember that these tests were not performed in-depth
enough to answer if vehicles are a problem to the
method. There seem to be some changes in the magnetic
field near the vehicles.
To work the measurement device needs a constant
internet access; this is possible in the Kemi mine but not
necessary in all other mines. To make system work in
all possible scenarios underground we feel that the
magnetic field positioning calculations should be
performed on the phone. This also would make system
safer for disturbances like fire and landslide not to
mention everyday problems every wireless network
goes through time to time.
The positioning system is also, at the moment, only
optimized for people walking. When the distances are
great, like in mines, this is not convenient. However,
when it’s now proven that geomagnetic positioning with
mobile phone works underground, there shouldn’t be
any technical or financial obstacles to overcome this
limitation. There could be for example a fixed unit on
the vehicle completed with odometry.
Currently there is no support for automatic floor
selection, this could be solved in future versions and
should be solvable already by combining positioning
techniques like wireless base station selection for the
map.
We hope that this positioning technique will be seen
in mines in the near future in daily usage.
Figure 9. Visual inspection of the positioning.
References
[1] S. Zhong, Proceedings of the 2012
International Conference on Cybernetics and
Informatics. Springer, 2013.
[2] S. van Duin, L. Meers, and G. Gibson, “HARD
AUTOMATION TRENDS IN AUSTRALIAN
UNDERGROUND COAL MINES,” in 30th
International Symposium of Automation and Robotics in
Construction and Mining (ISARC 2013) Proceedings,
2013.
[3] X. Huang, W. Zhu, and D. Lu, “Underground
miners localization system based on ZigBee and
WebGIS,” in 2010 18th International Conference on
Geoinformatics, 2010, pp. 1–5.
[4] L. Zhang, X. Li, L. Chen, S. Yu, and N. Xiao,
“Localization system of underground mine trackless
facilities based on Wireless Sensor Networks,” in IEEE
International Conference on Mechatronics and
Automation, 2008. ICMA 2008, 2008, pp. 347–351.
[5] A. F. C. Errington, B. L. F. Daku, and A. F.
Prugger, “Initial Position Estimation Using RFID Tags:
A Least-Squares Approach,” IEEE Trans. Instrum.
Meas., vol. 59, no. 11, pp. 2863–2869, Mar. 2010.
[6] M. Dunn, P. Reid, and J. Ralston, “A practical
inertial navigation solution for continuous miner
automation,” Coal Oper. Conf., Jan. 2012.
[7] J. Haverinen and A. Kemppainen, “A
geomagnetic field based positioning technique for
POSTER PAPER

underground mines,” in 2011 IEEE International
Symposium on Robotic and Sensors Environments
(ROSE), 2011, pp. 7–12.
[8] J. Haverinen and A. Kemppainen, “Global
indoor self-localization based on the ambient magnetic
field,” Robot. Auton. Syst., vol. 57, no. 10, pp. 1028–
1035, Oct. 2009.
[9] “A Compass in Every Smartphone - IEEE
Spectrum.” [Online]. Available:
http://spectrum.ieee.org/semiconductors/devices/a-
compass-in-every-smartphone. [Accessed: 14-Feb-
2014].
[10] B. Li, T. Gallagher, A. G. Dempster, and C.
Rizos, “How feasible is the use of magnetic field alone
for indoor positioning?,” in 2012 International
Conference on Indoor Positioning and Indoor
Navigation (IPIN), 2012, pp. 1–9.
[11] J. Chung, M. Donahoe, C. Schmandt, I.-J. Kim,
P. Razavai, and M. Wiseman, “Indoor Location Sensing
Using Geo-magnetism,” in Proceedings of the 9th
International Conference on Mobile Systems,
Applications, and Services, New York, NY, USA, 2011,
pp. 141–154.
[12] S. Suksakulchai, S. Thongchai, D. M. Wilkes,
and K. Kawamura, “Mobile robot localization using an
electronic compass for corridor environment,” in 2000
IEEE International Conference on Systems, Man, and
Cybernetics, 2000, vol. 5, pp. 3354–3359 vol.5.
[13] J. Haverinen and A. Kemppainen, “A global
self-localization technique utilizing local anomalies of
the ambient magnetic field,” in IEEE International
Conference on Robotics and Automation, 2009.
ICRA ’09, 2009, pp. 3142–3147.
[14] D. Navarro and G. Benet, “Magnetic map
building for mobile robot localization purpose,” in IEEE
Conference on Emerging Technologies Factory
Automation, 2009. ETFA 2009, 2009, pp. 1–4.
[15] E. Georgiou and J. Dai, “Self-localization of an
autonomous maneuverable nonholonomic mobile robot
using a hybrid double-compass configuration,” in 2010
7th International Symposium on Mechatronics and its
Applications (ISMA), 2010, pp. 1–8.
[16] Zureks, “English: Schematic representation of
Earth’s magnetic field lines,” 12-Jan-2012. [Online].
Available:
http://commons.wikimedia.org/wiki/File:Earth%27s_ma
gnetic_field,_schematic.png. [Accessed: 27-May-2014].
[17] K. Grinderud, H. Rasmussen, and S. Nilsen,
GIS: The Geographic Language of Our Age. Tapir
Academic Press, 2009.
[18] T. A. Reddy, Applied Data Analysis and
Modeling for Energy Engineers and Scientists. Springer,
2011.
The 31st International Symposium on Automation and Robotics in Construction and Mining (ISARC 2014)