Indoor Pedestrian Localization for Mobile Devices
The Model
Jonáš Ševčík
Faculty of Informatics, Masaryk University, Botanická 68a, Brno, Czech Republic
Keywords: Indoor Localization, Android, 802.11 Fingerprinting, Mobile Phone Tracking.
Abstract: Indoor localization using mobile devices is one of the emerging areas of today’s interest. This paper
presents a model for indoor localization based on 802.11 Wi-Fi fingerprinting in combination of inertial
navigation running in parallel. We introduce a novel model scheme, where we use a state-of-the-art
Compass system. The system is enhanced by clustering and the position calculation is influenced by the
distance travelled between each fingerprinting, allowing us to eliminate improbable location estimations.
Proposed system is supposed to be resilient to received signal strength blocking caused by a human body,
and also to be more accurate than other up-to-date solutions.
1 INTRODUCTION
Navigation has become inherent part of human lives.
The ability to successfully navigate oneself in an
unknown environment has lead into constructing
various means of navigation. These techniques and
technologies were gradually enhanced resulting into
more precise and reliable navigating systems. In
1994 United States Department of Defense finished
fully operational Global Positioning System (GPS)
providing the Standard Positioning Service for
everyone. With the decrease of prices of consumer
electronics in which was GPS frequently integrated,
it became widely spread technology used in
navigation systems. Location has proven to be an
important source of contextual information. If a
device can determine its own location then it can
infer its surroundings and adapt accordingly
(Woodman, 2010). One of the features of navigation
systems is to track position of pedestrians. But
despite the usage of GPS in open spaces, it is almost
impossible to use it indoors. Tracking GPS signals
indoors typically requires a receiver capable of
tracking signals with power levels ranging from
160dBW to 200dBW, however a typical receiver has
a noise floor of around 131dBW (Dedes, 2005).
Furthermore, to be fully functional, GPS relies on
direct visibility of at least 4 satellites (Kumar, 2008),
which is hardly attainable in closed spaces.
Multipath effects (signal multiplication caused by
reflections) are likely to cause degradation in
accuracy even if a receiver is able to track signals
from a sufficient number of satellites. Unlike in
outdoor environments, reflected signals are often
stronger than those received via direct line-of-sight
when indoors. Hence it is difficult for a receiver to
identify and track the correct (i.e. direct) signals
(Woodman, 2010).
It came into realization that for indoor use, GPS
has to be substituted by a different system. Emerged
solutions were based on a range of technologies
including lasers (Gutmann, 1996), ultrasound,
wireless networks, even a magnetic field
(Woodman, 2010; Storms, 2009). Unfortunately,
those solutions are depending on preinstalled
infrastructure, thus making them unusable in
ordinary environments.
2 RELATED WORK
Radiolocation based on 802.11 localization has been
an active area of research for the past two decades.
Originally, the research was focused on robot
navigation. Over the years, area of interest became
focused on pedestrians. Wi-Fi APs are now densely
deployed in many urban environments. A device can
estimate its position by searching a database
containing the locations of such APs for ones that
are visible from its current location. The accuracy of
488
Šev
ˇ
cík J..
Indoor Pedestrian Localization for Mobile Devices - The Model.
DOI: 10.5220/0005104604880493
In Proceedings of the 9th International Conference on Software Engineering and Applications (ICSOFT-EA-2014), pages 488-493
ISBN: 978-989-758-036-9
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
such approach is highly dependent on the density of
APs. However, positioning errors are at worst equal
to the range of an AP (typically 50 100 m)
(Woodman, 2010). Recent development has brought
focus on cheap Bluetooth LE sensors, which can
substitute more expensive Wi-Fi APs. Possible
disadvantages of radiolocation are that a database of
Wi-Fi APs/Bluetooth sensors must be constructed
beforehand and it must be frequently updated, since
there may be changes in the infrastructure.
2.1 Techniques Using AP Location
This approach is based on constructing a radio map,
in which Wi-Fi signals are recorded at locations of
interest along with a vector of received signal
strength (RSS). This technique is known as
fingerprinting. Location estimation than can be
reproduced by matching measured RSS readings
against previously established database. One of the
first techniques using deterministic fingerprinting
(computes a single best guess of a position of a
device) was RADAR (Gutmann, 1996; Bahl, 2000).
This technique shows error less than 9 m 95% of the
time. RADAR was enhanced over the years. One of
the successors is Horus (Youssef, 2005), which uses
a stochastic description of the RSS map with the
combination of maximum likelihood probabilistic
based approach. The Horus system reports accuracy
of 1.4 m 95% of the time. Also, Horus was enhanced
in the technique Compass by using digital
compasses and measuring RSS in 8 directions in one
location. There is also other technique, such as
(Varshavsky, 2007), which presents possibilities of
using GSM signal, or (Azizyan, 2009), presenting
map building, based on notable environment features
such as sound, light, colors etc.
All these presented approaches are depending on
detailed environment measurements. Some sources
(Haeberlen, 2004) state that 28 man-hours were
required to construct a radio map covering a 12,000
m
2
building. Moreover, this process must be
repeated whenever an AP is moved/added/removed,
to ensure reliable results. To avoid such time
consuming operations, there have been made several
efforts. For example DAIR (Bahl, 2005) does not
involve mapping at all, but it assumes, that used
environment is densely equipped with APs. We have
to realize, that all presented accuracy results were
made in areas housing multiple APs (typically not
family houses or apartments). In order to achieve
usable results it is required to have installed at least
3 APs (Woodman, 2010).
2.2 Inertial Positioning and Dead
Reckoning
Inertial positioning uses MEMS, particularly
magnetometers, accelerometers and gyroscopes, to
calculate change in mobile device’s location in time.
These systems (Cho, 2006) are based on dead
reckoning. We can use course heading from
magnetometer and readings from accelerometer can
serve as a step detector. With known course and
traveled distance, it is possible to calculate position
of the measuring device. When tracking a
pedestrian, we can use information from sensors
along with an estimate of a stride length to track his
approximate position relative to his starting point.
Over time, the calculated position will become less
accurate. This is caused by adding fixed stride
lengths (which differ from the actual ones), so that
the overall error accumulates. The longer path the
pedestrian walks the bigger error it accumulates.
This accumulation of error is commonly referred to
as drift (Ojeda, 2007). Drift is acceptable if the
target is tracked only for a short period of time. It is
apparent that if we want to build a system, which is
capable of tracking for a long period of time, to
avoid drift, position must be regularly calibrated by
absolute positioning system.
Pedestrian dead reckoning can be divided into 2
groups. One group is based on step detection, the
other on inertial navigation. Step based pedestrian
dead reckoning is based on detecting steps and
estimating stride length. As mentioned above, when
the step is detected, heading and step length is used
to update position of the tracked pedestrian. Main
advantage of this approach is that the tracking
device can be mounted almost anywhere on the
pedestrian’s body. Usual locations are foot, waist or
head. Since, 1 axis accelerometers are used for this
method; step is detected by peaks in the received
signals (Steinhoff, 2010). This also have a
disadvantage, because having only 1 axis
accelerometer means that this system works with the
assumption, that pedestrian is moving only forwards.
Step length estimation is also problematical, since
step length varies from step to step. Several
algorithms have been developed for estimating step
length. As presented in (Alvarez, 2006) step length
can be estimated using user-specific constants or
neural networks. Problem is that neural networks
must be trained for each individual user, which can
be time consuming. The last drawback is that such
systems cannot differentiate vertical displacement,
thus it is impossible to recognize for example
movement up on the stairs.
IndoorPedestrianLocalizationforMobileDevices-TheModel
489
The second group of pedestrian dead reckoning
is based on inertial navigation. It is a navigation
technique in which measurements provided by
accelerometers and gyroscopes are used to track the
position and orientation of an object relative to a
known starting point, orientation and velocity. Thus
it must use full 3 dimensional inertial navigation unit
(Foxlin, 2005). Actions must be taken to prevent
accumulation of drift namely by placing measuring
device exclusively on the foot of the pedestrian and
then correcting the inertial navigation system when
the foot is grounded. At the time, sensors report
near-zero velocity. As shown in (Godha, 2006)
Bayesian filters prove to offer an ideal framework to
track such correlations that exist between
accumulating errors in different components and to
apply zero velocity updates in a way that allows
correlated errors in all components of the state of the
inertial navigation system to be corrected.
Particularly, Kalman filters are widely used
(Woodman, 2010). The main advantage of this
group of pedestrian dead reckoning is that the
system does not assume only forward movement.
Hence, it can correctly handle moving backwards or
strafing.
2.3 Hybrid Models
Hybrid models are based on combining various
localization techniques to achieve better accuracy
results. It is important to note that combining more
localization techniques also results into higher inner
complexity, thus overall position error may be
influenced by more factors. For the purposes of our
research we are going to focus only for those based
on 802.11 and inertial localization.
The work (Evennou, 2006) introduces solution
based on Wi-Fi, inertial navigation system, and
particle filtering. This system is using 6-degrees-of-
freedom zero velocity updating. Main supporting
feature of this system is particle filtering. Possible
disadvantage is relatedness to known precise
building layout (needed for particle filtering) and
placement of the measuring unit to the foot of the
user.
Another most advanced hybrid system (Frank,
2009) presents Wi-Fi localization system supported
by inertial navigation. Emphasis of this system is
given to sparse Wi-Fi fingerprint sampling. Authors
used fingerprinting points with a span of 5 m. As
opposed to the previously mentioned system
(Evennou, 2006), inertial navigation implements
acquiring a heading data from the additional
magnetometer. The inertial system is processed by
its own Bayesian filter to estimate individual steps
of the user; these estimates are then combined with
the estimate of the location from fingerprinting. This
approach allows precise data processing at sensors’
local sampling rates, which reduces overall
complexity without suffering from significant loss of
final estimation accuracy (Frank, 2009). This system
shows arithmetic mean error of 3.1777 m for pure
Wi-Fi fingerprinting and 1.6468 m for the fusion of
fingerprinting and shoe data.
The goal of presented hybrid systems is to use as
few calibration locations as possible and to rely on
the short-term accuracy of foot mounted inertial
dead-reckoning (for instance zero velocity update
based techniques) among these points. The role of
the Wi-Fi positioning here is to provide long term
accuracy in the area of interest (Frank, 2009).
3 PROPOSED MODEL
We have identified, that current known solutions are
either not mobile-based (Frank, 2009; Ojeda, 2007
etc.), they are dependent on construction of support
infrastructure (Lukianto, 2011) or the operating
hardware is directly mounted to the foot or leg
(Serra, 2010). Even in the cases when a mobile
device is used, the solutions are entirely based on
Wi-Fi or Bluetooth localization (Martin, 2010), or a
combination of visual tracking and step detection
based systems (Kothari, 2012; Lukianto, 2011;
Serra, 2010 etc.). Author of (Woodman, 2010)
claims that step detection based inertial navigation is
suitable for usage with mobile devices, but we
believe otherwise. From studies (Serra, 2010; Shala,
2011) it is visible, that approach based on step
detection is not optimal. In combination with Wi-Fi
tracking, it is possible to achieve accuracy
improvement (comparing to pure Wi-Fi localization
system), but results in (Ševčík, 2013) show gradual
increase of drift over the travelled distance, which
had grown in the worst case by 7 m. This actually
made fused position estimate worse than using just
plain Wi-Fi positioning. The study (Serra, 2010)
reveals comparable results and claims that using a
particle filter might improve drift accumulation,
which was proved wrong in (Ševčík, 2013), where
particle filter was already used. The same results
were confirmed in (Shala, 2011). After taking into
consideration all above mentioned systems and
techniques, our objective is to create universally
applicable hybrid system for mobile indoor
localization. Proposed novel solution, as presented
by Figure 1, will be composed of 2 localization
ICSOFT-EA2014-9thInternationalConferenceonSoftwareEngineeringandApplications
490
techniques combined together for achieving better
accuracy. We have decided to adopt parallel
approach in order to process data locally and to
reduce overall complexity. This separation also
allows estimation filters to run at their local
sampling rates, helping to process data without
significant loss of final destination accuracy.
Location estimation will be based on probabilistic
Wi-Fi localization, using techniques proposed in
Horus system (Youssef, 2005), but for dealing with
problem of blocking effects caused by a human
body, we will extend this system by introducing
Compass system orientation based Wi-Fi
fingerprinting as proposed in (King, 2006).
Sampling of the signal strength for selected
orientations at each reference point during the
fingerprint gathering phase and combining a subset
of these values to histograms in the location
estimation phase, allows orientation specific signal
strength distribution to be computed and utilized to
increase the accuracy of position estimates. The
Compass system indicates average error distance of
less than 1.6 m. Therefore, it will be used for initial
position estimation and further position calibrations.
Furthermore, Compass is optimal for addressing the
influence of human body as an obstacle, which is
blocking Wi-Fi signal.
Figure 1: Localization system abstraction.
The second localization technique will feature
inertial navigation. Inertial navigation will be used
for improving accuracy of Wi-Fi localization. We
propose a system based on gyroscope,
accelerometer, and magnetometer fusion. The fusion
will help to reduce sensor flaws. Data from
accelerometer and gyroscope will be filtered,
normalized and processed by step detection
algorithm, to be consequently used for zero velocity
updating (calibration technique reducing drift
accumulation). Subsequently, normalized linear
acceleration vector and course will be processed by
extended Kalman filter (EKF). We have chosen EKF
to estimate the instantaneous state of a linear
dynamic system, with measurements linearly related
to the state but influenced by white noise. The
estimator uses incomplete and noise-corrupted
measurements to estimate the tracking object’s state,
and the resulting estimation is statistically optimal
with respect to any quadratic function of estimation
error. Thus parallel EKF will be used to estimate
position errors, velocity errors, and hand movement
errors in terms of a Gaussian probability density
function. Integration of inertial navigation system
and Wi-Fi localization system will be performed via
the main particle filter, which will keep track of
position and course (heading). With a particle filter,
more information than RSS location and inertial
navigation data can be fused. In particular, a
building map is another very useful information
source, since a lot of location related data can be
extracted from the building structure information.
For the tracking problem, this information helps to
reduce the uncertainty of the walking trajectory.
Using a particle filter, the estimation can be
improved by deleting impossible particles, i.e. the
particles which would have crossed a wall.
The Compass as it is introduced in (King, 2006)
does not integrate any clustering algorithm, therefore
when calculating user’s position, all database data
need to be evaluated. This is not needed for small
data sets, but for large data sets, it is necessary to
select only a subset of fingerprint data, for the
system to be capable of running in real time. We
propose usage of Incremental Triangulation
algorithm as it is presented in (Youssef, 2006). If
during the location determination phase APs are
used incrementally, one after the other, then starting
with the first AP (with the highest value of RSS), the
algorithm restrict our search space to the locations
covered by this AP. The second AP chooses only the
locations in the range of the rst AP and covered by
the second AP and so on, leading to a multi-level
clustering process. Location clustering can be further
enhanced using mobile sensors. We suggest
additional AP filtering using heuristics. From the
data obtained from inertial localization we can
improve the AP selection. Knowing user’s course
and approximate travelled distance from the last
known location, we can constrain the final location
estimation only to those points, which are in the
evaluated travel distance.
IndoorPedestrianLocalizationforMobileDevices-TheModel
491
4 CONCLUSIONS AND FUTURE
WORK
This paper presented hybrid pedestrian localization
system for mobile devices. The key concept is
parallel combination of inertial navigation and Wi-Fi
localization, where both parts are mutually
beneficial. Inertial navigation can be calibrated by
data obtained by Wi-Fi localization. On the other
hand, Wi-Fi localization accuracy can be enhanced
by restraining a selection of improbable results,
when taking sensor data into account mainly the
digital compass and the approximated travelled
distance.
In our previous research, we have focused on
developing inertial navigation based on step
counting. We have implemented techniques for step
detection, which proved to be nearly 100% positive
in detecting steps while walking continuously. Now,
we are implementing the Compass system enhanced
by clustering. Current prototype is capable of
localizing a pedestrian with an error close to 4
meters. We believe that combination with inertial
navigation will reduce this error and provide better
results. Scaling will also affect the overall speed of
the system.
REFERENCES
Alvarez, D., González, R. C., López, A., Alvarez, J. C.,
2006. Comparison of step length estimators from
weareable accelerometer devices. In Engineering in
Medicine and Biology Society.
Azizyan, M., Constandache, I., Choudhury, R. R., 2009.
Surroundsense: mobile phone localization via
ambience fingerprinting. In Proceedings of the 15th
annual international conference on Mobile computing
and networking. ACM.
Bahl, P., Padhye, J., Ravindranath, L., Singh, M.,
Wolman, A., Zill, B., 2005. Dair: A framework for
managing enterprise wireless networks using desktop
infrastructure. In HotNets IV.
Bahl, P., Padmanabhan, V. N., 2000. Radar: An inbuilding
rf-based user location and tracking system. In
INFOCOM 2000. Nineteenth Annual Joint Conference
of the IEEE Computer and Communications Societies.
Proceedings. IEEE.
Cho, S. Y., Park, C. G., 2005. Mems based pedestrian
navigation system. In Journal of Navigation.
Dedes, G., Dempster, A. G., 2005. Indoor GPS
positioning. In Proceedings of the IEEE Semiannual
Vehicular Technology Conference.
Evennou, F., Marx, F., 2006. Advanced integration of Wi-
Fi and inertial navigation systems for indoor mobile
positioning. In Eurasip journal on applied signal
processing.
Foxlin, E., 2005. Pedestrian tracking with shoe-mounted
inertial sensors. In Computer Graphics and
Applications. IEEE.
Frank, K., Krach, B., Catterall, N., Robertson, P., 2009.
Development and evaluation of a combined wlan &
inertial indoor pedestrian positioning system. In ION
GNSS.
Godha, S., Lachapelle G., Cannon, M. E., 2006. Integrated
gps/ins system for pedestrian navigation in a signal
degraded environment. In ION GNSS.
Gutmannm J., Schlegel, C., 1996. Amos: Comparison of
scan matching approaches for self-localization in
indoor environments. In Advanced Mobile Robot.
IEEE.
Haeberlen, A., Flannery, E., Ladd, A. M., Rudys, A.,
Wallach, D., Kavraki, L., 2004. Practical robust
localization over large-scale 802.11 wireless networks.
In Proceedings of the 10th annual international
conference on Mobile computing and networking.
ACM.
King, T., Kopf, S., Haenselmann, T., Lubberger, C.,
Effelsberg, W., 2006. Compass: A probabilistic indoor
positioning system based on 802.11 and digital
compasses. In Proceedings of the 1st international
workshop on Wireless network testbeds, experimental
evaluation & characterization. ACM.
Kothari, N., Kannan, B:, Glasgwow, E. D., Dias, B., 2012.
Robust indoor localization on a commercial smart
phone. In Procedia Computer Science.
Kumar, B. H., Reddy, K. C., Namassivaya, N., 2008.
Determination of gps receiver position using
multivariate newton-raphson technique for over
specified cases. In International Journal of Applied
Engineering Research.
Lukianto, C., Sternberg, H., 2011. Stepping: smartphone-
based portable pedestrian indoor navigation. MMT.
Martin, E., Vinyals, O., Friedland, G., Bajcsy, R., 2010.
Precise indoor localization using smart phones. In
Proceedings of the international conference on
Multimedia. ACM.
Ojeda, L., Borenstein, J., 2007. Non-gps navigation with
the personal dead-reckoning system. In Defense and
Security Symposium. International Society for Optics
and Photonics.
Serra, A., Carboni, D., Marotto, V., 2010. Indoor
pedestrian navigation system using a modern
smartphone. In Proceedings of the 12th international
conference on Human computer interaction with
mobile devices and services. ACM.
Shala, U., Rodriguez, A., 2011. Indoor positioning using
sensor-fusion in android devices. Kristianstad
University.
Steinhoff, U., Schiele, B., 2010. Dead reckoning from the
pocket-an experimental study. In Pervasive Computing
and Communications. IEEE.
Storms, W. F., 2009. Magnetic field aided indoor
navigation. Technical report, DTIC Document.
ICSOFT-EA2014-9thInternationalConferenceonSoftwareEngineeringandApplications
492
Varshavsky, A., de Lara, E., Hightower, J., LaMarca, L.,
Otsason, V., 2007. Gsm indoor localization. In
Pervasive and Mobile Computing.
Ševčík, J., 2013. Pedestrian Localization in Closed
Environments. In Environmental Software Systems.
Fostering Information Sharing. Springer.
Woodman, O., 2010. Pedestrian localisation for indoor
environments. St Catharines College.
Youssef, M., Agrawala, A., 2005. The horus wlan location
determination system. In Proceedings of the 3rd
international conference on Mobile systems,
applications, and services. ACM.
Youssef, M., Agrawala, A., 2006. Location-clustering
techniques for wlan location determination systems.
International Journal of Computers and Applications.
IndoorPedestrianLocalizationforMobileDevices-TheModel
493