Smartphone pedestrian navigation by foot

Smartphone Pedestrian Navigation
by Foot-IMU Sensor Fusion
Tobias G¨adeke, Johannes Schmid, Marc Zahnlecker, Wilhelm Stork, Klaus D. M¨uller-Glaser
Karlsruhe Institute of Technology (KIT)
Institute for Information Processing Technologies (ITIV)
Engesserstr. 5, 76131 Karlsruhe, Germany
Email: {tobias.gaedeke, johannes.schmid, marc.zahnlecker, wilhelm.stork, klaus.mueller-glaser}@kit.edu

Abstract—Determining one’s own position by means of a
smartphone is an important issue for various applications in the
fields of personal navigation or location-based services. Places
like large airports, shopping malls or extensive underground
parking lots require personal navigation but satellite signals
and GPS connection cannot be obtained. Thus, alternative or
complementary systems are needed.
In this paper a system concept to integrate a foot-mounted
inertial measurement unit (IMU) with an Android smartphone
is presented. We developed a prototype to demonstrate and
evaluate the implementation of pedestrian strapdown navigation
on a smartphone. In addition to many other approaches we also
fuse height measurements from a barometric sensor in order to

stabilize height estimation over time. A very low-cost single-chip
IMU is used to demonstrate applicability of the outlined system
concept for potential commercial applications.
In an experimental study we compare the achievable accuracy
with a commercially available IMU. The evaluation shows very
competitive results on the order of a few percent of traveled
distance. Comparing performance, cost and size of the presented
IMU the outlined approach carries an enormous potential in the
field of indoor pedestrian navigation.

I. I NTRODUCTION AND P ROBLEM S TATEMENT
With the more or less ubiquitously available smartphones,
location-based services and personal navigation have become
important areas of interest for research and industries. Most
of the time, positioning of a person carrying a smartphone
is achieved by means of the global positioning system (GPS).
Then, based on this position, the user is provided with spatially
related services such as information during a sightseeing
trip or museum visit, or with directions on how to find his
destination. However, GPS has the fundamental disadvantage

of usually not functioning sufficiently inside buildings. Also
”urban canyons” between skyscrapers in larger cities often
pose a problem. Both phenomena result in unavailable or at
least inaccurate position information. One possible way to
deal with this is to add a redundant localization system that
can take over navigation for a period of time. For example
automotive GPS systems usually rely on the cars velocity
input and map information to bridge the time a car is in a
tunnel. For pedestrian localization various approaches exist
employing other smartphone sensors. On one hand, a possibility is to make use of WiFi signals and to determine one’s
position from a comparison of the current WiFi fingerprint
with a predetermined database [1], [2]. These fingerprinting

Fig. 1. Implementation of IMU prototype, top and bottom view. Casing with
clip on the backside for foot mounting.

systems require, however, a database that has to be established
beforehand. The performance of such systems heavily depends
on the quality of the database and the WiFi access points
in range. On the other hand, completely infrastructure-free

systems exist that make use of inertial sensors to perform
dead reckoning [3], [4]. These systems usually rely on step
detection in combination with step length and direction estimation to sequentially determine a new position from the
previous position. A common extension to this method is the
inclusion of map knowledge to limit localization to possible
paths within the area of interest [5]. The major issue of these
inertial dead reckoning systems is, however, the obtainable
accuracy and long term stability due to the sensor drift of the
smartphone’s inertial sensors. To cope with this, the usual way
is to use a foot mounted inertial measurement unit (IMU) for
inertial strapdown navigation allowing to make use of human
motion patterns in an adapted system model [6], [7]. However,
most previous work did not set high value on the actual
system integration and implementation issues but limited their
assessment to a PC based evaluation of the concept. Also,
most of the time expensive sensor equipment was used that
does not allow a mass market adoption.
In this paper, we present a system level concept and prototype implementation of a foot mounted IMU that connects
to an Android smartphone via Bluetooth. Figure 1 shows the
prototype printed circuit board (PCB) and the casing that

allows simple fixing to a person’s shoestring with a special
clip on the backside of the box. A size of 2 ∗ 3 cm2 of the
PCB and the low weight of the IMU allows for a comfortable
placement. In addition to inertial and magnetic sensors the
developed hardware also incorporates a barometer and we
outline an approach to fuse height information in order to

improve 3D localization accuracy.
The structure of the paper is as follows: in Section II the
current state of the art is outlined and relevant previous work is
described. In Section III the proposed overall system concept
and the hardware concept of the sensor box is presented.
In Section IV the system design in terms of the developed
hard- and software and algorithmic approach is outlined. A
focus has been put on additional fusion of height information
from a barometric pressure sensor. After an evaluation of
the achievable accuracies and smartphone performance in
Section V the results are discussed in Section VI. The paper
is concluded and a short outlook on future research directions
is given in Section VII.

II. R ELATED W ORK
A. Foot-mounted Pedestrian Navigation
During the last decade interest in pedestrian inertial navigation has grown with the rapid price decline of micro-electromechanical systems (MEMS). Two major research directions
can be identified by categorizing the placement of inertial sensors: on one hand mounting the sensor on the foot and on the
other carrying it at an arbitrary position on the person’s body
(e.g. hip, pocket, chest, rucksack). From the algorithmic view,
both methods initially started by employing step counting
and step length estimation [4]. With further improvements in
inertial sensor technology foot mounted methods started using
strapdown navigation techniques with zero-velocity updates
(ZUPT) [6]. The ZUPT resets the velocity to zero each time
the foot comes to rest on the ground and thus limits the
error growth of the system to a linear function. Similarly
to ZUPT zero angular rate updates (ZARU) were introduced
to correct drift from gyro sensors [8]. Other methods have
been presented in order to reduce the errors resulting from
the integration of the sensor readings. The main focus of
these methods lies on yaw angle drift reduction since this
error is not observable with inertial sensors only (inclinometer
use of accelerometer can correct errors in the roll and pitch

angles). Therefore, the inclusion of a compass sensor has been
considered extensively along with software algorithms like
heuristic drift reduction (HDR) and others [9]–[11]. Another
very important process for foot mounted pedestrian navigation
is the reliable detection of the stance and swing phase of
the foot. Many different methods based on acceleration and
angular rates or combinations of them have been studied. It
has been shown that most information about the steps phase
is carried by the angular rates [12]. When acceleration or
acceleration variance is also considered, results can be slightly
improved [10].
The errors of current foot mounted approaches are typically
on the order of a few percent of the traveled distance. With the
mentioned HDR techniques or by means of improved sensor
qualities these figures can be lowered down to less than 1 %
of traveled distance depending on the scenario.
On the other hand, mounting the inertial sensors somewhere
else on the body and performing step counting and step length
estimation is still the method of choice when processing power


or practicability are important issues [13]. The advantage of
this method lies in the possible single device integration in
other products like smartphones, cameras, etc. and also allows
navigation based on low cost sensors. However, achievable
accuracies are typically on the order of 10 % of traveled
distance with such approaches. These figures can be further
improved by employing activity classification (e.g. walking,
running), resulting in higher processing costs [14]. Still, the
accuracy performance of foot mounted methods cannot be
matched. Especially if the system is to be used for longer runs
without any correcting system like GPS, better accuracies are
needed. This could be the case for smartphone navigation in
large shopping areas, airports or underground parking.
B. Inertial Smartphone Navigation
With the widespread market penetration of smartphones one
of the major applications is their use for personal navigation.
Many applications provide location based services (LBS) like
”the nearest coffee shop, hotel” etc. All these services are
based upon GPS or WiFi connectivity and are not available
with high precision in many indoor or underground environments.

The inertial sensors employed in today’s smartphones are
typically low cost sensors mainly intended for games and
enhanced user experience. An external IMU for the use with
a smartphone is considered in [15] but this work focuses on
the sensor and hardware selection mostly. Some algorithmic
approaches are mentioned but an overall system concept has
not yet been worked out. An activity classification based on
a smartphone carried on the foot is presented in [16]. In
this work localization is mentioned but only evaluated on a
very coarse level (detection of floor changes). More promising
results in the field of smartphone navigation are obtained by
step counting and activity classification algorithms [3]. A new
interesting approach is to use the smartphones camera for step
detection by comparing a previously acquired image of the
foot with the current image of the smartphone’s camera [17].
As a complementary system, GPS and map matching is often
used to further improve the localization accuracy [3], [18],
[19].
When summarizing the related work, the importance of
indoor smartphone localization is evident but the achievable

accuracies with inertial approaches are not yet sufficient for
navigation in large buildings. On the other hand accuracy of
foot mounted inertial navigation has reached a state which
allows for very precise localization even for longer runs. Thus,
in this paper we try to combine the strengths of easy and low
cost sensor deployment with higher accuracies by using foot
mounted sensors connected to a smartphone for navigation.
III. S YSTEM C ONCEPT
We propose to solve the problem of pedestrian localization
and navigation within buildings by means of a tiny low-power
and low-cost foot-mounted device incorporating all necessary
sensors and being compatible to smartphones. This sensor box
wirelessly connects to the smartphone and transmits sensor

Visualization

Possible: wireless
sensor network

Processing


User interface
(2 Buttons, 2
LEDs)

Power supply
(LTC3554 step down converter +
Li-Ion battery)

Bluetooth
(PAN1321)

Micro USB
(UART / Charging)

UART

Data recording

Microcontroller

(STM32F103)

6D IMU
(MPU6050)
I²C

Magnetometer
(HMC5883L)

UART

Fig. 2.

IMU and smartphone based pedestrian navigation.

data from the foot as illustrated in Figure 2. For the use with
other ultra low power sports devices or for the integration with
other on body or external sensors (wireless sensor network,
WSN) an additional low power wireless connection is possible. In comparison to state of the art smartphone navigation
systems that make use of the smartphone’s internal sensors, the
use of an additional sensor box mounted on the foot of the
person allows to exploit strapdown inertial navigation. This
enables accurate position tracking even under severe indoor
conditions. The heterogeneous wireless connectivity allows
for the integration of the IMU in various other application
scenarios. This also includes other sensor placements on the
human body and other technical use cases (e.g. stabilization
of robots or autonomous vehicles). One such use case could
be person localization in ad-hoc scenarios (e.g. localization of
firefighters, patients and doctors) where a wireless network for
long term localization is established on-site by means of dead
reckoning [13].
A. Choice of Sensors and Hardware Design
Figure 3 gives a design overview of the sensors and
peripherals considered in the sensor box. A microcontroller
handles the sensor data, communication links and controls
the power supply. A low-cost single chip IMU is used for
a very small and cheap overall design. To allow for attitude
correction a magnetometer is considered as well as a barometer
for additional height correction. The communication between
IMU and smartphone for the transmission of raw or preprocessed data is established either by Bluetooth or an ANT+
interface. The ANT+ standard is widely used in the field of
sports electronics and is characterized by its very low power
consumption. Some of today’s smartphones are also equipped
with ANT+ interfaces (e.g. Sony Ericson Xperia arc). To allow
for an integration of the IMU into a ZigBee WSN, the ANT+
chipset can be exchanged by an 802.15.4 chipset as a hardware
design alternative.
B. Foot-mounted Inertial Pedestrian Navigation
The basic idea of the strapdown calculation is to determine
the position by double integration of the 3D-acceleration signal
in the appropriate direction. Therefore, the attitude of the
inertial system needs to be known and can be obtained by
an integration of the angular rates. These angular rates are

ANT +
(CC2571)

Fig. 3.

ZigBee
(CC2530)

Antenna
(Chip)

Barometer
(MS5611)

Hardware concept of the foot mounted sensor box.

measured by– a set of orthogonal gyroscopes. The sensors
measure in body-frame, which is the coordinate system of the
IMU, with respect to the inertial frame. In general, the sensors
also measure the earths turn rate, the location depended
gravity and transport rate [20]. However, the effects of these
influences are usually below the noise level of today’s MEMS
sensors. This means that these influences can be neglected for
pedestrian navigation.
This leads to a simplified model of the strapdown calculation as shown in the upper part of Figure 4. It is basically
reduced to the integration process of angular rates ω
~ and
double-integration of accelerations ~a. Due to the inherent drift
introduced by the integration step a stochastic filter is typically
used to correct this drift based on an underlying system model.
Additionally, for foot mounted inertial pedestrian navigation,
ZUPT and ZARU mechanisms correct the velocity and angular
rates in the system model. These inputs are pseudo measurements as no additional sensors are used, but a step detection
algorithm decides whether to apply the correction. For further
long term stabilization of the yaw attitude a magnetic sensor
is used as a compass usually.
IV. I MPLEMENTATION
A. Hardware
The hardware for the foot-mounted IMU was designed with
the intention of a very small, low-cost and low-power device.
The developed hardware together with the battery is built into
a casing measuring less than 14 cm3 with a weight below 15 g.
With the demand of motion enabled consumer systems fully
integrated single-chip IMUs are on the market now. Typically,
the performance of these IMUs is lower compared to single
axis sensors. We chose to use the Invensense MPU6050 IMU
for our hardware design. It is a fully integrated single chip 6axis device also offering an internal motion processor which
can process measurements for attitude calculation. However, as
one of our goals is to evaluate and optimize the parameters for
the requirements of pedestrian inertial navigation, we do not
use the motion processor. The hardware is complemented by a

changing weather conditions. Obtaining an absolute height
over sea level is not accurate enough under these circumstances. However, differential small pressure changes are very
accurate and allow for a sub-meter resolution. Therefore,
some assumptions on the atmospheric parameters are made
commonly. Such assumptions consider an average standard
atmosphere in which the pressure is only dependent from the
current height. This relation is described by the barometric
formula which is given in (1).
 g·M

L · h R·L
(1)
p = p0 · 1 −
T0
Fig. 4.

Strapdown algorithm with additional correction mechanisms.

Honeywell HMC5883 magnetometer and the MS5611 barometer from Measurement Specialities. The MS5611 allows for
very precise pressure measurements which result in a height
resolution of about 15 cm according to the datasheet. This
resolution allows for the detection of single stairs, but is also
useful in other contexts like human activity classification (e.g.
sitting, standing with hip mounted IMU). The communication
link between the hardware and the smartphone is realized with
a Bluetooth module as this is probably the most widespread
communication standard in smartphones. Additionally, another
wireless low power and lower data rate communication module
is implemented for increased interoperability with various
other systems. This can be either a Texas Instruments ANT+
chipset (CC2571) or an 802.15.4 chipset (CC2530) for the
use in a ZigBee network. These chipsets share a very similar
footprint and are implemented as a design alternative. Beside
the sensors and communication the hardware provides a user
interface in the form of two buttons and LEDs. The power
supply is realized with a small lithium-ion battery which can
be charged by a micro-USB connector. A voltage converter
supplies the sensors and digital components independently so
that influences on the measured data are kept to a minimum.

The barometric formula describes the pressure p in dependence
of the current height h. The other parameters can be assumed
constant for standard atmospheric conditions and the values
are given in Table I [21].
In the basic Kalman filter implementation the position error
in z-direction is predicted from the integration of the turnrate and acceleration error. Thus, it drifts with erroneous
measurements and the error growth is limited by the ZUPT
to a linear increase. The differential height information is
considered as an additional measurement input to the Kalman
filter. It is thus fused in the correction step of the filter. In our
implementation we fuse height updates only in stance phases
of the foot. This reduces influences from high frequency and
other noise in the signal. The difference from strapdown height
estimation hk and hk−1 between successive steps and the
measured height difference from the barometer hk,dif f are
then fused as a height error in the correction step of the
Kalman filter. The measurement matrix H is extended with
an additional line corresponding to the additional element mh
in the measurement vector m.
Hpz = [01×3 01×3 [001] 01×3 01×3 ]

(2)

mh = (hk − hk−1 ) − hk,dif f

(3)

C. Sensor Calibration
B. Strapdown Mechanization
The strapdown mechanization in our work is based on the
implementations presented in [6] and [10]. We did not consider
any HDR algorithms from [10] because the assumption of
mostly straight or rectangular paths does not hold in arbitrary
scenarios. The lower part in Figure 4 shows the Error State
Space Kalman filter and the implemented correction mechanisms based on step detection and the additional sensors.
The systems state vector δ~x contains the errors of attitude
δϕ
~ , angular rates δ~
ω , position δ~
p, velocity δ~v and acceleration
δ~a. It turned out in previous studies that small errors in the
measured data can lead to significant errors in z-direction for
longer runs. Thus, a pressure sensor (barometer) is integrated
to correct height measurements additionally. Obtaining height
from atmospheric pressure measurements is a fairly complex
procedure as it depends on various unknown parameters in the
atmosphere. That means that the pressure greatly varies with

One very important issue in the context of inertial navigation
is the calibration of the sensor system. Often, a major problem
is the lack of a high quality reference calibration system. This
means that other calibration methods need to be considered.
These have to be robust under difficult conditions and carry
the potential of an in-field calibration. For the accelerometer,
one calibration method is to solve a set of linear equations
TABLE I
C ONSTANTS FOR STANDARD ATMOSPHERIC CONDITIONS .
Symbol [Unit]
p0 [hP a]
L [K/m]
T0 [K]
M [kg/mol]
R [J/(mol · K)]
g [m/s2 ]

Value
1013.25
0.0065
288.15
0.0289644
8.31447
9.81

Description
Sea level pressure
Temperature lapse rate
Sea level standard temperature
Molar mass of dry air
Universal gas constant
Earths gravitation

Fig. 5.

Visualization of normalized calibration data of magnetic sensor.

from a set of different attitudes [22]. Another method is to use
maximum likelihood estimation. We implemented the method
described in [22] for the calibration of the accelerometers.
For the magnetometer we adopted the procedure presented
in [23]. The parameters of an ellipsoid are estimated based
on multiple measurement points vxi , vyi , vzi from different
directions. Such an ellipsoid is described by (4) with the six
unknown parameters vox , voy , voz , bkx , bky , bkz .


vxi − vox
bkx

2

+



vyi − voy
bky

2

+



vzi − voz
bkx

2

Fig. 6.

Trajectory visualization on Android smartphone in Google Maps.

E. Software Design
= 1 (4)

These parameters can be determined by at least six independent equations from measurements. Figure 5 shows the
measurement points or the trajectory of the described calibration procedure in 3D space. Based on these measurements the
algorithm tries to fit an ideal ellipsoid. This gives the 3D center
point and the semi-axis length of the ellipsoid. For calibration
this means that the obtained ellipsoids center is the estimated
sensor bias and the semi-axis length represents the scale factor.
D. Smartphone Navigation
The smartphone has two major tasks in our setup: First,
the smartphone is used as a navigation device visualizing
the user’s path on Google Maps. Therefore, an intuitive user
interface and position calculation is provided by the smartphone. Second, it is possible for further detailed studies to
collect raw measurements from the IMU on the phones SDcard. Typically, data collection for later offline processing is a
cumbersome but necessary task for algorithm development and
optimization. Often, laptop computers and a lot of equipment
has been used for this task. With today’s smartphones and their
large processing and memory capabilities data collection with
mobile devices proofs to be a very handy method. Walking
with a smartphone for data collection also allows for more
accurate walking patterns as no other equipment carried by
the person influences their walking behavior.

Based on Google’s Android platform applications are developed in Java. A very common tool for algorithm development
and complex matrix computations has been MATLAB for a
while. In order to establish a consistent tool chain MATLAB’s
code generation capabilities are used to generate valid C-code.
This C-code is also validated against the original MATLAB
code through again integrating it into the model. A nice side
effect of such generated code is a MATLAB performance
speedup on the order of 6 times. To import the external
C-code into the Android Java application, Google’s NativeDevelopment-Kit (NDK) is used. A further software design
method to avoid blocking the application waiting for input
or output operations is splitting up the Android software into
more than one thread. The user interface to navigate through
the software and map visualization is realized as a context
menu and allows to use the full screen for position visualization as shown in Figure 6. At the moment, the Google Maps
API only provides outdoor position visualization but they
recently released their indoor navigation functionalities. As
soon as Google opens it indoor navigation for developers the
same visualization will also be possible for indoor scenarios.
In this, a 2.5D representation is used which allows selecting
different floors of the building. Similar to Google’s indoor
maps functionalities the OpenStreetMap project followed this
approach quickly. This also means that the availability of
indoor maps will increase in the near future. That carries
further potential for map building and application of map
matching algorithms.

Fig. 8. 3D performance evaluation to analyze height information fusion from
barometer.

Fig. 7. Indoor trajectory over two floors in typical office building recorded
with our own IMU compared to the Xsens MTi-G IMU.

V. E XPERIMENTAL E VALUATION
For the purpose of a performance evaluation we compared
the results obtained with our IMU with the widely used Xsens
MTi-G IMU. This IMU from Xsens is often considered lowcost but the price is still not yet in the consumer range.
Considering the size of more than 100 cm3 and a weight of
68 g it is useful for research applications but not applicable for
real world deployment. Figure 7 shows an example trajectory
from our experimental evaluation. The trajectory was recorded
inside our institute building which is considered a typical office
building. Both IMUs, MTi-G and our own (ITIV BT IMU)
were fixed on the same foot of the walking person and sent
data with a rate of 100 Hz. The trajectory started on the second
floor and the test person took the stairs down to the first floor,
walked around and went back upstairs to the starting point.
A. Accuracy Analysis
In Figure 7 it can be seen, that the position estimation
of both IMUs exhibits a drift over the distance. Comparing
the resulting errors of both systems shows that they are on
the order of a few percent of the traveled distance and very
comparable. On an absolute scale, the error adds up to 2.2 m
(own ITIV BT IMU) respectively 1.8 m (Xsens MTi-G) for a
total traveled distance of about 82 m.
When taking a closer look on the 3D performance evaluation
in Figure 8 it can be seen that the position estimation in
z-direction drifts over time. The figure shows the estimated
height over time starting with an initial value of zero on the
second floor. Especially, the estimation from the MTi-G IMU
drifts away towards the end of the trajectory as no additional
height information is fused. When fusing height measurements
from the barometer the height is corrected to a certain degree
as can be seen in the red graph in Figure 8. It can also be
seen that at the beginning of the run a strong drift occurs

for both estimations. We assume this drift occurs from initial
values of the system as no dedicated initialization procedure
is considered. Here, this error adds up to about 0.5 m. After
this drift, it can be seen that the test person walked down to
the first floor resulting in a height change of about 3 m which
corresponds well to the story height of the building. For the
height estimation based on the Xsens data it can be seen that a
strong drift occurs still after initialization over the whole run.
This adds up to an additional error of about 0.5 m even after
approximately 100 s. The developed information processing
with our IMU still exhibits the drift from initialization but the
height after 100 s corresponds well to the height at about 10 s
from the starting point of the trajectory.
B. Real-time Performance Evaluation on a Smartphone
Considering that today most smartphones are equipped with
fast processors and comparatively large memory resources it is
likely that the smartphone handles the strapdown calculation in
real time. We evaluated the performance on a Sony Ericsson
Xperia arc with a 1 GHz Qualcomm Snapdragon processor
and 512 M B RAM. Figure 9 shows the processor load for
the strapdown calculation on the Sony Ericson Xperia arc. On
the left side the phone carries out the strapdown calculation
only and on the right side the trajectory is visualized on
Google maps. It can be seen that less than half of the time
is used for navigation. When the visualization is not used,
the processor load decreases from 40 % down to 37 %.
This shows that strapdown pedestrian navigation on today’s
smartphones can be carried out even while other processes
run in the background. Operating on a lower processor load
also implies longer battery lifetimes of the phone. With the
upcoming new generations of smartphones employing even
more capable processors (dual and quad cores and higher clock
rates) this figure will be brought down even more.
VI. D ISCUSSION
The presented results show that the combination of a footmounted IMU with a smartphone for navigation in indoor
scenarios is a promising application area. The developed
sensor box can easily be attached to a person’s shoestring and
then transmits data to the phone via Bluetooth. An Android

VII. C ONCLUSION AND F UTURE W ORK
A. Conclusion

Fig. 9.

Real-time processing performance.

application provides a user interface for configuration of the
IMU and position visualization of the trajectory walked. State
of the art smartphone processors are capable of handling the
strapdown calculations efficiently while still providing enough
processing capabilities for other tasks.
To allow for 3D or at least 2.5D navigation (determination
of discrete floors) a barometer is a very useful supplement.
Determining changes between floors from the evaluation of the
atmospheric pressure is fairly easy. However, long-term effects
on the atmospheric pressure, e.g., from weather changes do
not allow for an absolute height measurement over sea level.
This information has to be obtained from another data source,
e.g., correction from GPS whenever available (outdoors). For
further improvements or when no barometer is available we
propose to introduce a Zero Height Update which makes the
assumption of planar floors within a building. The drawback
of this method is that some floor changes will be hard to be
recognized, e.g., if the person moves too slowly or over a
ramp.
The outlined practical tests show feasibility and applicability
to indoor navigation with accuracy on the order of a few
percent of traveled distance making the approach sufficient
for a wide range of applications. Especially, when looking into
potential indoor location based services where precise indoor
position information is necessary, the presented concept carries
an enormous potential. Accuracy of foot-mounted strapdown
navigation cannot be outperformed by step detection with the
smartphone’s internal sensors. Considering the low costs of
the sensory equipment used for the presented evaluation mass
market adoption becomes even more likely. A drawback of the
system is, however, that the user needs a small sensor box on
the foot in addition to the smartphone.
However, the results presented in this paper can serve
as a starting point only and therefore open a wide range
of possible further research topics in the various details of
the systems components. Here, for example, the integration
and possibilities of indoor maps generation and matching
techniques or the combination with sparsely available WiFi
access points carry a huge potential.

In this paper we proposed a foot-mounted IMU combined
with a smartphone for pedestrian navigation in indoor scenarios. The good performance in terms of accuracy of footmounted approaches is combined with the ease of use and
visualization capabilities of today’s smartphones. Therefore,
we presented a hardware design which is based on a very
low cost and low power single-chip IMU. A custom casing
with a volume of less than 14 cm3 was designed to fix the
IMU on the shoestring of a person’s foot. Comparing the
performance, cost and size of the developed IMU sensor-box
it is one of the smallest ever developed wireless IMUs to the
best of our knowledge. Also, in terms of power consumption
the developed IMU is very competitive by employing an
intelligent power management and sensors with low power
requirements. On the algorithmic side, a basic strapdown
algorithm is implemented and further extended with sensor
fusion capabilities of relative height information measured
with a barometer. To verify real-world applicability of our
presented approach we carried out an experimental study in a
typical office environment and analyzed real-time requirements
and capabilities. It has been shown that the obtainable localization accuracies are comparable with commercially available
IMUs. Especially, by fusing measurements from an additional
barometer the cumulative errors in z-direction of the strapdown
navigation can be corrected. Thus, the overall performance of
the system is increased. We also showed the feasibility and
performance of a smartphone performing the computational
complex strapdown calculation with a processor load lower
than 50 %. For navigation purposes the visualization of the
trajectory has been integrated into the Google Maps interface.
Additionally, our presented hardware setup allows for raw
IMU data collection on the smartphone which also enables
easy evaluation of other IMU positions, algorithms and sensor
fusion approaches.
B. Future Work
This paper presented a first study of the concept of footmounted inertial sensors combined with a smartphone. The
experimental evaluation was done on a relatively small data set
so far. Thus, further evaluation of the system under different
environmental conditions is planned for the next future. Along
with a more comprehensive evaluation of the system further
algorithmic aspects will be covered to gain more insight in
the error behavior at initialization and possible correction
mechanisms will be developed.
A promising research will also be map building and matching especially with the availability of indoor maps provided
by Google or OpenStreetMap. In this context fusion of further
information sources (e.g. localization based on WiFi or Near
Field Communication) carries a huge potential.
Another possibility on the algorithmic side is to perform
the strapdown calculation on the IMUs microcontroller itself.
This would reduce communication overhead but results in a

higher processor load of the microcontroller in the sensor
box. Therefore, it has to be analyzed how to achieve further
optimizations, e.g., by efficient fixed point implementation of
the algorithms.
ACKNOWLEDGMENT
This work was supported by the German Research Foundation (DFG) within the Research Training Group GRK 1194
”Self-organizing Sensor-Actuator Networks”. We would like
to thank the DFG for supporting our work.
R EFERENCES
[1] M. Azizyan, I. Constandache, and R. Roy Choudhury, “Surroundsense:
mobile phone localization via ambience fingerprinting,” in Proceedings
of the 15th annual international conference on Mobile computing and
networking. ACM, 2009, pp. 261–272.
[2] J. Schmid, D. Curtis, T. Gaedeke, and J. Ledlie, “Improving sparse
organic wifi localization with inertial sensors,” in Proceedings ot the
9th Workshop on Positioning, Navigation and Communication 2012
(WPNC’12), 2012.
[3] D. Gusenbauer, C. Isert, and J. Kr¨osche, “Self-contained indoor positioning on off-the-shelf mobile devices,” in Proc. Int Indoor Positioning
and Indoor Navigation (IPIN) Conference, 2010, pp. 1–9.
[4] R. G. Stirling, “Development of a pedestrian navigation system using
shoe mounted sensors,” Master Thesis, University of Alberta, 2004.
[5] P. Pombinho, A. Afonso, and M. Carmo, “Point of interest awareness
using indoor positioning with a mobile phone,” in Proceedings of the 1st
International Conference on Pervasive and Embedded Computing and
Communication Systesm, 2011.
[6] E. Foxlin, “Pedestrian tracking with shoe-mounted inertial sensors,”
IEEE Comput. Graph. Appl., vol. 25, no. 6, pp. 38–46, 2005.
[7] O. Woodman and R. Harle, “Pedestrian localisation for indoor environments,” in UbiComp ’08: Proceedings of the 10th international
conference on Ubiquitous computing. New York, NY, USA: ACM,
2008, pp. 114–123.
[8] S. Rajagopal, “Personal dead reckoning system with shoe mounted
inertial sensors,” Master Thesis, KTH Royal Institute of Technology,
Stockholm, Sweden, 2008.
[9] J. Borenstein, L. Ojeda, and S. Kwanmuang, “Heuristic reduction of gyro
drift for personnel tracking systems,” Journal of Navigation, vol. 62, pp.
41–58, 2009.
[10] A. R. Jimenez, F. Seco, C. Prieto, and J. Guevara, “Indoor pedestrian
navigation using an ins/ekf framework for yaw drift reduction and a footmounted imu,” in WPNC 2010: 7th Workshop on Positioning, Navigation
and Communication 2010., 2010.

[11] A. Jim´enez, F. Seco, F. Zampella, C. Prieto, and J. Guevara, “Improved
heuristic drift elimination (ihde) for pedestrian navigation in complex
buildings,” in International conference on indoor Positioning and Indoor
Navigation - IPIN, 2011.
[12] I. Skog, J.-O. Nilsson, and P. Handel, “Evaluation of zero-velocity
detectors for foot-mounted inertial navigation systems,” in Proc. Int
Indoor Positioning and Indoor Navigation (IPIN) Conference, 2010, pp.
1–6.
[13] T. G¨adeke, J. Schmid, W. Stork, and K. D. M¨uller-Glaser, “Pedestrian
dead reckoning as a complementary method for wireless sensor network
ad-hoc person localization,” in Proceedings of the 1st International
Conference on Sensor Networks (SENSORNETS), Rome, Italy, 2012.
[14] Z. Sun, X. Mao, W. Tian, and X. Zhang, “Activity classification
and dead reckoning for pedestrian navigation with wearable sensors,”
Measurement Science and Technology, vol. 20, no. 1, 2009. [Online].
Available: http://stacks.iop.org/0957-0233/20/i=1/a=015203
[15] C. Lukianto, C. Honniger, and H. Sternberg, “Pedestrian smartphonebased indoor navigation using ultra portable sensory equipment,” in
Proc. Int Indoor Positioning and Indoor Navigation (IPIN) Conf, 2010,
pp. 1–5.
[16] A. Parnandi, K. Le, P. Vaghela, A. Kolli, K. Dantu, S. Poduri, and
G. S. Sukhatme, “Coarse in-building localization with smartphones,” in
Mobile Computing, Applications, and Services, ser. Lecture Notes of the
Institute for Computer Sciences, Social Informatics and Telecommunications Engineering. Springer Berlin Heidelberg, 2010, vol. 35, pp.
343–354.
[17] F. Aubeck, C. Isert, and D. Gusenbauer, “Camera based step detection
on mobile phones,” in Indoor Positioning and Indoor Navigation (IPIN),
2011 International Conference on, 2011, pp. 1 –7.
[18] Y. Jin, H.-S. Toh, W.-S. Soh, and W.-C. Wong, “A robust dead-reckoning
pedestrian tracking system with low cost sensors,” in Proc. IEEE Int
Pervasive Computing and Communications (PerCom) Conf, 2011, pp.
222–230.
[19] J. A. B. Link, P. Smith, N. Viol, and K. Wehrle, “Footpath: Accurate
map-based indoor navigation using smartphones,” in Proc. Int Indoor
Positioning and Indoor Navigation (IPIN) Conf, 2011, pp. 1–8.
[20] J. L. Titterton, David H. ; Weston, Strapdown inertial navigation
technology, 2nd ed., ser. Progress in astronautics and aeronautics ; 207.
Stevenage: Institution of Electrical Engineers, 2004.
[21] U.S. Government Printing Office, “U.s. standard atmosphere,” Washington, D.C, 1976.
[22] S. P. Won and F. Golnaraghi, “A triaxial accelerometer calibration
method using a mathematical model,” vol. 59, no. 8, pp. 2144–2153,
2010.
[23] D. Campolo, M. Fabris, G. Cavallo, D. Accoto, F. Keller, and
E. Guglielmelli, “A novel procedure for in-field calibration of sourceless
inertial/magnetic orientation tracking wearable devices,” in Proc. First
IEEE/RAS-EMBS Int. Conf. Biomedical Robotics and Biomechatronics
BioRob 2006, 2006, pp. 471–476.