A LOCATION SERVICE TO DETERMINE LOCATIONS OF MOVING OBJECTS

safetroubledΚινητά – Ασύρματες Τεχνολογίες

24 Νοε 2013 (πριν από 3 χρόνια και 8 μήνες)

113 εμφανίσεις



A LOCATION SERVICE TO DETERMINE LOCATIONS OF MOVING
OBJECTS








Submitted by

Michael Loh Sheng Yang


Department of Electrical & Computer Engineering







In partial fulfillment of the

Requirements for the Degree of

Bachelor of Engineering

National
University of Singapore



B.Eng Dissertation





A LOCATION SERVICE TO DETERMINE LOCATIONS OF MOVING OBJECTS




CA1 Interim Report




By

Michael Loh Sheng Yang

National University of Singapore

2012/2013







Project ID: H0561040

Project Supervisor: Dr Pung Hung Keng

i


ABSTRACT


In this paper I will try to address the problem of accurately locating users inside buildings using
mobile phones. Besides
analyzing the different methods that can be used for indoor localization,
the work done by a previous FYP student will also be mentioned. Based on the work that has
been already done and the research I did, I will then propose an improvement to the current

system to further increase the accuracy for indoor localization.

Work will be done on eclipse
using the java language to program and create an Android application for the purpose of indoor
localization.












ii


ACKNOWLEDGEMENTS

I would like to thank Dr

Pung for his valuable help and direction throughout this semester. The
resources and advice that he provided was indispensible.

I would also like to thank Ishen
for his

help
with my understanding of the different algorithms.

I would also like to thank
Marcus for his patience in helping me out with android programming.
















CONTENTS

ABSTRACT

ACKNOWLEDGEMENTS

CHAPTER 1

INTRODUCTION


1.1

Introduction

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1


1.2

Motivation

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . 2

CHAPTER 2

RELATED WORK



2.1

Infrastructure based methods
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3


2.2

Infrastructure
-
less based methods

. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . 4


2.2.1

Radio Propagation Techniques

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2.2

Fingerprinting

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . 4


2.2.2.1

Limitations and Solutions to

Fingerprinting

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5


2.3

Sensors

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. 7


2.3.1

Accelerometers

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . 7


2.3.2

Gyroscopes

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8


2.3.3

Magnetometer

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . 8


2.3.4

Sensor Fusion

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9


2.3.4.1

Finding Orientation

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

`

2.3.4.2

Existing Solutions

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10


2.3.4.3

Limitations to Sensor Fusion

. . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10


2.3.4.4

Solutions to Sensor Fusion

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

CHAPTER 3

PREVIOUS WORK


3.1

Work done

b
y previous FYP student . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12


3.2

Analysis of work done by previous FYP student

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

CHAPTER 4


IMPROVEMENTS

. . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

CHAPTER 5

CONCLUSION
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

ii


CHAPTER 6

FUTURE
WO
RK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

BIBLIOGRAPHY

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.

. . . .18

















1


1.1

Introduction

As mobile devices get more powerful, the number of different functions that a mobile device
can

perform increases. One such f
unction is indoor localization.
As technology advances and mobile
phones are fitted with many different types of sensors, indoor localization via a mobile phone is
now possible.
The uses of indoor localization are manifold, such as locating a mobile user,
guiding a user in a building and

printing a document from the closest printer.
The purpose of this
paper is twofold. Firstly, to
attempt to analyze the
methods used for indoor localization.

Secondly, to analyze the work done by a previous student working on this project and improve
on it
.

Some of the

techniques that will be explored
are

Wi
-
Fi

fingerprinting, radio s
ignal
propagation as well as

sensor fusion o
f accelerometers, gyroscopes
. In order to take advantage of
a certain
method

and to compensate
for the inaccuracies of other method
s, a weighted algorithm
will be proposed
to help provide the most reliable and stable source of information for that
period of time.










2


1.2

Motivation

Global Positioning System (GPS)
is

generally not suitable to establish indoor locations as the
signals lose a significant portion of their power when passing through walls. Moreover, the
multiple reflections on buildings’ surfaces will result in multi
-
path propagation which leads to
high i
naccuracy of the user’s location. Therefore a different system needs to be used to track and
locate a user in a building more effectively and accurately. Today there are many systems that
are able to effectively track users indoors. However, most of these
require the use of specialized
sensors and equipment that needs to be installed in the building.
T
his method is not scalable
as

it
will be expensive to implement.
A better way is to

make use of existing infrastructure to identify
the user’s location. With
Wi
-
Fi access points being installed in almost all buildings, it is possible
to make use of Wi
-
Fi signal strengths to calculate a user’s location indoors.
With
the popularity
of
Wi
-
Fi enabled
smart phone
s

in modern society,
it is possible to create an indoo
r localization
application for smart phones
.
A previous FYP student has already
worked on and
created

an
indoor localization
application for Android devices.

The motivation of this project is to improve
the accuracy of this system. We will first look at
the different

methods and techniques that
are
used for indoor localization before attempting to improve the existing system.







3


2

Related work

2.1

Infrastructure based methods

Localization m
ethods that require installation of new hardware into building
s are known as
infrastructure based methods. Though
such a system

is capable of locating indoor positions with
high accuracy,
it is

not a scalable and is normally very costly. One such method is Active Badge

(Want)
.

An individual carries a badge that emits a unique code that identifies a user via pulse
-
width modulated infrared (IR) signals. Additional sensors will be deployed around the building
which picks up the signal and transmit this information to a master stat
ion, which processes the
information and displays the current location of said individual. It is important to note that
sensing range from badge to sensor is at maximum 6m. Thus the number of sensors which will
be needed for a single building is enormous.
Moreover, badges have to be obtained for every
single person in the building, further increasing the costs of implementing such a system.
Zigbee

(Liu)

is another system which follows such a design. It works by calculating

the rad
io signal
strength (RSS
) which is proportional to the distance the signal has travelled.

Therefore, these systems
are not scalable and are usually expensive though they
provide a high
accuracy for indoor localization.





4


2.2

Infrastructure
-
less based me
thods

There are a number of different methods that to do not require any additional installation of
hardware in a building.
Methods discussed below include
Radio Propagation,
Fingerprinting and
Sensing.

Radio propagation and fingerprinting methods both
depend on Wi
-
Fi signal strength for
tracking a user’s location, while sensing uses the built in sensors in a smart phone for indoor
localization.

2.2.1

Radio Propagation

Radio propagation techniques work by finding out certain characteristics on how radio
waves are
received by the smart phone. Examples are triangulation and received signal phase methods

(Lui,
November 2007)
. By estimating the time of arrival (TOA), time difference of arrival (TDOA) or
angle of arrival (AOA), one

is able to use triangulation to locate the user. The received signal
phase method works by calculating the carrier phase (or phase difference) to estimate the range.
Radio propagation techniques face many challenges due to severe multipath, low probabilit
y for
line
-
of
-
sight (LOS) paths and numerous reflecting surfaces. It is also affected by the floor layout
and environmental changes such as moving bodies. Thus, these methods are more suited for use
outdoors rather than indoors.

2.2.2

Fingerprinting

F
inger
printing is one of the more

popular Wi
-
Fi
techniques

being used today due to its accuracy
.
Fingerprinting consist of two different phases: the training phase and the measurement phase.
The training phase measures the Wi
-
Fi signal strength at different loca
tions in the building by
nearby access points. As Wi
-
Fi signal strength changes depending on the user’s orientation, four
5


readings at different orientations at the same location will be taken. The fingerprints of that
location is then stored in a database.

During the measurement phase, there are a number of different techniques which can be used to
locate the user.
Examples of p
robabilistic
techniques are Bayesian Inference and Gaussian
distribution. Bayesian Inference works by calculating the different pro
babilities over all
locations, given a particular signal strength. By assuming that the likelihood of each location
candidate is a Gaussian distribution, one is able to calculate the probability of all locations given
the measured signal strength on the sm
art phone

(Lui, November 2007)
.

Other methods include

t
he k
-
nearest neighbor algorithm
, neural networks and Support Vector
Machine (SVM). K
-
nearest neighbor

works by selecting k number of closest known locations in
signal spac
e from the information stored in the database during the measurement phase. An
estimated location is calculated by averaging the k different values.
For neural networks,
RSS
and corresponding location coordinates are adopted as inputs and targets respectiv
ely. Thus,
during the measurement phase, the system will be able to use the correct weights generated
during the training phase to provide the user’s location. SVM works in the same way as neural
networks, where appropriate values are fed into the system i
n order for the system to generate the
expected answer during the measurement phase.

2.2.2.1

Limitations to fingerprinting

Although fingerprinting provides a high accuracy for indoor locali
zation, it has its limitations.
Wi
-
Fi signal strengths va
ry due to

environmental changes. Thus indoor localization using the
fingerprinting method will not be accurate when

th
e environments in the training phase and
measurement phase differs
.

One such example is crowdedness of

the building
. During the day
6


the number of p
eople in the building is higher as compared to night time.

It is found that this
factor
affects the accuracy
of fingerprinting
by as much as 6meters

(Bahl, February 2000)
.
Another example is that Wi
-
Fi sensors being used to sen
se the signal strengths will differ during
the training phase and the measurement phase due to the usage of different smart phones, which
might cause the system to display an incorrect location due to the different fingerprint.

To address this issue, Hype
rbolic Location Fingerprinting

(Kjaergaard, 2011)

proposes a system
that takes ratios of signal strengths between pairs of access points instead of the absolute signal
strength. The results showed that taking the ratios helps i
mproves the stability of the system.
This method has already been implemented by the previous FYP student
and will not be
discussed here.

A
nother major

disadvantage
of fingerprinting

is that it is very labor intensive during the training
phase. Recalibration is also required if the location of access points changes.
This has led to
people creating indoor localization solutions without the need of a training phase. One such
method
, EZ,

(Chintalapudi, September 2010)

does no
t require any fingerprinting at
all for indoor
localization
.
EZ works on the fact that
knowing enough distances between the APs and mobile
devices allows unique determination of their
relative locations. With the occasional GPS lock to
give a “true” location, EZ is then able to calculate the user’s location indoors
. This method has
also been discussed but not implemented by the previous FYP student
.


Lastly, fingerprinting is unable to
differentiate which room the user is located in.
This is because
signal space does not take into account the Euclidean distance between points.
Therefore,
two
points that are far from each other in Euclidean space could be next to each other in signal spac
e.

RADAR
(Bahl, February 2000)

and
(Lassabe, 2007)
,

attempts to solve this problem by
using the
7


Viterbi algorithm or Viterbi
-
like algorithm
s
.

Histories of previous locations are used for
continuous user tracking

by feeding these locations into the Viterbi algorithm
.
By using

an
algorithm that is able to tell th
e location’s neighboring points

(Lassabe, 2007)
,
the system will
skip the computation of points that
are not in tagged as neighbors.

Another solution to this
problem is Acoustic Background Spectrum (ABS)
(Tarzia, 2011)
, which measures the
background sound of the room to diffe
rentiate different rooms. This method has been discussed
already by a previous student.

2.3

Sensors

Smart phone these days contain state of the art sensors that could be used to help locate a user’s
location. The sensors that are
commonly found in smart ph
ones

are accelerometers, gyroscopes
and magnetometers.
We shall take a look at the different sensors and
see if its

possible to build
an Inertial Navigation System (INS)

for indoor localization
.

2.3.1

Accelerometers

An accelerometer is a device that
measures the acceleration of the device. It measures
acceleration by sensing how much a mass press on something when a force acts on it. Therefore,
an accelerometer at rest relative to the Earth’s surface will indicate 1G upwards due to the
Earth’s gravita
tional force.


Using only accelerometers,
(Shala, September 2011)

and
(Serra, 2010)

were able to sense the
steps taken by a person.
By combining
step detection technique with the user’s orientation
,

one is
able

to calculate the user’s location

(Serra, 2010)
. However, this method is inaccurate as
8


everyone has a unique gait and step length and is thus useful for short distance tracking only.
It is
also possible to calculate

the distance travelled by double integrating the accelerometer’s value.
However, this means that the encountered errors will accumulate if we use this value to track the
user location. If the accelerometer readings are off even a little, the calculated sp
eed will be off,
which will make the position even further off for that estimate and all future estimates

(Shala,
September 2011)
. This means that relying solely on accelerometer readings to calculate distance
will not be accur
ate as errors will accumulate and the user’s location will become less accurate
overtime.

2.3.2

Gyroscopes

Gyroscope is a device that measures orientation, based on the principles of conservation of
angular momentum.
G
yroscopes measure orientation accurately but its drawback is that its
output drifts over time. If not compensated for the drift error accumulates and the gyroscope
output is
inaccurate

(Shala, September 2011)
.

2.3.4

Magnetomet
er

A magnetometer is an instrument used to measure the strength and direction of the magnetic
field in the surrounding area of the instrument. Magnetometer is useful for providing information
about the relative direction/angular distance from magnetic north,
but it is susceptible to changes
in the magnetic field made by electronic devices.

9


2.
3.4

Sensor Fusion

Sensor fusion

is the combining of

sensory
data such that the resulting information is in some
sense

better

than would be possible when these sources wer
e used individually
.
By
employing
Sensor Fusion,
we are now able to sense and calculate a user’s movement and location more
accurately.


2.3.4.1

Orientation

One of the uses of sensor fusion is the calculation of the user’s location.
By combining
the data
from
accelerometers and magnetometers

(Shala, September 2011)
,
we are able to compensate for
the drift of the gyroscope and the changes in magnetic field.

This
function, also known as compass, has already been implemented in
Android and there is no need to do additional calculation.
There are
many uses for finding the orientation of the user.
Linear acceleration is
calculated by using the magnetometer and ac
celerometer outputs to get
the orientation of the device
by
deduct
ing

the gravity components on
the respective axis. By reading the y
-
axis as shown in the picture

on the right
,
the forward
acceleration can be obtained

(Shala, Septem
ber 2011)
.
Another use of knowing the orientation is
by combining this wit
h the step detection algorithm to update the user’s location

as already
mentioned above
.


Another paper
(Lawitzki, March 2012)

mentions that it is possible to fuse all three sensors to
produce cleaner orientation data.
T
he
current orientation data obtained from the accelerometer
and magnetometer is inaccurate. This is due to the noise generated by both sensors and the slow
10


reacti
on time of the magnetometer. The gyroscope was able to accurately calculate rotation at
high frequency but is susceptible to the accumulation of small errors. By fusing all three sensors,
one is now able to the orientation of a user with higher accuracy t
han before.

2.3.4.2

Existing solutions

Using these sensors, UnLoc
(Wang, 2012)

has been able to develop an indoor localization
system based on smart phone sensors alone. It uses sensor fusion to track the user’s movement
throughout the building. Using sensors to track user movements will undoubtedly accumulate
error over time, leadin
g the inaccuracies in positioning. This is solved by fingerprinting locations
that exhibit unique sensor readings. For example, the magnetometer might experience unique
readings when a user passes an elevator due the magnetic field generated. Therefore, wh
en the
user passes a fingerprinted location, the system corrects itself as it knows its exact location.
However, the weakness of such a system is that fingerprinting still has to be done. The problem
of fingerprinting is also magnified as one would have to

search for the unique locations where
sensors have a special reading.

2.3.4.3

Limitations to sensor fusion

One major problem with using smart phone sensors for indoor localization is
the accumulation of
error as time passes. As with every INS, there is a need for an external system to correct it
occasionally and set it back on track. Thus, it is almost impossible to just use only smart phone
sensors for indoor localization.


11


Another w
eakness of using such a system is noise. Sensors are sensitive instruments and it also
captures unwanted information such as noise. This makes it hard to obtain the correct
information without some form of filter.
To calculate orientation accurately, one
h
as to use both
high pass and low pass filter

on different sensors

(Lawitzki, March 2012)
.
A K
alman filter
can
also be used
to reduce the random noise produced by an accelerometer to calculate the distance
travelled.
(Liu H. , 1999)


It

is

also

hard to determine the initial location of the
user with any external help.
A paper
(Serra,
2010)

solves this problem by prompting a user to take a photo of a QR code, which also contains
the map of that floor. This solves the initial location problem as the user has to be at a certain
location to take a photo. They then use a transformation matrix on
the photo taken to determine
the initial location and orientation of the user when taking the photo.


Lastly, the problem

with using sensors is that
s
ensor
s usually come with some offset

error

and
there is a
need to calibrate the sensors before use

(Shala, September 2011)
.
A smart phone has to
be placed on in a stationary position for optimal calibration. However, mobile users are often
impatient and on the move, making calibration challenging.

2.3.4.4

Solutions to sensor
fusion

It is quite apparent it is almost impossible to use sensor fusion as the only instrument for indoor
localization. Sensor fusion, as with all INS, requires an external system to keep it on track due to
the errors accumulated over time. Thus, sensor f
usion would be most appropriately used as a
system that compliments the findings of other systems.

12



By

creating

three different indoor localization measurement s
ystems that worked individually,
the results can then be compared at the end for congruency.

The three different systems are
linear
acceleration, step detection and Wi
-
Fi fingerprinting

(Shala, September 2011)
.
It also uses a
simple weighted simple to increase/decrease the priority of values based on inaccuracies. For
example, if the Wi
-
Fi signal is not stable, the weight given for the Wi
-
Fi value will be smaller
(lower priority).



Another way to use sensor fusion is to use an external system to correc
t the errors occasionally.
UnLoc

(Wang,
2012)

does this by fingerprinting locations with unique sensor data. This can also
be done via Wi
-
Fi fingerprinting, where the location is updated based on received signal strength.


3.1

Work done by previous FYP student

A student has already been w
orking on this project for about a year now. A few different Wi
-
Fi
fingerprinting methods
have already been
implemented. They are RADAR, HORUS and
hyperbolic location fingerprinting. RADAR uses a deterministic method called nearest neighbor

(Bahl, February 2000)

to locate the user. HORUS is an imp
rovement over RADAR, using a
Gaussian distribution over the average signal strength received to calculate the user’s location

(Rehim, 2004)
.



Another fingerprinting method that was being implemented in this project is the Hyperbolic
Location Fingerprinting (HLF). This method uses ratios of signal strengths instead of absolute
13


values to overcome environmental and hardware changes. HLF used the

nearest neighbor and
Bayesian Inference methods to find the user’s location

(Kjaergaard, 2011)
.


Clustering
,

which reduces the number of computations
executed

is also implemented.
There are
2 types of clustering, explicit and

implicit

(Rehim, 2004)
. Explicit clustering looks for the
strongest access points at that location. The search space is now being reduced to the area that
the signal transmitted by that access point is visible in. Implicit clustering works by
finding the
different access points

that is visible to the smart phone. By iterating over the different access
points, the search space is reduced as only the area overlapped by the different access points will
be brought forward for the next iteration till a definite location is found.
Exp
licit clustering was
implemented for this project.


Another method used to increase the accuracy is called continuous space estimator

(Rehim,
2004)
. As fingerprinting fingerprints an

area, user’s location tends
to jump from one area to another. This method tries to resolve this
by finding the different probabilities
of a user being in that area.
The higher the probability the closer the user is to that cell.



Small scale compensatio
n is a simple technique use to reduce the errors of Wi
-
Fi fingerprinting.
As signal strength can vary significantly over a distance as small as three inches, this can cause
the system to think that the user has moved by a large distance when that is not th
e case

(Rehim,
2004)
. This technique works by keeping a record of the previous user location. By comparing the
14


distance between the previous and current location and making sure it does not pass a certain
threshold, this techni
que makes sure that the user does not travel a distance that is not possible in
real life.


Besides that a method known as Acoustic Background S
pectrum

(ABS)

(Tarzia, 2011)

was also
implemented. This method is a solution to a problem co
mmonly found in fingerprinting; the
user’s location could “jump” to another room as the fingerprint does not contain information
about the room it is in. By recording an audio sample of the ro
om during fingerprinting, one is
able to figure out which room he is in. Thus, this method is a useful assistive method that can be
used in conjunction with fingerprinting.

3.2

Analysis
of current work done by previous student

The implementation of differe
nt Wi
-
FI fingerprinting techniques has ensured that this system
performed accurately consistently. Most of the weaknesses of fingerprinting
have

already been
addressed
.
The inconsistencies of environmental and hardware changes has been taken care of by
HLF. Computational load has been decrease by the implementation of clustering.
However, there
are certain disadvantages for using ABS:



ABS requires a one to one mapping of
background sound to room so as to accurately
determine which room the user is. Therefore, as the number of rooms increase, the
probability that two room
s

share the same background sound increase, leading to low
scalability
.



ABS has a poor performance when
placed in a room which has a lot of chatter (many
people present, talking and moving noisily)

15




ABS has a high computational load which is not ideal for mobile devices with limited
battery and processing power.



ABS requires quality audio equipment for measur
ing background sound. However, the
microphones installed currently on smart phones are small, inexpensive and crowded by
other electronics, making accurate background sound detection difficult.

Thus, the use of ABS might not be well sui
ted for use on a mob
ile device and this is an
area that
could be improved on.

4

Improvements

To further improve on the fingerprinting techniques would result in negligible increase in
accuracy

due to the extensive improvements made by the previous student
.
As there is little

work
being done with regards to
sensor related systems, this would

be a good direction to take

to
further increase the accuracy of
the current system
.


One possible way to make use of sensor fusion in this system is to implement this as a separate
system as a confirmation for the result obtained by the Wi
-
Fi fingerprinting methods. By testing
the two results for congruency, this reduces the amount of erro
r the localization system makes in
case one system
contains erroneous results. A weighted system can also be used to prioritize
accurate readings instead of giving all readings equal priority to reduce the number of errors.
Another reason for implementing
this as a separate system is the usefulness of
a backup system
when fingerprinting fails e.g. no connection to Wi
-
Fi access points.
Besides that, implementing
step detection would be great for tracking users when changing floors as well as for general
indo
or localization purposes.

16



Another possible improvement would be to use
the Viterbi Algorithm

rather than ABS to prevent
the movement of user’s through walls.
Many papers as mentioned above have used Viterbi or
Viterbi
-
like algorithms to solve this problem

by finding

the point’s neighboring nodes. S
ensor
fusion
would also be able to

assist in this problem as sensors would never s
how a user going
through walls, signifying that the result generated from fingerprinting is wrong.


5

Conclusion

This paper has covered various methods that can be used for indoor localization. Based on the
work that a previous student has done, Acoustic Background Spectrum has been found to be
inaccurate due to hardware limitations on smart phones. A Viterbi algorit
hm is proposed to
replace ABS as a solution for users walking through walls. Additionally, smart phone sensory
data has not been harnessed by the previous student and it would be a good direction to work on
sensor fusion to increase the accuracy of the exi
sting system. Sensor fusion techniques will be
deployed separately as another system running simultaneously with Wi
-
Fi fingerprinting. A
weighted
system will then be used to decide on the integrity of the results and the user’s location
will be calculated
based on that.




17


6

Future Work

The project implementation will span a total of
6 months till the end of semester 2 of year
2012/2013.

December

Create a working sensor fusion application that
is able to track linear acceleration, orientation
and distance
. Understand Viterbi Algorithm

January

Fine tune distance measurement using sensor
fusion techniques. Implement step motion
detector.

February

Integration with existing methods, replace
ABS with Viterbi Algorithm.

March

Integration with existing methods, replace
ABS with Viterbi Algorithm.


April

Fine tuning and improving the performance of
overall system. Report writing.








18


Bibliography

Bahl, P. (February 2000). Enhancements to the RADAR User Location and Tracking System.
Microsoft
Research

.

Chintalapudi, K. (September 2010). Indoor Localization Without the Pain.
Microsoft Research India

.

Cypriani, M. (January 2011). Wi
-
Fi
-
Based Indoor
Positioning: Basic Techniques, Hybrid Algorithms and
Open Software Platform.
Université de Franche
-
Comté, LIFC, Montbéliard, France

.

Kjaergaard, M. B. (2011). Indoor location fingerprinting with heterogeneous clients.
Pervasive and
Mobile Computing

.

Lassabe, F. (2007).
Numerica
-

Multimedia Developpement Center

.

Lawitzki, P. (March 2012). Application of Dynamic Binaural Signals in Acoustic Games.
Stuttgart Media
University

.

Liu, C.
-
Y. (n.d.). Location Tracking by Zigbee.
Computer Science and Informa
tion Engineering Department,
Cheng Shiu University

.

Liu, H. (1999). Evaluation of a Low Cost Solid
-
State Accelerometer as a Distance Measuring Sensor for
Vehicle Positioning System.
Dept. of Electrical and Electronic Engineering, The University of Hong Ko
ng

.

Lui, H. (November 2007). Survey of Wireless Indoor Positioning.
IEEE TRANSACTIONS ON SYSTEMS, MAN,
AND CYBERNETICS

, PART C: APPLICATIONS AND REVIEWS, VOL. 37.

Rehim, M. A. (2004). HORUS: A WLAN
-
BASED INDOOR LOCATION DETERMINATION SYSTEM.
Department
o
f Computer Science

.

Serra, A. (2010). Inertial Navigation Systems for User
-

Centric Indoor Applications.
University of Cagliari,
Italy

.

Shala, U. (September 2011). Indoor Positioning using Sensor
-
fusion in Android Devices.
Department
Computer Science an
d Embedded Systems

.

Tarzia, S. P. (2011). Indoor Localization without Infrastructure using the Acoustic Background Spectrum.
Northwestern University and University of Michigan

.

Wang, H. (2012). Unsupervised Indoor Localization.
Duke University

.

Want, R. (n.d.). The Active Badge Location System.
Olivetti Research LTD (ORL) Cambridge, England

.