An Ofﬂine Outdoor Navigation System with Full Privacy

Prakhar Kaushik

, Vishal Saraswat

and Francesco Buccafurri

Amity University, Delhi, India

Indian Statistical Institute, Kolkata, India

DIIES, University of Reggio Calabria, Reggio Calabria, Italy

Keywords:

Privacy, Positioning Systems, Smartphone.

Abstract:

GPS navigation systems are a potential threat to user privacy in case of curious providers, espionage and many

other aspects. Users tend to place blind trust into GPS applications without realizing the ease at which the GPS

can be spoofed or their position compromised via either the hardware or software. Thus, when a high level

of privacy assurance is required, the GPS should be completely switched off. This paper presents an efﬁcient

method, a smartphone-based alternative solution, for an outdoor ofﬂine navigation system, which works in

the absence of GPS, wireless, and cellular signals. The proposed approach exploits the various digital and

mathematical resources present to use DEM data and sensor data to minimize errors in the calculated position

data.

1 INTRODUCTION

The development of Global Positioning Systems

(GPS), Mobile triangulation, etc. have led a revolu-

tion in outdoor navigation. There are several tech-

niques employed to ﬁnd the position of the subject

with satellite signals. Inertial navigation has been

used since some time due to its applications in avi-

ation and military purposes with speciﬁc reference to

missile navigation. In inertial navigation, measure-

ments provided by the inertial sensors such as gyro-

scopes and accelerometers. Pedometer systems count

steps by monitoring the vertical acceleration using a

piezo electric accelerometer. Despite the voluminous

research in personal and pedestrian navigation (Kim

et al., 2004; Feliz Alonso et al., 2009; Ho et al., 2016),

a personalized positioning system which works with-

out any type of wireless, GPS or cellular signal is still

not publicly available despite the obvious demand.

We encounter such conditions while trekking and hik-

ing through mountains and forests. Another use case

is whenever a user (consider for example the intel-

ligence domain) wants to have fully secret naviga-

tion and needs to turn-off all the wireless functions

of the smartphone. In those cases, even the use of

only the GPS receiver could be not a solution be-

cause even though GPS is a self-positioning system,

its availability in telematics systems enables various

privacy abuses both in real-time and retrospect (Iqbal

and Lim, 2010; Alzantot and Youssef, 2012). Even a

passive GPS antenna with path logging off is suscep-

tible to spooﬁng attacks as well as remote GPS data

acquisition as long as they are connected to the net-

work. Further, the hardware itself can be manipulated

in order to leak position data. With increasing num-

ber of mobile applications and software directly or

indirectly collecting positioning data from our smart-

phones, it has become relatively easy for a third party

to track our positions.

We introduce a new algorithm which may be capa-

ble of resolving the aforementioned problem in terms

of an off-line navigation system. There have been ef-

forts to solve a similar problem, but in the context

of an indoor environment or a dense human interac-

tive environment (for example, a trade fair). This pa-

per presents a solution for outdoor environment with

sparse human activity.

We propose a positioning system which can work

with off-line smartphones, provided, the maps of the

territory are previously downloaded. Moreover, col-

lected data are only differential w.r.t. a given posi-

tion, so they are inherently less sensitive than GPS

data. The positioning is then really performed when-

ever an initial checkpoint (starting point set in the ap-

plication) is mapped to a point of a map. We believe

that a smartphone which is not transmitting or receiv-

ing any kind of wireless signal is certainly compara-

tively more safe than a GPS in terms of privacy.

Kaushik, P., Saraswat, V. and Buccafurri, F.

An Ofﬂine Outdoor Navigation System with Full Privacy.

DOI: 10.5220/0006473800950101

In Proceedings of the 14th International Joint Conference on e-Business and Telecommunications (ICETE 2017) - Volume 6: WINSYS, pages 95-101

ISBN: 978-989-758-261-5

Figure 1: Inertial Navigation System.

2 IMPORTANT DEFINITIONS

2.1 Inertial Navigation

Inertial navigation (Woodman, 2007) (Figure 1) is

a self-contained navigation technique in which mea-

surements provided by accelerometers and gyro-

scopes are used to track the position and orientation

of an object relative to a known starting point, orienta-

tion and velocity. Inertial measurement units (IMUs)

typically contain three orthogonal rate-gyroscopes

and three orthogonal accelerometers, measuring an-

gular velocity and linear acceleration respectively.

For tracking the position, we use the accelerome-

ter data A

(t) = (a

(t),a

(t))

and project it

to global frame of reference: A

= CA

. This data is

then integrated to get velocity and displacement.

(t) = v

(0) +

(t) −g

(t) = s

(0) +

(t)dt .

2.2 Directional Trail

The directional trail (Constandache et al., 2010) is de-

ﬁned as a series of last N compass readings and an

associated set of displacements between them. The

compass readings are collected with a time separa-

tion of 1 second. A compass trail is a tuple C

...c

) where c

is the compass reading at

the user’s estimated current location, c

is the pre-

vious reading, and so on. The accelerometer based

displacements between compass readings is the tuple

D = δ

,δ

,...δ

−N+1

where the user walked a distance

of δ

between two compass measurements, c

j−1

and

2.3 Sensor Fusion

Sensor Fusion (Elmenreich, 2002) refers to the com-

bination of two or more data sources, to produce bet-

ter data, better representation, improve signal to noise

ratio, etc. Gyroscope data may be fused with that of

magnetometer to smoothen the noise in the gyroscope

data. Similarly, gravity data may be fused with ac-

celerometer data in order to get only acceleration val-

ues for each axis. Sensor data may also be fused with

an absolute positioning system to provide a better and

accurate data.

2.4 Error Sources and Analysis

There are ﬁve major error types (Woodman, 2007)

while dealing with accelerometer and gyroscope —

white noise, temperature, calibration, bias and bias

instability. For accelerometer, double integration of

bias leads to a quadratically increasing positional er-

ror while in gyroscope there is steadily increasing an-

gular error.

2.5 Countermeasures

Two popular countermeasures against the above-

mentioned errors are:

Sensor Fusion. Raw data from two or more sensors

is processed through sensor fusion methodology

to obtain relatively more accurate readings.

Modelling and Subtraction of Bias. The system is

kept stationary for a period of time and the bias

observed is then modelled and subtracted from the

ﬁnal readings.

3 RELATED WORK

(Zhou, 2016) studied pedestrian dead reckoning and

its exploitation of the cyclical movements of the hu-

man body and provided algorithms to detect the foot

fall and estimate the stride length of the subject. A

reference to a relationship between z-axis acceleration

and step length is also provided. A Simple Moving

Average (SMA) preprocessing algorithm is proposed

for smoothing the acceleration data.

(Yun et al., 2007) provided a method for tracking

WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems

2D and 3D position of human movement using a self-

contained inertial/magnetic sensor module and pre-

liminary experimental results for various human mo-

tion including straight line walking, circular walking,

side stepping, backward walking, running, and climb-

ing stairs. They use a time heuristic and z-axis accel-

eration data to detect stance phase (two phases dur-

ing walking: stance phase (60%) and swing phase

(40%)). For angular measurements also, the x-axis

angular rate is near zero (for foot placed sensor) dur-

ing the stance phase. For straight line walking, the

average measurement error was recorded to be 5.5%

which was attributed to the presence of magnetic in-

terference in indoor environments. For outdoor con-

ditions, an average measurement error of 1.3% was

observed. The maximum error for outdoor straight

line running was recorded to be within 4.75%. Circu-

lar walking gave a distance estimation error of 1.8%.

(Constandache et al., 2010) presents a scheme,

CompAcc (Figure 2), to counter the cons of position-

ing systems like GPS, War Driving, GSM triangula-

tion, etc. The authors of this paper also had similar

motivations and objectives as us and seem to have had

similar train of thought, but they do not provide a fully

ofﬂine system as we do.

Figure 2: CompAcc.

Basically, a map tile containing the path informa-

tion of the area is acquired from the server. Using the

AGPS initial location, the system creates path signa-

tures with the initial location as its centre. A direc-

tional trail for the subject is created using (distance,

orientation) tuple. As a fall-back mechanism, Com-

pAcc allows the search window to cover at most a sin-

gle intersection at a point of time in the map tile. The

errors which aggregate during this period are reset by

the detection of directional change by the sensors. In

case the error grows too big, an AGPS request is sent

for location data reset.

(Link et al., 2013) is similar to our topic, though,

in an indoor navigation concept. They employ a two-

stage approach: (1) step detection and step heading

estimation, and (2) match detected steps onto the ex-

pected route from the source to the destination using

sequence alignment algorithms.

Basically, a pre-designed ﬂoor plan is used to

identify probable paths between the source and the

Figure 3: Step Detection.

destination. OpenStreetMap is used as a map ma-

terial. For step detection (Figure 3), the z-axis ac-

celerometer data is used with a time out window to

avoid false step detections. A step is detected if there

is a difference of at least p = 2m/s

on the low pass

ﬁltered z-axis of the accelerometer.

4 EXPERIMENTAL SETUP

All the sensor and raw data was collected initially on a

Samsung Galaxy Note 3 Neo Smartphone (Model no.

SM-N900) running Android version 5.0. The speciﬁ-

cations are as follows:

• Samsung Exynos 5420,

• Processor clock: 1.90 GHz,

• Number of cores: 8,

• GPU: ARM Mali-T628 MP6 @533 MHz.

The coordinate system, relative to the device, that

is used by the Sensor API is shown in the Figure 4.

Figure 4: Coordinate system relative to the device.

5 OUR MODEL

Our work in subdivided into the following parts:

An Ofﬂine Outdoor Navigation System with Full Privacy

5.1 Acquisition of Topographic Data

Taking reference from our initial GPS referenced po-

sition (it is not necessary to take an initial GPS posi-

tion - we can manually mark an initial point using a

known physical reference which will take the appli-

cability of the security notion by a notch), a relevant

area around the initial point is to be selected and data

points containing tuples:

Map(latitude,longitude,elevation)

are to be created. Taking an initial position from an

already compromised connection or system will cer-

tainly bring down the security level of the applica-

tion, but the application will still block any further

data leakage.

The elevation data with respect to the coordinate

data is extracted employing a digital elevation model

(DEM) (Figure 5). DEMs store the terrain informa-

tion as data using a Raster Data Model which repre-

sent the world as a surface divided into regular grid

of cells. Raster models are useful in storing data

which varies continuously. There are various DEM

data sources available free of cost all with compara-

tive degrees of resolution and accuracy.

Figure 5: Extracted Elevation proﬁle on MATLAB.

MATLAB itself provides various DEM sources and

pre-existing functions to manipulate them (Figure 6).

Google Earth/Maps also provides elevation data and

there are various online resources which can be used

to extract the relevant geo-referenced elevation data.

In our work, we used all these methods to collect the

relevant elevation and coordinate data.

5.2 Preprocessing of Sensor Data

In order to reduce errors and computation time, the

derived sensor data is preprocessed. For acceleration

data, the effect of gravity is subtracted and a smooth-

ing ﬁlter such as averaging ﬁlter and a FFT based

smoothing ﬁlter is applied to reduce the effects of

noise. A bandpass ﬁlter with appropriate cut-off fre-

quencies will provide a better result.

Figure 6: Extracted topographic map from a DEM source

in MATLAB.

As noted, most of the drift error in position data

is due to the gyroscope readings. The gyroscope data

can be fused with that of magneto scope. Also, the

smartphone can be kept still on a solid surface after

moving it for a while. This will allow us to calculate

and thereafter subtract a deﬁnite bias from our accel-

eration and angular velocity data. For barometer, the

readings should be appropriately modiﬁed with re-

spect to temperature, humidity and gravitational er-

rors.

5.3 Sensor Data Acquisition

Figure 7: Acquired Smartphone Sensor Data.

The Android sensor framework (Android.Com,

2009) is used to acquire raw sensor data and to mon-

itor for changes in the value of the sensor (Figure 7).

A sensor event provides you with four pieces of in-

formation: the name of the sensor that triggered the

event, the timestamp for the event, the accuracy of the

event, and the raw sensor data that triggered the event.

Two call back methods exist on SensorEventListener

WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems

interface: when the accuracy of sensor changes and

when sensor reports a new value.

5.4 Probable Path Patterns

In the case, no direct path trails exist between the

start position and the destination, probable path pat-

terns are calculated from the positional elevation data.

Imagining the map is divided into N number of cells,

with each cell consisting its positional and elevation

data, we will apply the 8-adjacency concept in the

neighbourhood of the cell.

The basic idea is to ﬁnd an adjacent cell in the

heading of the destination whose elevation is less than

a predetermined value k (probably 2 metres). Path

ﬁnding algorithms, greedy algorithm, etc. can be used

for this. In our case, we leaned towards the use of Ant

Colony Algorithm for ﬁnding the probable paths.

1. S is a set of all data points/cells consisting the lo-

cational and altitude data

2. ω is a set of constraints: (1) cell chosen should lie

in the present 8-adjacency neighbourhood of the

present cell (2) the difference between the eleva-

tions is less than k metres (3) the next cell chosen

should lie between −95

◦

and 95

◦

3. Objective function, f : Minimize(N), where N ⊂S

and is the number of cells used for path to desti-

nation.

Depending on the value of N, a ranking system is

provided to the probable path with the highest rank-

ing given to the path having the lowest value. In the

unusual circumstance of the subject not following any

of the probable paths, the elevation proﬁle can be used

to ﬁnd paths with similar proﬁle in the map.

5.5 Step and Heading Detection

In order to count the number of steps taken, we need

to ﬁrst calculate the (dynamic) stride length as well

detect. Due to the highly erroneous nature of data

received otherwise, the relationship between z-axis

acceleration data and the stride length is exploited.

Kalman ﬁlter is used with one of the equations

√

max

−a

min

∑

i=1

| or

∑

i=1

|−a

min

max

−a

min

for a better grasp on the dynamic nature of stride

length. For heading estimation, the angular displace-

ment data is obtained from the accelerometer, gyro-

scope and magnetometer readings.

5.6 Step Sequence Detection

After creating of several probable paths, step se-

quence arrays consisting of the azimuthal angle differ-

ence between each step is created. The same is done

for the actual subject movement. Taking a k width

window from this array, the values are correlated with

the values of the probable paths in order to identify

the probable path being used by the subject.

5.7 Pattern Correlation

The elevation proﬁle of the path that the subject is tak-

ing is created along with the motion. This elevation

proﬁle is then correlated (pattern or map correlation)

with the precomputed elevation proﬁle. The correla-

tion is done on a frequent basis, taking parts of the

elevation proﬁle as well as the whole. Pattern corre-

lations can be computed directly (uncentered) or by

computing anomalies from a central mean (centred).

This correlation process allows us to reduce the po-

sitional ambiguity and bypass the necessity to over-

come noise by providing us landmarks for position

identiﬁcation.

5.8 Algorithm

Finally, the overall algorithm (Flow chart in Figure 8)

is as follows:

1. The initial point data is collected from GPS and

cellular connection, etc. Coordinates and eleva-

tion data for the present position is collected.

2. DEM data server is accessed and a DEM tile of

adequate map area is downloaded. Tuples of Map

data (latitude, longitude, altitude) are created. Ad-

ditional information such as weather forecast, etc.

(for barometer reading correction) is also col-

lected and sampled at this juncture.

3. Acquisition and pre-processing of sensor data to

reduce errors in the reading.

4. Calculation of step length using z-axis accelerom-

eter data and Extended Kalman Filter. Step Detec-

tion and Heading Detection using the sensor data.

5. Creation of probable path patterns (with ranking)

via the Probable Path Algorithm. Data sets of (lat-

itude, longitude, altitude) for each of these pat-

terns are created.

6. Step heading sequence and altitude pattern is also

created for each of these paths.

7. Create positional approximates by dead reckoning

using step counts, step length and heading infor-

mation. Cross reference these positions to saved

DEM data and create tuples of Calculated Data

An Ofﬂine Outdoor Navigation System with Full Privacy

Figure 8: The ﬂow chart of the proposed Personalised Navigation System.

(latitude, longitude, altitude). Create the altitude

pattern and step sequence array.

8. A window of adequate length is moved back and

forth and is used to correlate the measured step

sequence with that of the probable paths (going

down the ranking).

9. The measured elevation proﬁle is correlated with

that of the expected paths. Given appropriate error

allowance δ, the elevation patterns are correlated

piecewise as well on an overall basis and the po-

sition corrected periodically.

10. In case a centre server is set up helping in sharing

the users’ data(the users choose to send the data

voluntarily). The data can be collected anony-

mously, without any reference to the identity of

the user who had used the path, etc. Probable path

patters and their position and elevation data will

be averaged out and the ranking of the particular

path will be decided by the number of users using

that path. This will greatly increase the overall

accuracy of the system.

6 CONCLUSION AND FUTURE

WORK

In this short paper, we have presented an ofﬂine nav-

igation system and some possible applications. Our

proposed algorithm creates a framework in which fur-

ther study can be done to achieve the creation of

a highly accurate ofﬂine positioning system. We

build our case centering on technologies available at

present. The paper argues that despite the obvious

failings of current MEMS and positioning technol-

ogy, we can use better algorithms to counter these

failings to achieve our objectives. We identify the ba-

sic building blocks and error sources of the existing

algorithms, which allows us to create a clear path-

way for further study. Privacy is a strong point of

our proposal and many secrecy/privacy-aware appli-

cations in the intelligence and military ﬁelds can be

developed. Moreover, our technique may have im-

plications in a large set of possible domains like ad-

venture sports, civil protection, location-base gaming,

speleology, and so on.

REFERENCES

Alzantot, M. and Youssef, M. (2012). Uptime: Ubiquitous

pedestrian tracking using mobile phones. In WCNC,

pages 3204–3209. IEEE.

Android.Com (2009). Sensors overview.

Android Developers’ API Guides.

https://developer.android.com/guide/topics/sensors/

sensors overview.html.

Constandache, I., Choudhury, R. R., and Rhee, I. (2010).

Towards mobile phone localization without war-

driving. In Infocom, pages 1–9. IEEE.

Elmenreich, W. (2002). An introduction to sensor fusion.

Vienna University of Technology, Austria.

Feliz Alonso, R. ’u. l., Zalama Casanova, E., and G mez

Garc

ı a Bermejo, J. (2009). Pedestrian tracking using

inertial sensors. Journal of Physical Agents, 3:35–42.

Ho, N.-H., Truong, P. H., and Jeong, G.-M. (2016).

Step-detection and adaptive step-length estimation for

pedestrian dead-reckoning at various walking speeds

using a smartphone. Sensors, 16(9):1423.

Iqbal, M. U. and Lim, S. (2010). Privacy implications of

automated gps tracking and proﬁling. Technology and

Society Magazine, 29(2):39–46.

Kim, J. W., Jang, H. J., Hwang, D.-H., and Park, C.

(2004). A step, stride and heading determination

for the pedestrian navigation system. Positioning,

1(08):0.

Link, J.

A. B., Smith, P., Viol, N., and Wehrle, K. (2013).

Accurate map-based indoor navigation on the mobile.

Journal of Location Based Services, 7(1):23–43.

WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems

100

Woodman, O. J. (2007). An introduction to inertial nav-

igation. Technical report, University of Cambridge,

Computer Laboratory.

Yun, X., Bachmann, E. R., Moore, H., and Calusdian, J.

(2007). Self-contained position tracking of human

movement using small inertial/magnetic sensor mod-

ules. In ICRA, pages 2526–2533. IEEE.

Zhou, R. (2016). Pedestrian dead reckoning on smartphones

with varying walking speed. In ICC, pages 1–6. IEEE.

An Ofﬂine Outdoor Navigation System with Full Privacy

101