ArticlePDF Available

SubwayPS: Towards Enabling Smartphone Positioning in Underground Public Transportation Systems

  • University of St.Gallen
SubwayPS: Towards Smartphone Positioning in
Underground Public Transportation Systems
Thomas Stockx
Expertise Ctr. For Digital Media
Hasselt University - tUL - iMinds
Brent Hecht
GroupLens Research
University of Minnesota
Johannes Schöning
Expertise Ctr. For Digital Media
Hasselt University - tUL - iMinds
Thanks to rapid advances in technologies like GPS and Wi-
Fi positioning, smartphone users are able to determine their
location almost everywhere they go. This is not true,
however, of people who are traveling in underground public
transportation networks, one of the few types of high-traffic
areas where smartphones do not have access to accurate
position information. In this paper, we introduce the
problem of underground transport positioning on
smartphones and present SubwayPS, an accelerometer-
based positioning technique that allows smartphones to
determine their location substantially better than baseline
approaches, even deep beneath city streets. We highlight
several immediate applications of positioning in subway
networks in domains ranging from mobile advertising to
mobile maps and present MetroNavigator, a proof-of-
concept smartphone and smartwatch app that notifies users
of upcoming points-of-interest and alerts them when it is
time to get ready to exit the train.
Categories and Subject Descriptors
H.5.m. Information interfaces and presentation (e.g., HCI):
mobile navigation, positioning, mobile devices,
underground public transport, GPS, accelerometer
A smartphone’s ability to detect its location is the
cornerstone of mobile applications in domains ranging from
location-based services to mobile crowdsourcing. While
technologies like GPS and Wi-Fi positioning have made
location detection possible in most places smartphone users
go, this is not true everywhere. Underground public
transportation systems, which are largely inaccessible to
GPS signals and often out of range of Wi-Fi networks,
represent a particularly important type of space in which
smartphones cannot reliably determine their location. As a
result, the millions of locals and tourists who ride subways
around the world each day [21] cannot take full advantage
of popular location-aware smartphone applications
including mobile map apps while traveling to and from
their destinations.
In this paper, we present SubwayPS, the first positioning
technique that allows modern smartphones to determine
their location while traveling in an underground public
transportation network. Our technique, which we
demonstrate can achieve reasonablethough not perfect
accuracy, does not depend on any instrumentation of the
environment. Instead, SubwayPS merely requires an
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. Copyrights
for components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to
post on servers or to redistribute to lists, requires prior specific
permission and/or a fee. Request permissions from
SIGSPATIAL '14, November 04 - 07 2014, Dallas/Fort Worth, TX, USA
Copyright is held by the owner/author(s). Publication rights li censed to
ACM. ACM 978-1-4503-3131-9/14/11...$15.00
Figure 1: A screenshot of the Google Maps iPhone
application used on the Blue Line subway in Chicago. The
gray dot shows the smartphone’s estimated position, while
the green dot shows the actual position of the user’s train.
Red dots are underground stations. One could easily miss
their stop if they relied solely on current positioning
technology. SubwayPS works towards addressing this issue by
providing improved positioning while traveling in
underground public transportation networks (Base map ©
Google Maps 2014).
accelerometer and gyroscope and an input origin location
and end destination. Accelerometers and gyroscopes are
included in most recent smartphones. Origins and
destinations can be easily extracted from routing requests
(especially if this type of positioning technology is built
into a mobile OS) or inferred from user behavior (as in
context-aware technologies like Google Now [6]).
Smartphone positioning in underground public
transportation networks affords wide-ranging benefits. The
most basic such benefit is that SubwayPS simply puts
subway riders “back on the map”. Figure 1 shows the result
of using Google Maps for iOS while traveling through an
underground portion of the Chicago Blue Line. Because the
app cannot obtain position information, it is rendered
effectively useless for navigation and orientation purposes.
Our technique can make mobile maps as useful in
underground transportation networks as they are above
ground. For example, a SubwayPS-enabled navigation app
could much more effectively help tourists navigate foreign
subway systems, especially when they do not speak and
cannot read the native language. Such an app could, for
instance, inform tourists who cannot understand the
announcements of when to prepare to exit the train, an
especially important concern when the train is crowded.
SubwayPS can also enable location-based applications
beyond basic navigation and orientation. For instance, if our
technique were built into the location API of a mobile OS,
smartphone users would be able to use cached content from
apps like Yelp to easily see the restaurants that would be
available if they were to get off at the next stop. In addition,
mobile advertising becomes possible underground (e.g.
“Exit at the next stop to get to restaurant X and receive $5
The core contribution of this paper is our underground
public transportation positioning technique SubwayPS,
along with three evaluations that show that our technique
can achieve reasonable accuracy and substantially
outperforms baseline approaches on multiple subway
systems. Below, we first put this core contribution in the
context of related work. Next, we present SubwayPS in
detail and introduce MetroNavigator, a proof-of-concept
application based on SubwayPS. We then demonstrate the
accuracy of SubwayPS using log data and in-the-wild user
studies, showing that it performs substantially better than
baseline approaches. We also highlight qualitative feedback
from participants, which included I want to have it
integrated in Google MapsandI feel safer when I can see
that we have stopped close to the next station when we’re
stopped in the middle of a tunnel(P11). Finally, we close
by highlighting in more detail the many applications of
subway positioning and discussing several means by which
accuracy can be improved in future work.
The work presented here is informed by research from three
areas: inertial navigation, transportation mode detection,
and other approaches to positioning in public transportation
networks. Below, we discuss each of these areas in turn.
Inertial navigation
SubwayPS uses an approach to positioning motivated by
the literature on accelerometer- and gyroscope-based
inertial navigation. Although inertial navigation approaches
have not been adapted to the context of public
transportation positioning as they are here, they have been
shown to be successful for other purposes, especially
pedestrian indoor navigation (e.g. [1,5,13,14]). For
instance, Robertson et al. [13] presented a simultaneous
localization and mapping (SLAM) approach for pedestrians
using only foot-mounted inertial sensors. Their method,
FootSLAM, can be used to generate an approximate 2D map
of an indoor layout with just 10 minutes of walking.
Other accelerometer-based approaches have combined
inertial navigation with GPS and other positioning signals
for improved accuracy. The NavMote Experience [4], for
example, includes a dead reckoning module, self-organizing
wireless ad hoc networks and an information center with a
map database. Another example is the Embedded
GPS/RFID/Self-contained Sensor System by Kourogi et al.
[9]. As described in the name of their system, Kourogi and
colleagues integrated self-contained dead reckoning sensors
(accelerometers, gyrosensors and magnetometers) with GPS
and an RFID tag system to adjust for errors in position and
direction that occur in the dead reckoning process.
A number of inertial navigation-based positioning
techniques for indoor navigation and related purposes have
been developed for smartphones. These techniques
generally use the detection of walking steps (“step
detection”) together with a method for heading
determination. Such techniques have seen huge accuracy
improvements in recent research (e.g. [8,10,11,15,16,17])
but are unsuitable for localization in underground public
transport systems where step detection is not an option.
Transportation Mode Detection
Wang et al. [19] showed that accelerometers on mobile
devices can be used to detect a user’s mode of transport
(e.g. walking vs. biking vs. driving). In this research, they
used a formula for acceleration synthesization to learn the
patterns of different transportation modes. We implement
their formula in our SubwayPS technique by using it to
detect if a train is moving or not. Hemminki et al. [7] have
recently achieved improved accuracy on transport mode
detection by providing a new algorithm for estimating the
gravity component and key characteristic pattern detection,
which involves the detection of mode- specific patterns
such as step detection for walking and smooth accelerations
for vehicle movements.
Positioning in Public Transportation Systems
The final area of work related to SubwayPS comes from the
literature on positioning in public transportation systems.
For instance, the EasyTracker system [2] provides a
location tracker for public transport that uses GPS traces
from above-ground transit vehicles (busses), but requires all
vehicles to be instrumented with equipment such as GPS
receivers and inertial navigation sensors. Zhou et al. [20]
developed a positioning technique that combines a variety
of sensor measurements from mobile devices owned by the
people on an above-ground transit vehicle to calculate an
average location. Thiagarajan et al. [18] describe a crowd-
sourced transit tracking system intended to improve real-
time arrival and departure estimates. As part of this
research, Thiagarajan and colleagues looked at subway
systems, but their approach is not designed for positioning
on mobile devices and requires a connection to a tracking
server, which is often not available while on a subway.
To summarize, in this paper we present the first smartphone
positioning technique (SubwayPS) for underground public
transportation systems that relies solely on the user’s
mobile device. As a result, our technique does not require
the infrastructure [2], connectivity [18], or the critical mass
of users [20] needed by existing approaches in this area.
Instead, SubwayPS merely requires a single accelerometer
and gyroscope, both of which are built into most modern
smartphones. In addition, SubwayPS has important privacy
benefits: because all calculations are done locally on the
user’s mobile device, SubwayPS does not share a user’s
position with any central server or third parties.
With the SubwayPS technique, we show that it is possible
to achieve reasonable positioning accuracy on underground
public transportation networks. Our results suggest that
SubwayPS is usable as-is on most subway systems in the
world. We also show how accuracy can be improved by
fine-tuning certain parameters to the characteristics of a
specific underground public transport system.
Collecting accelerometer data in a pretest on multiple
underground public transport systems showed that station
locations overlapped with periods in our data collection
when there was generally less acceleration on all axes of the
accelerometer. This is due to the fact that train movement
results in multiple small accelerations in all directions (i.e.
the small amount of shaking back and forth experienced by
passengers when a train is at high speed). As such, by
measuring the amount of variance on all axes, it is possible
to detect whether the train is moving or not.
Figure 2, which shows data collected on the London
Underground subway system, demonstrates this
phenomenon in more detail. Figure 2a depicts the
accelerations measured on each axis of the accelerometer
over a small period of time. Stations are clearly visible in
the data in that they have accelerations closer to zero and
less variation (around minute 1 and 2.7 in Figure 2a). By
calculating a general acceleration value based on the three-
axis input of the linear accelerometer data, it is possible to
derive a statistic a that has a large value if there is a large
amount of acceleration on any (or all) of the axes. To
calculate this value, we use the following formula for
acceleration synthesization [19]:
The resulting a values are smoothed by a rolling average
with a window of n samples. We found n = 100 samples at
50 Hz to be an appropriate general window. Lower values
would confuse user movement of the smartphone as false
train movement (see Study 2 for more details), while higher
values would increase the delay between events (e.g.
subway stops) and their detection by SubwayPS. Plotting
out a results in a graph like that shown in Figure 2b.
By comparing the smoothed data with a certain threshold, it
is possible to conclude if a train is moving or not. The
SubwayPS technique uses multiple parameters to parse the
data and uses states to define if a train is moving or not,
Figure 2: In Figure 2a, a visualization of accelerometer
values measured on each axis is shown. The blue dotted lines
indicate time in minutes while the red intervals indicate
stations. The data was collected on the Central line in
London. Figure 2b shows the result of the acceleration
synthesization used in SubwayPS, smoothed by a moving
average. This result is compared to the threshold to detect
train movement.
SubwayPS makes use of a smartphone’s accelerometer and
gyroscope by combining the accelerometer measurements
on all three axes to detect underground train movement.
The gyroscope is used to filter the gravity factor out of the
accelerometer measurements. As such, the accelerometer
reads (x = 0 m/s², y = 0 m/s², z= 0 m/s²) when the mobile
device is at rest and the coordinate system is defined
relative to a default orientation of the accelerometer sensor.
The “axes are not swapped when the device’s
screen orientation changes” [22]. In our implementation
of SubwayPS, we made use of the virtual linear
accelerometer provided by the Android operating system,
which handles the gravity removal internally.
based on a window of samples. The parameters considered
n = window size of moving average
𝛾 = threshold
𝛿!"#$% = min. amount of samples below threshold
𝛿!"#$% = min. amount of samples above threshold
The values for each of the four parameters was determined
empirically using data from four different subway systems
as described below in the evaluation section. SubwayPS
requires at least 𝛿!"#$% samples below the threshold before
it marks the current accelerometer measurements as a stop,
while at least 𝛿!"#$% samples above the threshold are
required for SubwayPS to conclude that the train is moving.
These values are used to prevent flagging user movement as
train movement (Study 3).
The accuracy of SubwayPS on individual public
transportation networks can be improved by customizing
the values of 𝛿!"#$% and 𝛿!"#$% for each network. For
instance, increased accuracy for the London and Cologne
subways can be achieved by using the values in the second
and third columns of Table 1, respectively. Trains in
London accelerated more slowly, resulting in a longer
period required to reach 𝛿!"#$% , while in Cologne, trains
accelerated rather quickly. Because the threshold was
reached much faster in Cologne, 𝛿!"#$% could be increased
to decrease the chance of false positives.
While the “world-wide” parameters were empirically
determined to work reasonably well on all subway systems
tested (see below), the parameters specifically tuned to the
data we collected on each subway system can increase local
performance. For example, using the world-wide
parameters for SubwayPS on data collected in London
results in an accuracy of 74.2%, while using the London
parameters increases the accuracy percentage to 85%
(Study 1).
Our stop detection method is complemented by a linear
interpolation of the current position based on the time
intervals between stations as defined in official timetables.
By interpolating over time, we can estimate the absolute
location between the previous and the next station. If
SubwayPS detects a stop that is not scheduled, SubwayPS
marks it as an “in-between” stop and uses the interpolation
of location over estimated time to show a position to the
user. Using this method we can provide the users with a
position estimate of where exactly the train stopped
between two stations. We consider any stop that occurs
with less than 70% of the estimated time between stations
to be an “in-between” stop, a number that was determined
empirically from our Study 1 data (see below). This
percentage can be set to a higher value if a timetable is
available with seconds-level precision, which was not the
case with the four subway systems considered here, all of
which list timetables at the precision of minutes.
We implemented a working version of SubwayPS on the
Android smartphone platform, as well as a proof-of-concept
SubwayPS-based Android application called
MetroNavigator, which has both smartphone and
smartwatch versions. We first describe our implementation
of core SubwayPS and then discuss MetroNavigator.
Implementation of SubwayPS
MetroNavigator Smartphone Application
MetroNavigator (Figure 3) is a proof-of-concept application
we built that uses SubwayPS as its positioning technology
(by implementing an intent listener). The application’s UI
consists of two main parts. The upper part contains a
miniature map with an indicator of the user’s current
position between the previous and next station, as well as
the origin and destination station, the estimated time of
arrival and the number of stops to the end station. The
lower part of MetroNavigator’s UI is used to display event
cards. Currently our MetroNavigator application supports
multiple different types of event cards:
0.2 m/s²
0.2 m/s²
0.2 m/s²
250 samples
250 samples
250 samples
350 samples
250 samples
500 samples
100 samples
at 50 Hz
100 samples
at 50 Hz
100 samples
at 50 Hz
Table 1: General, “world wide” parameters for
SubwayPS, as well as those specifically tuned for London
and Cologne.
SubwayPS makes use of the virtual linear accelerometer
class provided by the Android OS. It “measures the
acceleration force in m/s2 that is applied to a device on all
three physical axes (x, y, and z)” [22] without the
influence of gravity. These accelerometer measurements are
interpreted by the SubwayPS technique, which is
implemented as a background service and sends intents
(messages) such as StationDetected or MovingDetected to
the Android OS, which then forwards them to all
applications that implement a listener for these intents. Any
application (e.g. MetroNavigator) could implement an
intent listener and make use of our detection algorithm.
This background service is just a proof of concept.
Preferably, the Subway P S techni q u e w o u l d b e
implemented on an OS level such as within the
location services of Android.
1. The first, and most-often-used type is the
notification event card shown in Figure 3b. When
the train approaches the next station, it shows the
name of the next station and can indicate possible
connections. When the train approaches the final
station, this is prominently indicated.
2. The application also triggers event cards when the
train undergoes unplanned stops in between
stations. This is shown in Figure 3c.
3. Another event card type notifies the user of “rude”
driving by the train conductor based on measured
accelerations. This is shown in figure 4 and one
can share this event on Twitter (once connectivity
is obtained) with a direct mention of the public
transportation company.
4. MetroNavigator also contains point-of-interest
(POI) event cards that contain information about
POIs that are located above the user or at the next
station (see Figure 3e) and provide a link to more
information about the POI on Wikipedia, which is
locally stored on the device.
The MetroNavigator Smartwatch Extension
We also implemented a smartwatch version of
MetroNavigator that can be linked to an Android device
running SubwayPS via Bluetooth. Due to screen size
constraints, the smartwatch version has a much simpler UI
than the smartphone application. It displays to users only
the current or next station, as well as the estimated time of
arrival. A screenshot of the smartwatch application is
shown in Figure 5.
The smartwatch version also has a feature not yet available
in the smartphone version: when users reach the final
subway station of their trip, the smartwatch version tells
users in which direction to go to reach their destination
(Figure 6). In addition, when selecting a POI in the
smartphone app, the smartwatch will show the direction to
go to reach the POI once the user has reach the next station.
Figure 3: MetroNavigator application showing different “event cards”.
Figure 4: Sharing an event card of MetroNavigator with a
few clicks.
To evaluate SubwayPS, we conducted three separate
studies with the help of the MetroNavigator application.
The first study was focused on assessing the technical
soundness of the positioning technique in a variety of
public transportation systems around the globe using log
data. The second study was designed to compare SubwayPS
against a timetable-only baseline and to evaluate the effect
of arbitrary smartphone movements by users. The third
evaluation was focused on understanding the accuracy and
user experience of the MetroNavigator application in the
context of actual subway journeys made by locals and
Study 1
The goal of our first study was to collect accelerometer data
from a variety of diverse subway systems and use this data
to inform and evaluate SubwayPS.
Participants & Apparatus
For the data collection process, we developed an Android
application that captures and stores the data measured by
smartphone accelerometer sensors. The data collection
application supports the public underground transportation
network of four major cities, namely Brussels, London,
Cologne, and Minneapolis. Each of these subway systems
has unique characteristics: they use different trains, some
go above ground, and others merge with vehicular traffic.
We wanted to study a diverse set of subway systems in
order to support broader generalizability.
To record data, participants first selected a city, a line and
an origin and destination station. Participants were advised
to select the start and end station of their journey before
boarding the train and were told to mainly place their phone
in their pockets.
We recruited twelve participants that downloaded the data
collection application to their mobile device and asked them
to record data “whenever they use an underground public
transport system”. We also asked them to record very long
tracking segments to get tracks that are longer than normal
underground rides so our algorithm could be tested
exhaustively (21.51 minutes trip length vs. about 9 minutes
in the follow up studies). Most of the participants were
Android developers and computer science students and
collected the data on a voluntary basis on their daily
commutes or on business trips. These participants collected
a total of about 70 tracks in a period of around two months
and we received tracks from every underground transport
system supported by the application.
The average number of stations per track was 9.23 and the
average track length was 21.51 minutes. We analyzed this
data to determine the values for each of the SubwayPS
parameters described above (e.g. 𝛿!"#$%, 𝛿!"#$%). Using the
subway system-specific values shown in Table 1, we found
that for London, 103 out of 120 stops (105 stations and 15
“in-between” stops) were classified correctly (85.8%
excluding the start stations). 10 stations and 7 “in-between”
stops were missed. No false positives (detection of a stop
where there was none) were recorded.
This can be directly compared to the results of the same
tracks tested with the “world-wide” parameters, which
caused SubwayPS to correctly detect and classify about
74.2% of these stops.
Similar results (around 85% in comparison to around 75%
for world-wide” parameters) were measured in the other
three subway systems.
These results are somewhat comparable to those of a study
done by Thiagarajan et al. [18]. Using four tracks in the
subway system of the Chicago public transportation
network (as opposed to our 70 tracks across 4 subway
systems), Thiagarajan et al. were able to achieve around
55% accuracy (compared to our 85.8%), although it is
important to note that their focus was on station detection
Figure 5: Screenshot of the SubwayPS smartwatch
application that provides basic feedback to the user without
needing to take the smartphone out of their pocket.
Figure 6: Screenshot of the SubwayPS smartwatch
application that provides guidance on in which direction to
exit the final station to reach a certain POI.
rather than location prediction (which involves detecting in-
between stops). In Study 3, we show that SubwayPS also
strongly outperforms a “timetable” baseline.
Examining where SubwayPS failed, we noticed that trains
can run smoother” on certain lines (such as parts of the
Hammersmith & City line in London) than others in a given
subway network. Because SubwayPS’s parameters were
tuned on a subway system-wide basis, this variation
resulted in errors. We are working to implement track-level
parameters, which should increase accuracy by a substantial
Another issue arose out of huge variations in travel time for
trips between the same two stations. We recorded multiple
readings over multiple days for some lines and noticed a
standard deviation of about 7.12 minutes for a trip that
normally takes 29 minutes. We looked into this particular
case, and found out it was due to extensive waiting times at
stations when trains were overcrowded. These delays could
cause misclassification due to the integrated time-tables, as
“in-between” stops could be classified as a station if the trip
takes too long. An example of these variations in trip length
is visible in Figure 7. The graphs depict two trips between
the same stations and differ greatly in the in-between
Finally, another source of error was the quality of
accelerometer data. For example, badly calibrated
gyroscopes in all data captured by Samsung devices
resulted in incorrect measurements by the virtual linear
accelerometer (e.g. accelerations of above 1 m/s² on some
axes while at rest). It would be straightforward to correct
for these errors on a device-by-device basis if these errors
are known a-priori.
Study 2
A second user study was conducted with the
MetroNavigator application to directly compare the use of
SubwayPS against an approach that merely interpolated
from an official timetable.
The study was conducted across two consecutive days
between 8 and 6 pm in May 2014 on the Cologne, Germany
subway. One experimenter engaged in 50 trips in which he
randomly boarded a train, travelled with it for exactly seven
stations and then left the train and boarded the next train
arriving at that station. In order to compare against a
timetable baseline, the experimenter recorded the exact time
at which the train he was riding arrived at each station. No
restrictions were placed on the placement of the smartphone
in this study. The Study 2 experimenter performed common
activities normally executed on a smartphone as reported by
Böhmer et al. [3], which involved periodically having the
phone in his hands and storing the phone in his pocket
during the trips.
Across all 50 trips, 282 of 335 stops (300 stations and 35
“in-between” stops) were classified correctly (82.7%,
excluding the start stations; 85.0% including the start
stations). 35 stations and 18 “in-between” stops were
missed. Again, no false positive stops were recorded. These
results are in line with our previous results from Study 1
and Study 3 presented later in this paper.
To evaluate SubwayPS against the timetable baseline, we
compared the actual arrival time at each stop on all 50 trips
to (1) the arrival time indicated by SubwayPS and (2) the
arrival time indicated by the official subway schedule.
Using a 30-second “tolerance” window, we found that
SubwayPS tracked 39 entire trips with perfect accuracy
(78% of trips), while using the official timetable, only 21
entire trips (42%) were accurate.
The challenges presented to timetables from the “ripple
effects of a single delay have led some public transport
systems to abandon timetables for an “interval”-based
approach. In this approach, which was recently examined
by Pritchard et al. [12], service is guaranteed every n
minutes rather than at specific times. To understand
SubwayPS’s performance in this type of subway network,
Figure 7: Example of day-to-day variability measured on the Piccadilly line in London. Both graphs picture the data
calculated by SubwayPS. Stations A, B, and C are clearly visible. These graphs provide an example of how much variability
there is in the measurements. Times between the same two stations can vary greatly (see Track A-B) with a difference of
almost one minute. The period during which a train stands still at a station can also vary greatly (see B).
we also evaluated SubwayPS against a relative time
baseline, which uses just the reported travel time between
stations. This baseline was able to track 25 out of 50 trips
correctly (50%), as opposed to SubwayPS’s 39 of 50.
The trip-level accuracy of SubwayPS is directly influenced
by trip length and accumulation errors (if one station is
missed, the trip is not tracked correctly), which are known
issues for inertial navigation techniques (i.e. “drift” errors).
Fortunately, there has been extensive research on solutions
for recalibration in inertial navigation. We are currently
working to adapt one of these solutions to a public
transportation network context.
Study 3
To test the “in the wild” feasibility and appeal of SubwayPS
and its proof-of-concept implementation in
MetroNavigator, a user study was conducted.
MetroNavigator was pre-installed on a Google Nexus 4
device and the device was handed out to randomly selected
passengers waiting at different stations in the Cologne,
Germany subway system (lner U-Bahn). Due to novelty
effect concerns, we focused on the smartphone version of
MetroNavigator for the evaluation (instead of the
smartwatch version) and restricted our participant
population to subway riders who were already using an
Android device.
Sixteen participants (8 male, 8 female) with an average age
of 39 years took part in the study. Eight participants were
employees at different local companies in Cologne, seven
participants were tourists, and one participant was a student
at a local university. The study was conducted across two
days between 8 and 10am (peak travel time) within the
period of a week in February 2014.
The experimenter approached passengers that were using
Android devices at one of the selected subway stations to
invite them to participate in the study. If they agreed, the
experimenter read a quick script that explained the purpose
and functionality of MetroNavigator and informed the
participant that the experimenter would be riding with
her/him to their final destination. Participants were then
asked to enter their start and end subway stops and were
also requested to “think aloud when interacting with the
application. The script was kept short as we did not want to
cause our participants to miss their train. The experimenter
boarded the train with the participants and recorded their
reaction during the ride.
Due to the relatively short nature of subway trips and our
desire to have participants engage with the application,
participants were asked to hold the device in their hands
during the trip and not store it in their pockets. The
robustness of SubwayPS to normal user movement was
established in Study 2. After leaving the train, participants
filled out a background questionnaire about their age,
gender and occupation, and the experimenter conducted a
semi-structured interview. The experimenter also explained
how the system derives position information by using the
accelerometer of the smartphone and the smartwatch
extension was explained to the participants to get feedback
on its possible uses.
The participants took the train for an average of 6 stations
(including the start and end station) and the average time on
the train was about 8.75 minutes. 78 out of 91 stops (80
stations and 11 “in-between” stops) were classified
correctly (85.7 % excluding the start stations; 87.9 % also
including the start stations). Eight stations and 5 “in-
between” stops were missed. No false positives were
recorded. Out of the 16 trips, 12 were tracked completely
correctly (75%).
ResultsQualitative Feedback
During the interviews, 15 participants expressed a positive
opinion of SubwayPS and were interested in using
MetroNavigator (and other applications using its SubwayPS
engine) in the future. We received comments like “As my
GPS does not work here, this is like a GPS for
undergrounds trains I want to have it integrated in
Google Maps” (P3) or It is just cool to see the train
moving and stopping on the map I feel safer when I can
see that we have stopped close to the next station when
we’re stopped in the middle of a tunnel (P11). The main
advantage of MetroNavigator for most participants was the
ability to be guided to their destination stop (14 out of 16
participants) and general location awareness (12 out of 16
participants). “Wow - This is like magic” (P3) was the
comment of one participant when the visualization of her
current position stopped between two stations as the train
did. She commented at first, I though this app is a bit
boring as it just moves a metro along with the schedule
from station to station, but now I see, that it has GPS”.
Participants also enjoyed specific features of the
MetroNavigator app. In particular, several of the tourists
found value in MetroNavigator’s POI information cardsas
[they] can easily build local knowledge of a city(P2). The
smartwatch was generally thought to be a useful extension
for such an application. One participant remarked as I do
not want look at my smartphone all the time, this
(SubwayPS) could be the killer app for smartwatches in the
future” (P13, an IT consultant).
During the ride of one participant the MetroNavigator
application missed three stops in a row. She commented on
that by saying Oops, it seems that the system missed a stop
that is not good. I would also be happy to help the system
to detect the stops, if the system could help me to get out at
the right station. That would be totally fine with me(P3, a
tourist). Other critical comments were targeted at the
limited set of POIs in the app (e.g. Can it also show shops
and café places?”, P3) . In our prototype, we stored just a
few POIs and the dataset could be easily extended.
In this paper, we presented SubwayPS, a smartphone
positioning system that puts users back on the map” when
they are traveling in underground public transportation
networks. SubwayPS does not require any instrumentation
of the environment, meaning it can be implemented without
any expense to the often-cash-strapped operators of public
transportation systems. SubwayPS merely requires an
accelerometer and a gyroscope both of which are standard
on many modern smartphones and a start and an end
destination as input, which can be inferred from user
behavior (as in Google Now) or extracted from routing
Our evaluation showed that SubwayPS works “out of the
box” with four subway systems from two different
continents. However, we also showed that SubwayPS’s
accuracy can be increased if its four parameters are tuned
specifically for an individual subway network. This tuning
is simple, and merely requires a single user to collect
accelerometer data as they travel on the subway.
In the short term, SubwayPS is straightforward enough to
be implemented directly into a mobile app as we have done
with MetroNavigator, our proof-of-concept SubwayPS
application. We have taken care to ensure that any
developer who wishes to implement SubwayPS can do so
by following the instructions (and using the parameters)
laid out in the “SubwayPS Technique section above.
However, if SubwayPS (or a technology like it) were built
into a mobile OS, it would have the greatest impact,
allowing all location-aware mobile apps including mobile
map apps to function while their users traveled in subway
That said, before OS integration can occur, several
limitations of SubwayPS must be addressed. In its current
implementation, accuracy at a trip level is highly dependent
on trip length due to accumulation errors. Fortunately, there
are various solutions to this issue. For instance, in many
subway networks, subway stations have localized Wi-Fi
coverage. This means that traditional Wi-Fi positioning
techniques can be used to correct any errors before they
accumulate. In addition, per-stop accuracy can likely be
improved by using relatively straightforward machine
learning approaches rather than the simple, simple subway
system-wide approaches considered here.
Future work is proceeding along three directions. First, we
are working to implement Wi-Fi-based positioning
correction directly into the SubwayPS technique to correct
for drift errors. This should greatly increase per-trip
accuracy rates. Second, we are working on developing more
sophisticated stop detection approaches using trained
models as well as making small changes to support per-
track parameters. Finally, we are considered the means by
which crowdsourcing may be used to collect training data,
as suggested by P3 in Study 2.
This research was supported in part by a Google Faculty
Research Award, a 3M Non-Tenured Faculty Award, a
Yahoo! ACE Award, and BOF project R-5209. The authors
would like to thank our anonymous reviewers for their
1. Ascher, C., Kessler, C., Wankerl, M., and Trommer,
G.F. sing OrthoSLAM and aiding techniques for precise
pedestrian indoor navigation. (2009), 743749.
2. Biagioni, J., Gerlich, T., Merrifield, T., and Eriksson, J.
EasyTracker: Automatic Transit Tracking, Mapping, and
Arrival Time Prediction Using Smartphones.
Proceedings of the 9th ACM Conference on Embedded
Networked Sensor Systems, ACM (2011), 6881.
3. Böhmer, M., Hecht, B., Schöning, J., Krüger, A., and
Bauer, G. Falling Asleep with Angry Birds, Facebook
and Kindle A Large Scale Study on Mobile
Application Usage. MobileHCI ’11: 13th International
Conference on Human-Computer Interaction with
Mobile Devices and Services, (2011).
4. Fang, L., Antsaklis, P.J., Montestruque, L.A., et al.
Design of a wireless assisted pedestrian dead reckoning
system - the NavMote experience. IEEE Transactions
on Instrumentation and Measurement 54, 6 (2005),
5. Fischer, C., Muthukrishnan, K., Hazas, M., and
Gellersen, H. Ultrasound-aided Pedestrian Dead
Reckoning for Indoor Navigation. Proceedings of the
First ACM International Workshop on Mobile Entity
Localization and Tracking in GPS-less Environments,
ACM (2008), 3136.
6. Google, Inc. Control location settings for Google Now -
Search Help. Android Help.
7. Hemminki, S., Nurmi, P., and Tarkoma, S.
Accelerometer-based Transportation Mode Detection on
Smartphones. Proceedings of the 11th ACM Conference
on Embedded Networked Sensor Systems, ACM (2013),
8. Keally, M., Zhou, G., Xing, G., Wu, J., and Pyles, A.
PBN: Towards Practical Activity Recognition Using
Smartphone-based Body Sensor Networks. Proceedings
of the 9th ACM Conference on Embedded Networked
Sensor Systems, ACM (2011), 246259.
9. Kourogi, M., Sakata, N., Okuma, T., and Kurata, T.
Indoor/Outdoor Pedestrian Navigation with an
Embedded GPS/RFID/Self-contained Sensor System.
Proceedings of the 16th International Conference on
Advances in Artificial Reality and Tele-Existence,
Springer-Verlag (2006), 13101321.
10. Löchtefeld, M., Gehring, S., Schöning, J., and Krüger,
A. PINwI: Pedestrian Indoor Navigation Without
Infrastructure. Proceedings of the 6th Nordic
Conference on Human-Computer Interaction: Extending
Boundaries, ACM (2010), 731734.
11. Muehlbauer, M., Bahle, G., and Lukowicz, P. What Can
an Arm Holster Worn Smart Phone Do for Activity
Recognition? 2011 15th Annual International
Symposium on Wearable Computers (ISWC), (2011),
12. Pritchard, G., Vines, J., Briggs, P., Thomas, L., and
Olivier, P. Digitally Driven: How Location Based
Services Impact the Work Practices of London Bus
Drivers. Proceedings of the SIGCHI Conference on
Human Factors in Computing Systems, ACM (2014),
13. Robertson, P., Angermann, M., and Krach, B.
Simultaneous Localization and Mapping for Pedestrians
Using Only Foot-mounted Inertial Sensors. Proceedings
of the 11th International Conference on Ubiquitous
Computing, ACM (2009), 9396.
14. Ruiz, A.R.J., Granja, F.S., Prieto Honorato, J.C., and
Rosas, J.I.G. Accurate Pedestrian Indoor Navigation by
Tightly Coupling Foot-Mounted IMU and RFID
Measurements. IEEE Transactions on Instrumentation
and Measurement 61, 1 (2012), 178189.
15. Schöning, J., Cheverst, K., Löchtefeld, M., Krüger, A.,
Rohs, M., and Taher, F. PhotoMap: Using
Spontaneously taken Images of Public Maps for
Pedestrian Navigation Tasks on Mobile Devices. Mobile
HCI ’09: 11th International Conference on Human-
Computer Interaction with Mobile Devices and Services,
16. Serra, A., Carboni, D., and Marotto, V. Indoor
Pedestrian Navigation System Using a Modern
Smartphone. Proceedings of the 12th International
Conference on Human Computer Interaction with
Mobile Devices and Services, ACM (2010), 397–398.
17. Susi, M., Renaudin, V., and Lachapelle, G. Motion
Mode Recognition and Step Detection Algorithms for
Mobile Phone Users. Sensors (Basel, Switzerland) 13, 2
(2013), 1539–1562.
18. Thiagarajan, A., Biagioni, J., Gerlich, T., and Eriksson,
J. Cooperative Transit Tracking Using Smart-phones.
Proceedings of the 8th ACM Conference on Embedded
Networked Sensor Systems, ACM (2010), 85–98.
19. Wang, S., Chen, C., and Ma, J. Accelerometer Based
Transportation Mode Recognition on Mobile Phones.
2010 Asia-Pacific Conference on Wearable Computing
Systems (APWCS), (2010), 44–46.
20. Zhou, P., Zheng, Y., and Li, M. How Long to Wait?:
Predicting Bus Arrival Time with Mobile Phone Based
Participatory Sensing. Proceedings of the 10th
International Conference on Mobile Systems,
Applications, and Services, ACM (2012), 379–392.
21. Metro systems by annual passenger rides. Wikipedia, the
free encyclopedia,
22. Sensors Overview. Android Documentation. https://
Note: This version of the paper contains a fix for a
reference issue that appeared in the original version.
... . The proliferation of the Internet and mobile technologies contributes to the application of virtual-mobile application to the public transportation sector. A number of mobile and web applications associated with public transportation available in the market include: InfoBus [2], OneBusAway [1] NavMetro [3], SubwayPS [4], StopFinder [5] and web application for Alynysse Bus [6]. These applications were developed as tools to assist travellers in planning their journey within the most convenient way by providing accurate real-time information and fast mobile ticketing services. ...
... This approach will be more real compare to lab test. Based on the literature review of the usability methods used in public transportation mobile application research, 11 usability methods were identified in this paper with the occurrences of 41 in total -(1) field testing [1,[3][4]12,15,27], (2) laboratory testing [4,24], (3) remote usability test [5,6,22,], (4) observation [6,13,21,24], (5) self-reporting logs [6], (6) [6]. The result of usability methods occurrences in the literature is depicted in Fig. 2 with the top four methods as: survey, field testing, interview, and user testing in the order of occurrence beginning from the highest. ...
... This approach will be more real compare to lab test. Based on the literature review of the usability methods used in public transportation mobile application research, 11 usability methods were identified in this paper with the occurrences of 41 in total -(1) field testing [1,[3][4]12,15,27], (2) laboratory testing [4,24], (3) remote usability test [5,6,22,], (4) observation [6,13,21,24], (5) self-reporting logs [6], (6) [6]. The result of usability methods occurrences in the literature is depicted in Fig. 2 with the top four methods as: survey, field testing, interview, and user testing in the order of occurrence beginning from the highest. ...
Full-text available
The existence of information systems and applications has changed the lifestyle of people around the world. The applications developed to serve specific purposes in the area of public transportation have made peoples' lives much easier. The usability aspects of the applications have enabled users to use computers or mobile devices to find information about the services offered such as in the booking and ticket purchasing system. In this paper, a systematic review on the metrics and methods used to evaluate the usability of the applications related to public transportation systems are presented. The objective of this paper is to identify the metrics and methods employed by researchers and developers in conducting the usability tests and evaluation on applications related to public transportations. In total, 144 research papers were reviewed and out of the number, 22 most related research papers were selected. The results indicate that satisfaction, effectiveness and efficiency are the most frequently employed usability metrics. Meanwhile, survey seems to be the most popular usability method amongst researchers, followed by field testing and interview. The paper provides insights for professionals and researchers.
... When the vehicle is stationary, for example at traffic lights, there are lower vibrations as compared to the moving state, making TMD difficult [13]. Therefore, we detect and remove these stops/pauses based on thresholds which are calculated from the average value of the energy harvesting signals [21]. ...
Conference Paper
Conventional systems for motion context detection rely on batteries to provide the energy required for sampling a motion sensor. Batteries, however, have limited capacity and, once depleted, have to be replaced or recharged. Kinetic Energy Harvesting (KEH) allows to convert ambient motion and vibration into usable electricity and can enable batteryless, maintenance free operation of motion sensors. The signal from a KEH transducer correlates with the underlying motion and may thus directly be used for context detection, saving space, cost and energy by omitting the accelerometer. Previous work uses the open circuit or the capacitor voltage for sensing without using the harvested energy to power a load. In this paper, we propose to use other sensing points in the KEH circuit that offer information-rich sensing signals while the energy from the harvester is used to power a load. We systematically analyze multiple sensing signals available in different KEH architectures and compare their performance in a transport mode detection case study. To this end, we develop four hardware prototypes, conduct an extensive measurement campaign and use the data to train and evaluate different classifiers. We show that sensing the harvesting current signal from a transducer can be energy positive, delivering up to ten times as much power as it consumes for signal acquisition, while offering comparable detection accuracy to the accelerometer signal for most of the considered transport modes.
... Kumareson, Danis and Cabarkapa come up with technologies using Bluetooth in smart phones or Bluetooth beacon to position indoors respectively (Fig. 4) [11]- [13]. In 2014, Stockx et al. in [14] proposed SubwayPS, based on accelerometer and gyroscope equipped in smartphones, to calculate the metro's position in the subway tunnels as well as the remaining time to a specific stop. In 2015, H. Xia et al. in [15] proposed a technology which could detect which floor one is located for indoor positioning by smart phones with built-in barometric sensors (Fig. 5). ...
Conference Paper
Full-text available
With the proliferation and development of mobile devices, various mobile apps have appeared to satisfy people's needs. Metro, with the feature of convenient, punctuality and economic, is one of the most popular modes of transportation in cities. Yet, there are still some inconveniences brought by various factors, impacting tourists' riding experience. The aim of this study is to help tourists to shorten the time of purchasing tickets, to provide them clear metro information and direct navigation, detailed schedule as well as a way to collect metro cards as souvenir. The study collects data through three phases, including observation, survey and test. Data collected from 106 tourists totally in Wuhan metro stations are discussed in the study. The result reflects tourists' demand when they take the metro. It also indicates the feasibility of using mobile technology to improve passenger's experience.
... slope of sidewalks or different surface properties like sand or cobblestone, which are difficult to overcome in a wheelchair). Various approaches analyzed the accessibility of sidewalks with sensors and technical equipment such as depth cameras or acceleration sensors [3,4,5,6,9,12,16,19,20]. Besides these (semi-) automated approaches, Hara et al. [2] explored the use of Amazon Mechanical Turk (AMT) workers for classifying the accessibility of sidewalks for wheelchair users. ...
Conference Paper
Full-text available
Platforms like Google Maps or Bing Maps are used by a large number of users to find the shortest path to their destinations. While these services mainly focus on supporting drivers and pedestrians, first services exist that support wheelchair users. Routing algorithms for wheelchair users try to avoid obstacles like stairs or bollards and optimize on criteria like surface properties and slope of the route. In this study, we undertake the first controlled examination of wheelchair routing approaches. By analyzing three routing platforms, including two wheelchair routing algorithms and three pedestrian routing algorithms, across fifteen major cities in Germany, our results highlight that the routes for wheelchair users are significantly longer and partially also more complex than those for pedestrians. In addition, we show that today's pedestrian routing algorithms also output very diverse routes.
... There are several existing researches about positioning in subway and other types of linear networks. Stockx et al. [2014] resolved the problem of subway train positioning using Inertial Measurement Unit (IMU) integrated in mobile device. They pointed that IMU only based application is inaccurate for train movement detection and other reference networks like Wi-Fi should also be used. ...
Full-text available
Location of passengers on subway lines could be applied in different areas. To overcome the problem of GNSS unavailability in subway I developed the prototype of application for positioning based on alternative methods. The only requirement for using this application is availability of cellular signal in subway, that could be received by passenger's mobile phone. Accuracy test, based on mathematical model of passengers movement shows mean 400 meters accuracy after 50 seconds of user's waiting time. The author point, that final accuracy depends on coverage of cellular network, model of smartphone and also the type of train movement.
Intelligent city transportation systems are one of the core infrastructures of a smart city. The true ingenuity of such an infrastructure lies in providing the commuters with real-time information about citywide transport like public buses, allowing them to pre-plan their travel. However, providing prior information for transportation systems like public buses in real-time is inherently challenging because of the diverse nature of different stay-locations where a public bus stops. Although straightforward factors like stay duration extracted from unimodal sources like GPS at these locations look erratic, a thorough analysis of public bus GPS trails for 1335.365 km at the city of Durgapur, a semi-urban city in India, reveals that several other fine-grained contextual features can characterize these locations accurately. Accordingly, we develop BuStop , a system for extracting and characterizing the stay-locations from multi-modal sensing using commuters’ smartphones. Using this multi-modal information BuStop extracts a set of granular contextual features that allow the system to differentiate among the different stay-location types. A thorough analysis of BuStop using the collected in-house dataset indicates that the system works with high accuracy in identifying different stay-locations like regular bus stops, random ad-hoc stops, stops due to traffic congestion, stops at traffic signals, and stops at sharp turns. Additionally, we also develop a proof-of-concept setup on top of BuStop to analyze the potential of the framework in predicting expected arrival time, a critical piece of information required to pre-plan travel at any given bus stop. Subsequent analysis of the PoC framework, through simulation over the test dataset, shows that characterizing the stay-locations indeed helps make more accurate arrival time predictions with deviations less than 60 seconds from the ground-truth arrival time.
Intelligent city transportation systems are one of the core infrastructures of a smart city. The true ingenuity of such an infrastructure lies in providing the commuters with real-time information about citywide transports like public buses, allowing her to pre-plan the travel. However, providing prior information for transportation systems like public buses in real-time is inherently challenging because of the diverse nature of different stay-locations that a public bus stops. Although straightforward factors stay duration, extracted from unimodal sources like GPS, at these locations look erratic, a thorough analysis of public bus GPS trails for 720km of bus travels at the city of Durgapur, a semi-urban city in India, reveals that several other fine-grained contextual features can characterize these locations accurately. Accordingly, we develop BuStop, a system for extracting and characterizing the stay locations from multi-modal sensing using commuters' smartphones. Using this multi-modal information BuStop extracts a set of granular contextual features that allow the system to differentiate among the different stay-location types. A thorough analysis of BuStop using the collected dataset indicates that the system works with high accuracy in identifying different stay locations like regular bus stops, random ad-hoc stops, stops due to traffic congestion stops at traffic signals, and stops at sharp turns. Additionally, we also develop a proof-of-concept setup on top of BuStop to analyze the potential of the framework in predicting expected arrival time, a critical piece of information required to pre-plan travel, at any given bus stop. Subsequent analysis of the PoC framework, through simulation over the test dataset, shows that characterizing the stay-locations indeed helps make more accurate arrival time predictions with deviations less than 60s from the ground-truth arrival time.
Rail transit systems are important parts of modern cities such as the urban subways and intercity trains. During the travelling time, many passengers access the networks via cellular communications for working or entertainment. However, intermittent connection of LTE signals usually decreases the user experiences, especially due to the poor LTE link quality of underground subways. On the other hand, we observe that there are usually multiple cellular operators covering a city. For example, AT&T and T-Mobile in American cities, China Mobile and China Unicom in Chinese cities. Based on the different eNodeB distributions of different operators, we propose to study a new problem cross-operator switch . Unlike the handover between eNodeBs from one operator, cross-operator switch can switch to an eNodeB belonging to other operators. Although current mobile phones support dual SIM cards and Android supports the quick swap between two cards, it is still challenging to determine which eNodeB to switch because of no information exchange between different operators. To address this challenge, we propose a robust Crowdsensing based Switch (CrowdSwitch) solution dedicated for rail transit systems. CrowdSwitch utilizes mass mobile devices of passengers to sense and collect LTE signals along subway tracks, builds the signal heat map (S-Map) in the cloud servers, and finally recommends the optimal switch taking advantages of the known tracks. We implement CrowdSwitch in off-the-shelf mobile phones and conduct extensive experiments on Shanghai Metro. The results show that CrowdSwitch can depict the accurate LTE distribution and always recommend the optimal operator for users to connect.
Conference Paper
The authors have developed a mobile application for word reordering problems (WRPs) and its learning management system (LMS) based on the WRP web application in order to collect learners' study logs. Various parameters such as "total distance of movements" and "answer time" are generated by learners' trajectories to detect their hesitations in answering WRPs. The main objective of this research is to exploit the practical use of new data types provided by physical sensors embedded in mobile devices, with the belief that specific new data types possibly benefit learning analytics. The authors used acceleration data and force data to detect abnormal trajectories - a series of trajectories caused by unstable environments resulting in low accuracy of detecting hesitations. The authors conducted an experiment to determine the most suitable pair of thresholds (acceleration and force parameter values). By setting this pair, f-measure to identify abnormal trajectories has reached approximately 0.727. This means that we are able to detect abnormal trajectories at a certain rate and cut some of them off to get more accurate results to detect hesitations using mobile devices.
Full-text available
The ways in which geographic information are produced have expanded rapidly over recent decades. These advances have provided new opportunities for geographical information science and spatial analysis—allowing the tools and theories to be expanded to new domain areas and providing the impetus for theory and methodological development. In this light, old problems of inference and analysis are rediscovered and need to be reinterpreted, and new ones are made apparent. This article describes a new typology of geographical analysis problems that relates to uncertainties in the relationship between individual‐level data, represented as point features, and the geographic context(s) that they are associated with. We describe how uncertainty in context linkage (uncertain geographic context problem) is also related to, but distinct from, uncertainty in point‐event locations (uncertain point observation problem) and how these issues can impact spatial analysis. A case study analysis of a geosocial dataset demonstrates how alternative conclusions can result from failure to account for these sources of uncertainty. Sources of point observation uncertainties common in many forms of user‐generated and big spatial data are outlined and methods for dealing with them are reviewed and discussed.
Full-text available
This paper examines how an occupational group has adapted to the demands of working with a Location Based Service (LBS). Instead of following a rigid timetable, London's bus drivers are now required to maintain an equal distance between the bus in front and the one behind. Our qualitative study employs ethnographic fieldwork and in-depth semi-structured interviews to elicit drivers' perspectives of the new system and show how it has modified their driving and general work conditions. We explore how passengers influence the movement of the bus and how the technology frames bus drivers' relationships to their managers and commuters. This work contributes to our understanding of the impact of LBS in the workplace and shows how technological imperatives can be established that cause unanticipated consequences and gradually undermine human relationships.
Full-text available
Microelectromechanical Systems (MEMS) technology is playing a key role in the design of the new generation of smartphones. Thanks to their reduced size, reduced power consumption, MEMS sensors can be embedded in above mobile devices for increasing their functionalities. However, MEMS cannot allow accurate autonomous location without external updates, e.g., from GPS signals, since their signals are degraded by various errors. When these sensors are fixed on the user's foot, the stance phases of the foot can easily be determined and periodic Zero velocity UPdaTes (ZUPTs) are performed to bound the position error. When the sensor is in the hand, the situation becomes much more complex. First of all, the hand motion can be decoupled from the general motion of the user. Second, the characteristics of the inertial signals can differ depending on the carrying modes. Therefore, algorithms for characterizing the gait cycle of a pedestrian using a handheld device have been developed. A classifier able to detect motion modes typical for mobile phone users has been designed and implemented. According to the detected motion mode, adaptive step detection algorithms are applied. Success of the step detection process is found to be higher than 97% in all motion modes.
Conference Paper
Full-text available
In this paper we describe a new Bayesian estimation approach for simultaneous mapping and localization for pedestrians based on odometry with foot mounted inertial sensors. When somebody walks within a constrained area such as a building, then even noisy and drift-prone odometry measurements can give us information about features like turns, doors, and walls, which we can use to build a form of a map of the explored area, especially when these features are revisited over time. Our initial results for our novel scheme which we call “FootSLAM” are very surprising in that true SLAM with stable relative positioning accuracy of 1-2 meters for pedestrians is indeed possible based on inertial sensors alone without any prior known building indoor layout. Furthermore, the 2D maps obtained even for just 10 minutes of walking converge to a good approximation of the true layout forming the basis for future automated collaborative mapping of buildings.
Conference Paper
Full-text available
While applications for mobile devices have become extremely important in the last few years, little public information exists on mobile application usage behavior. We describe a large-scale deployment-based research study that logged detailed application usage information from over 4,100 users of Android-powered mobile devices. We present two types of results from analyzing this data: basic descriptive statistics and contextual descriptive statistics. In the case of the former, we find that the average session with an application lasts less than a minute, even though users spend almost an hour a day using their phones. Our contextual findings include those related to time of day and location. For instance, we show that news applications are most popular in the morning and games are at night, but communication applications dominate through most of the day. We also find that despite the variety of apps available, communication applications are almost always the first used upon a device's waking from sleep. In addition, we discuss the notion of a virtual application sensor, which we used to collect the data.
Conference Paper
Full-text available
In this work we present a pedestrian navigation system for indoor environments based on the dead reckoning positioning method, 2D barcodes, and data from accelerometers and magnetometers. All the sensing and computing technologies of our solution are available in common smart phones. The need to create indoor navigation systems arises from the inaccessibility of the classic navigation systems, such as GPS, in indoor environments.
Conference Paper
Full-text available
Navigation in larger unfamiliar buildings like town halls, airports, shopping malls or other public indoor locations is often difficult for humans. Due to the high amount of infrastructure needed for indoor positioning, just a few navigation services for indoor environments exist. Therefore in many of these buildings 'YOU-ARE-HERE' (YAH) maps are provided, often located at the entrance or other key places, to facilitate orientation and navigation within the building, but they have the disadvantages of being stationary. In this paper, we try to overcome these problems by presenting PINwI (Pedestrian Indoor Navigation without Infrastructure), an application that allows the user of a mobile camera device with integrated compass and accelerometer to utilize a photo of such an indoor YAH-map to navigate through the corresponding building. Using a dead reckoning approach, we enrich stationary analog YAH-maps with basic location functionality and turn them into a digital and dynamic medium that can help decision making while taking turns or estimating distances.
An Integrated Pedestrian Navigation System (IPNS) for precise indoor localization is presented in this paper. A laser scanner enables mapping and long-term stability of the navigation solution. Reliable information about personnel positions and escape routes is a crucial factor in many security or first responder missions. Especially in indoor scenarios in previously unknown environments, precise localization within the explored surroundings is essential for effective coordination. Therefore, a robust and accurate indoor IPNS with mapping capabilities enables fast and successful missions and reduces risk for the deployed personnel. However, the mentioned applications make high demands on navigational hard- and software. Tracking highly dynamic motions is required as well as providing a long-term accurate navigation estimation, even without reliable satellite navigation signals. The pedestrian navigation system has to be portable, light and small, which leads to limited computational power being available. Deployment in unknown environments requires use of exteroceptive sensors and mapping techniques. Our proposal is an Integrated Pedestrian Navigation System comprising two inertial measurements units (IMUs), mounted at the foot and the torso, a magnetometer and a barometer, aided by a two-dimensional laser scanner. The navigation solution is obtained by a Strapdown approach, with Zero Velocity Updates used as aiding technique. The laser scanner enables mapping of the surrounding structure and guarantees long-term accuracy of the position and attitude estimation. The scan data is processed using an Orthogonal Simultaneous Localization and Mapping (OrthoSLAM) technique. In this approach, a reduction of required computational power is reached by considering only straight and perpendicular structures. OrthoSLAM exploits the fact that man-made buildings consist of orthogonal structures to a large extent and provides long-term stable navigation aiding at low computational cost. Therefore, this method is targeted to the special conditions of indoor navigation. First results indicate that the combined Dual-IMU/magnetometer/laser system can meet the challenging requirements of pedestrian indoor navigation. The capabilities of our approach will be demonstrated using post-processed data obtained in real-world test scenarios.
Conference Paper
We present novel accelerometer-based techniques for accurate and fine-grained detection of transportation modes on smartphones. The primary contributions of our work are an improved algorithm for estimating the gravity component of accelerometer measurements, a novel set of accelerometer features that are able to capture key characteristics of vehicular movement patterns, and a hierarchical decomposition of the detection task. We evaluate our approach using over 150 hours of transportation data, which has been collected from 4 different countries and 16 individuals. Results of the evaluation demonstrate that our approach is able to improve transportation mode detection by over 20% compared to current accelerometer-based systems, while at the same time improving generalization and robustness of the detection. The main performance improvements are obtained for motorised transportation modalities, which currently represent the main challenge for smartphone-based transportation mode detection.
The bus arrival time is primary information to most city transport travelers. Excessively long waiting time at bus stops often discourages the travelers and makes them reluctant to take buses. In this paper, we present a bus arrival time prediction system based on bus passengers' participatory sensing. With commodity mobile phones, the bus passengers' surrounding environmental context is effectively collected and utilized to estimate the bus traveling routes and predict bus arrival time at various bus stops. The proposed system solely relies on the collaborative effort of the participating users and is independent from the bus operating companies, so it can be easily adopted to support universal bus service systems without requesting support from particular bus operating companies. Instead of referring to GPS enabled location information, we resolve to more generally available and energy efficient sensing resources, including cell tower signals, movement statuses, audio recordings, etc., which bring less burden to the participatory party and encourage their participation. We develop a prototype system with different types of Android based mobile phones and comprehensively experiment over a 7 week period. The evaluation results suggest that the proposed system achieves outstanding prediction accuracy compared with those bus company initiated and GPS supported solutions. At the same time, the proposed solution is more generally available and energy friendly.
Conference Paper
In many mid- to large-sized cities public maps are ubiquitous. One can also find a great number of maps in parks or near hiking trails. Public maps help to facilitate orientation and provide special infor- mation to not only tourists but also to locals who just want to look up an unfamiliar place while on the go. These maps offer many advantages compared to mobile maps from services like Google Maps Mobile or Nokia Maps. They often show local landmarks and sights that are not shown on standard digital maps. Often these 'YOU ARE HERE' (YAH) maps are adapted to a special use case, e.g. a zoo map or a hiking map of a certain area. Being designed for a fashioned purpose these maps are often aesthetically well de- signed and their usage is therefore more pleasant. In this paper we present a novel technique and application called PHOTOMAP that uses images of 'YOU ARE HERE' maps taken with a GPS-enhanced mobile camera phone as background maps for on-the-fly navigation tasks. We discuss different implementations of the main challenge, namely helping the user to properly georeference the taken image with sufficient accuracy to support pedestrian navigation tasks. We present a study that discusses the suitability of various public maps for this task and we evaluate if these georeferenced photos can be used for navigation on GPS-enabled devices.