Available via license: CC BY 4.0
Content may be subject to copyright.
Sensafety: Crowdsourcing the Urban Sense of Safety
Sandro Rodriguez Garzon, Bersant Deva
Service-centric Networking, Technische Universit¨
at Berlin, Telekom Innovation Laboratories, Berlin, Germany,
[sandro.rodriguezgarzon, bersant.deva]@tu-berlin.de
Abstract: Today, community initiatives to improve the urban quality of life can be conducted in a more focused way be-
cause local authorities and urban planners are able to reveal urban hotspots through the investigation of location-annotated
crime and accident data. However, urban areas, which according to well-recorded incident data are characterized by a
high level of public safety, but which are generally perceived by citizens as unsafe, remain undiscovered and therefore un-
treated. This work presents Sensafety, a citizen-centric crowdsourcing approach that enables users by means of a mobile
application to report their personal feeling of safety anytime and at any site. Sensafety’s goal is to reveal a comprehensive
and complete picture of the perceived safety in urban environments in order to identify blind spots that have not been fur-
ther investigated due to lack of data. To encourage citizens to participate and contribute, Sensafety’s mobile application
offers different ways to explore and experience the collected data depending on the user’s location. This paper gives a
detailed description of Sensafety’s integrated concept and outlines the major technical and non-technical findings.
Keywords: crowdsourcing, participatory sensing, public safety, location-based services, geofencing
1. Introduction
Studies on the quality of life (QoL) in urban spaces have
lately become increasingly popular, especially to compare
the QoL of different cities (Mercer, 2019). Most stud-
ies are based upon purpose-driven QoL indices that take
different combinations of subjective and objective indica-
tors into consideration (McCrea et al., 2006). Objective
indicators usually encompass but are not limited to crime
statistics, public facilities, public transportation, and state
of pollution. Subjective indicators, on the other hand, try
to capture the satisfaction and dissatisfaction with the ur-
ban living situation through citizen surveys. The fear of
crime, as a major subjective indicator and partially deter-
minable by capturing the perceived safety in public spaces,
has been proven to have a non-negligible impact on the ur-
ban QoL (Michalos and Zumbo, 2000)(Møller, 2005). An-
other important aspect of the perceived safety is related to
the fear to become a victim of an accident. This subjec-
tive indicator has so far caught more attention in municipal
crowdsourcing campaigns1then in QoL research.
To get a holistic picture of the perceived public safety, con-
sisting of, among others, crime- and traffic-related aspects,
within a public space, at least a representative sample of
the population needs to be continuously interviewed. Even
though this is already practically difficult to achieve, it is
aggravated by the fact that perceived safety not only de-
pends on demographics and social factors but also on the
location (Kitchen and Williams, 2010) and the situation on
site (Doran and Burgess, 2012). In participatory sensing,
the required information is not actively requested through
cost and time-intensive telephone and street surveys, but is
provided voluntarily by the citizens in a proactive way via
their personal mobile devices (Burke et al., 2006). Each
contribution can thereby be automatically enriched with
data originating from smartphone sensors, such as the ge-
1The Traffic Agent, http://www.trafikkagenten.no
ographical location at which the contribution was submit-
ted. Participatory sensing has been successfully applied in
related research to capture the personal feeling of crime-
related safety on site (Christin et al., 2013) and the dif-
ferentiated location-dependent emotions (Hamilton et al.,
2011) via mobile applications or the perceived level of dan-
ger with respect to traffic via a web (Jiˇ
r´
ı, 2018) or mobile
application (Aubry et al., 2014). The location-annotated
data obtained through participatory sensing does not only
help to determine the urban QoL in a more differentiated
and location-specific manner, but it allows in particular to
uncover spots in a city that are perceived to be unsafe with
respect to crime or traffic. This knowledge can be used to
carry out urban development measures in a more focused
way or to increase the presence of law enforcement agen-
cies where and when required.
However, without proper incentives (Restuccia et al., 2016),
participatory sensing will not lead to the amount of data re-
quired to make meaningful statements about the perceived
site-specific safety. In order to encourage volunteers to
stay on board and continuously share their personal feeling
of safety, the mobile application has to offer free-of-charge
value-added services, based on already collected data. This
work introduces Sensafety, a citizen-centric mobile appli-
cation that gives citizens the opportunity to proactively re-
port their personal feeling of safety on site. Sensafety is
designed to minimize the required effort for users to share
their rating at the expense of the rating’s information con-
tent. It provides, in addition, multiple value-added ser-
vices for users to explore and experience the commonly
perceived safety in different ways such as via an interac-
tive map or augmented reality view.
The paper starts with a discussion of similar approaches to
capture the perceived safety in Section 2. Section 3 intro-
duces Sensafety’s integrated concept and describes partic-
ular design decisions while Section 4 presents major as-
pects of the prototypical implementation. The paper fin-
Advances in Cartography and GIScience of the International Cartographic Association, 2, 2019.
15th International Conference on Location Based Services, 11–13 November 2019, Vienna, Austria. This contribution underwent
double-blind peer review based on the full paper | https://doi.org/10.5194/ica-adv-2-12-2019 | © Authors 2019. CC BY 4.0 License
2 of 8
ishes with concluding remarks and possible use cases in
Section 5 and a description of a potential extension and
remaining investigations in Section 6.
2. Related Work
The urban quality of life with respect to the perception of
public safety can be improved, beside many others, by the
installation of additional street lights (Painter, 1996) or by
an increased presence of law enforcement agencies (Zhao
et al., 2002). Where and when to take what kind of action
is today decided upon by officials, in most cases, based on
data that is already collected for legal reasons, such as for
crime or traffic incidents. However, sites that are neither
insecure with respect to the local crime rate nor danger-
ous with respect to the traffic but that are perceived to be
unsafe for whatever reason are not considered. In citizen
science (Mueller et al., 2012), all citizens of an urban area
are encouraged to participate in science projects. Partici-
patory sensing is a concrete variant of it in which humans
are considered to act as sensors for geographic information
(See et al., 2016) (Berntzen et al., 2018). Due to the om-
nipresence of mobile devices in everyday life, it is thereby
possible to collect location-annotated data on a large scale,
area-wide and inexpensively. It is applied in a variety of
urban research projects to map urban concerns in general
(Ruiz-Correa et al., 2017) or to investigate specific aspects
of an urban environment such as accessibility (Prandi et
al., 2014), transport network’s state (Chen et al., 2013) or
infrastructure (Berntzen et al., 2018), or social night life
patterns (Santani et al., 2016), to mention just a few.
Crowdsourcing urban human emotions in general or the
perceived safety in particular has caught less attention
compared to the collection of geo-annotated incidents.
Transafe, a mobile platform for crowdsourcing crime and
safety perceptions for the city of Melbourne, is a mobile
application to report crimes (incidents) but also four dif-
ferent emotions while being in public space (Hamilton et
al., 2011). Optionally, comments, pictures or files can be
attached to the reports. After aggregation, a map is shown
with the aggregated mood at the granularity level of city
blocks. The EmoMap mobile application follows a similar
path and allows the user to report location-specific emo-
tions via Likert scales and to attach the social context to it
(Klettner et al., 2013) (Huang and Gartner, 2016). Instead
of providing aggregation results, ratings are published and
displayed on a map. The mobile application uSafe allows
the user to report the safety feeling via a slider that is an-
notated with ”I am feeling safe”, ”I am worried” and ”I am
feeling unsafe” (Christin et al., 2013). In a subsequent step,
the user has to choose from eight different crime-related
context types such as pickpockets. Reports are aggregated
based on a predefined grid and finally being visualized with
different granularity levels on a map. uSafe also allows
to create retrospective reports for sites the user has visited
during the day. The People as Sensors app captures urban
emotions as a ground truth for further processing by letting
volunteers express their emotions via a mobile application
(Resch et al., 2015). For a rating, users select the emotion,
e.g. afraid or pleased, the aspect, e.g. safety or traffic, and
then submit it after verifying the correctness of the entries.
While Transafe and uSafe put their focus on crime-related
perceived safety, the People as Sensors app is more flex-
ible and lets users select from different safety aspects.
EmoMap puts the rating in its social context since the rat-
ing level can depend upon whether the safety was per-
ceived while being alone or in company. But EmoMap falls
short when it comes to privacy because single ratings are
published and become visible to all users. Despite the flex-
ibility of EmoMap and People as Sensors app to address
different aspects of perceived safety, they lack required in-
centives, besides a map for single ratings in EmoMap or a
city emotions map immediately after a rating submission
within the People as Sensors app, to attract users beyond
intrinsically motived volunteers. Transafe and uSafe, on
the other hand, offer an always-accessible map view for
aggregated results, plus a list view in Transafe. However,
it’s not clear how Transafe aggregates the emotions on a
city block level. uSafe provides, in addition, an alert ser-
vice based on the ratings, but - similarly - it remains un-
clear how it determines areas that are rated to be dangerous
and how it monitors the user’s location with respect to the
marked areas in a reliable and energy efficient way.
3. Concept
This paper proposes a new integrated concept to leverage
the power of crowdsourcing to uncover hotspots in an ur-
ban environment that are commonly perceived to be par-
ticularly safe or unsafe. The Sensafety concept comprises
the citizen-centric mobile application Sensafety (App) and
a Sensafety backend (Server). Equipped with the App, cit-
izens are able to proactively and pseudonymously submit
their personal site-specific feeling of safety to the Server.
The ratings of all users are then centrally processed and
the aggregated results being provided back to the App.
Location-based services (LBS) within the App are used to
present and experience the results in different forms by tak-
ing the position of the user into consideration.
3.1 Rating
Since the perceived public safety is made up of multiple
aspects such as the fear of crime, fear to get involved into
a traffic accident or fear to get lost, to mention just a few,
it is required for a participatory sensing application to ei-
ther give the user the opportunity to express a differentiated
opinion or to focus only on a single aspect of the public
safety. However, the possibility to capture different aspects
of the perceived public safety with the help of a single mo-
bile application inevitably leads to a higher cognitive effort
for the user, an increased amount of interaction steps and,
as a consequence, a longer capturing process. If each user
is supposed to capture the aspects of the perceived public
safety only once, this is a reasonable approach. If a user
should be encouraged to voluntarily share the perceived
safety at multiple sites and at different times of the day in
order to get a holistic picture of perceived safety across a
larger urban environment, this approach is likely to attract
lesser long-term motivated users. Capturing only a single
aspect of the perceived safety simplifies the user interac-
tion and minimizes the required steps for the user to share
the perceived safety. But this goes along with the drawback
that the site-dependent perceived safety is not captured in
its full complexity. Sensafety’s concept follows a differ-
ent path than the aforementioned approaches and provides
users via the App a means to report their general feeling of
safety for a site, without specifying the considered aspects
(required in uSafe and People as Sensors app) nor its in-
tensities (required in EmoMap, uSafe and Transafe). This
Advances in Cartography and GIScience of the International Cartographic Association, 2, 2019.
15th International Conference on Location Based Services, 11–13 November 2019, Vienna, Austria. This contribution underwent
double-blind peer review based on the full paper | https://doi.org/10.5194/ica-adv-2-12-2019 | © Authors 2019. CC BY 4.0 License
3 of 8
is intended to simplify and speed up the rating procedure
in order to increase the chance that users stay engaged. To
keep it even simpler and to ease the on-boarding process,
there is no need for users to register (required in EmoMap,
Transafe and uSafe) nor the possibility to enter any addi-
tional user information, such as age or gender. Although
the data collected by Sensafety does not contain any demo-
graphics nor contextual indicators for the commonly per-
ceived safety at a site, it reveals out of a macroscopic per-
spective which sites should be investigated more closely.
Sensafety does therefore not intend to replace quantitative
and qualitative street interviews but to pinpoint sites that
require further investigation.
The perceived safety at a location is subject to constant
change. Even slight changes of the environment, e.g. re-
placement of a broken street lamp, might have a not neg-
ligible impact on the perceived safety. Hence, Sensafety
allows participants to assess and rate the perceived safety
more than once for a given site. With the possibility to rate
multiple times, there comes also the danger that users make
several deliberately incorrect assessments at a site in order
to either encourage officials to take actions although the
site is commonly perceived to be safe or to upvalue places
that are commonly perceived to be unsafe. Even though the
concept of Sensafety is not capable to prevent this fraudu-
lent practice without losing the inherent reevaluation fea-
ture, it makes this practice at least more difficult by allow-
ing users to report the perceived safety only once in a given
period. Since the perceived safety may significantly vary
between day and night, users are allowed to rate the per-
ceived safety at a site twice a day, once during daylight and
once at night. Sites are defined as same-sized, predefined,
rectangular and non-overlapping geographic areas that are
not bound to any administrative entities such as counties or
districts (further details on Sensafety’s concept of a site are
given in Subsection 4.2). If a user already rated for a site,
then the user has to either wait for the sunset or sunrise to
rate again at the same site or to move and rate at another
site. In order to prohibit remote ratings without a proper
situational assessment, users are only allowed to rate on
site. This design decision prevents users from making any
decisions based on hearsay or speculation. It also prevents
potentially biased ratings based on human memory, as it is
possible in uSafe through retrospective assessments. Once
a rating has been submitted, it cannot be withdrawn. This
restriction is intended to prevent results from being tem-
porarily falsified by targeted rating and subsequent with-
drawal of ratings. The implications on the compliance with
the European General Data Protection Regulation (GDPR)
are further discussed in Section 4.5.
3.2 Metrics
Defining a site as a geographic area rather than a geograph-
ical point makes it possible to determine a site-specific
metric for the perceived safety, denoted as the Sensafety
Index. As a single key value, the Sensafety Index is meant
to represent the commonly perceived safety at a site, across
the ratings of all participants. The ratings, as opposed to all
the aforementioned approaches, are distinguished by the
local time of submission, leading to three site-specific out-
comes of the Sensafety Index: one for the day, one for the
night and another overall assessment. A rating is declared
to be a day rating if the local time of submission lies be-
tween sunrise and sunset or otherwise, a night rating. De-
pending on these time ranges, variants of the subsequent
Sensafety Index In+1 for each site are determined through
exponential smoothing given by
In+1 =α·rnew + (1 −α)·In(1)
where rnew is a numerical representation of a new rating
for a site, Inthe current Sensafety Index and αthe smooth-
ing factor. As opposed to a simple average calculation in
uSafe, more recent ratings are considered in Sensafety with
a higher weight than older ratings to better reflect the cur-
rent situation. Since the Sensafety Index is a product of
aggregation, it can finally be shared with all users with-
out revealing a single user rating, as opposed to EmoMap
where single ratings are published. This property of Sen-
safety is crucial when it comes to privacy because even
though users rate anonymously without providing any de-
mographics, in some cases, the exact position at which a
rating was made can be traced back to a certain group of
people who, for example, have exclusive access to the site.
3.3 Incentives
Besides the goal to provide a simple and fast way to re-
port the perceived safety, the App needs to offer additional
incentives in order to foster engaged citizens, with an in-
trinsic motivation to improve their neighborhood, to rate
on a regular basis and to encourage less active citizens to
start contributing. For that purpose, the App offers build-
in value-added LBS’s that let users explore and experience
the aggregated results from a global or local point of view.
The global view is provided by visualizing the different
Sensafety Indices for all sites on an interactive map, sim-
ilar to the approach taken by uSafe. Users are then able
to examine the Sensafety Index for all sites for which at
least one rating was submitted. A local view onto the re-
sults is provided by an augmented reality (AR) view and a
safety compass. Both views take the current position of the
user into consideration to visualize Sensafety Index infor-
mation from within the vicinity of the user. In the AR view,
the camera view is annotated with Sensafety Index infor-
mation while the safety compass indicates how the safety
is commonly perceived in all cardinal directions, with the
reference position given by the user’s location. While the
global and local views are intended to be used for explo-
ration purposes only, an alert service is provided in addi-
tion to experience the hyper-local Sensafety Indices while
on the go. It proactively alerts the user when she/he is
about to enter a hot spot where the average level of per-
ceived safety falls below a user-defined threshold.
4. Implementation
The App and Server are the key components of the Sen-
safety concept. The App contains all user-facing elements
and is implemented for both leading mobile operating sys-
tems Android and iOS in order to ensure the availability for
all citizens with access to these mobile ecosystems. Its de-
sign and features has been developed in close collaboration
with potential users. Sensafety’s server-sided implementa-
tion enables the aggregated computation of the Sensafety
Index and its provision to the App as a common knowl-
edge base. This section provides a more detailed view on
the implementation of Sensafety’s core aspects, including
the realization of the rating view, the computation of the
Advances in Cartography and GIScience of the International Cartographic Association, 2, 2019.
15th International Conference on Location Based Services, 11–13 November 2019, Vienna, Austria. This contribution underwent
double-blind peer review based on the full paper | https://doi.org/10.5194/ica-adv-2-12-2019 | © Authors 2019. CC BY 4.0 License
4 of 8
Figure 1. Sensafety’s rating view with the dichotomous
question on perceived safety
Sensafety Index and its aggregation, as well as the presen-
tation of featured incentives, such as the compass and AR
view and the adaptive alerts. It concludes with a discussion
on measures taken to protect the user’s privacy and to ease
the App’s use within the sensitive topic of perceived safety.
4.1 Rating View
The App uses a distinct view for the rating of the perceived
safety. It is quickly accessible throughout the App via a
centrally positioned button. The rating view (shown in
Figure 1) contains the single dichotomous question ”Do
you feel safe at your location right now?”, which has been
carefully constructed to encapsulate context-sensitive as-
pects. Its wording contains the spatial (at your location)
and temporal (right now) context information in which
the question is expected to be answered. In addition to
the time, which is visible on the status bar of mobile de-
vices, the human-readable address of the current location
is shown to make the user aware to which site the rating is
accounted for before it is submitted. Since the question
is quite complex, the answer option is kept binary with
colored ”YES” and ”NO” buttons. A binary is favored
over a gradual selection (uSafe, EmoMap and Transafe use
gradual selection), because it reduces the mental effort to
answer the question on the go. The user should immedi-
ately give a rating based on gut feeling without worrying
about how intense the feeling is and where to locate it on a
scale. This type of rating is intended to encourage users to
quickly express their safety feeling in different situations,
places and times during the day. Potential situations should
explicitly include circumstances in which quick respond-
ing is essential, e.g. while passing through crime-affected
neighborhoods. Sensafety requires explicit assessments by
the user, so that the question is only answered if the user is
willing and has no ambiguities in his or her safety feelings
for the site. If the user is in doubt, the rating process can
be canceled easily at any time. As emphasized before, the
rating view for a specific site can only be accessed and an-
swered when the user is located on that site and only twice
per day during daylight and night. Once the question has
been answered by a user, the rating is sent to the Server and
contains a hashed pseudonomous identifier, a local times-
tamp, a daylight flag, the geographical location, the site
identifier, and the answer.
4.2 Sensafety Index Computation
The Sensafety Index for a site is computed on the Server
based on ratings as presented in equation 1. Sensafety
L
L+1
L+2
Figure 2. Grid-based aggregation of the Sensafety Index
for different levels of precision at different zoom levels
uses the spatial, hierarchical and rectangular geohash
(Niemeyer, 2008) data structure for the representation of
sites as grid cells. Figure 3a shows the interactive map
with the calculated index at different sites alongside UI ele-
ments to show the number of collected ratings for the view
area and a switch to change between day, night and over-
all Sensafety Indices. Users are able to immediately grasp
the perceived safety of their city through the colorization
of sites from green to red. By pressing on a rated site, it re-
veals a colored gauge with the corresponding Sensafety In-
dex and the number of ratings. As mentioned above, Sen-
safety favors early and more recent ratings. The smoothing
factor αis dependent on the number of ratings nfor a par-
ticular site as shown in Table 1. This ensures that early
ratings at a site have a higher impact on the overall result
of the index, encouraging users to express their perception
of safety at an unrated site. With a significant number of
ratings (≥100), the smoothing factor stabilizes at a value
of 0.2 and thus continues to emphasize the most recent rat-
ing with 20% of the overall index.
n0<5<10 <25 <40 <60 <80 <100 ≥100
α1 0.7 0.5 0.4 0.35 0.3 0.25 0.22 0.2
Table 1. Exponential smoothing αparameters depending
on the number of ratings nper site
Geohashes use the Z-order curve and a variant of base 32
encoding for its computation, thus, each geohash (repre-
senting a grid cell) consists of 32 sub-cells with a higher
precision at every lower hierarchical level. The Sensafety
Index is initially calculated for the precision of geohash
length 7. A site is therefore defined in Sensafety to be a
grid cell with a dimension of around 152.9 m x 152.4 m
at the equator and is uniquely identified by a geohash of
length 7. This extent of a site has been carefully selected
because of two reasons. Due to the anonymous nature of
Sensafety and the sensitivity of information, grid cell di-
mensions should be large enough so that it is unlikely that
ratings can be attributed to individual users, however, at the
same time small enough so that grid cells depict the actual
size of an urban site properly. The interactive map adapts
the visualization of the rated sites based on the zoom level.
When zoomed out, the map shows an aggregated level of
the Sensafety Index using higher level grid cells with a
geohash of shorter length (see Figure 2). The aggregation
of higher level grid cells is computed by the weighted sum
IL+1 =
32
X
i=1
IL[i]·wL[i](2)
where IL+1 is the aggregated Sensafety Index for a higher
level grid cell, IL[i]is the Sensafety Index of the i-th sub-
Advances in Cartography and GIScience of the International Cartographic Association, 2, 2019.
15th International Conference on Location Based Services, 11–13 November 2019, Vienna, Austria. This contribution underwent
double-blind peer review based on the full paper | https://doi.org/10.5194/ica-adv-2-12-2019 | © Authors 2019. CC BY 4.0 License
5 of 8
(a) Map View (b) Compass View (c) AR View (d) Alert Service View
Figure 3. Sensafety App Views
cell, and wL[i]is the weight of the i-th sub-cell calculated
by the number of ratings in the sub-cell divided by the total
number of ratings on the higher level L+ 1. It should be
noted that unrated sites at the lowest level 7 are not consid-
ered for the aggregation as the weight for these cells will
be 0.
4.3 Vicinity Views
Sensafety uses hyper-local views to let the users investigate
the perceived safety information within their vicinity. This
is accomplished by using modern means of context-aware
visualizations and interactions within a compass (see Fig-
ure 3b) view and an AR (see Figure 3c) view.
Compass View: The safety compass view provides an
immediate overview on the perceived safety by aligning
the Sensafety Index with the cardinal directions. It depicts
the previously used colored gauge containing the Sensafety
Index and the number of ratings for the site the user is cur-
rently located in at the center of the screen. Indicators
of the northern and current cardinal direction the device
is facing towards are placed to provide users with direct
orientation. The varying sized areas at the verge of the
compass show aggregated ratings at eight directions of sur-
rounding sites within a user-defined visibility range. This
verged area computation is based on the compass sensor of
the mobile devices and on angle-, distance- and number-
of-ratings-based weighted averages. For this, the distance
and angle between the user’s location and the centers of
encompassing sites within the visibility range are calcu-
lated. Closer distanced sites with higher number of rat-
ings and angles pointing closer to the eight directions are
weighted linearly higher for the aggregated rating. Sizes of
the verged areas are determined by the number of ratings,
thus, the larger the areas are, the more ratings were given
in the respective direction providing users with a measure
of confidence in the rating. For example, a large red area in
a certain direction would indicate a high number of ratings
with a low perceived safety in that direction.
Augmented Reality View: The AR view allows users to
experience the location-specific insights using the built-in
camera. It dynamically overlays Sensafety’s colored gauge
(displaying Sensafety Index and amount or ratings) on the
real world camera view based on the position, the tilt and
direction of the mobile device. By looking down through
the camera, insights for the currently visited site are dis-
played while looking at the horizon reveals the insights of
directly neighboring sites. In order to retrieve a rotation
matrix Rof the horizontal and vertical orientation of the
device the gyroscope sensor is used. The AR view uses
systematic matrix transformations through several coordi-
nate systems to be able to overlay the markers of the sites
on the mobile device’s screen. At first, a world space ma-
trix W(i), depicting the relative coordinate space between
the i-th sites on the lowest level 7 and the user is computed,
by converting the closest sites to Mercator coordinates and
subtracting each by the user’s geographic location. Ad-
ditionally, a screen space matrix Sand perspective space
matrix Pare introduced to adapt to the screen size and for-
mat as well as the perspective of projected entities on the
screen. The resulting 4D matrix transformation is com-
puted as follows:
AR(i) = S×P×R×W(i)
To extract the point coordinates for the AR view, a 2D pro-
jection is used to position the gauge elements on the user’s
screen. The AR view also contains a tilted map in the lower
third of the screen using the rotation matrix to pan along-
side the user’s movement. This screen introduces an en-
hanced experience and gives the user orientation and in-
formation from two different perspectives.
4.4 Adaptive Alerts
While the App maintains a holistic set of views for explor-
ing the Sensafety Index in the vicinity of users, chances are
that areas perceived as unsafe might be missed when the
Advances in Cartography and GIScience of the International Cartographic Association, 2, 2019.
15th International Conference on Location Based Services, 11–13 November 2019, Vienna, Austria. This contribution underwent
double-blind peer review based on the full paper | https://doi.org/10.5194/ica-adv-2-12-2019 | © Authors 2019. CC BY 4.0 License
6 of 8
getfiltered&
sortednearby
geohashedsites
getsite
exploreneighbors
ofsite
Yes
computegeofence
fromcluster
splitcluster
No
No
selectclosest
geofencesas
alertareas
addsitetocluster
Is neighbor among
filtered sites?
Coverage
ratio
sufficient?
Sites left
exploring?
clusters left
computing?
calculatecoverage
ratiobetween
geofence&cluster
Yes
Yes
No
YesNo
Figure 4. Activity diagram depicting the alert area compu-
tation process from sites to geofences.
App is not actively used. For this reason, adaptive alerts are
introduced that enable users to get proactively notified as
they approach areas that are perceived to be unsafe. Figure
3d shows the In-App interaction screen in the notifications
settings. Users can enable or disable the alerts, as well as
choose a duration for its operation. A sensitivity threshold
can be set by the familiar colored gauge that is being used
throughout the various means of visualization. Once the
alert function has been activated, the computed alert areas
are shown on a map view below. The continuous moni-
toring of these alert areas is carried out via the geofencing
APIs, which are specific to the respective mobile operat-
ing system and vendor. Geofencing (K ¨
upper et al., 2011)
describes the process by which events are proactively trig-
gered when a geographically defined area (geofence) is
entered or left. To provide reliable and battery efficient
geofencing, vendors’ APIs allow only circular shapes and
a limited number of simultaneously monitored geofences.
These limitations need to be taken into consideration for
the computation process of the alert areas, which is de-
picted in Figure 4.
In an initial step, a set of sites that surpassed the user-
defined threshold in the vicinity of the user’s location are
queried. This filtered set of nearby sites is sorted by the
current distance between each site and the user’s location,
thus limiting the computation to relevant close-by sites
only. Based on this set, cohesive clusters of areas that are
perceived to be unsafe are computed. For this purpose, the
site set is iteratively explored and expanded towards their
neighbors to form common clusters. If neighbors of a site
are among the nearby filtered site set, they are integrated
into a common cluster and further expanded upon. Once
all sites have been explored and the initial set of clusters
has been formed, geofences for each cluster are computed.
Each geofence consists of a point and a radius. The point
is calculated as the centroid of all sites within a cluster,
whereas the radius is taken as the maximum distance be-
tween all elements of a cluster and its centroid. This ap-
proach makes sure that all sites perceived to be unsafe are
Figure 5. Computed alert areas surrounding sites perceived
to be unsafe in the area of Berlin-Charlottenburg
covered by a geofence, however, it also leads to large sized
alert areas when cluster elements are mostly at the edge
of the cluster, for example, in long consecutive chains. In
order to reduce the size of these large geofences, the cov-
erage ratio between the surfaces of the clustered sites and
the computed geofence is calculated. In case the cover-
age ratio is not sufficient the cluster is split into multiple
ones. The selection process for the split is determined by
finding sites within the cluster that have the most neighbor-
ing sites. These sites are then selected together with their
neighbors as the new split cluster. After all clusters have
been computed to geofences and the sufficient coverage ra-
tio has been reached, the geofences closest to the user are
selected as alert areas to be monitored. Figure 5 depicts
the resulting alert areas in circular shape surrounding the
clusters of sites perceived to be unsafe.
The computation of alert areas takes place not only when
the alert service is activated through the user interface, but
also when certain contextual conditions are met in order
to keep the state of the indices and the alert areas around
the user up-to-date at all times. As illustrated above, the
alert area computation process is strongly based on Sen-
safety’s underlying dynamic data model and only consid-
ers the closest areas to the user. Therefore, to stay relevant
for users with the most recent alert areas, the service re-
initiates the computation process with current parameters,
and thus adapts the alert areas to three different kind of
contextual changes. First, Sensafety triggers recalculations
when users move to other locations that are beyond the
range of the currently computed set of alert areas. Second,
the Sensafety Index updates itself periodically for nearby
sites when alerts are activated, potentially leading to the
recalculation of alert areas by updating the Sensafety In-
dices. Finally, alerts adapt to day- and night-time ratings
and follow sunrise and sunset intervals. The computations
and the continuous background tracking used by the mo-
bile operating systems to provide the geofencing functions
are exclusively executed on the users’ mobile devices. The
alerts appear as notifications on the respective mobile op-
erating systems when entering the alert area and disappear
once the area has been left.
Advances in Cartography and GIScience of the International Cartographic Association, 2, 2019.
15th International Conference on Location Based Services, 11–13 November 2019, Vienna, Austria. This contribution underwent
double-blind peer review based on the full paper | https://doi.org/10.5194/ica-adv-2-12-2019 | © Authors 2019. CC BY 4.0 License
7 of 8
4.5 Privacy
Once the App is installed, the user can start to rate im-
mediately, without the need to register with personal cre-
dentials. However, the significant simplified onboarding
process makes it difficult for Sensafety to prevent mali-
cious users bypassing the restriction to rate at most twice
per day for a single site. A user might, for example,
try to uninstall and re-install the mobile application to
reset its state. Assuming ratings are anonymous, Sen-
safety would then not be able to detect after a reinstall
whether a user has already rated a particular site before. So
instead of introducing non-anonymous ratings by means
of a mandatory user registration, Sensafety makes use of
unique pseudonomous identifiers for mobile devices that
remain the same even after resetting or reinstalling the mo-
bile application. Since each submitted rating contains a
hashed version of the pseudonomous identifier, Sensafety
is able to detect and discard maliciously submitted ratings.
Even though the hashed pseudonomous identifier within
a rating does not reveal the identity of the user nor the
device type, it can be used to track a device across mul-
tiple ratings in order to infer the user’s personal identity
(Krumm, 2007). Although the user is not able to withdraw
a rating as outlined above, she/he is offered the opportu-
nity to request a full anonymization of all past user ratings
given the pseudonomous identifier. This complies with a
required data deletion policy as defined by the European
GDPR since the personal information stored by Sensafety,
the hashed version of the pseudonomous identifier, can be
removed on demand. Malicious use is therefore still possi-
ble, but the process to bypass the restrictions requires more
time and effort. The Sensafety Index computation and its
aggregation complies with the GDPR as no personal data
in form of the hashed pseudonomous identifier is attached
to these site-specific indices.
Despite this privacy-preserving property of the Sensafety
Index, it still remains, as indicated above, the possibility to
trace back all ratings within a rectangular geohash-based
area of about 152.9m x 152.4m (at equator) in case that
area lies entirely within a greater area where only a limited
set of people have access to. While this scenario is not a
big issue for dense urban areas, it is fairly probable in ru-
ral areas. It must be emphasized that Sensafety is aiming
to determine the perceived safety in public spaces and not
within private properties. Sensafety has no build-in sup-
port to detect whether a rating has been submitted for a
public space or a private property. All submitted ratings
are treated the same way. Therefore, it is the responsibility
of the user to not submit ratings for private spaces. These
ratings would be, anyhow, of no interest to public authori-
ties or urban planners.
5. Conclusion
This work presented an integrated approach to capture per-
sonal site-specific perceived safeties within urban environ-
ments through the citizen’s mobile devices. Since citizens
are supposed to proactively report their personal perceived
safety voluntarily and more than once by means of a mo-
bile application, a simple rating interaction concept with
a single dichotomous question was favored over capturing
all potential aspects of the perceived safety, their intensities
and weighting via a more complex and time-consuming
user interface. Although the captured data does not allow
conclusions to be drawn about the reasons behind a par-
ticular rating, it does help to narrow down the search for
unknown hotspots within a city that are, with respect to the
perceived safety, worth to be investigated in more detail
by traditional means. The visualization of site- and time-
specific Sensafety Indices by means of an interactive map,
augmented reality view and a safety compass, and a corre-
sponding alert service should not only make the common
urban sense of safety accessible to users but it should also
encourage them to contribute on a regular basis.
However, Sensafety’s main task remains to quantify the
general urban feeling of safety, without the need to rely
on word-of-mouth recommendations. But it remains to be
seen, whether Sensafety’s integrated concept and its con-
crete implementation is able mobilize a critical mass of
contributers so as to provide a sufficient and sound data
basis for further investigations. The results might be used
in many ways to improve, facilitate or understand life in
urban areas. Depending on the point of view, residents can
take Sensafety as a means to participate in urban planning
processes, visitors can inquire Sensafety about the situa-
tion on site, law enforcement agencies can identify hidden
hotspots, politicians can address local specificities in their
campaigns, realtors can consider the Sensafety Indices dur-
ing the estimation of real estate prices, insurances can offer
more differentiated rates and scientists can make use of the
data to investigate the factors that influence the perceived
urban safety. This is only a small selection of the wide
range of potential applications and by no means a complete
list.
6. Future Work
From the conceptional point of view, the proposed ap-
proach can be extended with the possibility for Sensafety’s
mobile application to automatically initiate an interview
with the user based on the user’s location, rating history
and commonly perceived safety at a site. This application-
initiated interview can then be used to learn more about the
user’s attitude, without the need to extend the simplified
user-initiated rating process. Such an unsolicited inquiry
by Sensafety should, however, be carefully selected and
only take place when the user enters a site under investiga-
tion for which the personal ratings show a clear tendency
that match with the commonly perceived safety at a site.
For example, instead of proactively alerting the user when
entering an alert area, Sensafety could automatically ask
the user to name the aspect, e.g. crime or accident, that the
user mainly addresses when she/he reports a low perceived
safety for that site.
In general, Sensafety’s concept is based upon the funda-
mental assumption that volunteers contribute in a truthful
manner. However, if the personal rating shared with Sen-
safety indirectly leads to a deterioration of the local quality
of life, then volunteers may be encouraged to stop rating
in a truthful manner. For example, visitors equipped with
Sensafety may avoid to visit a neighborhood that is per-
ceived by its residents to be unsafe. This could even con-
tribute to a worsening of the situation on site which may,
in turn, trigger residents to report untruthful ratings. Sen-
safety concept should therefore be further examined as to
whether and how such behavior can be detected, whether
Advances in Cartography and GIScience of the International Cartographic Association, 2, 2019.
15th International Conference on Location Based Services, 11–13 November 2019, Vienna, Austria. This contribution underwent
double-blind peer review based on the full paper | https://doi.org/10.5194/ica-adv-2-12-2019 | © Authors 2019. CC BY 4.0 License
8 of 8
incorrect assessments have a significant influence on the
overall perceived safety at a site and how Sensafety can po-
tentially deal with incorrect assessments. It must, in addi-
tion, be investigated, whether citizens are willing to rate in
alleged dangerous situations and how visitor ratings might
bias the overall results for a neighborhood.
Acknowledgments
We would like to thank Fabian Puch, Michael Raring,
Jonathan Zimmermann, Nino Filiu, Victor Morel, Mar-
cel Reppenhagen, Michał Zwolak, Mikołaj Robakowski,
Yared D. Dessalk, Martin Kachev and Denis Rangelov for
their valuable contribution to the Sensafety project.
References
Aubry, E., Silverston, T., Lahmadi, A. and Festor, O.,
2014. CrowdOut: a Mobile Crowdsourcing Service for
Road Safety in Digital Cities. In: 2014 IEEE Int. Conf.
on Pervasive Computing and Communication Workshops
(PERCOM WORKSHOPS), pp. 86–91.
Berntzen, L., Johannesen, M. R., B¨
ohm, S., Weber, C. and
Morales, R., 2018. Citizens as Sensors - Human Sensors
as a Smart City Data Source. In: Proc. of the 7th Int. Conf.
on Smart Cities, Systems, Devices and Technologies.
Burke, J., Estrin, D., Hansen, M., Parker, A., Ramanathan,
N., Reddy, S. and Srivastava, M. B., 2006. Participatory
Sensing. In: First Workshop on World-Sensor-Web: Mo-
bile Device Centric Sensory Networks and Applications
(WSW’06).
Chen, K., Lu, M., Tan, G. and Wu, J., 2013. CRSM:
Crowdsourcing based Road Surface Monitoring. In: 2013
IEEE 10th Int. Conf. on High Performance Computing and
Communications & 2013 IEEE Int. Conf. on Embedded
and Ubiquitous Computing, pp. 2151–2158.
Christin, D., Roßkopf, C. and Hollick, M., 2013. ”uS-
afe: A Privacy-aware and Participative Mobile Application
for Citizen Safety in Urban Environments”. Pervasive and
Mobile Computing 9(5), pp. 695 – 707.
Doran, B. J. and Burgess, M. B., 2012. Putting Fear of
Crime on the Map: Investigating Perceptions of Crime Us-
ing Geographic Information Systems. Springer Series on
Evidence-Based Crime Policy, 1 edn, Springer-Verlag New
York.
Hamilton, M., Salim, F., Cheng, E. and Choy, S. L., 2011.
Transafe: a Crowdsourced Mobile Platform for Crime and
Safety Perception Management. In: 2011 IEEE Int. Sym-
posium on Technology and Society (ISTAS), pp. 1–6.
Huang, H. and Gartner, G., 2016. Using mobile crowd-
sourcing and geotagged social media data to study people’s
affective responses to environments. European Handbook
of Crowdsourced Geographic Information pp. 385–399.
Jiˇ
r´
ı, P., 2018. Participatory mapping in community partic-
ipation – case study of Jesen´
ık, Czech Republic. Quaes-
tiones Geographicae 37(3), pp. 151–162.
Kitchen, P. and Williams, A., 2010. Quality of Life and
Perceptions of Crime in Saskatoon, Canada. Social Indi-
cators Research 95(1), pp. 33–61.
Klettner, S., Huang, H., Schmidt, M. and Gartner, G.,
2013. Crowdsourcing affective responses to space. Kar-
tographische Nachrichten 63, pp. 66–73.
Krumm, J., 2007. Inference Attacks on Location Tracks.
In: A. LaMarca, M. Langheinrich and K. N. Truong (eds),
Proc. of the 5th Int. Conf. on Pervasive Computing, Perva-
sive 2007, Springer Berlin Heidelberg, pp. 127–143.
K¨
upper, A., Bareth, U. and Freese, B., 2011. Geofencing
and Background Tracking - The Next Features in LBSs.
In: H.-U. Heiß, P. Pepper, H. Schlingloff and J. Schneider
(eds), INFORMATIK 2011, Lecture Notes of Informatics,
Vol. P192, Gesellschaft f ¨
ur Informatik e.V., Bonn.
McCrea, R., Shyy, T.-K. and Stimson, R., 2006. What is
the Strength of the Link Between Objective and Subjective
Indicators of Urban Quality of Life? Applied Research in
Quality of Life 1(1), pp. 79–96.
Mercer, 2019. Quality of Living Ranking. https:
//mobilityexchange.mercer.com/Insights/
quality-of-living-rankings. [Accessed 2019-09-
16].
Michalos, A. C. and Zumbo, B. D., 2000. Criminal Victim-
ization and the Quality of Life. Social Indicators Research
50(3), pp. 245–295.
Møller, V., 2005. Resilient or Resigned? Criminal Victim-
isation and Quality of Life in South Africa. Social Indica-
tors Research 72(3), pp. 263–317.
Mueller, M. P., Tippins, D. and Bryan, L., 2012. The Fu-
ture of Citizen Science. Democracy and Education.
Niemeyer, G., 2008. Geohash. http://geohash.org, Ac-
cessed: 2019-09-16.
Painter, K., 1996. The influence of street lighting im-
provements on crime, fear and pedestrian street use, after
dark. Landscape and Urban Planning 35(2), pp. 193 –
201. Streets Ahead.
Prandi, C., Salomoni, P. and Mirri, S., 2014. mPASS: Inte-
grating People Sensing and Crowdsourcing to Map Urban
Accessibility. In: 2014 IEEE 11th Consumer Communica-
tions and Networking Conf. (CCNC), pp. 591–595.
Resch, B., Summa, A., Sagl, G., Zeile, P. and Exner, J.-P.,
2015. Urban Emotions—Geo-Semantic Emotion Extrac-
tion from Technical Sensors, Human Sensors and Crowd-
sourced Data. Springer International Publishing, Cham,
pp. 199–212.
Restuccia, F., Das, S. K. and Payton, J., 2016. Incentive
Mechanisms for Participatory Sensing: Survey and Re-
search Challenges. ACM Transactions on Sensor Networks
12(2), pp. 13:1–13:40.
Ruiz-Correa, S., Santani, D., Ram´
ırez-Salazar, B.,
Ruiz-Correa, I., Rend´
on-Huerta, F. A., Olmos-Carrillo,
C., Sandoval-Mexicano, B. C., Arcos-Garc´
ıa, A. H.,
Hasimoto-Beltr´
an, R. and Gatica-Perez, D., 2017.
SenseCityVity: Mobile Crowdsourcing, Urban Awareness,
and Collective Action in Mexico. IEEE Pervasive Com-
puting 16(2), pp. 44–53.
Santani, D., Biel, J.-I., Labhart, F., Truong, J., Landolt,
S., Kuntsche, E. and Gatica-Perez, D., 2016. The Night
is Young: Urban Crowdsourcing of Nightlife Patterns. In:
Proc. of the 2016 ACM Int. Joint Conf. on Pervasive and
Ubiquitous Computing, UbiComp ’16, ACM, New York,
NY, USA, pp. 427–438.
See, L., Mooney, P., M. Foody, G., Bastin, L., Comber,
A., Estima, J., Fritz, S., Kerle, N., Jiang, B., Laakso,
M., Liu, H.-Y., Milˇ
cinski, G., Nikˇ
siˇ
c, M., Painho, M.,
P˝
od¨
or, A., Olteanu Raimond, A.-M. and Rutzinger, M.,
2016. Crowdsourcing, Citizen Science or Volunteered
Geographic Information? The Current State of Crowd-
sourced Geographic Information. ISPRS Int. Journal of
Geo-Information.
Zhao, J. S., Schneider, M. and Thurman, Q., 2002. The
effect of police presence on public fear reduction and satis-
faction: A review of the literature. The Justice Professional
15(3), pp. 273–299.
Advances in Cartography and GIScience of the International Cartographic Association, 2, 2019.
15th International Conference on Location Based Services, 11–13 November 2019, Vienna, Austria. This contribution underwent
double-blind peer review based on the full paper | https://doi.org/10.5194/ica-adv-2-12-2019 | © Authors 2019. CC BY 4.0 License
