MODELLING MICROSCOPIC PEDESTRIAN MOBILITY USING

BLUETOOTH

Thomas Liebig

and Armel Ulrich Kemloh Wagoum

Fraunhofer IAIS, Schloss Birlinghoven, 53754 Sankt Augustin, Germany

Forschungszentrum J

ulich GmbH, JSC, 52425 J

ulich, Germany

Keywords:

Bluetooth tracking, Event monitoring, Pedestrian dynamics, Route choice.

Abstract:

Emergence of Bluetooth tracking technology for event monitoring is currently applied to extract individual

pathways, movement patterns or to rank popularity of locations by their visitor quantities. The next steps

are to achieve short term movement predictions, to understand people’s motivations and to come up with

microscopic trafﬁc values. This work proposes a solution for these questions, namely, the combination of

recorded values with a microsimulation. In our presented framework, simulated pedestrians move from one

decision area to the next one in a navigation graph. The graph is automatically generated from the facility

based on the inter-visibility of the exits. Intermediate areas are inserted if needed. With the data obtained

from the Bluetooth scanners, individual pathways of pedestrians are determined. The routing algorithm will

then use those information to adjust the pathways of the agents in the simulation. An accurate reproduction of

pedestrian route choice in a complex facility is expected.

1 INTRODUCTION

Major public events as concerts and soccer matches

which attract thousands or billions of visitors are on

one hand a great chance for street marketers and ad-

vertisement companies but on the other hand also a

growing ﬁnancial risk for the organizers and a safety

hazard for the guests themselves, due to huge ex-

penses and high visitor densities. Understanding the

movement behaviour, identiﬁcation of attractors and

distractors, determination of waiting times, as well as

localization of congestions and bottle-necks gives in-

sights on visitor preferences and motivations during a

particular event. Knowing such detailed information

on visitor behaviour helps not just at the next sim-

ilar event, but is also a location-based performance

indicator for the event itself. Various locations and

attractions can be ranked by their popularity, safety

or frequency. Currently used technologies to mea-

sure these highly needed microscopic movement val-

ues are surveys and video surveillances. Whereas the

ﬁrst solution (surveys) is expensive and hardly repre-

sentative due to the non-random sampling among all

visitors the second one (video surveillance) depends

on weather, brightness and density of the people and

does not seldom require special scaffoldings to carry

the cameras.

Within this work we propose a novel four stage

approach to monitor microscopic pedestrian move-

ment during events. Our system is cheap, fast de-

ployable and it is independent from the pre-existing

technical infrastructure. The sensor technology we

use, can be applied seamlessly to monitor indoor and

outdoor movement, which offers the chance to track

peoples movement during their whole stay (e.g. start-

ing at the arrival by car, during the event and leav-

ing again by car after the event). We utilize recently

evolved, Bluetooth-scanners (Bruno and Delmastro,

2003; Fuller, 2009) to record people’s presence (step

1). This is basically a mesh of radio frequency sensors

of certain diameters (depending on the used Bluetooth

hardware 10 m, 20m or 100m). Whenever a person

with a Bluetooth enabled device (e.g. a mobile phone

or an intercom) passes the footprint of a sensor, an en-

try is attached to a data-log storing the time-stamp, the

position and a unique id for this person. Each sensor

itself generates pedestrian counts. By use of multiple

sensors, movement patterns and transition times are

recorded. Expected representativeness is about 7 per-

cent of the people (Leitinger et al., 2010). The tech-

nology is already widely used for performance moni-

toring (Hagemann and Weinzerl, 2008; Stange et al.,

2011) which just depends on macroscopic movement

values. To extend this as well to microscopic features,

270

Liebig T. and Kemloh Wagoum A..

MODELLING MICROSCOPIC PEDESTRIAN MOBILITY USING BLUETOOTH.

DOI: 10.5220/0003833802700275

In Proceedings of the 4th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2012), pages 270-275

ISBN: 978-989-8425-96-6

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

we adjust and apply (step 2) an agent simulation to the

recorded values. The analysis returns highly granular

values on pedestrian speeds, duration of stay, direc-

tion of movement and pedestrian counts. We apply

the proposed methods during a soccer match at Stade

des Costi

eres in N

ımes, the achieved results will be

used later on by the local ﬁreﬁghters and forces.

The remainder of the paper is structured as fol-

lows: Section 2 gives a brief overview on event mon-

itoring and Bluetooth tracking systems. In section 3

our use case and the empirical data collection is de-

scribed. Based on this our novel microscopic mod-

elling method is introduced and explained by means

of the soccer match use case in section 4. The paper

ends with a summary and outlook in section 5.

2 RELATED WORK

Existing indoor tracking technologies are surveys and

video surveillances. Whereas the ﬁrst solution (sur-

veys) is expensive and hardly representative due to the

non-random sampling among all visitors, the second

one (video surveillance) depends on weather, bright-

ness and density of the people and does not seldom

require special scaffoldings to carry the cameras. The

need for further robust passive localization technolo-

gies pushed development of sensors that are capa-

ble to monitor people’s movement. First choice is

to track most popular digital gadgets: mobile phones

and intercoms. Analysis of mobile network GSM

(Global System for Mobile Communications) log ﬁles

(Giannotti and Pedreschi, 2008) causes strong pri-

vacy objections. Besides, Bluetooth technology is an

emerging technology for monitoring tasks. Recently

evolved Bluetooth based mobility sensors have been

used for event monitoring at the Sziget festival in Bu-

dapest (Leitinger et al., 2010). There, a mesh of six

sensors has been placed at carefully selected places,

having an intermediate distance from 50m to 200m.

This work extracts the route choices and number of

people at particular locations but does not consider

any extraction of microscopic values. Besides event

monitoring, also other successful indoor applications

of Bluetooth scanners are described in literature. In

(Pels et al., 2005) various scanners were placed at

Dutch train stations to record transit travellers. In

(Hagemann and Weinzerl, 2008) not only the tran-

sit travellers are monitored but, by placing sensors

in public busses, also the performance of the public

transport network itself. Accurate locating and fol-

lowing of objects within complex facilities is as well

an important research topic (Hallberg et al., 2003).

So far Bluetooth tracking is used to monitor a

sample of visitors and extract their route choices. In

few works time-geography and movement patterns

are addressed (Stange et al., 2011). The next data

mining step is to combine recorded values with a

microsimulation to (1) achieve short term movement

predictions, (2) to understand people’s motivations

and (3) to come up with microscopic trafﬁc values.

This challenge has not been addressed in mobility

mining and Bluetooth tracking literature, yet. The

next sections describe our novel approach to achieve

these goals.

3 PEDESTRIAN MONITORING

USE CASE

Arbitrary events require a ﬂexible, robust and easy de-

ployable monitoring technology as they are varying in

space and time. Thus, we decide for recently evolved

Bluetooth-scanners (Fuller, 2009). The particular real

word scenario we focus within this work is a soccer

match at Stade des Costi

eres in N

ımes, France. The

stadium has 18.364 seats and expects many visitors

since the local football team happened to play in the

second league. We deploy a mesh of 17 sensors at the

stadium in order to monitor people’s behaviour and

route choices. Afterwards, see section 4, we are going

to build a pedestrian model using the collected data,

which represents accurately route choices and pedes-

trian behaviour.

For microscopic event monitoring we propose a

four stage process (ﬁgure 1). In step 1, Survey De-

sign, basic parameters and decisions (e.g. number of

sensors, sensor placement, sensor form factor and an-

tennas) are made according to the particular applica-

tion needs. Step 2 is the Data Collection phase dur-

ing the event. Afterwards, step 3, Data Preparation

is required. In this step recorded values become tem-

porally aggregated to compensate asynchronous data

entries. During the ﬁnal step 4, Microscopic Pedes-

trian Modelling the pedestrian simulation, which is

capable of route choice representation (see section 4),

becomes adjusted by the recorded values.

Figure 1: Workﬂow for microscopic event monitoring.

3.1 Sensorplacement and Data

Collection

The sensors we use contain multiple Bluetooth anten-

nas which search simultaneously for visible Bluetooth

devices within the sensor footprint. Thus, a complete

MODELLING MICROSCOPIC PEDESTRIAN MOBILITY USING BLUETOOTH

271

scan of the frequency band is accelerated and mov-

ing people are more likely to become detected while

they are crossing the footprint. Each time a Blue-

tooth device (e.g. smartphone or intercom) is seen

a data entry is stored in a ﬁle. This log-entry consists

of timestamp, sensor identiﬁer, unique scrambled de-

vice identiﬁer and the signal strength. The need to

scramble the device identiﬁer results from the fact that

Bluetooth sensors collect privacy sensitive data. Ev-

ery Bluetooth chip is identiﬁable by its unique Media-

Access-Control-address (MAC). Hence, a Bluetooth

device (respectively a person) is detectable (and there-

fore trackable) beyond the spatial-temporal bound-

aries of an event. Hence, our Bluetooth-scanners save

just an anonymized identiﬁer, valid for the time of

the monitored event. To scramble the MAC-address,

we embed the irreversible SHA-256 encryption al-

gorithm (National Institute of Standards and Tech-

nology, 2002) with an event speciﬁc random seed

into the sensor software. Thus, privacy of the mon-

itored pedestrians is preserved from the very ﬁrst data

recording.

We use multiple of these Bluetooth-sensors,

which allow re-identiﬁcation of the persons at various

locations. This allows recording of transition times,

stay times, movement patterns and movement prefer-

ences. The sensor locations were chosen carefully.

We placed one sensor at each entrance in order to

record visitors at their arrival and leaving of the sta-

dium. Additionally, sensors were place at the shops in

the uppermost ﬂoor. In intermediate ﬂoor levels, we

placed sensors as well at junctions. Finally, sensors

which monitor people’s presence during the match

were place in the corridors of the tribunes. In total we

placed 17 sensors but before the start of the match,

two sensors were removed by vandalism. The three-

dimensional sensor placement strategy is depicted in

ﬁgure 2.

3.2 Data Preparation

Every single sensor runs asynchronous and the max-

imum time for each Bluetooth scan may easily ex-

ceed the theoretical upper bound of 10.24s (Woodings

et al., 2001; Bruno and Delmastro, 2003). One rea-

son is the noisy environment during the event. Thus,

temporal ﬁltering and spatial aggregation is necessary

to get a pure dataset. Duplicate entries are removed

from the dataset as well as devices which were only

detected once (spot readings). The sensors also record

by chance devices of non-interest (e.g. navigation

systems of passing cars or pedestrians at the border of

the event area). These artifacts are removed by ven-

dor ﬁltering as well as spatial-temporal ﬁltering. Fi-

Figure 2: Three-dimensional sensorplacement at Stade des

Costi

eres in N

ımes, France.

nally, arbitrary jumps among sensor locations, result-

ing from overlapping sensors are also removed. For

this action, the spatial distances between sensor foot-

prints, time-stamp and duration of the stay are taken

into account to calculate speed and position changes

per time. This step was recently introduced in (Stange

et al., 2011).

After the data has been puriﬁed the dataset con-

tains sequences of positions visited per device en-

riched with the time-stamp and stay-time duration.

Thus, for every sensor location quantity of detected

visitors can be determined within a particular time-

interval. Additionally, movement patterns and dy-

namic information on popularity of footprint transi-

tions remained preserved in the data. In total we

recorded 553 distinct persons. Press

states that 3.898

spectators have been watching the match. This re-

sults in a coverage of about 14 percent which is by far

higher than the expectation given by (Leitinger et al.,

2010). These empirical recordings are utilized in the

subsequent micro simulation step (section 4) for ac-

curate route choice modelling.

4 MICROSCOPIC PEDESTRIAN

MODELLING

The framework used for describing pedestrian trafﬁc

can be divided in a three-tier structure. One distin-

guishes between the strategic, the tactical and the op-

erational level (Hoogendoorn et al., 2002). The start

and the end trip for each pedestrian is usually known

in advance. At the strategic level pedestrians choose

http://www.foot-national.com/match-foot-nimes-

vannes-32912.html, last accessed 25.11.2011

ICAART 2012 - International Conference on Agents and Artificial Intelligence

272

their self estimated best route, among a collection of

different alternatives. This can be done based on ex-

perience. Examples could be the global shortest path

or the familiar path to a given destination. Short-terms

decisions are taken at the tactical level, avoiding jams

or switching to a faster route for instance. Basic rules

for motions are deﬁned at the tactical level, these in-

clude accelerating, decelerating, stopping. The differ-

ent tiers have local interactions with each others.

There are mainly three different classes of models

for pedestrian dynamic at the operational level: cellu-

lar automata models (Blue and Adler, 2001; Kirch-

ner and Schadschneider, 2002), rule based models

(Thompson, 1994; Galea et al., 2004; Korhonen et al.,

2008; Raney and Nagel, 2006) and force based mod-

els (Helbing and Moln

ar, 1995; Yu et al., 2005). Cel-

lular automata have the advantage of being computa-

tionally efﬁcient, but the resolution of the simulated

geometry is limited by the size of the cells. Force

based models usually operate on a continuous geom-

etry. They need more computations. For more about

the advantages and disadvantages of the individual

models we refer to (Schadschneider et al., 2009).

4.1 Force based Model

In our simulation the operational level of the pedes-

trian walking is described by the Generalized Cen-

trifugal Force Model (GCFM) (Chraibi et al., 2011)

which operates in continuous space. In the GCFM

at the operational level pedestrians are described with

ellipses with velocity dependent semi-axes. Faster el-

lipses (pedestrians) need more space in the moving di-

rection. The motion is ruled by the social forces (Hel-

bing and Moln

ar, 1995; Moln

ar, 1995). At each simu-

lation step the forces between the pedestrians and the

obstacles (e.g. walls) are computed. Given a pedes-

trian i with coordinates

−→

, the equation of motion is:

−→

−−→

drv

∑

j∈N

−−→

rep

i j

∑

w∈W

−−→

rep

, (1)

where

−−→

rep

i j

denotes the repulsive force from pedes-

trian j acting on pedestrian i,

−−→

rep

is the repulsive

force emerging from the obstacle w and

−−→

drv

is a driv-

ing force. m

is the mass of pedestrian i. N

is the set

of all pedestrians that inﬂuences pedestrian i and W

the set of walls or borders that acts on pedestrian i.

They are within a certain cut-off radius r

= 2m. This

model has been validated in corridors and bottlenecks

using the fundamental diagram. This model has al-

ready been used to perform simulations of a multipur-

pose arena (Holl and Seyfried, 2009; Seyfried et al.,

2010).

4.2 Pedestrians Route Choice

The route choice for pedestrians can be done using

navigation ﬁeld (Hartmann, 2010; Guo and Huang,

2011). This approach is spread in the cellular au-

tomata area. Continuous models usually work on a

visibility graph, where the driven force of the simu-

lated agent points towards a node of the graph. The

strategies used are usually the shortest path combined

with the quickest path (Kretz, 2009; Kirik et al., 2009;

Helivaara et al., 2011; Kemloh Wagoum and Seyfried,

2011). These strategies are in most of the case vali-

dated using a visual assessment on some screenshots

taken from the simulation. Some experiments have

been conducted to determine pedestrians route choice

using video surveillance, but only on simple scenar-

ios, reducing the problem to an exit selection prob-

lem (Guo and Huang, 2011; Helivaara et al., 2011;

Lo et al., 2006). This is partially due to the fact that

in complex facilities pedestrians have to be tracked

across many rooms.

Figure 3: Example of a navigation graph generated from a

section of a stadium considering which exits are closed.

In the framework used here, pedestrians move

from one decision area to the next one. A decision

area is a place where the pedestrian decides which

way to go or change the current destination. Ideally

the decision areas are around the exits, which might

be relevant for an evacuation scenario. The navigation

network is automatically generated from the facility

based on the inter-visibility of the exits, intermediates

areas are inserted if needed. Visibility graphs can be

constructed using different algorithms (de Berg et al.,

2008; H

ocker et al., 2010). In the case of an evacua-

tion scenario, the navigation graph can be limited to a

visibility graph. A sample navigation graph for a sec-

tion of a stadium is presented in Fig. 3. Pedestrians

MODELLING MICROSCOPIC PEDESTRIAN MOBILITY USING BLUETOOTH

273

are routed to the outside in this graph using four algo-

rithms: the local shortest path, the global shortest path

and a combination with the the quickest path (Kem-

loh Wagoum and Seyfried, 2011). This example is

suitable for an evacuation scenario where the pedestri-

ans might prefer the shortest or quickest path to reach

the outside. This approach is insufﬁcient for nor-

mal day life situations, where the individual trips of

pedestrians are subjected to other motivations. Some

pedestrians might choose to go out the shortest way,

whereas others might feel more comfortable walking

along the promenade to get to some other points.

With the data obtained from the Bluetooth-

tracking system, it is possible to assign individual

destination and to calibrate the complete process of

the route choice. The routing algorithm will then use

the provided information to adjust the pathway of the

agents in the simulation. Accurate reproduction of

pedestrian route choice in complex facilities is ex-

pected.

5 CONCLUSIONS AND FUTURE

WORK

In this contribution Bluetooth tracking was applied at

events to monitor a sample of visitors and extract their

route choices. Individual pathways could be tracked

across many rooms. By sensor-design, the whole pro-

cess is at any time privacy preserving.

As part of the spatial knowledge discovery pro-

cess related works, analysing Bluetooth based mobil-

ity data, focussed on time-geography and movement

pattern analysis. Driven by major goals of mobility

data analysis (to achieve short term movement predic-

tions, to understand people’s motivations and to come

up with microscopic trafﬁc values), we proposed a

solution for the next step, namely, the combination

of the recorded values with a microsimulation. The

routing algorithm in the microsimulation will then use

the recorded information to adjust the pathways of

the simulated pedestrians. An accurate reproduction

of pedestrian route choice patterns in complex facili-

ties is expected. As Bluetooth-scanners work seam-

lessly indoors and outdoors, further research needs

to focus on outdoor models and the integration with

existing mobility models. We conclude that use of

Bluetooth-scanners for event monitoring is not just

feasible for pattern extraction but by utilizing our

novel approach also for understanding microscopic

movement behaviour and to expose people’s motiva-

tion.

ACKNOWLEDGEMENTS

This work has been performed within the program

Research for Civil Security in the ﬁeld ”Protecting

and Saving Human Life” funded by the German Gov-

ernment, Federal Ministry of Education and Research

(BMBF). The project is supported under the grant no.

13N9952 and 13N9960. Additionally, the work was

partly supported by the European Emergency Sup-

port System project grant no. 217951. We thank our

project partner CEREN and the local stadium admin-

istration for their support during the data collection.

REFERENCES

Blue, V. J. and Adler, J. L. (2001). Cellular automata

microsimulation for modeling bidirectional pedestrian

walkways. Transportation Research Part B, 35:293–

312.

Bruno, R. and Delmastro, F. (2003). Design and analysis

of a bluetooth-based indoor localization system. In

In Personal Wireless Communications, IFIP-TC6 8th

International Conference, PWC 2003, pages 711–725.

Chraibi, M., Kemloh, U., Seyfried, A., and Schadschneider,

A. (2011). Force-based models of pedestrian dynam-

ics. Networks and Heterogeneous Media, 6(3):425–

442.

de Berg, M., Cheong, O., van Kreveld, M., and Overmars,

M. (2008). Computational Geometry: Algorithms and

Applications. Springer, 3rd edition.

Fuller, R. (2009). Mobile Entity Localization and Track-

ing in GPS-less Enviroments. In: Tutorial on Location

Determination by RF Means. Springer-Verlag Berlin

Heidelberg, Berlin Heidelberg.

Galea, E. R., Gwynne, S., Lawrence, P., Filippidis, L.,

Blackspields, D., and Cooney, D. (2004). buildingEX-

ODUS V 4.0 - User Guide and Technical Manual.

www.fseg.gre.ac.uk.

Giannotti, F. and Pedreschi, D., editors (2008). Mobility,

Data Mining and Privacy - Geographic Knowledge

Discovery. Springer.

Guo, R.-Y. and Huang, H.-J. (2011). Route choice in pedes-

trian evacuation: formulated using a potential ﬁeld.

Journal of Statistical Mechanics: Theory and Experi-

ment, 2011(04):P04018.

Hagemann, W. and Weinzerl, J. (2008). Automatische Er-

fassung von Umsteigern per Bluetooth-Technologie.

In: Nahverkerspraxis. Springer-Verlag Berlin Heidel-

berg, Berlin Heidelberg.

Hallberg, J., Nilsson, M., and Synnes, K. (2003). Position-

ing with bluetooth. In Telecommunications, 2003. ICT

2003. 10th International Conference on, volume 2,

pages 954–958.

Hartmann, D. (2010). Adaptive pedestrian dynamics based

on geodesics. New Journal of Physics, 12:043032.

Helbing, D. and Moln

ar, P. (1995). Social force model for

pedestrian dynamics. Phys. Rev. E, 51:4282–4286.

ICAART 2012 - International Conference on Agents and Artificial Intelligence

274

Helivaara, S., Kuusinen, J.-M., Rinne, T., Korhonen, T., and

Ehtamo, H. (2011). Pedestrian behavior and exit se-

lection in evacuation of a corridor an experimental

study. Safety Science.

ocker, M., Berkhahn, V., Kneidl, A., Borrmann, A., and

Klein, W. (2010). Graph-based approaches for sim-

ulating pedestrian dynamics in building models. In

8th European Conference on Product & Process Mod-

elling (ECPPM), University College Cork, Cork, Ire-

land. http://zuse.ucc.ie/ECPPM/.

Holl, S. and Seyfried, A. (2009). Hermes - an Evacuation

Assistant for Mass Events. inSiDe, 7(1):60–61.

Hoogendoorn, S. P., Bovy, P., and Daamen, W. (2002). Mi-

croscopic Pedestrian Wayﬁnding and Dynamics Mod-

elling. In Schreckenberg, M. and Sharma, S., ed-

itors, Pedestrian and Evacuation Dynamics, pages

123–155.

Kemloh Wagoum, A. U. and Seyfried, A. (2011). Modelling

dynamic route choice of pedestrians to assess the crit-

icality of building evacuation. arXiv:1103.4080.

Kirchner, A. and Schadschneider, A. (2002). Simulation of

evacuation processes using a bionics-inspired cellular

automaton model for pedestrian dynamics. Physica A,

312:260–276.

Kirik, E. S., Yurgel’yan, T. B., and Krouglov, D. V. (2009).

The Shortest Time and/or the Shortest Path Strategies

in a CA FF Pedestrian Dynamics Model. Journal of

Siberian Federal University. Mathematics & Physics,

2(3):271–278.

Korhonen, T., Hostikka, S., Heli

ovaara, S., and Ehtamo, H.

(2008). FDS+Evac: An Agent Based Fire Evacuation

Model. Proceedings of the 4th International Confer-

ence on Pedestrian and Evacuation Dynamics.

Kretz, T. (2009). Pedestrian trafﬁc: on the quickest path.

Journal of Statistical Mechanics: Theory and Experi-

ment, P03012.

Leitinger, S., Gr

ochenig, S., Pavelka, S., and Wim-

mer, M. (2010). Erfassung von Personenstrmen mit

der Bluetooth-Tracking-Technologie. In: Angewandte

Geoinformatik 2010. Addison Wesley Longman Inc.,

New York, 15 edition.

Lo, S., Huang, H., Wang, P., and Yuen, K. (2006). A game

theory based exit selection model for evacuation. Fire

Safety Journal, 41(5):364–369.

Moln

ar, P. (1995). Modellierung und Simulation der Dy-

namik von Fußg

angerstr

omen. Dissertation, Univer-

sit

at Stuttgart.

National Institute of Standards and Technology (2002). Se-

cure Hash Standard. National Institute of Standards

and Technology, Washington. Federal Information

Processing Standard 180-2.

Pels, M., Barhorst, J., Michels, M., Hobo, R., and Barendse,

J. (2005). Tracking people using Bluetooth. Implica-

tions of enabling Bluetooth discoverable mode. Tech-

nical report, University of Amsterdam.

Raney, B. and Nagel, K. (2006). An improved framework

for large-scale multi-agent simulations of travel be-

havior. Towards better performing European Trans-

portation Systems, pages 305–347.

Schadschneider, A., Klingsch, W., Kluepfel, H., Kretz, T.,

Rogsch, C., and Seyfried, A. (2009). Encyclopedia

of Complexity and System Science, volume 5, chapter

Evacuation Dynamics: Empirical Results, Modeling

and Applications, pages 3142–3176. Springer, Berlin,

Heidelberg.

Seyfried, A., Chraibi, M., Kemloh, U., Mehlich, J., and

Schadschneider, A. (2010). Runtime Optimization

of Force Based Models within the Hermes Project.

In Pedestrian and Evacuation Dynamics 2010, pages

363–373. Springer, 2011.

Stange, H., Liebig, T., Hecker, D., Andrienko, G., and An-

drienko, N. (2011). Analytical workﬂow of mon-

itoring human mobility in big event settings using

bluetooth. In Proceedings of the 3rd International

Workshop on Indoor Spatial Awareness, pages 51–58.

ACM.

Thompson, P. A. (1994). Developing new techniques for

modelling crowd movement. Phd thesis, University of

Edinburgh.

Woodings, R., Joos, D., Clifton, T., and Knutson, C. D.

(2001). Rapid heterogeneous connection establish-

ment: Accelerating bluetooth inquiry using irda.

Yu, W. J., Chen, R. Dong, L., and Dai, S. (2005). Centrifu-

gal force model for pedestrian dynamics. Phys. Rev.

E, 72(2):026112.

MODELLING MICROSCOPIC PEDESTRIAN MOBILITY USING BLUETOOTH

275