c©2019 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media,including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers

or lists, or reuse of any copyrighted component of this work in other works. DOI: 10.1109/MCOM.2019.1800611

Toward Understanding Crowd Mobilityin Smart Cities through the Internet of Things

Gurkan Solmaz1, Fang-Jing Wu2, Flavio Cirillo1,3, Erno Kovacs1,Juan Ramon Santana4, Luis Sanchez4, Pablo Sotres4, and Luis Munoz4

1NEC Laboratories Europe, Heidelberg, Germany2TU Dortmund University, Dortmund, Germany3University of Napoli Federico II, Napoli, Italy

4University of Cantabria, Santander, Spain

Abstract— Understanding crowd mobility behaviors would bea key enabler for crowd management in smart cities, benefitingvarious sectors such as public safety, tourism and transportation.This article discusses the existing challenges and the recentadvances to overcome them and allow sharing informationacross stakeholders of crowd management through Internet ofThings (IoT) technologies. The article proposes the usage of thenew federated interoperable semantic IoT platform (FIESTA-IoT),which is considered as “a system of systems”. The platformcan support various IoT applications for crowd management insmart cities. In particular, the article discusses two integrated IoTsystems for crowd mobility: 1) Crowd Mobility Analytics System, 2)Crowd Counting and Location System (from the SmartSantandertestbed). Pilot studies are conducted in Gold Coast, Australia andSantander, Spain to fulfill various requirements such as providingonline and offline crowd mobility analyses with various sensorsin different regions. The analyses provided by these systemsare shared across applications in order to provide insights andsupport crowd management in smart city environments.

I. INTRODUCTION

Sustainable development of cities is a major global chal-lenge as more than half of the world population is living inurban areas. The smart city concept allows optimizing servicesfor urban areas because or as a result of the advancementof the new technologies ranging from very small devices tobig data centers. These technologies can be considered in thecontext of IoT, where many objects, devices, machines, anddata centers are connected. The usage of IoT technologies forcrowd management in urban environments is promising for thefuture of smart cities.

IoT technologies can enable many improvements for crowdmanagement, which spans sectors such as transportation ser-vices (e.g., operating public transport or directing pedestriantraffic), public safety (e.g., detection of fighting incidents),and tourism (e.g., event management for enhanced visitorexperience). For instance, movement behaviors of crowdsmay indicate situations such as traffic congestion, emergencyincidents, and panic situations during certain events such aslarge gatherings in city squares.

While cities aim to achieve smart urban services, newchallenges arise due to the limitations and deficiencies ofthe current systems and technologies in terms of scalabilityof the connected systems, information transparency between

Dashboard

Santander Spain

Gold Coast Australia

Federated and Interoperable IoT Platform

Crowd Mobility Analytics System

Smart City Magnifier

Crowd management stakeholders

3rd party applications

Safety services

City councils

Public transport

Other smart cities

Fig. 1. Federated and interoperable IoT platform supporting crowd manage-ment stakeholders of smart cities.

different systems (i.e., semantic interoperability) or stakehold-ers, data federation, and information privacy. When mobil-ity information must be shared across multiple stakeholders,a proprietary infrastructure cannot fulfill all the differentrequirements that they impose. For example, some of thestakeholders expect real-time mobility monitoring service forevent detection while others require historical mobility dataanalytics to analyze efficiency of services in different urbanenvironments (e.g., train station, stadium, city square). It isvery difficult to re-design a one-size-fit-all IoT system whennew requirements arise for various environments or differenttime periods. A better solution is to provide “a system ofsystems” in which a new service can be easily developed orsetup to handle any new requirement by leveraging existingtechnologies and infrastructure. To make such a system of sys-tems useful, semantic models based on an appropriate ontologyare needed for transparently exchanging data, analytics results,and allowing to share new insights from different crowdmanagement applications. The federation of data, results, and

arX

iv:1

909.

0852

2v1

[ee

ss.S

P] 1

7 Se

p 20

19

learned insights is the key technical enabler to understandthe crowd mobility behaviors in a smart city. Finally, privacypreservation is a problem of utmost importance for smartcities. While various data from vast deployment of sensorstravel through the IoT systems, preserving privacy at a levelcloser to the data contributors (providers) is an importantchallenge.

This article describes the recent advances in IoT for un-derstanding crowd mobility in smart cities. The federatedand interoperable semantic IoT (FIESTA-IoT) platform forsmart cities is introduced for the specific perspective of crowdmanagement applications. Fig. 1 illustrates the outlook of thesmart city applications leveraging the smart city platform forsharing information across various stakeholders. While theplatform is currently in use for several smart city testbeds,the article focuses on two IoT systems for crowd mobility,namely Crowd Mobility Analytics System (CMAS) and CrowdCounting and Location System (CCLS) and discusses theaspects related to the aforementioned limitations.

Two pilot studies are conducted in Gold Coast, Australiaand Santander, Spain, where various sensors are deployed inurban areas. The first pilot study uses CMAS in Gold Coastfor a medium-scale smart city deployment. The requirementsof the pilot include analyzing heavy or light pedestrian trafficat streets with or without vehicles. The second pilot study usesCCLS in an indoor market in Santander. The requirements in-clude detecting people (crowd size) and locating their positionsat public buildings of a city and other critical infrastructures. Inboth pilots, data anonymisation limits tracking devices for longtime periods. On the other hand, online and offline analyticsinformation needs to be shared across various stakeholderssuch as city councils and visualized in several interfaces usingIoT technologies and infrastructure to provide insights forcrowd management in smart cities.

II. CROWD MOBILITY ANALYTICS USING THE SMARTCITY PLATFORM

A. Federated and Interoperable IoT Platform

Smart city data is often gathered by solutions where dedi-cated networks of sensors or data sources produce observationsto be consumed by specific applications. The systems usuallydiffer from each other, serving for distinct purposes, and theyare mostly not interoperable [1], [2]. In this regard, creatingcrowd management services that harness the abundant datafrom a smart city (e.g., environmental data, road traffic infor-mation) would require either ad-hoc integration or creation ofnew systems. This situation raises a new requirement of anintegrated “system of systems” or “container of systems”.

To overcome this challenge, we propose a crowd mobility-based instantiation of the FIESTA-IoT platform [3] and pro-vide semantic interoperability from IoT deployments to theservices (shown in Figure 2). The heterogeneous IoT deploy-ments on the IoT Devices and Systems (bottom layer) are in-tegrated to the Cloud and data is anonymised with salting andhashing. In this layer, in addition to the two crowd mobilitysystems (CMAS and CCLS), IoT Cloud Data from externalplatforms can be connected. Furthermore, there exist other IoT

Managing Cloud

Infrastructure

Managing IoT Deployments

Cro

wd

Man

age

me

nt

Ap

plic

atio

ns

Fed

era

ted

Clo

ud

In

fras

tru

cutu

re

FIESTA-IoT Crowd Mobility Semantic Model

IoT

Dev

ice

s an

d S

yste

ms

Crowd Mobility Linked-Data

Storage

Managing IoT Services and

Applications

Federated Context Management

IoT Broker Discovery of

IoT Resources

Privacy management

Salting Hashing

Go

ld C

oas

t O

per

atio

n C

ente

r

Smar

tSan

tan

der

M

aps

Smar

t C

ity

Mag

nif

ier

Situ

atio

n

Ass

essm

ent

Serv

ice

Oth

er Io

T Se

rvic

es

Co

nte

xtu

aliz

atio

n

Serv

ice

Cro

wd

Mo

bili

ty

An

alyt

ics

Syst

em

Smar

tSan

tan

der

Sy

stem

Oth

er Io

T Sy

stem

s

IoT

Clo

ud

D

ata

Fig. 2. Crowd mobility-based instantiation of the federated and interoperableIoT platform.

FIESTA-IoT systems that can be leveraged. Currently morethan 5000 sensors (from 11 integrated testbeds [4]) reportenvironmental data (e.g., temperature, humidity, illuminance,noise level), road traffic information (e.g., vehicle speed,traffic intensity), car and bike parking spots, estimated arrivaltimes of buses, and smart building information (e.g., humanoccupancy, power consumption).

At the Federated Cloud Infrastructure (middle layer), thedata from the bottom layer is modelled using the FIESTA-IoT Semantic Model and stored in the Linked-Data Storage.In particular, the semantic model for crowd mobility data isdescribed in Section II-B. The data in the Cloud infrastructureis accessible through the Federated Context Managementwhich exposes NGSI and SPARQL interfaces. Our open sourceIoT Broker (Aeron Broker) component provides scalable fed-eration for the context management, whereas IoT Discovery(NEConfMan) enables easy registration and discovery of re-sources with features such as geo-discovery.

The crowd management-related IoT data is harnessed byCrowd Management Applications (top layer) which containIoT services provided by the platform and crowd mobilityapplications. These services enhance the crowd mobility datathrough reasoning by aggregating the semantic data and as-sessing the situations related to physical objects (i.e., Contex-tualization Service) at different levels of abstraction such asbuildings level or street level. Assessment of the situations canbe performed through; a) pre-defined thresholds, b) anomalydetection, c) time-series analysis, d) artificial intelligence. Theobtained situations are displayed on the dashboard in Figure1, named Smart City Magnifier, which reports alerts regardingtraffic status, crowd flows, critical events (e.g., fire bursting),and so on. Moreover, crowd mobility applications such as GoldCoast Operation Center and SmartSantander Maps receive the

results (generated by CMAS, CCLS or other IoT services)from the Cloud and provide visualizations.

B. Crowd Mobility Semantic Model

In order to provide seamless interoperability and informa-tion transparency from IoT systems to the crowd managementapplications, the crowd mobility outcomes are semanticallyannotated following the FIESTA-IoT ontology [5] as shownin Fig. 3 (with a stress on the specific taxonomy of M3-litefor crowd mobility).

Rich and complex knowledge is represented with an ontol-ogy as things are connected to each other through relation-ships. Things are not identified as individuals, but as classesof individuals. Moreover, a class might have sub-classes. Forexample, peopleCounterX is an instance of PeopleCountSensorclass which is a subclass of Counter (see Fig. 3). The classescan be defined and described in taxonomies and an ontologymay use classes from different ontologies or taxonomies.Relationships between classes are known as properties and itis possible to define properties’ cardinality. Each class andproperty used in an ontology is uniquely identified by anamespace prefix and the class or property specific name. Forexample, m3-lite:PeopleCountSensor is a class defined in theM3-lite ontology. For the sake of readability, in this paragraphwe are omitting the namespace prefix while they are shownwith prefix in Fig. 3.

The core concept is the SensingDevice, representing a sensorthat produces Observation, which is a measurement (or com-putation) of a phenomenon related to an object happened at aspecific Instant. For example, a crowd mobility detector can beseen as a Device composed of multiple SensingDevices. In thissense, such a detector can have one PeopleFlowCountSensorand one StayingPeopleCountSensor, which are subclasses ofPeopleCountSensor. The Observation(s) is expressed with aQuantityKind having a Unit. Following our example, theQuantityKind associated to the data generated by the People-FlowCountSensor is CountPeopleMoving (subclass of Quanti-tyKind) with Item as its Unit and with the Direction propertyexpressed either in geodetic DirectionAzimuth or as a genericDirectionHeading. The directions start from the Point that isthe location of the physical Platform. Platform is meant as thesupporting dock to which the Device is attached. The Staying-PeopleCountSensor generates CountPeopleStaying values ex-pressed in Item. The system also consists of PeopleStayDura-tionSensor that generates PeopleStayDurationAverage valuesmeasured in SecondTime. Each SensingDevice might have aCoverage, specified either as Polygon/Rectangle/Circle or asa simple Point. This indicates the geographic extent of theObservation.

C. Integrated IoT Systems

1) Crowd Mobility Analytics System: The CMAS (extendedfrom our system in [6]) is integrated with the platform viasemantic annotation of the outcome. The developed systemconsists of Wi-Fi sniffers, stereoscopic cameras, IoT gateways,and data analytics modules. The Wi-Fi sniffers are capableof capturing wireless probes broadcasted by mobile devices.

Based on the captured Wi-Fi probes, the system can count themobile devices in these sensing areas. The cameras are co-located with specific Wi-Fi sniffers deployed at the dedicatedcalibration choke points. A built-in people counting softwareruns in the cameras. Both Wi-Fi device detection and peoplecounting results are reported to to the Cloud, where dataanalytics modules reside, through the IoT gateways. Threeanalytics modules are developed: crowd estimation, peopleflows, and stay duration. The crowd estimation module outputsnumber of people by correlating the stereoscopic cameracounts and the number of Wi-Fi enabled devices at thecalibration points. Based on the correlation between the twodata modalities, the calibration of the data analytical results areapplied in other sensing areas without cameras. The modulemonitoring people flows infers crowd movement in these areas.Finally, the stay duration module estimates the waiting timesand the number of waiting people. All analytics results areexported to the Federated Cloud Infrastructure so the crowdanalytics results are discoverable and available for applicationsin the smart city platform.

2) Crowd Counting and Location System: Different fromCMAS, CCLS aims at analysing crowd behaviour in publicbuildings of a city, as well as critical infrastructures. Thesystem relies in the analyses of IEEE802.11 frames to discoverdevices in the surroundings of the deployment, normallywithin the monitored areas. Similar to CMAS, the deployednodes capture “Probe Request” frames sent by smartphones,which include a Wi-Fi interface in “active search” mode,incorporated in most of them. However, CCLS does not onlyaim at detecting people, but also aims at locating them. Forthis, the system stores the RSSI and sequence number from thecaptured frames. It is possible to locate people by processingthis information using RSSI-based algorithms. All the post-processing is performed in an edge server, where all themeasurements are sent after the corresponding anonymisationtechniques are applied. Once the anonymised raw measure-ments are analyzed and the counting and location analyticsapplied over them (i.e., the estimated crowd size and positionsare obtained), these observations are semantically annotatedand pushed to the Federated Cloud Infrastructure. For thesemantic modelling, each crowd estimator is modelled as anPeopleCountSensor, with a specific Coverage (representing thearea to which the estimations apply), that generates CountPeo-ple observations expressed in Item.

D. Privacy Considerations

One of the essential requirements is dealing with trackeddevices’ privacy. Nowadays, privacy is one of the majorpublic concerns. In this sense, data protection laws have tobe observed when handling data that could be personal. Quiterestrictive rules apply in most countries of the world, beingthe countries from the European Union (EU) some of the mostrestrictive ones. These rules are recently updated through theGeneral Data Protection Regulation (GDPR) [7] enforcement.

The Wi-Fi sensors in CMAS and CCLS deal with MACaddresses, which are considered personal data under the newEU regulation. As it is stated in the GDPR [7], “The principles

SSN (Sensors/Devices)

Time

M3-lite (Taxonomy)

QU (Quantity Kinds/Units)

WGS84(position)

ssn:Sensor

ssn:SensingDevice

ssn:Device

iot-lite:Coverage

ssn:Observation

ssn:SensorOutput ssn:ObservationValue

qu:QuantityKind

qu:Unit

m3-lite:Direction

ssn:o

bserved

By

ssn:m

adeO

bservatio

n

ssn:observationResult

ssn:hasValue

iot-lite:hasUnit

iot-lite:hasCoverage

geo:Point

ssn:System ssn:Platform ssn:onPlatform

ssn:attachedSystem

geo:location geo:location

time:Instant

ssn:observationSamplingTime

IoT-lite (Resources/ Entities/Services)

m3-lite:Counter m3-lite:PeopleCountSensor

m3-lite:CountPeople

m3-lite:PeopleStayDurationSensor

m3-lite:PeopleStayDurationAverage

Ontology Core Concepts

subClassOf

Object Property Cardinality: Some

Object Property Cardinality: Max1 Object Property

Cardinality: Exactly1

Legend

Crowd Mobility Concepts

iot-lite:Rectangle

iot-lite:Circle

iot-lite:Polygon

Other specific Crowd Mobility subclasses SubClasses

m3-lite: CountPeople

m3-lite:CountPeopleStaying

m3-lite:CountPeopleMoving

m3-lite: PeopleCountSensor

m3-lite:PeopleFlowCountSensor

m3-lite:StayingPeopleCountSensor

m3-lite: Direction

m3-lite:DirectionAzimuth

m3-lite:DirectionHeading

qu: Unit

m3-lite:SecondTime

m3-lite:Item

Namespaces: IoT-lite: http://purl.oclc.org/NET/UNIS/fiware/iot-lite M3-lite: http://purl.org/iot/vocab/m3-lite SSN: http://purl.oclc.org/NET/ssnx/ssn Time: http://www.w3.org/2006/time QU: http://purl.org/NET/ssnx/qu/qu WGS84: http://www.w3.org/2003/01/geo/wgs84_pos

Fig. 3. Modeling crowd mobility information based on FIESTA-IoT ontology.

of data protection should apply to any information concerningan identified or identifiable natural person”. Therefore, Wi-Fi-based tracking services in public or private spaces canbe performed only if the service obtains the user’s opted-inpermission, or data is anonymised in such manner that theuser is no longer identifiable, as mentioned in the 26th articlefrom the aforementioned regulation. The Article 29 WorkingParty, recently replaced by the European Data ProtectionBoard (EDPB), is in charge of analysing the complianceof the privacy rules. In a document released to analyze theePrivacy regulation compliance with GDPR [8], the DataProtection Working Party states that Wi-Fi tracking can onlybe performed either if there is consent or the personal data isanonymised. Within the same document, four conditions arementioned for the latter case to be compliant with the GDPR:

• The purpose of the data collection from terminal equip-ment is restricted to mere statistical counting.

• The tracking is limited in time and space to the extentstrictly necessary for this purpose.

• The data will be deleted or anonymised immediatelyafterwards.

• There exist effective opt-out possibilities.

Considering that user’s permission request is impossibleto obtain in normal conditions within the subject of the

experimentation, the only option is to anonymise data re-garding to MAC addresses. Thus, experimentation securitymeasures must be undertaken to address both, data integrityand anonymisation. Therefore, any type of experimentation orservice provision must take into account this concern, whichis usually underestimated by system developers.

CCLS in Santander is based on the Spanish Personal DataProtection Laws and the Spanish Law Protection Office recom-mendations for data anonymisation [9]. The recommendationconsists on the use of a cryptographic hash function withrandomly generated hash keys. More precisely, the HMACprotocol, which provides such mechanisms, is recommended.In the SmartSantander deployment, we implement the HMACalgorithm along with the SHA256 hashing function, with a12-bytes randomly generated key. Finally, in order to ensure anon-reversible process, this implementation also comprises aprocedure to destroy and renew the key during specific sessionperiods. For CMAS in Gold Coast, the hashed and salted Wi-Fi probe data is sent to the Cloud. The stereoscopic camerasdo not record video or perform face detection. The camerassimply count the passage of people through predefined lines atthe choke points. The outputs of the camera are people count-in and -out values. The main drawback of this procedure isthe limitation of tracking devices throughout long periods (as

Fig. 4. Heatmap from Mercado del Este in Santander.

in [10]) or longer travels within the city, but it is the price thatmust be paid to meet the privacy requirements.

III. PILOT STUDIES IN AUSTRALIA AND SPAIN

A. Pilot Deployment in Gold Coast

1) Pilot Setup: The deployments in Gold Coast include17 Wi-Fi sensors and 2 stereoscopic cameras. The Wi-Fisensors are custom-built devices for outdoor deployments. Twocameras are used at the calibration choke points, where there isa camera and a Wi-Fi sensor deployed together. The camerasare the Hella Advanced People Sensor APS-90E deployed at aheight about 3.6 meters. Each camera is configured to capturethe entire choke point for accurate counting.

The deployments target two regions. These sensors deployedin these areas are considered as Cluster 1 for (expected)heavy pedestrian traffic and Cluster 2 for light traffic places.Each cluster has a stereoscopic camera for the calibration.The collected data is sent to the Cloud where two virtualmachines are created for the clusters. Clustering the areasallows applying CMAS to city-scale by sharing the raw dataload.

2) Pilot Operation: The pilot study activities started inSeptember 2017 and CMAS has been in use starting fromNovember 2017. Various types of pilot tests are conducted onthe field during the operation of the pilot. Manual counting isperformed using video footages taken from different deploy-ment areas. In comparison to manual counting, the camerasprovide an accuracy between 88% and 98%, which mainlydepends on the weather and lightning conditions. Furthermore,field tests for heavy and light traffic areas resulted in 93% and89% crowd size accuracy compared to manual counting. Theresults obtained from outside the choke points give furtherconfidence to treat stereoscopic camera results as near groundtruth as proposed in [6].

Gold Coast pilot successfully tests the crowd mobilityanalytics services by leveraging federation of clusters andinteroperability using the semantic model to share the resultswith stakeholders. This shows that similar systems can bedeveloped and leveraged by future crowd management appli-cations using the smart city platform.

B. Pilot Deployment in Santander

1) Pilot Setup: CCLS is deployed in the “Mercado delEste” market, a restored symmetric building that containsshops, restaurants, a regional tourist office, and a museum.This building is particularly interesting as it usually receivessignificant numbers of visitors due to its central location, withexceptionally crowded periods.

The system is composed of 8 devices installed within themarket building. These devices include a Wi-Fi interfaceaimed at detecting surrounding Wi-Fi enabled visitors’ de-vices. Internet connectivity is provided through the Municipal-ity Network, and the devices are powered using Power overEthernet connected to the market’s electrical grid. In additionto the wireless interfaces, half of the devices also includeenvironmental sensors measuring temperature and humidity.Device deployment is carried out with the collaboration andsupervision of the municipality and the market managers.Considering the main goal of monitoring people within themarket, two parameters are considered in order to get marketstatus snapshots over the time. First, the number of visitorswithin the market in different time frames and second, thelocation of the visitors in the different areas of the market.

2) Pilot Operation: Firstly, in order to monitor the visitorswithin the market, we follow a deterministic approach, inwhich we consider that a device is inside the building if aminimum of 6 sensor nodes detect it with a certain level ofRSSI. In our deployment, this solution is feasible consideringthe particular symmetric distribution of the building and thelocation of the sensor nodes, covering the external wall ofthe building. Secondly, device locations are estimated usingthe Weighted Centroid Algorithm [11], which provides areasonable approximation of 5 meters to the ground-truthmeasurements without any ad-hoc calibration. For the casesthat require more precision, these positioning methods areable to introduce less than 2 meters error if the system iscalibrated in advance. Synthesized information including real-time visitor location and detected number of visitors per unit oftime is provided through a web portal to the market managersand municipality responsibles. Fig. 4 shows the heat map ofthe market in a specific moment. Other parameters, such asthe visitors’ dwell time in different long-term periods, arenot analysed due to the privacy safeguards that have to beaddressed.

IV. CITY-SCALE EXPERIMENTS

This section discusses some of the experimental observa-tions from the Gold Coast pilot with CMAS. Specifically, itincludes the variance in the crowd estimation for the Wi-Fisensors and cameras. Our focus in the experimental study isto observe the dynamic changes in the number of unique Wi-Fi devices detected and the correlation coefficient (or simplycamera/Wi-Fi ratio), which is a dynamic parameter that iscomputed by the Adaptive Linear Calibration Algorithm [6].The coefficient basically indicates the proportion of the num-ber of people (count-in and count-out events) detected by thecamera to the number of devices detected by the Wi-Fi sensorsevery time interval. We analyze the hourly results for the two

20/03/2018

21/03/2018

22/03/2018

23/03/2018

24/03/2018

25/03/2018

26/03/2018

27/03/2018

Measurement dates

0

50

100

150

200

250

300

350

400Nu

mbe

r of W

i-Fi d

evice

sCluster 1Cluster 2

20/03/2018

21/03/2018

22/03/2018

23/03/2018

24/03/2018

25/03/2018

26/03/2018

27/03/2018

Measurement dates

0.0

0.5

1.0

1.5

2.0

2.5

Cam

era/

Wi-F

i rat

io

Cluster 1Cluster 2

(a) The number of unique Wi-Fi devices detected (hourly). (b) Hourly changes of the camera/Wi-Fi ratios.

Fig. 5. Weekly measurements from Gold Coast. Cluster 1: Pedestrian ways with heavy pedestrian traffic, Cluster 2: Roads including vehicles and lightpedestrian traffic.

clusters, where 5 minute time intervals are aggregated andaveraged for 1 hour.

Figure 5-a shows the average number of Wi-Fi devicesdetected for one-week period. There exists an increased activ-ity in Cluster 1 region especially during Friday (23/03/2018)and the following weekend. This can be due to crowdednessin the shopping street and the beach area contained in thisregion. Moreover, there is a peak in Saturday that can bedue to an event or gathering. Figure 5-b shows the changeof the coefficients (ratios). The ratios are computed at thecalibration choke points (providing near-ground truth for themeasurements). The hourly ratio is computed such that num-ber of people count-in and count-out events are divided tothe number of Wi-Fi probes. First, for Cluster 2 with lighttraffic, correlation coefficient is mostly (almost all days) highercompared to Cluster 1. Second, correlation coefficient valueslie mostly in the range of (0.2, 2), whereas the peak value isabout 2.8. This indicates that the results based on Wi-Fi-onlymeasurements are likely to have less accuracy most of thetimes of the days and the correlation changes throughout thedays. Lastly, there exists certain regularity in the correlationfrom one day to another, which can be learned through atime period and then applied to other time periods wherecamera is temporarily inactive or removed. On the otherhand, as seen in the peak hours of Cluster 2, the ratios donot lie within a narrow range. One reason can be eventsaffecting the volume of pedestrians. Lastly, Fig. 5-b showsrelatively higher variance of the coefficient for areas with lightpedestrian traffic. Calibration could be necessary for shortertime intervals.

Overall, it is observed that effective use of Wi-Fi sensingand combining them with sensing by stereoscopic camerasproduce accurate sensing in large scale for both the heavy andlight pedestrian traffic areas. Moreover, the variance betweenheavy and light traffic shows the usefulness of the clusteringapproach which treats these regions separately.

V. RELATED WORK

There are recent studies that focus on understanding ofhuman mobility through IoT devices such as wireless sensors.Jara et al. [12] observed the relation between traffic behaviorand temperature conditions as a smart city application throughdeployment of IoT devices in Santander. Tong et al. [13]propose usage of Wi-Fi sensors to understand passenger flows.Evaluation through simulation results shows high accuracy.Zhao et al. [14] survey the recent advances in understandinghuman mobility in urban environments. The study lists someof the existing urban human mobility datasets collected suchas GPS, GSM, Wi-Fi, and Bluetooth traces. Similarly, Zhouet al. [15] discuss the topic of human mobility in urbanenvironments and present a taxonomy of crowdsensed inputdata types and application outcomes such as crowd densityand flows within building, and people transportation modeidentification (cycling, running, bus riding). Lastly, Montjoyeet al. [10] focus on the privacy aspect by analyzing long periodWi-Fi traces and show that 95% of the individuals can beuniquely identified using spatiotemporal datasets.

VI. FUTURE WORK AND CHALLENGES

The current work focuses on finding insights behind crowdmobility such as detecting crowdedness. However, under-standing more complex crowd mobility behaviour in a large-scale city area such as movements of groups (e.g., family)could be helpful for crowd management and enhancing smartmobility in the cities. The collected mobility information canserve as input of human mobility simulations to further studyhow city dynamics are affected by crowd mobility patterns.With the combination of real mobility dataset in a simulatedenvironment, learning new mobility insights opens up newopportunities for new crowd management strategies (e.g., con-gestion avoidance, evacuation planning, demand management)that can further improve the public service and safety in smartcities.

In our future developments, the semantic interoperabilitythrough ontologies can be leveraged more extensively forcross-infrastructure communication and knowledge sharing.The new advancements of the NGSI protocol by the ETSIIndustry Specification Group (ISG) on Context InformationManagement (CIM) are centered around the concepts of linkeddata. This opens a new horizon where knowledge graphs areshared among various infrastructures and, while their adminis-trators own the produced data, it is still accessible seamlesslyand transparently by all actors in the multi-infrastructurefederation.

VII. CONCLUSIONS

This article discusses the new advancements towards under-standing crowd mobility in smart cities using IoT. While thereexist certain limitations, the CMAS and CCLS systems usingthe smart city platform offer improvements for more efficientcrowd management. The pilot studies in Gold Coast andSantander show the capability to fulfill various requirementsand share information across stakeholders by leveraging theIoT technologies and infrastructure.

Acknowledgment: The pilot study in GoldCoast is conducted in collaboration with NECAustralia. This work has been partially funded

by the Spanish Government (MINECO) under Grant Agree-ment No. TEC2015-71329-C2-1-R ADVICE (Dynamic Pro-visioning of Connectivity in High Density 5G Wireless Sce-narios) project and by the EU Horizon 2020 Programme un-der Grant Agreements No. 731993 AUTOPILOT (AutomatedDriving Progressed by Internet Of Things), 643943 FIESTA-IoT (Federated Interoperable Semantic IoT Testbeds and Ap-plications), and 643275 FESTIVAL (Federated InteroperableSmart ICT Services Development and Testing Platforms)projects and the joint project by NEC Laboratories Europe andTechnische Universitat Dortmund. The content of this paperdoes not reflect the official opinion of the Spanish Governmentor European Union. Responsibility for the information andviews expressed therein lies entirely with the authors.

REFERENCES

[1] A. Zanella et al., “Internet of things for smart cities,” IEEE Internet ofThings Journal, vol. 1, no. 1, pp. 22–32, Feb. 2014.

[2] M. Yannuzzi et al., “A new era for cities with fog computing,” IEEEInternet Computing, vol. 21, no. 2, pp. 54–67, Mar. 2017.

[3] J. Lanza et al., “Experimentation as a Service over Semantically Inter-operable Internet of Things Testbeds,” IEEE Access, vol. 6, no. 1, pp.51 607 – 51 625, Sept. 2018.

[4] L. Sanchez et al., “Federation of Internet of Things Testbeds for the Re-alization of a Semantically-Enabled Multi-Domain Data Marketplace,”Sensors, vol. 18, no. 10, p. 3375, Oct. 2018.

[5] R. Agarwal et al., “Unified IoT ontology to enable interoperability andfederation of testbeds,” in Proceedings of IEEE WF-IoT’18, Dec. 2016,pp. 70–75.

[6] F.-J. Wu and G. Solmaz, “CrowdEstimator: Approximating crowd sizeswith multi-modal data for internet-of-things services,” in Proceedings ofACM MobiSys’18, June 2018, pp. 337–349.

[7] G. D. P. Regulation, “Regulation (EU) 2016/679 of the EuropeanParliament and of the Council of 27 April 2016 on the protection ofnatural persons with regard to the processing of personal data and onthe free movement of such data, and repealing Directive 95/46,” OfficialJournal of the European Union (OJ), vol. 59, no. 1-88, p. 294, Apr.2016.

[8] Article 29 Working Party, “Opinion 01/2017 on the proposed regulationfor the ePrivacy regulation (2002/58/EC),” Apr. 2017, accessed on Jan.3, 2019. [Online]. Available: https://ec.europa.eu/newsroom/document.cfm?doc id=44103

[9] Agencia Espanola de Proteccion de Datos, “Orientacionesy garantıas en los procedimientos de anonimizacion dedatos personales,” p. 17, October 2016, accessed on Jan.3, 2019. [Online]. Available: https://www.aepd.es/media/guias/guia-orientaciones-procedimientos-anonimizacion.pdf

[10] Y.-A. D. Montjoye et al., “Unique in the crowd: The privacy bounds ofhuman mobility,” Scientific reports, vol. 3, p. 1376, Mar. 2013.

[11] I. N. Kosovic and T. Jagust, “Enhanced weighted centroid localizationalgorithm for indoor environments,” International Journal of Computer,Control, Quantum and Information Engineering, vol. 8, no. 7, July 2014.

[12] A. J. Jara, D. Genoud, and Y. Bocchi, “Big data for smart cities withKNIME a real experience in the SmartSantander testbed,” Software:Practice and Experience, vol. 45, no. 8, pp. 1145–1160, May 2015.

[13] H. Tong et al., “Modeling large passenger flow safety by simulation andtesting,” in Proceedings of IEEE ICCC’17, Dec. 2017, pp. 235–239.

[14] K. Zhao et al., “Urban human mobility data mining: An overview,” inProceedings of IEEE Big Data’16, Dec. 2016, pp. 1911–1920.

[15] Y. Zhou et al., “Understanding urban human mobility through crowd-sensed data,” IEEE Communications Magazine, vol. 56, no. 11, pp. 52–59, Nov. 2018.

GURKAN SOLMAZ (gurkan.solmaz@neclab.eu) is a seniorresearcher at NEC Laboratories Europe, Germany. His re-search interests are in mobile computing, human mobilityanalytics, and Internet of Things.FANG-JING WU (fang-jing.wu@tu-dortmund.de) is a juniorprofessor at TU Dortmund. Her research interests are pri-marily in pervasive computing, wireless communications andnetworks, and Internet of Things.FLAVIO CIRILLO (flavio.cirillo@neclab.eu) is a researchscientist at NEC Laboratories Europe. IoT platforms are hismain research focus, with a particular interest to scalability,interoperability, and federation aspects.ERNO KOVACS (ernoe.kovacs@neclab.eu) is the senior man-ager for the IoT Research Group, NEC Laboratories Europe.His context brokering, Cloud-Edge Computing, real-time sit-uation awareness, knowledge extraction, and smart cities.JUAN RAMON SANTANA (jrsantana@tlmat.unican.es) is asenior research fellow at the University of Cantabria. Amonghis research interests are wireless sensor networks, M2Mcommunications, and mobile phone application research.LUIS SACHEZ (lsanchez@tlmat.unican.es) is a professor atthe Department of Communications Engineering, Universityof Cantabria. He is active on IoT-enabled smart cities, meshednetworking on heterogeneous wireless scenarios, and opti-mization of network performance through cognitive network-ing techniques.PABLO SOTRES (psotres@tlmat.unican.es) is a senior re-search fellow at the University of Cantabria, Spain. He hasbeen involved in several international projects framed underthe smart city paradigm, such as SmartSantander; and relatedto inter-testbed federation, such as Fed4FIRE, Fed4FIRE+ andWise-IoT.LUIS MUNOZ (luis@tlmat.unican.es) is the head of theNetwork Planning and Mobile Communications Laboratory atthe University of Cantabria, Spain. His research focuses onadvanced data transmission techniques, heterogeneous wire-less multihop networks, and applied mathematical methods fortelecommunications.