A Cloud-Assisted Infrastructure for OccupancyTracking in Smart Facilities

Dimitrios Sikeridis and Michael DevetsikiotisDepartment of Electrical and Computer Engineering

The University of New MexicoAlbuquerque, NM, USA{dsike, mdevets}@unm.edu

Ioannis PapapanagiotouPlatform Engineering

NetflixLos Gatos, CA, USA

ipapapa@unm.edu

Abstract—The need of smarter and more interaction-friendlyindoor spaces compels the combination of Internet of Thingsarchitectures with Cloud Computing approaches to produceefficient and realizable infrastructure deployments. In this work,we discuss a Cloud-assisted infrastructure model for monitoringuser motion in smart buildings with reference to a real-subjectrealization trial. We focus on the model architecture and discussthe applicability of the generated data on Cloud-based potentialapplication scenarios.

I. INTRODUCTION

Advanced technologies, such as Internet-of-Things (IoT)enable creative ways of enhancing productivity and life qualityinside facilities. The limits are pushed beyond managingtemperature, door-locks and general security, to go as faras reducing energy costs, detecting and building knowledgebased on human patterns and improving the occupant-buildinginteraction.

One of the key features such intelligent facilities shouldpossess is the ability to efficiently track occupants’ mobility,either in order to take real-time actions or use the data tocalculate long-time patterns. This type of services are realisedeither by attempting to estimate the user’s 2D coordinatesin a given space, which is referred to as micro-location[1], [2], or by attempting to accurately place the user inthe vicinity of certain anchor points, known as proximitysensing [3]. Especially, for indoor spaces, while a numberof RF-based technologies has been proposed over the years,the unpredictability of signal propagation due to the variablephysical indoor environment, makes it still an ongoing researchtopic [2].

Alternative methods that utilize data from body-mountedInertial Measurement Units (IMUs) have yielded accurateresults concerning microlocation [4], [5]. However, the costof mass distribution to individuals along with the lack of acentral processing and decision making node (such as a Cloudfederation), prohibit their use in well-attented everyday-useenvironments, making them more appealing to first responderlocalization applications [6]. Sub-meter accuracy, scalable andlow-cost installation, as well as low energy demands areimportant factors for large scale indoor localization services.

In this paper, we discuss a Cloud-assisted infrastructuremodel for location-aware smart facilities as featured in Fig. 1.

Smart Facility Cloud

Building State

Analytics

UserTransitionTracking

Edge Devices(Computing

Infrastructure)

: Moving User : Periodic Transmission

Fig. 1. Location-aware intelligent building model

The model is based on proximity-sensing and a movingtransmitter-fixed scanner architecture that enables collectionof user-centric data. We also provide a proof of concept-reference to our real-subject trial of this model [7] alongwith an occupant mobility tracking approach that leveragesthe collected data.

This work is organized as follows: Section II discussesrelated work. Section III presents the model architecture.Our proof of concept deployment and trial are presented inSection IV, while Section V discusses the basic user movementtracking method. Finally, Section VI highlights the Cloudinvolvement and Section VII concludes this paper.

II. RELATED WORK

Localization services based on proximity detection havebeen developed around a variety of architectures. Similarto our approach, Martella et al. in [8] describe a sensinginfrastructure comprising of moving users equipped withtransmitters, predefined broadcasting anchors and a backbonesystem of sniffers. Their approach is based on face-to-faceand mobile-to-mobile proximity sensing representing movingusers as proximity graphs [9]. Other approaches deploy lesselaborate methods that utilize already installed infrastructurewith the first option in such cases being the use of WiFiequipment along with smartphone deployment [10]. However,such solutions suffer from low-accuracy in distance calcula-tions due to complex signal propagation as a WiFi beacon canbe detected occasionally from buildings away.

BLE Beacon Edge Device Cloud

Fig. 2. System architecture

In response to the this issue, proximity sensing approachesare increasingly utilising the Bluetooth Low Energy (BLE)technology for their services. The dominant approach in thatcase is utilizing BLE-beacons [11] as static short-range trans-mitters that mark certain areas [3], [12]–[14]. Such topologyrequires users equipped with scanning devices (usually smart-phones) that gather the broadcast packets and use the ReceivedSignal Strength Indicator (RSSI) values to place the user to theproximity of the anchored area. Also, these beacons are passivewireless devices without Internet connectivity or processingpower, and the proximity calculations are carried out either inthe user’s scanning device [13], or in most recent approachesdirectly to Cloud-based environments [3], [12].

III. MODEL ARCHITECTURE

Although, the aforementioned architectures can yield sig-nificant results in terms of accuracy, they heavily rely on theassumptions that users are equipped with smartphones andpossess the appropriate application. Our schematic model asdepicted in Fig. 1 avoids this assumption by supplying theoccupants with small-factor low-consumption devices that pe-riodically broadcast user-centric data. On the receiving end ofthese transmissions, we place dedicated edge devices installedthroughout the facility. These scanning devices partition, eitherby themselves of by forming clusters, the monitored smartspace into discrete cells.

At the next level of this IoT-inspired architecture, the edgedevices are connected to the facility’s networking infrastruc-ture and continuously forward the received user data to afederated Cloud level that allows centralized, but scalablecoordination. That way Cloud resources can be leveraged torun proximity algorithms and provide cell to cell occupantmovement detection utilizing the central supervision of thesmart space. This tiered architecture will reduce the responsetimes to the central processing Cloud deployment, resulting ina system better suited for real-time monitoring applications.Finally, the edge devices will be able to cache a significantamount of recent data before reporting to the Cloud. Thiscapability will enhance the accuracy of the proximity estima-tions by utilizing all the latest measurements and will preventdata losses in the case of a temporary disconnection from theCloud.

IV. DEPLOYMENT & PROOF OF CONCEPT TRIAL

In order to provide a proof of concept of the proposedinfrastructure, a simpler version was realized at three floorsof NC State University’s ECE Department as discussed in

: Edge Device - Scanner 1st Floor

CELL A

CELL B

CELL C

CELL D

CELL E

CELL G

CELL F

Fig. 3. Edge device and cell locations

[7]. The deployment used 30 Raspberry Pi modules as edgedevices and Gimbal BLE iBeacons Series 10 to equip themoving users. The edge devices in this case are used onlyas dedicated scanners that receive the beacons’ advertisingpackets. For every reception they report to a centralized serverinformation regarding the identity of the beacon, the time ofpacket reception and its RSSI value accompanied with thescanner’s ID. The server-scanner communication is achievedvia the MQTT protocol [15]. The system’s architecture isshown in Fig. 2.

The majority of edge devices was installed on the firstfloor of the facility and their positions were dictated bythe presence of power outlets. Fig. 3 shows their positionsalong with their cell clusterization. Our deployment utilizesa Geofencing strategy to define localization [16]. Each edgedevice is considered a Point of Interest (PoI) and defines avirtual barrier around it, where our system tracks incomingand outgoing users. In our case, the geofence radius for eachscanner was chosen to be seven meters to avoid overlappingareas, while for the same reason, closely installed edge devicesform unified cells as shown in Fig. 3.

In order to visualize the relationship between the RSSI val-ues and the user-scanner distances we collected and averaged30 RSSI samples at a number of distances starting from 0meters up to 10 meters. For this experiment the edge devicewas mounted on a wall at the height of 1.6 meters and theresults are shown in Fig. 4. The fitted curve is based on thelog-distance path loss model:

RSSI = −10 γ log10(d

d0) + C (1)

where γ represents the path loss exponent of the propagationchannel, d is the user-scanner distance, d0 is the referencedistance and C represents the average value of RSSI at d0

Finally, the installation was used to host a IRB-approvedreal-subject trial at the spaces of NC State University’s Cen-tennial Campus - Engineering Building II. The trial lasted for33 days while 46 students participated by carrying with theman iBeacon device at all times.

0 1 2 3 4 5 6 7 8 9 10 11

Distance [m]

-100

-95

-90

-85

-80

-75

-70

-65

-60A

vera

ge R

SS

I [d

Bm

]

RSSI vs. Distance

Fitted Curve

Fig. 4. Curve fitting for RSSI values at distances from 0 to 10 meters

1

3

2

Fig. 5. User cell transition

V. USER CELL OCCUPANCY & TRANSITIONS

Using the fixed edge devices to partition the facility spaces,proximity estimation methods can be utilised to place an indi-vidual occupant to a distinct cell. The continuous string of thisuser’s transmitted packets provide information regarding theidentity of the scanner in his vicinity and when the broadcasthappened along with its RSSI. Since after a transmission burstmultiple edge-devices will receive the signals, a comparingmethod can be used to select the cell where the user willbe classified to. Using the stronger RSSI value is one ofthe solutions. However, we should also consider the optimalrefresh period of the decision since the collection of multipleRSSI values can enable us to use RSSI smoothing filters forincreased accuracy. The area covered by each scanner can bedetermined by simple signal propagation models that relateRSSI with distance such as Equation 1. This parameter is alsotunable by considering signal strengths over a limit, reducingthat way the range an advertisement packet can travel.

In order to determine the applicability of this method inlocations near the cell limit we performed a series of staticexperiments. The classification criterion that assigns the staticuser to a cell is the ”naive” strongest RSSI approach. Theuser is periodically assigned to the cell that receives theadvertisement message with the strongest RSSI. The decisionperiod was investigated in these tests with Fig. 6 showing the

0 5 10 15 20 25

Decision period [s]

50

55

60

65

70

75

80

85

90

95

100

Avera

ge A

ccu

racy [%

]

Tests Average

Fitted Curve

Fig. 6. Accuracy of near-edge cell occupancy - Tests average

0 5 10 15 20 25 30 35

Day of Trial

104

105

106

107

Edge D

evic

es to C

loud T

raffic

RSSI Report Messages

Fig. 7. Generated traffic per trial day

average accuracy while the period duration ranges from 2 to25 seconds.

By assigning the occupant to a specific and unique cellfor each refresh time slot, the system will be able to trackmovements inside the facility as depicted in Fig. 5. It ishowever obvious from Fig. 6 that while higher periods yieldeventually accurate results, they are not able to detect rapiduser movements and therefore fast cell transitions. Therefore,we have to additionally investigate the optimal decision pe-riod to also accommodate this functionality. Over time, suchdata can generate useful motion patterns between cells andlocations along with user-centric preferences and habits.

VI. CLOUD ASSISTANCE

Depending on the beacon advertising period, the proposedsystem is able to yield a significant amount of facility to Cloudtraffic, especially if the facility size and user numbers areof large scale. Therefore, a Cloud federation is imperativeto provide real-time services to each individual and at thesame time collect the data loads for further processing. Fig. 7shows total daily messages received by the server from all theinstalled edge devices during our trial.

Regarding off-line processing, the gathered data can beused not only to produce mobility patterns for each occupant,

0

2

4

6

8

10

12

×104

Hour

Ed

ge t

o C

loud M

essag

es

00

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

Weekdays Average ( µ ± σ )

Fig. 8. Average weekday activity per hour

but also enable features that pertain to facility managementanalytics produced directly on the Cloud. Monitoring useractivity per location or per hour are cases in point. Fig. 8 showsthe average and standard deviation of total facility activity perhour during weekdays as resulted from our 1-month trial. Fig.9 shows the average and standard deviation of first floor cellactivity during weekdays. As activity metric we used the totalamount of edge-to-Cloud messages produced by each cell.

VII. CONCLUSION

In this paper we present a moving transmitter-fixed scannerarchitecture for future location aware facilities. The systemwas deployed in real-word conditions and was evaluated withindependent experiments and a real-subject, one-month trial.Also, we highlighted the applicability of the architecture tofacilitate user occupancy and mobility monitoring. Finally, wediscuss the Cloud-centered opportunities that can be exploitedby the amount of data generated by our IoT-enabled model.The results reveal a reliable solution for future smart facil-ities where occupant tracking is a necessity. This work is abasis to future develop our solution by optimizing functionalparameters and adding extra Cloud-based services.

ACKNOWLEDGMENT

This research was supported by IBM with a Faculty Award.The authors are grateful to Mahdi Inaya and Michael Meli forcarrying out the installation, the NC State ECE Departmentfor allowing us to host the trial in the EBII building, and tothe gracious ECE trial participants.

REFERENCES

[1] F. Zafari and I. Papapanagiotou, “Enhancing ibeacon based micro-location with particle filtering,” in 2015 IEEE Global CommunicationsConference (GLOBECOM), Dec 2015, pp. 1–7.

[2] F. Zafari, I. Papapanagiotou, and K. Christidis, “Microlocation forinternet-of-things-equipped smart buildings,” IEEE Internet of ThingsJournal, vol. 3, no. 1, pp. 96–112, 2016.

A B C D E F G

Cell ID

-1

0

1

2

3

4

5

6

Edg

e to

Clo

ud M

essag

es

×104

Weekdays Average ( µ ± σ )

Fig. 9. Average weekday activity per cell

[3] F. Zafari, I. Papapanagiotou, M. Devetsikiotis, and T. J. Hacker, “En-hancing the accuracy of ibeacons for indoor proximity-based services,”in (To appear in IEEE ICC 2017). IEEE, 2017.

[4] A. Petropoulos and T. Antonakopoulos, “Indoor location estimationusing a pair of wearable devices,” in IEICE Information and Communi-cation Technology Forum (ICTF). IEICE, 2016.

[5] Y. Liu, M. Dashti, and J. Zhang, “Indoor localization on mobile phoneplatforms using embedded inertial sensors,” in Positioning Navigationand Communication (WPNC), 2013 10th Workshop on. IEEE, 2013,pp. 1–5.

[6] R. Zhang, F. Hoflinger, and L. Reindl, “Inertial sensor based indoorlocalization and monitoring system for emergency responders,” IEEESensors Journal, vol. 13, no. 2, pp. 838–848, 2013.

[7] M. Inaya, M. Meli, D. Sikeridis, and M. Devetsikiotis, “A real-subjectevaluation trial for location-aware smart buildings,” in Computer Com-munications Workshops (INFOCOM WKSHPS), 2017 IEEE Conferenceon. IEEE, 2017.

[8] C. Martella, M. Cattani, and M. van Steen, “Exploiting density totrack human behavior in crowded environments,” IEEE CommunicationsMagazine, vol. 55, no. 2, pp. 48–54, 2017.

[9] C. Martella, M. van Steen, A. Halteren, C. Conrado, and J. Li, “Crowdtextures as proximity graphs,” IEEE Communications Magazine, vol. 52,no. 1, pp. 114–121, 2014.

[10] C. N. Klokmose, M. Korn, and H. Blunck, “Wifi proximity detectionin mobile web applications,” in Proceedings of the 2014 ACM SIGCHIsymposium on Engineering interactive computing systems. ACM, 2014,pp. 123–128.

[11] Getting started with ibeacon. [Online]. Available:https://developer.apple.com/ibeacon

[12] F. Zafari, I. Papapanagiotou, M. Devetsikiotis, and T. Hacker, “Anibeacon based proximity and indoor localization system,” arXiv preprintarXiv:1703.07876, 2017.

[13] M. Kouhne and J. Sieck, “Location-based services with ibeacon technol-ogy,” in Artificial Intelligence, Modelling and Simulation (AIMS), 20142nd International Conference on. IEEE, 2014, pp. 315–321.

[14] R. Faragher and R. Harle, “Location fingerprinting with bluetooth lowenergy beacons,” IEEE journal on Selected Areas in Communications,vol. 33, no. 11, pp. 2418–2428, 2015.

[15] Mq telemetry transport. [Online]. Available: http://mqtt.org[16] S. Rodriguez Garzon and B. Deva, “Geofencing 2.0: taking location-

based notifications to the next level,” in Proceedings of the 2014 ACMInternational Joint Conference on Pervasive and Ubiquitous Computing.ACM, 2014, pp. 921–932.