Implementing an iBeacon Indoor Positioning Systemusing Ensemble Learning Algorithm

Kuan-Wu Su1,He-Yen Hsieh2,Jen-Chieh Hsu3,Bo-Han Chen4,Che-Jui Chang5,Jenq-Shiou Leu6

Department of Electronic and Computer Engineering123456,National Taiwan University of Science and Technology, Taiwan(R.O.C.)

{d101021071,m104021242,m105021033,m105021434,m104021045,jsleu6}@mail.ntust.edu.tw

ABSTRACTNowadays, there is an increasing requirement for in-door positioning and navigation with Location BasedServices (LBSs). Many applications on smartphonesexploit different techniques and inputs for positioning.Most of the indoor wireless positioning systems rely onReceived Signal Strengths (RSSs) from indoor wirelessemitting devices. However, the accuracy of indoor po-sition is easily affected by servals signal interference. Inthis paper, we propose a LBS system using ensemblemachine learning with iBeacon RSSs fingerprint, andhope to achieve higher accuracy for indoor positioning.Preliminary experiments showed very promising resultsthat our approach can improve indoor positioning ac-curacy at shorter distance.

KeywordsIndoor positioning; iBeacon; ensemble learning; ma-chine learning

1. INTRODUCTIONIn recent years, Location Based Services (LBS) have

been widely applied in many applications and services inmobile environments, where location awareness is cru-cial to people’s daily lives. However, Global PositioningSystem (GPS), although is very convenient and accu-rate outdoor, but has significant signal loss indoor. Re-cently, more and more important applications requiredindoor LBS have being applied to places like universi-ties, shopping malls, and hospitals, with huge demandin precision and accuracy for people to find locationsinside buildings.

Permission to make digital or hard copies of all or part of this work for personalor classroom use is granted without fee provided that copies are not made ordistributed for profit or commercial advantage and that copies bear this noticeand the full citation on the first page. Copyrights for components of this workowned by others than ACM must be honored. Abstracting with credit is per-mitted. To copy otherwise, or republish, to post on servers or to redistribute tolists, requires prior specific permission and/or a fee. Request permissions frompermissions@acm.org.

c© 2017 ACM. ISBN .

DOI:

Many different kinds of signals can be used for in-door positioning, such as Wi-Fi[2], ZigBee, RFID, Ul-trasound, Laser, and Bluetooth Low Energy (BLE).Most popular indoor positioning algorithm is based onfingerprint database[3], where it uses Received SignalStrength Indicator (RSSI) as fingerprint basis, and canbe read easily with widely available portable devices.Here, we use iBeacon technology, which utilize BLEfor indoor LBS combined with machine learning algo-rithms. Bluetooth beacons are cost effective, easy todeploy and small in sizes[1], and iBeacon doesn’t rely onexternal power source and can last for months. There-fore, we implement our ensemble learning indoor posi-tioning algorithm with iBeacon and hopefully to achievehigh positioning accuracy as other methods.

2. INDOOR POSITIONING WITH EN-SEMBLE LEARNING

2.1 System Architecture

2.1.1 Data Collection SystemThe data collection system consists of two main com-

ponents - client side on mobile devices, and server sidewith database and services. Mobile application on theclient side collects RSSI signals, and then send its datato the server side. Ensemble learning algorithm runningon the server side calculate the results based on RSSIfingerprint stored in the database. The topology of oursystem is shown in Figure 1.

Figure 1: Data Collection System Architecture

2.1.2 Ensemble LearningWe have implemented a positioning application on

several different types of mobile devices gathering dataand connecting to our servers to evaluate the systemperformance. Beacon signals collected on the clientsides are send to multiple classifiers, and all the resultsfrom them are then ensemble into an aggregated pre-diction result. Several different ensemble learning algo-rithms were evaluated and algorithms with parametersthat generate the highest accuracy results in shorter dis-tance were picked. The ensemble learning process flowis shown in Figure 2.

Figure 2: Ensemble Learning Process Flow

2.2 Experiments and Result

2.2.1 Experiment Environment and SetupIn order to evaluate the proposed localization ap-

proach, several preliminary experiments were conducted.The test environment is located at the seventh floor ofthe Electrical and Computer Engineering Building inNational Taiwan University of Science and Technology,Taiwan. The tracking area has a dimension of 4.8m by7.2m by 3m within the red rectangular. Three beaconsare deployed on three different walls, two meters highoff the ground. The floor plan of our test environmentis shown in Figure 3.

Figure 3: Floor Plan of the Test Area

To evaluate our proposed system in a real environ-ment, we implemented an application for indoor posi-tioning with ASUS Zenfone3, an Android smartphoneequipped with Wi-Fi and BLE. Each machine learningalgorithm as well as the ensemble learning process areimplemented with python using scikit-learn, and deeplearning neural network using Keras with TensorFlowbackend.

2.2.2 ResultsThe results of estimated accuracy(ACC) over differ-

ent positioning displacement distance is shown in Fig-ure 4. And it shows the ensemble learning process canincrease the positioning accuracy at the displacementdistance around 2 meters, and achieve over 80% of ac-curacy at the distance less than 1.3 meter, and over 90%around 2.5 meter.

Figure 4: Positioning Accuracy over Distance

3. CONCLUSIONSOur prototype indoor positioning system using en-

semble learning algorithm has shown great promise tobe on par in accuracy with other indoor positioningmethods. And there’s still more room to improve oursystem’s performance like deploying more BLE beacons,or using other ensemble learning techniques like boost-ing to make shorter distance accuracy increased further.The low cost and cheap iBeacon also mean our systemcan be cost effective and energy efficient, and would bea solid foundation for other LBS applications to buildupon.

4. REFERENCES[1] M. Altini, D. Brunelli, E. Farella, and L. Benini.

Bluetooth indoor localization with multiple neuralnetworks. In Wireless Pervasive Computing(ISWPC), 2010 5th IEEE InternationalSymposium on, pages 295–300. IEEE, 2010.

[2] J.-S. Leu, M.-C. Yu, and H.-J. Tzeng. Improvingindoor positioning precision by using receivedsignal strength fingerprint and footprint based onweighted ambient wi-fi signals. ComputerNetworks, 91:329–340, 2015.

[3] J. Zhu, K. Zeng, K.-H. Kim, and P. Mohapatra.Improving crowd-sourced wi-fi localization systemsusing bluetooth beacons. In Sensor, Mesh and AdHoc Communications and Networks (SECON),2012 9th Annual IEEE Communications SocietyConference on, pages 290–298. IEEE, 2012.