location-based group clustering using ble beacons in an ... · scheme and readily available using...

4
2016 International Conference on Indoor Positioning and Indoor Navigation (IPIN), 4-7 October 2016, Alcalá de Henares, Spain Fig. 1. An example of layout-free lecture room. Location-Based Group Clustering Using BLE Beacons in an Unconstrained Indoor Space Akihiro Sato and Naohiko Kohtake Graduate School of System Design and Management, Keio University, Yokohama, Japan Email: [email protected] Abstract—Grouping of smart devices based on their locations enables context-aware features of pervasive computing in indoor layout-free spaces. Fingerprinting is a leading indoor positioning scheme and readily available using Bluetooth Low Energy (BLE) beacons on standard smart devices. However, the accuracy of positioning using fingerprinting of radio signal degrades by changing the layout of equipment in the room. In this paper, we propose a method to classify and identify groups of smart devices by clustering analysis for RSSI monitored on smart devices. Impact of fast fading cannot be bypassed on indoor RSSI measurement, but the experimental results show a good separation of the two groups in a distance of 3m. Keywords—Bluetooth Low Energy; RSSI; clustering analysis; layout-free space; smart classroom; I. INTRODUCTION In recent years, much attention has been given to utilization of information and communication technology (ICT) in indoor spaces for educational use, such as a classroom or a lecture room [1]. Practical use for educational site is a challenge of pervasive computing and requires to develop context-aware features [2]. On the contemporary educational activities, workshop-style lecture has been gradually spreading for hands- on learning program at a university [3] or community center [4]. Group work is one of the principal methods in such kind of lecture program, that has higher level demands to the pervasive computing than ever. Furthermore, as mobile devices such as smart phones and tablet computers have been popular, Bring Your Own Device (BYOD) solution is attracting interest not only for enterprises [5] but also for educational institutions [6]. To facilitate co-working of the team members on the group work sessions, it is effective sharing digital data such as photos, drawings and documents in the team by utilizing online groupware or social networking service (SNS). However, using groupware or SNS such as Facebook group or Google Drive requires to exchange personally identifiable information (PII) such as user ID or e-mail address with team members, and therefore it causes time-wasting processes. It is desirable to develop a system that enables to automatically identify team members and create online working areas in the context of the workshop program, without exchanging PII even if it is the first time the team members meet each other. On the other hand, demands of cost for device and maintenance are drawbacks when dedicated IT device are equipped in the lecture room. Our motivation of this study is to design a group identification system of workshop participants in a lecture room based on locations of mobile devices belonging to workshop participants, without interrupting participant’s activity. We supposed layout-free lecture room such as shown in Fig. 1, so that various styles of workshop can be supported. Nowadays, proximity detection standards using Bluetooth Low Energy (BLE) beacon represented by iBeacon [7] and Eddystone [8] are widely supported by smart devices. Thus, using the signal of BLE beacons makes the system readily available; however, in order to distinct smart devices regardless the distance from BLE beacon and identify the team that the owner of each smart device joins, we evaluated a method using values of Radio Signal Strength Indicator (RSSI) on smart devices. Fingerprinting is one of the leading indoor positioning schemes using RSSI, however, the positional accuracy of fingerprinting degrades by changing the layout of equipment in the room [9]. In this paper, we propose a method to classify and identify groups of smart devices by clustering analysis for RSSI monitored on smart devices. The smart devices send the RSSI data to a server via wireless network, then the server runs the clustering analysis using the instant measurement of RSSI whereby the method works without severe impact of layout change. This method allows the BLE beacons to setup by just putting in the lecture room and without acquiring the BLE beacon’s coordinate. We evaluated the proposed method by measurement of RSSI with off-the-shelf smart devices and clustering analysis by k-means algorithm. Experimental results show that the potentiality of proposed method for developing the group identification system in a indoor space in which the layout of equipment is flexible.

Upload: others

Post on 11-Oct-2020

5 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Location-Based Group Clustering Using BLE Beacons in an ... · scheme and readily available using Bluetooth Low Energy (BLE) workshop participants beacons on standard smart devices

2016 International Conference on Indoor Positioning and Indoor Navigation (IPIN), 4-7 October 2016, Alcalá de Henares, Spain

Fig. 1. An example of layout-free lecture room.

Location-Based Group Clustering Using BLE Beacons in an Unconstrained Indoor Space

Akihiro Sato and Naohiko Kohtake Graduate School of System Design and Management,

Keio University, Yokohama, Japan Email: [email protected]

Abstract—Grouping of smart devices based on their locations enables context-aware features of pervasive computing in indoor layout-free spaces. Fingerprinting is a leading indoor positioning scheme and readily available using Bluetooth Low Energy (BLE) beacons on standard smart devices. However, the accuracy of positioning using fingerprinting of radio signal degrades by changing the layout of equipment in the room. In this paper, we propose a method to classify and identify groups of smart devices by clustering analysis for RSSI monitored on smart devices. Impact of fast fading cannot be bypassed on indoor RSSI measurement, but the experimental results show a good separation of the two groups in a distance of 3m.

Keywords—Bluetooth Low Energy; RSSI; clustering analysis; layout-free space; smart classroom;

I. INTRODUCTION In recent years, much attention has been given to utilization

of information and communication technology (ICT) in indoor spaces for educational use, such as a classroom or a lecture room [1]. Practical use for educational site is a challenge of pervasive computing and requires to develop context-aware features [2]. On the contemporary educational activities, workshop-style lecture has been gradually spreading for hands-on learning program at a university [3] or community center [4]. Group work is one of the principal methods in such kind of lecture program, that has higher level demands to the pervasive computing than ever. Furthermore, as mobile devices such as smart phones and tablet computers have been popular, Bring Your Own Device (BYOD) solution is attracting interest not only for enterprises [5] but also for educational institutions [6].

To facilitate co-working of the team members on the group work sessions, it is effective sharing digital data such as photos, drawings and documents in the team by utilizing online groupware or social networking service (SNS). However, using groupware or SNS such as Facebook group or Google Drive requires to exchange personally identifiable information (PII) such as user ID or e-mail address with team members, and therefore it causes time-wasting processes. It is desirable to develop a system that enables to automatically identify team members and create online working areas in the context of the workshop program, without exchanging PII even if it is the first time the team members meet each other. On the other hand, demands of cost for device and maintenance are drawbacks when dedicated IT device are equipped in the lecture room.

Our motivation of this study is to design a group identification system of workshop participants in a lecture room based on locations of mobile devices belonging to workshop participants, without interrupting participant’s activity. We supposed layout-free lecture room such as shown in Fig. 1, so that various styles of workshop can be supported. Nowadays, proximity detection standards using Bluetooth Low Energy (BLE) beacon represented by iBeacon [7] and Eddystone [8] are widely supported by smart devices. Thus, using the signal of BLE beacons makes the system readily available; however, in order to distinct smart devices regardless the distance from BLE beacon and identify the team that the owner of each smart device joins, we evaluated a method using values of Radio Signal Strength Indicator (RSSI) on smart devices. Fingerprinting is one of the leading indoor positioning schemes using RSSI, however, the positional accuracy of fingerprinting degrades by changing the layout of equipment in the room [9].

In this paper, we propose a method to classify and identify groups of smart devices by clustering analysis for RSSI monitored on smart devices. The smart devices send the RSSI data to a server via wireless network, then the server runs the clustering analysis using the instant measurement of RSSI whereby the method works without severe impact of layout change. This method allows the BLE beacons to setup by just putting in the lecture room and without acquiring the BLE beacon’s coordinate. We evaluated the proposed method by measurement of RSSI with off-the-shelf smart devices and clustering analysis by k-means algorithm. Experimental results show that the potentiality of proposed method for developing the group identification system in a indoor space in which the layout of equipment is flexible.

Page 2: Location-Based Group Clustering Using BLE Beacons in an ... · scheme and readily available using Bluetooth Low Energy (BLE) workshop participants beacons on standard smart devices

2016 International Conference on Indoor Positioning and Indoor Navigation (IPIN), 4-7 October 2016, Alcalá de Henares, Spain

Fig. 2. Infrastructure to realize group identification method.

II. RELATED WORKS

A. Indoor Positioning with BLE Beacons Fast fading induces RSSI fluctuations of the Bluetooth

signal. Pei et al. [10] presented a solution using probability distributions for approximating the time domain RSSI variation. Faragher et al. [9] shows the difference of spatial domain RSSI fluctuations between 3 BLE advertise channels. These studies report fingerprinting-based positioning equipped with Bluetooth signal. Fingerprinting requires re-measurement of RSSI vector for reference database after layout change to maintain accuracy. This study aims to develop a group identification system which is not affected by changing layout of the room.

B. Smart Classroom Many ICT applications to support teaching and learning

have been studied with different approach. Choudhury et al. [6] proposed a real-time classroom system that enables to deliver the lecture contents to the student’s device. Shen et al. [11] proposes a smart classroom system based on near field communication (NFC) technology. This system provides functions of attendance management, locating students and real-time feedback from students. A static facility layout is presented for employing this system. This paper focuses on location based context-aware function for lecture room with tables and chairs that can be moved freely.

III. USE CASE SCENARIO The group identification system is utilized based on the

following use case. A workshop for hands-on learning takes place in a lecture room with around 10 to 20 participants. Organizer of the workshop consists of a moderator and assistants. In the venue, the tables and chairs are feasible to locate flexibly based on the purpose of group work and the number of teams made up from the participants. Before beginning the workshop, several BLE beacons are placed in the venue. The participants install a workshop-support application to on their own smart phone or tablet computer in advance.

When the participant arrives at the venue, he/she launches the workshop-support application, that equips a function to confirm attendance, and then the application begins to monitor signals from BLE beacons. The application continues to monitor in the background and send measured data to the system server.

After opening introduction by the moderator, the participants create teams voluntary; the moderator monitors the number of the team and controls it based on the plan for the workshop. Each team gathers in the respective place, and then begins their activities. A short time later, the workshop assistant controls the group identification system to run the grouping method; the number of the groups is given to the system. Then each smart device receives the result respectively from the server. The workshop-support application notifies the information of the access link to the working area on cyberspace. Each team promotes co-working and discussion, taking advantage of online working area, and gives a presentation to share their work with other participants at the end of the lecture.

IV. PROPOSED GROUP IDENTIFICATION METHOD

A. Physical Configuration Fig. 2 shows the infrastructure to realize the proposed

group identification method. The smart devices belonging to the workshop participants receive the signals from BLE beacons, and transmits the RSSI data to the server for grouping calculation by means of wireless LAN/Ethernet. BLE beacons are placed on the table at intervals of a few meters.

B. Operation The operation of the proposed group identification method

consists of two phases; data measurement phase and grouping calculation phase.

1) On the data measurement phase • The smart devices measure RSSI of all BLE beacons at

a sampling rate of 1Hz and send the data to the server for grouping calculation. In order to prevent interruption of participant’s work, these functions work background on the smart device.

• The server for grouping calculation records the RSSI data on its database.

2) On the grouping calculation phase • When the server for grouping calculation receives a

trigger to run the grouping method, it switches to the grouping calculation phase.

• The server calculates the average of recorded RSSI in definite duration of time window just before receiving the trigger, and generates the RSSI vector for each smart device.

• When the server receives the trigger, it also receives the number of the groups. The server runs clustering analysis using k-means algorithm and given RSSI vectors. The server sends the result of group identification to each smart devices.

Using clustering analysis allows to calculate grouping without measuring xy coordinate of the BLE beacon’s location. Although we choose the appropriate distances between BLE beacons, the BLE beacons does not require to locate the center of the group. These characteristics enable the simple setup of BLE beacons in the room.

Page 3: Location-Based Group Clustering Using BLE Beacons in an ... · scheme and readily available using Bluetooth Low Energy (BLE) workshop participants beacons on standard smart devices

2016 International Conference on Indoor Positioning and Indoor Navigation (IPIN), 4-7 October 2016, Alcalá de Henares, Spain

Fig. 4. RSSI variation over 30 seconds.

Fig. 3. The layout of the experiment environment.

C. Clustering Algorithm To obtain the group identification result directly from the

set of the RSSI vectors on each smart device with a simple algorithm, we choose k-means algorithm, the most common nonhierarchical clustering algorithm [12]. The k-means algorithm is a local optimal algorithm and the clustering result is affected by the configuration of initial centroid [13]. In this study, at first we randomly allocated initial cluster to objects i.e. the RSSI vectors, and then derived the initial centroid from the initial clusters. The clustering algorithm implemented on this study is as follows.

1. Each RSSI vector is allocated to one of K clusters so that each cluster has at least one object.

2. The centroids are calculated. The centroid of kth cluster µk is given by:

!" = $%"&

%=1(% $%"

&

%=1 (1)

where xn is the RSSI vector, N is the number of RSSI vectors, and rnk is 1 if xn is allocated to kth cluster and 0 for other cases.

3. The Euclidean distance between one RSSI vector and each centroid !"-$% is calculated. The RSSI vector is reallocated to the cluster having closest centroid. This step is repeated for all N RSSI vectors.

4. Steps 2 and 3 are repeated until no change of allocated cluster occurs on all RSSI vectors. The sum of squared error SSE is calculated from the result:

!!" = $%& '%-)& 2+

&=1

-

%=1 (2)

5. Steps 1 to 4 are repeated for all possible pattern of initial clusters. The clustering result having minimum SSE is selected as the final result.

V. EVALUATION

A. Experimental Environment We conducted measurement of BLE beacon’s RSSI in a

7.2m x 6.8m lecture room. BLE beacons and smart devices were placed on the table as shown in Fig. 3. This layout of smart devices represents that there are two teams of participants, and therefore each smart device should be classified into one of two groups. Three devices in the same group were placed in side by side. The distance between the locations of two smart device groups was approximately 3m.

2 BLE beacons were located on the table, and the signal strength was set to +4dBm; these BLE beacons are compliant

with iBeacon standard. We used a total of 6 iOS devices (iPhone 6s, iPhone 5s, iPhone 4s, iPad Air, iPad Air, and iPad), and the iOS version was 9.3.2.

B. Results Fig. 4 shows an example of RSSI fluctuations during 30

seconds monitord on a device; iOS devices recorded RSSI at a rate of 1 Hz. Since more than 10 dB of variation can be recognized, the real time values are no longer available for clustering of group identification.

As a countermeasure for the RSSI fluctuations, we evaluated a smoothing by taking the mean value in 10 seconds. Fig. 5 shows six sets of mean RSSI on each smart device. Even after smoothing is executed, variation of the mean RSSI vector on the same smart device is recognized in Fig. 5. Then we executed clustering analysis using the algorithm shown in the previous section for each of the six data sets. Although almost of classification were correct, one classification error was observed on the smart device F in the clustering result with the 2nd data set. Despite several tries of re-clustering by changing initial centroids, the result was not changed; remained as an error.

Next, we composed smoothing with 15 second time window from the same RSSI data set. Fig. 6 shows four sets of mean RSSI on each smart device, and this time the classification by k-means algorithm was perfect.

Page 4: Location-Based Group Clustering Using BLE Beacons in an ... · scheme and readily available using Bluetooth Low Energy (BLE) workshop participants beacons on standard smart devices

2016 International Conference on Indoor Positioning and Indoor Navigation (IPIN), 4-7 October 2016, Alcalá de Henares, Spain

Fig. 5. Clusters of the RSSI (10 seconds of time window). Fig. 6. Clusters of the RSSI (15 seconds of time window).

C. Discussion Obtaining suitable RSSI vectors is the key for reaching the

correct classification. Multipath fading and switching of the channel for beacon signal are possible factors to cause the RSSI fluctuations. Although smoothing RSSI value sequence is valid to resist the fading impact, for the real-time response of the system, an appropriate time window for smoothing should be found out.

The fading impact might be more severe condition with the distance between groups closer than this experiment, or the distance between smart devices in same group apart. Meanwhile, it is worth evaluating in case that the distance between BLE beacons is changed and three or more BLE beacons are used.

Since we did not execute any kind of calibration, individual variability between smart devices regarding receiving characteristic of BLE beacon’s signal is possible as a candidate of the cause of classification error. Having any calibration process in use case scenario may be effective to absorb the individual variability.

VI. CONCLUSION AND FUTURE WORK We have proposed a method to classify and identify groups

of smart devices in indoor space that layout changes frequently. BLE beacons and clustering analysis by k-means algorithm for RSSI monitored on smart devices are used. In future work, we plan to find out reasonable smoothing time window, the number of equipped BLE beacons, and the distance between BLE beacons. Furthermore, we also plan to examine if any hierarchical clustering algorithm is feasible to detect the number of the smart device’s group automatically.

REFERENCES [1] Y. Suo, N. Miyata, H. Morikawa, T. Ishida, and Y. Shi, “Open smart

classroom: Extensible and scalable learning system in smart space using web service technology,” IEEE Trans. Knowl. Data Eng., vol. 21, no. 6, pp. 814–828, 2009.

[2] X. Wang, J. S. Dong, C. Chin, R. Hettiarachchi, and D. Zhang, “Semantic Space : An Infrastructure for,” 2004.

[3] Keio University, “Enhancing Development of Global Entrepreneur Program.” Available: http://edge.keio.ac.jp/, [Accessed 2016]

[4] Edo-Tokyo Museum, “Traditional culture experience program for foreign visitors.” Available: http://www.edo-tokyo-museum.or.jp/en/event/, [Accessed 2016]

[5] Y. Wang, J. Wei, and K. Vangury, “Bring your own device security issues and challenges,” 2014 IEEE 11th Consum. Commun. Netw. Conf., pp. 80–85, 2014.

[6] N. Choudhury, V. Tamarapalli, and S. Bhattacharya, “An ICT-Based System to Improve the Learning Experience in a Large Classroom,” 2015 IEEE Seventh Int. Conf. Technol. Educ., pp. 27–30, 2015.

[7] Apple Inc., “iBeacon for Developers.” Available: https://developer.apple.com/ibeacon/, [Accessed 2016]

[8] Google Inc., “Google Beacons.” Available: https://developers.google.com/beacons/, [Accessed 2016]

[9] R. Faragher and R. Harle, “An Analysis of the Accuracy of Bluetooth Low Energy for Indoor Positioning Applications,” Proc. 27th Int. Tech. Meet. Satell. Div. Inst. Navig. (ION GNSS+ 2014), pp. 201–210, 2014.

[10] L. Pei, R. Chen, J. Liu, T. Tenhunen, H. Kuusniemi, and Y. Chen, “Inquiry-Based Bluetooth Indoor Positioning via RSSI Probability Distributions,” in 2010 Second International Conference on Advances in Satellite and Space Communications, 2010, pp. 151–156.

[11] C. W. Shen, Y. C. J. Wu, and T. C. Lee, “Developing a NFC-equipped smart classroom: Effects on attitudes toward computer science,” Comput. Human Behav., vol. 30, no. January 2014, pp. 731–738, 2014.

[12] J. MacQueen, "Some methods for classification and analysis of multivariate observations," Proceedings of the fifth Berkeley symposium on mathematical statistics and probability, vol. 1, no. 14, pp. 281-297, 1967.

[13] D. Steinley and M. J. Brusco, "Initializing k-means batch clustering: A critical evaluation of several techniques," Journal of Classification vol. 24, no. 1, pp. 99-121, 2007.