- Email: [email protected]
Tracking a moving user in indoor environments using Bluetooth low energy beacons
Journal of Biomedical Informatics 98 (2019) 103288
Contents lists available at ScienceDirect
Journal of Biomedical Informatics journal homepage: www.elsevier.com/locate/yjbin
Tracking a moving user in indoor environments using Bluetooth low energy beacons
T
Didi Suriana, , Vitaliy Kima, Ranjeeta Menonb,c, Adam G. Dunna, Vitali Sintchenkob,c, Enrico Coieraa ⁎
a
Centre for Health Informatics, Australian Institute of Health Innovation, Macquarie University, Sydney, Australia Sydney Medical School and Marie Bashir Institute for Infectious Diseases and Biosecurity, The University of Sydney, Sydney, Australia c Centre for Infectious Disease and Microbiology-Public Health, ICPMR, Westmead Hospital, Sydney, Australia b
ARTICLE INFO
ABSTRACT
Keywords: Bluetooth low energy Indoor positioning Location tracking BLE beacon
Background: Bluetooth low energy (BLE) beacons have been used to track the locations of individuals in indoor environments for clinical applications such as workflow analysis and infectious disease modelling. Most current approaches use the received signal strength indicator (RSSI) to track locations. When using the RSSI to track indoor locations, devices need to be calibrated to account for complex interference patterns, which is a laborious process. Our aim was to investigate an alternative method for indoor location tracking of a moving user using BLE beacons in dynamic indoor environments. Methods and Materials: We developed a new method based on the received number of signals indicator (RNSI) and compared it to a standard RSSI-based method for predicting a user's location. Experiments were performed in an office environment and a tertiary hospital. Both RNSI and RSSI were compared at various distances from BLE beacons. In moving user experiments, a user wearing a beacon walked from one location to another based on a pre-defined route. Performance in predicting user locations was measured based on accuracy. Results: RNSI values decreased substantially with distance from the BLE beacon than RSSI values. Moving user experiments in the office environment demonstrated that the RNSI-based method produced higher accuracy (80.0%) than the RSSI-based method (76.2%). In the hospital, where the environment may introduce signal quality problems due to increased signal interference, the RNSI-based method still outperformed (83.3%) the RSSI-based method (51.9%). Conclusions: Our results suggest that the RNSI-based method could be useful to track the locations of a moving user without involving complex calibration, especially when deploying within a new environment. RNSI has the potential to be used together with other methods in more robust indoor positioning systems.
1. Introduction Monitoring the locations or paths that individuals traverse in an indoor environment has become an important element in applications that involve contextual information, such as analysing the workflow in a clinical environment [1–4] and modelling the spread of infectious diseases/hospital acquired-infections [5–7]. Traditional methods such as direct observation or surveys are inefficient, expensive, and time consuming, especially where monitoring is done over time or where many people need to be tracked at once [8]. Despite the fact that there are many technologies already developed for indoor location tracking (localisation) [9], we know of no devices and methods that work as well as a global positioning system (GPS) does in outdoor settings [10]. Most
⁎
indoor tracking systems use radio frequency (RF) based technologies, but these tend to be unreliable because they are sensitive to the surrounding environment of the transmitters and the sensors. Previous solutions to this challenge have involved various technologies in the implementation, can have high deployment cost, and are often only tested in controlled and closed environments [11]. In dynamic environments where activities and RF signal interference can change over time, understanding the characteristics of RF signal may help us better design a more effective indoor localisation solution. Currently, many of the proposed RF-based methods use technologies that might not always be feasible for some environments. For example, methods that rely on Wi-Fi [12–14] may not be applicable for an environment like a hospital, as Wi-Fi signal may not be available in all
Corresponding author. E-mail address: [email protected] (D. Surian).
https://doi.org/10.1016/j.jbi.2019.103288 Received 2 April 2019; Received in revised form 29 August 2019; Accepted 7 September 2019 Available online 09 September 2019 1532-0464/ © 2019 Elsevier Inc. All rights reserved.
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
areas. Other RF methods that show promise for indoor localisation are based on RFID [15–17] and Bluetooth low energy (BLE) [18–21]. BLE devices (beacons) may represent a more practical solution than RFID in applications with many users and locations because of their low cost and low power consumption. Most RF-based indoor localisation methods—including methods developed for BLE—make use of signal information, such as the travel time of a signal from a transmitter to a receiver [22,23], the direction of propagation of a signal [24], and the relative strength of a signal (received signal strength indicator, RSSI) [25]. RSSI-based methods are considered simpler and lower cost to implement. Despite a growing literature on the use of RSSI for indoor localisation, previously reported methods tend to use algorithms that require time consuming and costly calibration. For example, fingerprint-based methods [26,27] require signal patterns from each location to be collected in the beginning and would not work well if new sensed signals have different patterns from the signal patterns used to train the algorithms. Methods that rely on the information from other beacons such as a weighted k-nearest neighbors [28,29] may suffer in performance if the variable k is not calibrated correctly. One common perception is that stronger signals will be sensed by a receiver from the closest BLE beacons. However, in environments where signal interference might cause problems, additional calibrations are needed to filter out signals from incorrect locations (outlier signals). The potential for BLE signal strength to vary over time even at the same distance is an additional challenge [30]. A naïve method to handle outlier signals would be to use a threshold value, so signals below the threshold will be filtered out. A threshold value for one location may be different from another location. Further steps would be needed to learn threshold values whenever a new location is encountered, which is less practical. Given the limitations of existing methods for indoor localisation, our aim was to develop and test a new method using BLE beacons that is practical in the deployment, yet effective for location tracking of moving users in environments where the interference changes over time. To do this, we made use of the number of signals received by a sensor—we refer to this as the received number of signals indicator (RNSI)—rather than the established standard RSSI. Our RNSI-based method is simple: a user is predicted to be at a location that gives the greatest number of signals within a period of time.
Fig. 1. BLE beacons: (a) the wearable beacon, and (b) the location beacon.
Fig. 2. System architecture. The arrow represents the data flow: (i) the data from location beacons (at each location) is sensed by a wearable beacon (worn by a user) and saved in its local memory, (ii) once the wearable beacon senses the signal from the base station, it sends the stored data to the base station, and (iii) the base station sends the data to the remote database server.
The system architecture for experiments in this study included a set of location beacons, a single wearable beacon worn by a user using a lanyard over the chest, and a single base station connected to a database server (Fig. 2). In our experiments, we configured location beacons to only broadcast signals, to be sensed by the wearable beacon worn by a user. We configured the location beacons to broadcast a signal once every 100 ms at −20 dBm. The base station also transmits BLE signal used for communication with the wearable beacons. The wearable beacons were configured to sense (scanning operation) a signal from the location beacons once every 125 ms and initiate a communication with the base station once every 20 s whenever the wearable beacons sensed the signal broadcast by the base station. The wearable beacon stored data about when it received the signal and RSSI from the location beacons in its local memory. A node.js [32] script ran on the base station to automatically send any data received from the wearable beacons to a remote MySQL (v5.5) database server for further processing.
2. Methods and data collection 2.1. Hardware and system configuration In this study, we used four hardware components: a wearable beacon, a location beacon, Raspberry Pi Zero W (a small single-board computer that has wireless LAN and Bluetooth connectivity) [31], and a database server. The wearable beacon was worn by a user in the experiments and received signals broadcast from location beacons, which were assigned to static locations in the environment. The Raspberry Pi Zero W (hereafter, the base station) facilitated the storage of data from the wearable beacons and transfer to the database server for further processing. We chose commercial BLE beacons from Shenzhen Minew Technologies Co., Ltd powered by 2.4 GHz wireless system on chip (SoC) nRF52-series chipset from Nordic Semiconductor (Fig. 1a) as the wearable beacons because they are battery-powered (CR2032 battery), small, light, affordable, and have a lanyard so they could be directly worn by a user without additional physical modifications. BLE beacons from Shenzhen Wellcore Technologies Co, Ltd. powered by 2.4 GHz wireless system on chip (SoC) nRF51-series chipset from Nordic Semiconductor were selected for location beacons (Fig. 1b). The location beacons are bigger in size than the wearable beacons, as they require a bigger battery (CR2477 battery) for a longer uptime, feature a power button, and are light and suitable for ceiling-mounted installation.
2.2. The RNSI-based method for indoor localisation using BLE beacons The RNSI-based method does not depend on the number of location beacons, but relies on the number of signals to determine the location of a user, i.e. the location that gives the highest total number of signals. The RNSI-based method was initiated by iteratively merging the presence of signals separated by five-second interval from the same location into the same group (Fig. 3a). Given that the distance between two locations were 5–8 m and the walking pace for humans is 1.4 m/s on average [33,34], the five-second interval was considered suitable to separate locations if a user moved from one location to another. It is possible that a wearable beacon senses signals from different locations at the same time. We aimed to remove signals that may trigger an incorrect prediction by comparing the total number of RNSI between signal groups that existed in the same period of time—i.e., the intersection of groups of signals. We removed the intersections with only one signal from one or more locations (e.g. the 3rd and 4th intersections in Fig. 3b). For example, the first intersection shown in Fig. 3b is labelled as location 1 because only signals from location 1 exist in the 2
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 3. The RNSI-based method for indoor localisation: (a) grouping the presence of signals separated by five-second interval from each location, (b) finding intersections among the signal groups. The black circle represents if a signal is present from the respective location, the background colour represents a group for each location, and the coloured dashed line rectangle represents a group intersection. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 4. The pre-defined routes for moving user experiments performed at (a) AIHI, and (b) CIDM.
intersection. The second intersection resulted location 2 as the predicted location as the total RNSI for location 2 is 12 (3 + 0 + 5 + 4) while the total RNSI for location 1 is only 8 (2 + 2 + 2 + 2).
At AIHI, we performed signal pattern analysis in a meeting room, a corridor, and at seven pre-defined locations. Similarly, signal patterns were collected from five pre-defined locations at CIDM. The location beacon was mounted on the ceiling (approximately 2.7 m from the floor) with the power button side facing the floor in all experiments. In the AIHI meeting room and corridor, we used one location beacon and one wearable beacon. The location beacon was mounted on the ceiling in the centre of the meeting room in order to capture the signal patterns from four directions. The wearable beacon and the base station were placed on a chair (approximately 1.1 m from the floor). For the experiments at several pre-defined locations, we used two location beacons per location and one wearable beacon was worn by a user. A detailed layout diagram for each location is provided in the Appendix.
2.3. Signal pattern analysis experiments The aim of signal pattern analysis is to understand BLE signal propagation characteristics with respect to the distance from a location beacon in an indoor environment. Since the signal propagation might be affected by the environment (e.g. due to shadowing, multipath propagation, etc.), we aimed to better understand and get more comparisons about RSSI and RNSI patterns from different environments. We performed signal pattern analysis in two different environments: the Australian Institute of Health Innovation (AIHI), Macquarie University; and the Centre for Infectious Diseases & Microbiology (CIDM) - Public Health, Westmead Hospital. During the data collection, we had no control over these environments, and dynamic elements included moving staff during the experiments and interference/noise from furniture, building structure, or electronic devices such as medical devices. We took the average value per second for RSSI and the actual count per second for RNSI.
2.4. Moving user experiments In the moving user experiments, one user (DS) wore the wearable beacon and moved from one location to another based on a pre-defined route (Fig. 4). The time when the user walked past the location beacons at each location was recorded, which was used as our ground truth. The predicted locations were considered correct if the method predicted the 3
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 5. Signal patterns for RNSI and RSSI in the AIHI meeting room. (a) The positions (shown in letters) where the signal patterns were collected (the blue circle in the middle of layout diagram represents the location beacon, where the protruding side shows the top side of the beacon—see inset on the upper left side), (b)-(e): the results for the four directions of the location beacon. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 6. Signal patterns for RNSI and RSSI in the AIHI corridor. (a)–(d): the results for the four directions of the location beacon.
user was at the respective locations at the recorded time. Both RNSI- and RSSI-based methods were applied to predict locations at AIHI and CIDM. For the RNSI-based method, the user was assigned to the location that gives the greatest number of signals within a period of time. For the RSSI-based method, the user was assigned to the location that gives the strongest signals within a period of time. We compared the performance of RSSI- and RNSI-based methods in predicting the user’s visited locations.
3. Results 3.1. Signal pattern analysis results In the AIHI meeting room (Fig. 5), RNSI showed a decreasing value at the 2 m distance (position B, E, H, and K) compared to at the 1 m distance (position A, D, G, and J). RNSI attained the same downtrend in value only at position F and I (3 m distance). Compared to RNSI, RSSI
4
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 7. Comparison of RNSI and RSSI from the seven pre-defined locations in AIHI: (a)–(g): signal patterns at location 1–7, where the horizontal axis represents the sensed signals from all locations that could still be sensed by the wearable beacon, (h) the pre-defined locations.
only showed a decreasing pattern at positions B, E, and H (2 m distance), and F and I (3 m distance). Both RNSI and RSSI increased at positions C and L (at 3 m distance), which may be the “error spot” in that area caused by multipath propagation problem [34]. In the AIHI corridor (Fig. 6), we found that RNSI consistently decreased as the distance increased. RNSI showed a significant drop in values at the distance between 4 m and 6 m from the location beacon. In general, RSSI also decreased after 4 m from the location beacon, but inconsistently. These results show that RNSI may be more reliable than RSSI for determining the location of a user in an indoor environment. From the seven pre-defined locations in AIHI, we found that RNSI from the user’s location was higher than RNSI from other locations (Fig. 7). While RSSI produced a similar general pattern of decreasing values with distance, the differences in values between the correct location and other locations were more distinct for RNSI compared to RSSI. For example, when the user was at location 1 (Fig. 7a), the RSSI values were still high from location 2, 3, and location 4 (RSSILocation_1: −82.9 dBm, RSSILocation_2: −87.4 dBm, RSSILocation_3: −86.7 dBm, RSSILocation_4: −87.1 dBm), while RNSI values from these locations showed significant differences in values (RNSILocation_1: 66, RNSILocation_2: 14, RNSILocation_3: 3, RNSILocation_4: 17). Similarly, when the user was at location 3, RSSI value from the location beacons at location 3 was slightly higher than from other locations, while the RNSI value at location 3 was notably high. These results suggest that RNSI is more sensitive to the distance than RSSI.
From the five pre-defined locations in CIDM, we also found that the RNSI value from the correct location was substantially higher than from the other locations. On the contrary, RSSI did not always show a similar pattern with RNSI. For example, when the user was at location 1, the RNSI value from location 1 was the highest but the RSSI value from location 2 was the highest. Also at the location 1 (a room), the RSSI from location 2 (outside the room) and location 5 (a different room) were still sensed with relatively high average RSSI values (RSSILocation_1: −83.3 dBm, RSSILocation_2: −81.5 dBm, and RSSILocation_5: −84.3 dBm). We noticed that location 4 gave different signal patterns (Fig. 8d). There was medical laboratory equipment near locations 4, and a WiFi router near the location beacons at location 4. These devices appear to have caused signal interference, though other aspects of the environment (e.g. the materials of the walls, furniture, etc.) may have also contributed to interference patterns [35]. 3.2. Moving user experiment results In the moving user experiments, we found that RNSI values were higher when a user stayed in one location for a while; but dropped when a user walked past the location beacons. The RSSI-based method correctly predicted 16 positions according to the pre-defined route and 5 locations were returned incorrectly (76.2% accuracy) from the moving user experiment at AIHI (Fig. 9a). The results showed that the 5
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 8. Comparison of RNSI and RSSI from the five pre-defined locations in CIDM: (a)–(e): signal patterns at location 1–5, where the horizontal axis represents the sensed signals from all locations that could still be sensed by the wearable beacon, (f) the pre-defined locations.
Fig. 9. Plots of the signals sensed by the wearable beacon and the predicted locations for the moving user experiment at AIHI using: (a) the strongest RSSI-based method, (b) the RNSI-based method. 6
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 10. Plots of the signals sensed by the wearable beacon and the predicted locations for the moving user experiment at CIDM using: (a) the strongest RSSI-based method, (b) the RNSI-based method.
and CIDM were only performed once. It will be necessary in the future to replicate the same experiments in the same and different environments to demonstrate the efficacy of RNSI-based method. The elimination of incorrect signals may result in missing location information for location prediction in certain conditions, such as when only the signals from incorrect locations were sensed but the signals from the correct location were blocked by the user’s body, building structure, or furniture. While this was expected, we did not attempt to maximise signal sensing by increasing the number of location beacons or measuring the strategic placement of location beacons in both of our moving user experiments. We mounted the location beacons on the ceiling to maintain a clear line of sight between the location beacons and the wearable beacon, but we did not compare the performance if the location beacons were mounted on the wall or other positions. An incorrect location prediction could also happen if there were location beacons that failed to work, resulting in other locations being falsely identified. We decided to use two location beacons for each location to minimise this risk of failure. Future work would benefit from considering approaches that can robustly handle these situations. A further limitation in this work is that we performed our moving user experiments with only one user wearing a beacon, in an experimental setting, and did not consider how the use of additional wearable beacons or location beacons might affect the reception of BLE signal [36]. Wireless devices that operate in the radio spectrum from 2.400 GHz to 2.485 GHz can potentially cause signal interference with the BLE beacons [35]. It will be important to study performance over a longer time period and involve more users and locations.
incorrectly detected locations using the RSSI-based method tend to have small RNSI values and small RSSI values (weaker than −85 dBm), although some correct locations might also give weak signals (weaker than −85 dBm). On the other hand, the RNSI-based method successfully removed all signals from the incorrect locations and correctly returned 12 locations according to the pre-defined route but missed location 3 and 6 as there were not enough signals to determine the locations (80% accuracy) (Fig. 9b). In the moving user experiment performed at CIDM, the strongest RSSI-based method identified 13 incorrect locations and returned 14 correct locations according to the pre-defined route (51.9% accuracy) (Fig. 10a). Like the results from the moving user experiments at AIHI, RNSI values from incorrect locations were lower than the RNSI values from the correct locations. However, compared to the moving experiments in AIHI, most of signals for incorrect locations were not all weak (stronger than −85 dBm). The RNSI-based method removed the incorrect signals and kept the correct ones (Fig. 10b). As a result of the signal removal process, the method missed location 2 and location 4 (orange labels) but correctly identified the users in 10 locations (83.3% accuracy). Further analysis is provided in the Appendix. 4. Discussion 4.1. Limitations As we were limited by the number of devices and time (especially at CIDM) to perform experiments, the moving user experiments at AIHI 7
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
4.2. Implications and future research directions [3]
This work has implications for many applications using indoor localisation. With the simplicity of RNSI-based method, we found that it was possible to predict a user’s location with higher accuracy by removing signals that may contribute to an incorrect prediction. As the RNSI-based method relies on the number of signals sensed within a period of time and not from the signal strength, the use of low signal strength may prolong battery life of the beacons. Future implementations could take the advantages of using different signal strengths and signal transmission rates depending on the location. The RNSI represents a relatively simple additional measure that can be used to complement any indoor localisation methods that currently rely on RSSI, especially in applications that involve human movements. Both RNSI and RSSI could be used together as part of a more complex method to determine a location of a user. For example, when the locations are quite close to one another and the location beacons at each location have different signal strengths and signal transmission rates, both RNSI and RSSI could be used to determine which location beacons the user is close to. Future studies may also benefit from contextual information to understand about how humans move and interact with each other in indoor environments using RFID signals or accelerometers [37–39].
[4] [5] [6]
[7]
[8] [9] [10]
[11]
5. Conclusion
[12]
In this work, we demonstrated a novel method to detect the location of a moving user in indoor and dynamic environments. We introduced RNSI as a new measurement, and proposed a practical method for removing signals that could lead to incorrect location prediction. Our results suggest that the RNSI-based method produced more consistent patterns than the RSSI-based method for determining locations. Our experiments showed that the RNSI-based method effectively filtered out signals from incorrect locations, outperforming a baseline method using the strongest RSSI. We showed that while the signals from incorrect locations are sometimes stronger than from correct locations, they tended to have smaller RNSI values. Our approach was also relatively robust in a healthcare setting where the environment is more dynamic and has more aspects that could cause signal interference than an office environment.
[13] [14] [15] [16]
[17]
[18]
Declaration of Competing Interest
[19]
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
[20]
Acknowledgement [21]
VS and EC report funding from the NHMRC Centre of Research Excellence in Emerging Infection Diseases (APP1102962). DS and VK designed the research, undertook the device installation, and experiments. DS designed the methods and drafted the manuscript. VK configured all devices. All authors critically reviewed the manuscript.
[22] [23]
Appendix A. Supplementary material Supplementary data to this article can be found online at https:// doi.org/10.1016/j.jbi.2019.103288.
[24]
References
[25]
[1] S. Malhotra, D. Jordan, E. Shortliffe, V.L. Patel, Workflow modeling in critical care: piecing together your own puzzle, J. Biomed. Inform. 40 (2007) 81–92, https://doi. org/10.1016/j.jbi.2006.06.002. [2] E.A. Fry, L.A. Lenert, MASCAL: RFID tracking of patients, staff and equipment to
[26]
8
enhance hospital response to mass casualty events, AMIA Annu. Symp. Proc. 261–5 (2005). A. Vankipuram, S. Traub, V.L. Patel, A method for the analysis and visualization of clinical workflow in dynamic environments, J. Biomed. Inform. 79 (2018) 20–31, https://doi.org/10.1016/j.jbi.2018.01.007. J. Frisby, V. Smith, S. Traub, V.L. Patel, Contextual Computing: a Bluetooth based approach for tracking healthcare providers in the emergency room, J. Biomed. Inform. 65 (2017) 97–104, https://doi.org/10.1016/j.jbi.2016.11.008. J. Stehlé, N. Voirin, A. Barrat, C. Cattuto, V. Colizza, L. Isella, et al., Simulation of an SEIR infectious disease model on the dynamic contact network of conference attendees, BMC Med. 9 (2011) 1–15, https://doi.org/10.1186/1741-7015-9-87. P. Vanhems, A. Barrat, C. Cattuto, J.-F. Pinton, N. Khanafer, C. Régis, et al., Estimating potential infection transmission routes in hospital wards using wearable proximity sensors, PLoS ONE 8 (2013) e73970, , https://doi.org/10.1371/journal. pone.0073970. T. Hornbeck, D. Naylor, A.M. Segre, G. Thomas, T. Herman, P.M. Polgreen, Using sensor networks to study the effect of peripatetic healthcare workers on the spread of hospital-associated infections, J. Infect. Dis. 206 (2012) 1549–1557, https://doi. org/10.1093/infdis/jis542. M. Vankipuram, K. Kahol, T. Cohen, V.L. Patel, Toward automated workflow analysis and visualization in clinical environments, J. Biomed. Inform. 44 (2011) 432–440, https://doi.org/10.1016/j.jbi.2010.05.015. G. Deak, K. Curran, J. Condell, A survey of active and passive indoor localisation systems, Comput. Commun. 35 (2012) 1939–1954, https://doi.org/10.1016/j. comcom.2012.06.004. D. Lymberopoulos, J. Liu, X. Yang, R.R. Choudhury, V. Handziski, S. Sen, A realistic evaluation and comparison of indoor location technologies: experiences and lessons learned, The 14th International Conference on Information Processing in Sensor Networks (IPSN). Seattle, Washington: ACM New York, NY, USA, 2015, pp. 178–189, , https://doi.org/10.1145/2737095.2737726. H. Liu, H. Darabi, P. Banerjee, J. Liu, Survey of wireless indoor positioning techniques and systems, IEEE Trans. Syst., Man, Cybernet. 37 (2007) 1067–1080, https://doi.org/10.1109/TSMCC.2007.905750. C. Wu, Z. Yang, Y. Liu, W. Xi, WILL: wireless indoor localization without site survey, IEEE Trans. Parallel Distrib. Syst. 24 (2013) 839–848, https://doi.org/10.1109/ TPDS.2012.179. Meng W, He Y, Deng Z, Li C. Optimized access points deployment for WLAN indoor positioning system. IEEE Wireless Communications and Networking Conference (WCNC). Shanghai, China: IEEE; 2012, p. 2457-61. 10.1109/WCNC.2012.6214209. C. Feng, W.S.A. Au, S. Valaee, Z. Tan, Received-signal-strength-based indoor positioning using compressive sensing, IEEE Trans. Mob. Comput. 11 (2012) 1983–1993, https://doi.org/10.1109/TMC.2011.216. H. Xu, Y. Ding, P. Li, R. Wang, Y. Li, An RFID indoor positioning algorithm based on Bayesian probability and k-nearest neighbor, Sensors. 17 (2017) E1806, https:// doi.org/10.3390/s17081806. L.M. Ni, Y. Liu, Y.C. Lau, A.P. Patil, LANDMARC: indoor location sensing using active RFID, in: Proceedings of the First IEEE International Conference on Pervasive Computing and Communications (PerCom). IEEE, Fort Worth, TX, USA, 2003. 10. 1109/PERCOM.2003.1192765. D. Lieckfeldt, J. You, D. Timmermann, Exploiting RF-Scatter: human localization with bistatic passive UHF RFID-systems, in: IEEE International Conference on Wireless and Mobile Computing, Networking and Communications. IEEE, Marrakech, Morocco, 2009. p. 179–184. 10.1109/WiMob.2009.39. Z. Jianyong, L. Haiyong, C. Zili, L. Zhaohui, RSSI based Bluetooth Low Energy indoor positioning, in: International Conference on Indoor Positioning and Indoor Navigation (IPIN). IEEE, Busan, South Korea; 2014. 10.1109/IPIN.2014.7275525. V. Chandel, N. Ahmed, S. Arora, A. Ghose, InLoc: an end-to-end robust indoor localization and routing solution using mobile phones and BLE beacons, International Conference on Indoor Positioning and Indoor Navigation (IPIN): IEEE, 2016, https://doi.org/10.1109/IPIN.2016.7743592. N. Kitbutrawat, S. Kajita, H. Yamaguchi, T. Higashino, Location identification of BLE-embedded HVACs for smart building management, in: The 14th International Conference on Intelligent Environments (IE), IEEE, Rome, Italy, 2018. 10.1109/IE. 2018.00019. F. Zafari, I. Papapanagiotou, T.J. Hacker, A novel Bayesian filtering based algorithm for RSSI-based indoor localization, in: IEEE International Conference on Communications (ICC). IEEE, Kansas City, MO, USA, 2018. 10.1109/ICC.2018. 8423012. N. Patwari, J.N. Ash, S. Kyperountas, A.O. Hero, R.L. Moses, N.S. Correal, Locating the nodes: cooperative localization in wireless sensor networks, IEEE Signal Process. Mag. 22 (2005) 54–69, https://doi.org/10.1109/MSP.2005.1458287. X. Li, K. Pahlavan, M. Latva-aho, M. Ylianttila, Comparison of indoor geolocation methods in DSSS and OFDM Wireless LAN systems, in: Proceedings of the 52nd Vehicular Technology Conference. IEEE, Boston, MA, USA, 2000, pp. 3015–3020. 10.1109/VETECF.2000.886867. R. Peng, M.L. Sichitiu, Angle of arrival localization for wireless sensor networks, in: Proceedings of the 3rd Annual IEEE Communications Society on Sensor and Ad Hoc Communications and Networks. IEEE, Reston, VA, USA, 2006, pp. 374–382. 10. 1109/SAHCN.2006.288442. K. Heurtefeux, F. Valois, Is RSSI a good choice for localization in wireless sensor network? In: Proceedings of the IEEE 26th International Conference on Advanced Information Networking and Applications, IEEE, Fukuoka, Japan, 2012, pp. 732739. 10.1109/AINA.2012.19. S. Subedi, J.-Y. Pyun, Practical fingerprinting localization for indoor positioning system by using beacons, Sensors (2017) 1–16, https://doi.org/10.1155/2017/ 9742170.
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al. [27] Z. Iqbal, D. Luo, P. Henry, S. Kazemifar, T. Rozario, Y. Yan, et al., Accurate real time localization tracking in a clinical environment using Bluetooth Low Energy and deep learning, PLoS ONE 13 (2018) e0205392, , https://doi.org/10.1371/journal. pone.0205392. [28] S. Lee, B. Cho, B. Koo, S. Ryu, J. Choi, S. Kim, Kalman filter-based indoor position tracking with self-calibration for RSS variation mitigation, Int. J. Distrib. Sens. Netw. 11 (2015) 1–10, https://doi.org/10.1155/2015/674635. [29] P. Kriz, F. Maly, T. Kozel, Improving indoor localization using Bluetooth Low Energy beacons, Mobile Inform. Syst. 2016 (2016) 1–11, https://doi.org/10.1155/ 2016/2083094. [30] K. Urano, K. Hiroi, K. Kaji, N. Kawaguchi, A location estimation method using BLE tags distributed among participants of a large-scale exhibition, Proceedings of the 13th International Conference on Mobile and Ubiquitous Systems: Computing Networking and Services (MOBIQUITOUS). Hiroshima, Japan: ACM New York, NY, USA, 2016, pp. 124–129, , https://doi.org/10.1145/3004010.3004051. [31] Raspberry Pi Zero W. [Available from https://wwwrapsberrypiorg/products/ raspberry-pi-zero-w]. [32] Node.js. [Available from: https://nodejsorg/]. [33] N. Carey, Establishing pedestrian walking speeds, Portland State University, 2005. [34] D.-Y. Kim, S.-H. Kim, D. Choi, S.-H. Jin, Accurate indoor proximity zone detection based on time window and frequency with Bluetooth Low Energy, Proc. Comput.
Sci. 56 (2015) 88–95, https://doi.org/10.1016/j.procs.2015.07.199. [35] J. Frisby, Contextual Computing: tracking healthcare providers in the Emergency Department via Bluetooth beacons, Arizona State University, 2015. [36] D.P.-Dd. Cerio, Á. Hernández, J.L. Valenzuela, A. Valdovinos, Analytical and experimental performance evaluation of BLE neighbor discovery process including non-idealities of real chipsets, Sensors 17 (2017) 499, https://doi.org/10.3390/ s17030499. [37] H. Li, C. Ye, A.P. Sample, IDSense: a human object interaction detection system based on passive, Proceedings of the 33rd ACM Conference on Human Factors in Computing Systems. Seoul, Republic of Korea: ACM New York, NY, USA, 2015, pp. 2555–2564, , https://doi.org/10.1145/2702123.2702178. [38] H. Li, P. Zhang, S.A. Moubayed, S.N. Patel, A.P. Sample, ID-Match: a hybrid computer vision and RFID system for recognizing individuals in groups, Proceedings of the 2016 ACM Conference on Human Factors in Computing Systems. ACM New York, NY, USA, San Jose, California, USA, 2016, pp. 4933–4944, , https://doi.org/ 10.1145/2858036.2858209. [39] A.R.J. Ruiz, F.S. Granja, J.C.P. Honorato, J.I.G. Rosas, Accurate pedestrian indoor navigation by tightly coupling foot-mounted IMU and RFID measurements, IEEE Trans. Instrum. Meas. 61 (2012) 178–189, https://doi.org/10.1109/TIM.2011. 2159317.
9
Contents lists available at ScienceDirect
Journal of Biomedical Informatics journal homepage: www.elsevier.com/locate/yjbin
Tracking a moving user in indoor environments using Bluetooth low energy beacons
T
Didi Suriana, , Vitaliy Kima, Ranjeeta Menonb,c, Adam G. Dunna, Vitali Sintchenkob,c, Enrico Coieraa ⁎
a
Centre for Health Informatics, Australian Institute of Health Innovation, Macquarie University, Sydney, Australia Sydney Medical School and Marie Bashir Institute for Infectious Diseases and Biosecurity, The University of Sydney, Sydney, Australia c Centre for Infectious Disease and Microbiology-Public Health, ICPMR, Westmead Hospital, Sydney, Australia b
ARTICLE INFO
ABSTRACT
Keywords: Bluetooth low energy Indoor positioning Location tracking BLE beacon
Background: Bluetooth low energy (BLE) beacons have been used to track the locations of individuals in indoor environments for clinical applications such as workflow analysis and infectious disease modelling. Most current approaches use the received signal strength indicator (RSSI) to track locations. When using the RSSI to track indoor locations, devices need to be calibrated to account for complex interference patterns, which is a laborious process. Our aim was to investigate an alternative method for indoor location tracking of a moving user using BLE beacons in dynamic indoor environments. Methods and Materials: We developed a new method based on the received number of signals indicator (RNSI) and compared it to a standard RSSI-based method for predicting a user's location. Experiments were performed in an office environment and a tertiary hospital. Both RNSI and RSSI were compared at various distances from BLE beacons. In moving user experiments, a user wearing a beacon walked from one location to another based on a pre-defined route. Performance in predicting user locations was measured based on accuracy. Results: RNSI values decreased substantially with distance from the BLE beacon than RSSI values. Moving user experiments in the office environment demonstrated that the RNSI-based method produced higher accuracy (80.0%) than the RSSI-based method (76.2%). In the hospital, where the environment may introduce signal quality problems due to increased signal interference, the RNSI-based method still outperformed (83.3%) the RSSI-based method (51.9%). Conclusions: Our results suggest that the RNSI-based method could be useful to track the locations of a moving user without involving complex calibration, especially when deploying within a new environment. RNSI has the potential to be used together with other methods in more robust indoor positioning systems.
1. Introduction Monitoring the locations or paths that individuals traverse in an indoor environment has become an important element in applications that involve contextual information, such as analysing the workflow in a clinical environment [1–4] and modelling the spread of infectious diseases/hospital acquired-infections [5–7]. Traditional methods such as direct observation or surveys are inefficient, expensive, and time consuming, especially where monitoring is done over time or where many people need to be tracked at once [8]. Despite the fact that there are many technologies already developed for indoor location tracking (localisation) [9], we know of no devices and methods that work as well as a global positioning system (GPS) does in outdoor settings [10]. Most
⁎
indoor tracking systems use radio frequency (RF) based technologies, but these tend to be unreliable because they are sensitive to the surrounding environment of the transmitters and the sensors. Previous solutions to this challenge have involved various technologies in the implementation, can have high deployment cost, and are often only tested in controlled and closed environments [11]. In dynamic environments where activities and RF signal interference can change over time, understanding the characteristics of RF signal may help us better design a more effective indoor localisation solution. Currently, many of the proposed RF-based methods use technologies that might not always be feasible for some environments. For example, methods that rely on Wi-Fi [12–14] may not be applicable for an environment like a hospital, as Wi-Fi signal may not be available in all
Corresponding author. E-mail address: [email protected] (D. Surian).
https://doi.org/10.1016/j.jbi.2019.103288 Received 2 April 2019; Received in revised form 29 August 2019; Accepted 7 September 2019 Available online 09 September 2019 1532-0464/ © 2019 Elsevier Inc. All rights reserved.
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
areas. Other RF methods that show promise for indoor localisation are based on RFID [15–17] and Bluetooth low energy (BLE) [18–21]. BLE devices (beacons) may represent a more practical solution than RFID in applications with many users and locations because of their low cost and low power consumption. Most RF-based indoor localisation methods—including methods developed for BLE—make use of signal information, such as the travel time of a signal from a transmitter to a receiver [22,23], the direction of propagation of a signal [24], and the relative strength of a signal (received signal strength indicator, RSSI) [25]. RSSI-based methods are considered simpler and lower cost to implement. Despite a growing literature on the use of RSSI for indoor localisation, previously reported methods tend to use algorithms that require time consuming and costly calibration. For example, fingerprint-based methods [26,27] require signal patterns from each location to be collected in the beginning and would not work well if new sensed signals have different patterns from the signal patterns used to train the algorithms. Methods that rely on the information from other beacons such as a weighted k-nearest neighbors [28,29] may suffer in performance if the variable k is not calibrated correctly. One common perception is that stronger signals will be sensed by a receiver from the closest BLE beacons. However, in environments where signal interference might cause problems, additional calibrations are needed to filter out signals from incorrect locations (outlier signals). The potential for BLE signal strength to vary over time even at the same distance is an additional challenge [30]. A naïve method to handle outlier signals would be to use a threshold value, so signals below the threshold will be filtered out. A threshold value for one location may be different from another location. Further steps would be needed to learn threshold values whenever a new location is encountered, which is less practical. Given the limitations of existing methods for indoor localisation, our aim was to develop and test a new method using BLE beacons that is practical in the deployment, yet effective for location tracking of moving users in environments where the interference changes over time. To do this, we made use of the number of signals received by a sensor—we refer to this as the received number of signals indicator (RNSI)—rather than the established standard RSSI. Our RNSI-based method is simple: a user is predicted to be at a location that gives the greatest number of signals within a period of time.
Fig. 1. BLE beacons: (a) the wearable beacon, and (b) the location beacon.
Fig. 2. System architecture. The arrow represents the data flow: (i) the data from location beacons (at each location) is sensed by a wearable beacon (worn by a user) and saved in its local memory, (ii) once the wearable beacon senses the signal from the base station, it sends the stored data to the base station, and (iii) the base station sends the data to the remote database server.
The system architecture for experiments in this study included a set of location beacons, a single wearable beacon worn by a user using a lanyard over the chest, and a single base station connected to a database server (Fig. 2). In our experiments, we configured location beacons to only broadcast signals, to be sensed by the wearable beacon worn by a user. We configured the location beacons to broadcast a signal once every 100 ms at −20 dBm. The base station also transmits BLE signal used for communication with the wearable beacons. The wearable beacons were configured to sense (scanning operation) a signal from the location beacons once every 125 ms and initiate a communication with the base station once every 20 s whenever the wearable beacons sensed the signal broadcast by the base station. The wearable beacon stored data about when it received the signal and RSSI from the location beacons in its local memory. A node.js [32] script ran on the base station to automatically send any data received from the wearable beacons to a remote MySQL (v5.5) database server for further processing.
2. Methods and data collection 2.1. Hardware and system configuration In this study, we used four hardware components: a wearable beacon, a location beacon, Raspberry Pi Zero W (a small single-board computer that has wireless LAN and Bluetooth connectivity) [31], and a database server. The wearable beacon was worn by a user in the experiments and received signals broadcast from location beacons, which were assigned to static locations in the environment. The Raspberry Pi Zero W (hereafter, the base station) facilitated the storage of data from the wearable beacons and transfer to the database server for further processing. We chose commercial BLE beacons from Shenzhen Minew Technologies Co., Ltd powered by 2.4 GHz wireless system on chip (SoC) nRF52-series chipset from Nordic Semiconductor (Fig. 1a) as the wearable beacons because they are battery-powered (CR2032 battery), small, light, affordable, and have a lanyard so they could be directly worn by a user without additional physical modifications. BLE beacons from Shenzhen Wellcore Technologies Co, Ltd. powered by 2.4 GHz wireless system on chip (SoC) nRF51-series chipset from Nordic Semiconductor were selected for location beacons (Fig. 1b). The location beacons are bigger in size than the wearable beacons, as they require a bigger battery (CR2477 battery) for a longer uptime, feature a power button, and are light and suitable for ceiling-mounted installation.
2.2. The RNSI-based method for indoor localisation using BLE beacons The RNSI-based method does not depend on the number of location beacons, but relies on the number of signals to determine the location of a user, i.e. the location that gives the highest total number of signals. The RNSI-based method was initiated by iteratively merging the presence of signals separated by five-second interval from the same location into the same group (Fig. 3a). Given that the distance between two locations were 5–8 m and the walking pace for humans is 1.4 m/s on average [33,34], the five-second interval was considered suitable to separate locations if a user moved from one location to another. It is possible that a wearable beacon senses signals from different locations at the same time. We aimed to remove signals that may trigger an incorrect prediction by comparing the total number of RNSI between signal groups that existed in the same period of time—i.e., the intersection of groups of signals. We removed the intersections with only one signal from one or more locations (e.g. the 3rd and 4th intersections in Fig. 3b). For example, the first intersection shown in Fig. 3b is labelled as location 1 because only signals from location 1 exist in the 2
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 3. The RNSI-based method for indoor localisation: (a) grouping the presence of signals separated by five-second interval from each location, (b) finding intersections among the signal groups. The black circle represents if a signal is present from the respective location, the background colour represents a group for each location, and the coloured dashed line rectangle represents a group intersection. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 4. The pre-defined routes for moving user experiments performed at (a) AIHI, and (b) CIDM.
intersection. The second intersection resulted location 2 as the predicted location as the total RNSI for location 2 is 12 (3 + 0 + 5 + 4) while the total RNSI for location 1 is only 8 (2 + 2 + 2 + 2).
At AIHI, we performed signal pattern analysis in a meeting room, a corridor, and at seven pre-defined locations. Similarly, signal patterns were collected from five pre-defined locations at CIDM. The location beacon was mounted on the ceiling (approximately 2.7 m from the floor) with the power button side facing the floor in all experiments. In the AIHI meeting room and corridor, we used one location beacon and one wearable beacon. The location beacon was mounted on the ceiling in the centre of the meeting room in order to capture the signal patterns from four directions. The wearable beacon and the base station were placed on a chair (approximately 1.1 m from the floor). For the experiments at several pre-defined locations, we used two location beacons per location and one wearable beacon was worn by a user. A detailed layout diagram for each location is provided in the Appendix.
2.3. Signal pattern analysis experiments The aim of signal pattern analysis is to understand BLE signal propagation characteristics with respect to the distance from a location beacon in an indoor environment. Since the signal propagation might be affected by the environment (e.g. due to shadowing, multipath propagation, etc.), we aimed to better understand and get more comparisons about RSSI and RNSI patterns from different environments. We performed signal pattern analysis in two different environments: the Australian Institute of Health Innovation (AIHI), Macquarie University; and the Centre for Infectious Diseases & Microbiology (CIDM) - Public Health, Westmead Hospital. During the data collection, we had no control over these environments, and dynamic elements included moving staff during the experiments and interference/noise from furniture, building structure, or electronic devices such as medical devices. We took the average value per second for RSSI and the actual count per second for RNSI.
2.4. Moving user experiments In the moving user experiments, one user (DS) wore the wearable beacon and moved from one location to another based on a pre-defined route (Fig. 4). The time when the user walked past the location beacons at each location was recorded, which was used as our ground truth. The predicted locations were considered correct if the method predicted the 3
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 5. Signal patterns for RNSI and RSSI in the AIHI meeting room. (a) The positions (shown in letters) where the signal patterns were collected (the blue circle in the middle of layout diagram represents the location beacon, where the protruding side shows the top side of the beacon—see inset on the upper left side), (b)-(e): the results for the four directions of the location beacon. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 6. Signal patterns for RNSI and RSSI in the AIHI corridor. (a)–(d): the results for the four directions of the location beacon.
user was at the respective locations at the recorded time. Both RNSI- and RSSI-based methods were applied to predict locations at AIHI and CIDM. For the RNSI-based method, the user was assigned to the location that gives the greatest number of signals within a period of time. For the RSSI-based method, the user was assigned to the location that gives the strongest signals within a period of time. We compared the performance of RSSI- and RNSI-based methods in predicting the user’s visited locations.
3. Results 3.1. Signal pattern analysis results In the AIHI meeting room (Fig. 5), RNSI showed a decreasing value at the 2 m distance (position B, E, H, and K) compared to at the 1 m distance (position A, D, G, and J). RNSI attained the same downtrend in value only at position F and I (3 m distance). Compared to RNSI, RSSI
4
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 7. Comparison of RNSI and RSSI from the seven pre-defined locations in AIHI: (a)–(g): signal patterns at location 1–7, where the horizontal axis represents the sensed signals from all locations that could still be sensed by the wearable beacon, (h) the pre-defined locations.
only showed a decreasing pattern at positions B, E, and H (2 m distance), and F and I (3 m distance). Both RNSI and RSSI increased at positions C and L (at 3 m distance), which may be the “error spot” in that area caused by multipath propagation problem [34]. In the AIHI corridor (Fig. 6), we found that RNSI consistently decreased as the distance increased. RNSI showed a significant drop in values at the distance between 4 m and 6 m from the location beacon. In general, RSSI also decreased after 4 m from the location beacon, but inconsistently. These results show that RNSI may be more reliable than RSSI for determining the location of a user in an indoor environment. From the seven pre-defined locations in AIHI, we found that RNSI from the user’s location was higher than RNSI from other locations (Fig. 7). While RSSI produced a similar general pattern of decreasing values with distance, the differences in values between the correct location and other locations were more distinct for RNSI compared to RSSI. For example, when the user was at location 1 (Fig. 7a), the RSSI values were still high from location 2, 3, and location 4 (RSSILocation_1: −82.9 dBm, RSSILocation_2: −87.4 dBm, RSSILocation_3: −86.7 dBm, RSSILocation_4: −87.1 dBm), while RNSI values from these locations showed significant differences in values (RNSILocation_1: 66, RNSILocation_2: 14, RNSILocation_3: 3, RNSILocation_4: 17). Similarly, when the user was at location 3, RSSI value from the location beacons at location 3 was slightly higher than from other locations, while the RNSI value at location 3 was notably high. These results suggest that RNSI is more sensitive to the distance than RSSI.
From the five pre-defined locations in CIDM, we also found that the RNSI value from the correct location was substantially higher than from the other locations. On the contrary, RSSI did not always show a similar pattern with RNSI. For example, when the user was at location 1, the RNSI value from location 1 was the highest but the RSSI value from location 2 was the highest. Also at the location 1 (a room), the RSSI from location 2 (outside the room) and location 5 (a different room) were still sensed with relatively high average RSSI values (RSSILocation_1: −83.3 dBm, RSSILocation_2: −81.5 dBm, and RSSILocation_5: −84.3 dBm). We noticed that location 4 gave different signal patterns (Fig. 8d). There was medical laboratory equipment near locations 4, and a WiFi router near the location beacons at location 4. These devices appear to have caused signal interference, though other aspects of the environment (e.g. the materials of the walls, furniture, etc.) may have also contributed to interference patterns [35]. 3.2. Moving user experiment results In the moving user experiments, we found that RNSI values were higher when a user stayed in one location for a while; but dropped when a user walked past the location beacons. The RSSI-based method correctly predicted 16 positions according to the pre-defined route and 5 locations were returned incorrectly (76.2% accuracy) from the moving user experiment at AIHI (Fig. 9a). The results showed that the 5
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 8. Comparison of RNSI and RSSI from the five pre-defined locations in CIDM: (a)–(e): signal patterns at location 1–5, where the horizontal axis represents the sensed signals from all locations that could still be sensed by the wearable beacon, (f) the pre-defined locations.
Fig. 9. Plots of the signals sensed by the wearable beacon and the predicted locations for the moving user experiment at AIHI using: (a) the strongest RSSI-based method, (b) the RNSI-based method. 6
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
Fig. 10. Plots of the signals sensed by the wearable beacon and the predicted locations for the moving user experiment at CIDM using: (a) the strongest RSSI-based method, (b) the RNSI-based method.
and CIDM were only performed once. It will be necessary in the future to replicate the same experiments in the same and different environments to demonstrate the efficacy of RNSI-based method. The elimination of incorrect signals may result in missing location information for location prediction in certain conditions, such as when only the signals from incorrect locations were sensed but the signals from the correct location were blocked by the user’s body, building structure, or furniture. While this was expected, we did not attempt to maximise signal sensing by increasing the number of location beacons or measuring the strategic placement of location beacons in both of our moving user experiments. We mounted the location beacons on the ceiling to maintain a clear line of sight between the location beacons and the wearable beacon, but we did not compare the performance if the location beacons were mounted on the wall or other positions. An incorrect location prediction could also happen if there were location beacons that failed to work, resulting in other locations being falsely identified. We decided to use two location beacons for each location to minimise this risk of failure. Future work would benefit from considering approaches that can robustly handle these situations. A further limitation in this work is that we performed our moving user experiments with only one user wearing a beacon, in an experimental setting, and did not consider how the use of additional wearable beacons or location beacons might affect the reception of BLE signal [36]. Wireless devices that operate in the radio spectrum from 2.400 GHz to 2.485 GHz can potentially cause signal interference with the BLE beacons [35]. It will be important to study performance over a longer time period and involve more users and locations.
incorrectly detected locations using the RSSI-based method tend to have small RNSI values and small RSSI values (weaker than −85 dBm), although some correct locations might also give weak signals (weaker than −85 dBm). On the other hand, the RNSI-based method successfully removed all signals from the incorrect locations and correctly returned 12 locations according to the pre-defined route but missed location 3 and 6 as there were not enough signals to determine the locations (80% accuracy) (Fig. 9b). In the moving user experiment performed at CIDM, the strongest RSSI-based method identified 13 incorrect locations and returned 14 correct locations according to the pre-defined route (51.9% accuracy) (Fig. 10a). Like the results from the moving user experiments at AIHI, RNSI values from incorrect locations were lower than the RNSI values from the correct locations. However, compared to the moving experiments in AIHI, most of signals for incorrect locations were not all weak (stronger than −85 dBm). The RNSI-based method removed the incorrect signals and kept the correct ones (Fig. 10b). As a result of the signal removal process, the method missed location 2 and location 4 (orange labels) but correctly identified the users in 10 locations (83.3% accuracy). Further analysis is provided in the Appendix. 4. Discussion 4.1. Limitations As we were limited by the number of devices and time (especially at CIDM) to perform experiments, the moving user experiments at AIHI 7
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al.
4.2. Implications and future research directions [3]
This work has implications for many applications using indoor localisation. With the simplicity of RNSI-based method, we found that it was possible to predict a user’s location with higher accuracy by removing signals that may contribute to an incorrect prediction. As the RNSI-based method relies on the number of signals sensed within a period of time and not from the signal strength, the use of low signal strength may prolong battery life of the beacons. Future implementations could take the advantages of using different signal strengths and signal transmission rates depending on the location. The RNSI represents a relatively simple additional measure that can be used to complement any indoor localisation methods that currently rely on RSSI, especially in applications that involve human movements. Both RNSI and RSSI could be used together as part of a more complex method to determine a location of a user. For example, when the locations are quite close to one another and the location beacons at each location have different signal strengths and signal transmission rates, both RNSI and RSSI could be used to determine which location beacons the user is close to. Future studies may also benefit from contextual information to understand about how humans move and interact with each other in indoor environments using RFID signals or accelerometers [37–39].
[4] [5] [6]
[7]
[8] [9] [10]
[11]
5. Conclusion
[12]
In this work, we demonstrated a novel method to detect the location of a moving user in indoor and dynamic environments. We introduced RNSI as a new measurement, and proposed a practical method for removing signals that could lead to incorrect location prediction. Our results suggest that the RNSI-based method produced more consistent patterns than the RSSI-based method for determining locations. Our experiments showed that the RNSI-based method effectively filtered out signals from incorrect locations, outperforming a baseline method using the strongest RSSI. We showed that while the signals from incorrect locations are sometimes stronger than from correct locations, they tended to have smaller RNSI values. Our approach was also relatively robust in a healthcare setting where the environment is more dynamic and has more aspects that could cause signal interference than an office environment.
[13] [14] [15] [16]
[17]
[18]
Declaration of Competing Interest
[19]
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
[20]
Acknowledgement [21]
VS and EC report funding from the NHMRC Centre of Research Excellence in Emerging Infection Diseases (APP1102962). DS and VK designed the research, undertook the device installation, and experiments. DS designed the methods and drafted the manuscript. VK configured all devices. All authors critically reviewed the manuscript.
[22] [23]
Appendix A. Supplementary material Supplementary data to this article can be found online at https:// doi.org/10.1016/j.jbi.2019.103288.
[24]
References
[25]
[1] S. Malhotra, D. Jordan, E. Shortliffe, V.L. Patel, Workflow modeling in critical care: piecing together your own puzzle, J. Biomed. Inform. 40 (2007) 81–92, https://doi. org/10.1016/j.jbi.2006.06.002. [2] E.A. Fry, L.A. Lenert, MASCAL: RFID tracking of patients, staff and equipment to
[26]
8
enhance hospital response to mass casualty events, AMIA Annu. Symp. Proc. 261–5 (2005). A. Vankipuram, S. Traub, V.L. Patel, A method for the analysis and visualization of clinical workflow in dynamic environments, J. Biomed. Inform. 79 (2018) 20–31, https://doi.org/10.1016/j.jbi.2018.01.007. J. Frisby, V. Smith, S. Traub, V.L. Patel, Contextual Computing: a Bluetooth based approach for tracking healthcare providers in the emergency room, J. Biomed. Inform. 65 (2017) 97–104, https://doi.org/10.1016/j.jbi.2016.11.008. J. Stehlé, N. Voirin, A. Barrat, C. Cattuto, V. Colizza, L. Isella, et al., Simulation of an SEIR infectious disease model on the dynamic contact network of conference attendees, BMC Med. 9 (2011) 1–15, https://doi.org/10.1186/1741-7015-9-87. P. Vanhems, A. Barrat, C. Cattuto, J.-F. Pinton, N. Khanafer, C. Régis, et al., Estimating potential infection transmission routes in hospital wards using wearable proximity sensors, PLoS ONE 8 (2013) e73970, , https://doi.org/10.1371/journal. pone.0073970. T. Hornbeck, D. Naylor, A.M. Segre, G. Thomas, T. Herman, P.M. Polgreen, Using sensor networks to study the effect of peripatetic healthcare workers on the spread of hospital-associated infections, J. Infect. Dis. 206 (2012) 1549–1557, https://doi. org/10.1093/infdis/jis542. M. Vankipuram, K. Kahol, T. Cohen, V.L. Patel, Toward automated workflow analysis and visualization in clinical environments, J. Biomed. Inform. 44 (2011) 432–440, https://doi.org/10.1016/j.jbi.2010.05.015. G. Deak, K. Curran, J. Condell, A survey of active and passive indoor localisation systems, Comput. Commun. 35 (2012) 1939–1954, https://doi.org/10.1016/j. comcom.2012.06.004. D. Lymberopoulos, J. Liu, X. Yang, R.R. Choudhury, V. Handziski, S. Sen, A realistic evaluation and comparison of indoor location technologies: experiences and lessons learned, The 14th International Conference on Information Processing in Sensor Networks (IPSN). Seattle, Washington: ACM New York, NY, USA, 2015, pp. 178–189, , https://doi.org/10.1145/2737095.2737726. H. Liu, H. Darabi, P. Banerjee, J. Liu, Survey of wireless indoor positioning techniques and systems, IEEE Trans. Syst., Man, Cybernet. 37 (2007) 1067–1080, https://doi.org/10.1109/TSMCC.2007.905750. C. Wu, Z. Yang, Y. Liu, W. Xi, WILL: wireless indoor localization without site survey, IEEE Trans. Parallel Distrib. Syst. 24 (2013) 839–848, https://doi.org/10.1109/ TPDS.2012.179. Meng W, He Y, Deng Z, Li C. Optimized access points deployment for WLAN indoor positioning system. IEEE Wireless Communications and Networking Conference (WCNC). Shanghai, China: IEEE; 2012, p. 2457-61. 10.1109/WCNC.2012.6214209. C. Feng, W.S.A. Au, S. Valaee, Z. Tan, Received-signal-strength-based indoor positioning using compressive sensing, IEEE Trans. Mob. Comput. 11 (2012) 1983–1993, https://doi.org/10.1109/TMC.2011.216. H. Xu, Y. Ding, P. Li, R. Wang, Y. Li, An RFID indoor positioning algorithm based on Bayesian probability and k-nearest neighbor, Sensors. 17 (2017) E1806, https:// doi.org/10.3390/s17081806. L.M. Ni, Y. Liu, Y.C. Lau, A.P. Patil, LANDMARC: indoor location sensing using active RFID, in: Proceedings of the First IEEE International Conference on Pervasive Computing and Communications (PerCom). IEEE, Fort Worth, TX, USA, 2003. 10. 1109/PERCOM.2003.1192765. D. Lieckfeldt, J. You, D. Timmermann, Exploiting RF-Scatter: human localization with bistatic passive UHF RFID-systems, in: IEEE International Conference on Wireless and Mobile Computing, Networking and Communications. IEEE, Marrakech, Morocco, 2009. p. 179–184. 10.1109/WiMob.2009.39. Z. Jianyong, L. Haiyong, C. Zili, L. Zhaohui, RSSI based Bluetooth Low Energy indoor positioning, in: International Conference on Indoor Positioning and Indoor Navigation (IPIN). IEEE, Busan, South Korea; 2014. 10.1109/IPIN.2014.7275525. V. Chandel, N. Ahmed, S. Arora, A. Ghose, InLoc: an end-to-end robust indoor localization and routing solution using mobile phones and BLE beacons, International Conference on Indoor Positioning and Indoor Navigation (IPIN): IEEE, 2016, https://doi.org/10.1109/IPIN.2016.7743592. N. Kitbutrawat, S. Kajita, H. Yamaguchi, T. Higashino, Location identification of BLE-embedded HVACs for smart building management, in: The 14th International Conference on Intelligent Environments (IE), IEEE, Rome, Italy, 2018. 10.1109/IE. 2018.00019. F. Zafari, I. Papapanagiotou, T.J. Hacker, A novel Bayesian filtering based algorithm for RSSI-based indoor localization, in: IEEE International Conference on Communications (ICC). IEEE, Kansas City, MO, USA, 2018. 10.1109/ICC.2018. 8423012. N. Patwari, J.N. Ash, S. Kyperountas, A.O. Hero, R.L. Moses, N.S. Correal, Locating the nodes: cooperative localization in wireless sensor networks, IEEE Signal Process. Mag. 22 (2005) 54–69, https://doi.org/10.1109/MSP.2005.1458287. X. Li, K. Pahlavan, M. Latva-aho, M. Ylianttila, Comparison of indoor geolocation methods in DSSS and OFDM Wireless LAN systems, in: Proceedings of the 52nd Vehicular Technology Conference. IEEE, Boston, MA, USA, 2000, pp. 3015–3020. 10.1109/VETECF.2000.886867. R. Peng, M.L. Sichitiu, Angle of arrival localization for wireless sensor networks, in: Proceedings of the 3rd Annual IEEE Communications Society on Sensor and Ad Hoc Communications and Networks. IEEE, Reston, VA, USA, 2006, pp. 374–382. 10. 1109/SAHCN.2006.288442. K. Heurtefeux, F. Valois, Is RSSI a good choice for localization in wireless sensor network? In: Proceedings of the IEEE 26th International Conference on Advanced Information Networking and Applications, IEEE, Fukuoka, Japan, 2012, pp. 732739. 10.1109/AINA.2012.19. S. Subedi, J.-Y. Pyun, Practical fingerprinting localization for indoor positioning system by using beacons, Sensors (2017) 1–16, https://doi.org/10.1155/2017/ 9742170.
Journal of Biomedical Informatics 98 (2019) 103288
D. Surian, et al. [27] Z. Iqbal, D. Luo, P. Henry, S. Kazemifar, T. Rozario, Y. Yan, et al., Accurate real time localization tracking in a clinical environment using Bluetooth Low Energy and deep learning, PLoS ONE 13 (2018) e0205392, , https://doi.org/10.1371/journal. pone.0205392. [28] S. Lee, B. Cho, B. Koo, S. Ryu, J. Choi, S. Kim, Kalman filter-based indoor position tracking with self-calibration for RSS variation mitigation, Int. J. Distrib. Sens. Netw. 11 (2015) 1–10, https://doi.org/10.1155/2015/674635. [29] P. Kriz, F. Maly, T. Kozel, Improving indoor localization using Bluetooth Low Energy beacons, Mobile Inform. Syst. 2016 (2016) 1–11, https://doi.org/10.1155/ 2016/2083094. [30] K. Urano, K. Hiroi, K. Kaji, N. Kawaguchi, A location estimation method using BLE tags distributed among participants of a large-scale exhibition, Proceedings of the 13th International Conference on Mobile and Ubiquitous Systems: Computing Networking and Services (MOBIQUITOUS). Hiroshima, Japan: ACM New York, NY, USA, 2016, pp. 124–129, , https://doi.org/10.1145/3004010.3004051. [31] Raspberry Pi Zero W. [Available from https://wwwrapsberrypiorg/products/ raspberry-pi-zero-w]. [32] Node.js. [Available from: https://nodejsorg/]. [33] N. Carey, Establishing pedestrian walking speeds, Portland State University, 2005. [34] D.-Y. Kim, S.-H. Kim, D. Choi, S.-H. Jin, Accurate indoor proximity zone detection based on time window and frequency with Bluetooth Low Energy, Proc. Comput.
Sci. 56 (2015) 88–95, https://doi.org/10.1016/j.procs.2015.07.199. [35] J. Frisby, Contextual Computing: tracking healthcare providers in the Emergency Department via Bluetooth beacons, Arizona State University, 2015. [36] D.P.-Dd. Cerio, Á. Hernández, J.L. Valenzuela, A. Valdovinos, Analytical and experimental performance evaluation of BLE neighbor discovery process including non-idealities of real chipsets, Sensors 17 (2017) 499, https://doi.org/10.3390/ s17030499. [37] H. Li, C. Ye, A.P. Sample, IDSense: a human object interaction detection system based on passive, Proceedings of the 33rd ACM Conference on Human Factors in Computing Systems. Seoul, Republic of Korea: ACM New York, NY, USA, 2015, pp. 2555–2564, , https://doi.org/10.1145/2702123.2702178. [38] H. Li, P. Zhang, S.A. Moubayed, S.N. Patel, A.P. Sample, ID-Match: a hybrid computer vision and RFID system for recognizing individuals in groups, Proceedings of the 2016 ACM Conference on Human Factors in Computing Systems. ACM New York, NY, USA, San Jose, California, USA, 2016, pp. 4933–4944, , https://doi.org/ 10.1145/2858036.2858209. [39] A.R.J. Ruiz, F.S. Granja, J.C.P. Honorato, J.I.G. Rosas, Accurate pedestrian indoor navigation by tightly coupling foot-mounted IMU and RFID measurements, IEEE Trans. Instrum. Meas. 61 (2012) 178–189, https://doi.org/10.1109/TIM.2011. 2159317.
9