Hardware and software design of BMW system for multi-floor localization
- Mu Zhou^{1, 2},
- Bin Wang^{1}Email author,
- Zengshan Tian^{1} and
- Liangbo Xie^{1}
https://doi.org/10.1186/s13638-017-0925-0
© The Author(s) 2017
Received: 26 May 2017
Accepted: 2 August 2017
Published: 16 August 2017
Abstract
Although the Micro Electro Mechanical System (MEMS) sensors are capable of providing short-term high positioning accuracy, every positioning result significantly depends on the historical ones, which inevitably leads to the long-term error accumulation. The Bluetooth Low Energy (BLE) is independent of the accumulative error, but the positioning accuracy is suffered by the irregular jump error resulted from the Received Signal Strength Indicator (RSSI) jitter. Considering the requirement of accurate, seamless, and consecutive positioning by the existing commercial systems, we propose a new integrated BLE and MEMS Wireless (BMW) system for multi-floor positioning. In concrete terms, first of all, the way of fingerprint database construction with the reduced workload is introduced. Second, the fingerprint database is denoised by the process of affinity propagation clustering, outlier detection, and RSSI filtering. Third, the robust M estimation-based extended Kalman filter is applied to estimate the two-dimensional coordinates of the target on each floor. Finally, the barometer data are used to calculate the height of the target. The extensive experimental results show that the proposed system can not only restrain the accumulative error caused by the MEMS sensors but also eliminate the irregular jump error from the BLE RSSI jitter. In an actual multi-floor environment, the proposed system is verified to be able to achieve the Root Mean Square (RMS) positioning error within 1 m.
Keywords
1 Introduction
At present, the indoor positioning has broad application such as searching the cars and elevators in underground parking lot and pushing advertisements or discounts to the customers in large shopping malls. The Global Navigation Satellite System (GNSS) [1–3] well meets the precision requirement of outdoor positioning, but in indoor environment, its performance may drastically deteriorate due to the serious signal blocking and multipath effect. In response to this compelling problem, a batch of researchers put forward a variety of indoor positioning systems based on the Bluetooth Low Energy (BLE) [4], Ultra Wideband (UWB) [5], Radio Frequency Identification (RFID) [6], Micro Electro Mechanical System (MEMS) sensors [7], and Wireless Local Area Network (WLAN) [8, 9].
The hardware cost, range limitation, and long-term error accumulation limit the development of the conventional positioning systems. Among them, the MEMS sensors can be used to perform the Pedestrian Dead Reckoning (PDR) [7, 10] by using the inertia and heading information though it contains the accumulative error. The BLE Received Signal Strength Indicator (RSSI)-based positioning system can meet the requirement of low power, low cost, and no accumulative error though the RSSI jitter caused by multipath effect may seriously decrease the positioning accuracy.
In recent decade, various integrated systems have been significantly concerned to improve positioning accuracy [11, 12]. The systems [13–15] using Kalman and particle filter to achieve the WLAN/MEMS fusion positioning reduce the accumulative error of MEMS sensors, but they fail to constrain the WLAN RSSI jitter. To solve the RSSI jitter problem, the authors in [16] build the observation equation of the RSSI according to the signal propagation model, which is difficult to be constructed in the complex indoor environment. In [17], the authors design an indoor positioning system by fusing the WLAN and Magnetic Angular Rate and Gravity (MARG) data to solve the problems of RSSI jitter and accumulative error. However, this system cannot achieve the three-dimensional positioning, which is significantly required by the current commercial applications.
This paper presents a new integrated BMW system for multi-floor positioning. First of all, the fingerprint database is optimized by using the affine propagation clustering, outlier detection, and RSSI filtering algorithms. Specifically, the fingerprint denoising is performed to reduce the probability of large errors of BLE positioning, and then, the gait detection approach is used to estimate the walking speed and heading angle of the target based on the extended Kalman filter (EKF). Second, the robust EKF is designed to restrain the accumulative error caused by the MEMS sensors, as well as eliminate the irregular jump error from the BLE RSSI jitter. Finally, according to the barometer data and geographical location information, the height of the target is calculated to achieve the multi-floor positioning.
2 Related work
In this section, we will briefly introduce some related work in three aspects of fast fingerprint database construction, fusion positioning, and multi-floor positioning and also address the corresponding limitations.
Fast fingerprint database construction Although the fingerprint positioning has been studied for a long time, it still cannot be applied widely since its offline phase generally spends a huge amount of time on fingerprint database construction. By using the conventional approaches, the target area is calibrated with a batch of equally spaced grids, and then, the fingerprints are collected at the grids point-by-point [18, 19]. The involved time cost rises significantly with the increase of environmental size, which hinders the development of fingerprint positioning. In [20], the authors propose to construct an incomplete fingerprint database with realistic coverage gaps, and meanwhile study the performance of several interpolation and extrapolation approaches used for recovering the missing fingerprints. In [21], according to the distribution of Reference Points (RPs) with respect to each Wi-Fi Access Point (AP), the signal propagation model is constructed as a function of spatial structure, which can be used to construct the fingerprint database quickly.
Fusion positioning At present, the common single positioning systems are difficult to adapt to the complex indoor environment. In response to this compelling problem, a variety of fusion positioning systems are designed to compensate for the shortcomings of each single one. In [22], the data fusion from both the proprioceptive and exteroceptive sensors, like the odometer, Global Positioning System (GPS), Light Detection and Ranging (LIDAR), and vision, as well as the knowledge of road map is considered to perform the fusion positioning. In [23], the authors conduct data fusion by integrating the RSSI and Time difference of Arrival (TDOA) measurements to estimate the superior locations of the target. Specifically, by employing the nonparametric estimation approach, which is robust to the variations of measurement noise and quantization, it is addressed that the fusion positioning is more robust and higher accurate and has lower implementation cost.
Multi-floor positioning The mainstream of indoor positioning systems mainly focuses on the horizontal coordinate estimation, whereas little research has been done on the vertical coordinate estimation. In [24], the authors propose a WiFi-based indoor positioning system that takes both the characteristics of trilateration and scene analysis into account. The authors in [25] rely on the path loss model to construct a light fingerprint radio map to find the target floor. However, these systems are applied to only determine the floor on which the target is most probably located, but they cannot estimate the accurate locations of the target.
3 System description
3.1 System framework
3.2 BLE fingerprint positioning
3.2.1 Fast fingerprint database construction
where \(({X_{0}},{Y_{0}})\phantom {\dot {i}\!}\) is the actual origin location, which is represented by (0,0) in the map. \(\phantom {\dot {i}\!}({x_{\text {map}{_{i}}}},{y_{\text {map}{_{i}}}})\) and \(({X_{i}},{Y_{i}})\phantom {\dot {i}\!}\) are the coordinates of the ith point in the map (with the length m _{mapx } and width m _{mapy }) and actual environment (with the length M _{ x } and width M _{ y }) respectively.
where L _{step_x } and L _{step_y } are the physical intervals in the X and Y directions, respectively, in the actual environment, which are normally set as L _{step_x }=L _{step_y }.
where v _{ n } and θ _{ n } are the walking speed and heading. \({\varepsilon _{xi}} = \frac {{{v_{i}}\sin ({\theta _{i}})}}{{\sum {{v_{i}}\sin ({\theta _{i}})} }}(L\sin ({\Psi _{1}}) - {L_{pdr}}\sin ({\Psi _{2}}))\) and \({\varepsilon _{yi}} = \frac {{{v_{i}}\cos ({\theta _{i}})}}{{\sum {{v_{i}}\cos ({\theta _{i}})} }}(L\cos ({\Psi _{1}}) - {L_{pdr}}\cos ({\Psi _{2}}))\) are the error compensation in the X and Y directions respectively. L and \({L_{pdr}} = \sum \limits _{i = 1}^{N} {{v_{i}}}\) are the actual distance and the estimated one by the PDR from the starting to ending locations. Ψ _{1} is the heading of (x _{end},y _{end}) relative to (x _{0},y _{0}). Ψ _{2} is the heading of (x _{ N },y _{ N }) relative to (x _{0},y _{0}).
where \({P_{{d_{0}}}}\) is the RSSI with 1-m distance from the anchor. d is the distance from the anchor to terminal. N _{ t } is the path loss exponent.
3.2.2 Outlier detection
Pseudo-code of sub-database construction
Algorithm: Sub-database construction |
---|
Data: Raw fingerprint database |
Result: sub-databases |
1 c l u s t e r←A P_C l u s t e r(i n p u t); |
2 for each c l u s t e r _{ i }∈c l u s t e r s do |
3 for each p o i n t _{ j }∈c l u s t e r _{ i } do |
4 if r e l a t i v e_d e n s i t y(p o i n t _{ j }) >Threshold |
5 Outliers are reassigned to the nearest neighbor; |
6 back to Step 1; |
7 end |
8 end |
9 Obtain the cluster center; |
10 end |
11 Construct each cluster as a sub-database; |
where distance(x,y) is the distance between x and y. N(x,k) is the set of k nearest neighbors with respect to x. |N(x,y)| is the number of RSSI data in N(x,k).The pseudo-code of sub-database construction is shown in Table 1.
c l u s t e r _{ i } is the ith cluster. point_{ j } is the jth point. Threshold is the threshold of relative density.
3.3 Speed and heading estimation
3.3.1 Speed estimation
3.3.2 Heading estimation
3.4 Robust EKF
3.4.1 EKF
where X _{ t }=[E _{ t } N _{ t } v _{ t } φ _{ t }]^{T}. E _{ t−1} and N _{ t−1} are the geographic east and north directions, respectively, at moment t−1. v _{ t−1} and φ _{ t−1} are the walking speed and heading, respectively, in the geographic reference coordinate system at moment t−1. W _{ t−1} is the Gaussian white noise with zero mean at moment t−1. \(E\left [ {{W_{i}}W_{j}^{\mathrm {T}}} \right ] = Q(i,j){\delta _{ij}},\;\;i,j = 1, \cdots,o\), in which o is the number of estimation variables, δ _{ ij } is the Kronecker function, and Q is the covariance matrix of process noise.
where \({Z_{t}} = {\left [ {E_{t}^{\text {ble}}\;N_{t}^{\text {ble}}\;v_{t}^{\text {mems}}\;\varphi _{t}^{\text {mems}}} \right ]^{\mathrm {T}}}\). \(E_{t}^{{\text {ble}}}\) and \(N_{t}^{{\text {ble}}}\) are the geographic east and north directions by the BLE fingerprint positioning at moment t. \(v_{t}^{\text {mems}}\) and \(\varphi _{t}^{\text {mems}}\) are the estimated speed and heading, respectively, by using the MEMS sensors at moment t. V _{ t } is the Gaussian white noise with zero mean at moment t. E[V(i)V ^{ T }(j)]=R(i,j)δ _{ ij }, i,j=1,⋯,n, in which n is the number of observation variables and R(R>0) is the covariance matrix of observation noise.
3.4.2 Robustness enhancement
where v _{ i } is the ith element in the n×1-dimensional residual vector of observation V. \(\sigma _{v_{i}} = \frac {\sigma _{0}}{\sqrt {q_{v_{i}}}}\), in which \(q_{v_{i}}\) is the reciprocal of v _{ i } and σ _{0} is the variance factor [32].
where \(\bar K\left (t \right)\) is the filter gain at moment t. P(t,t−1) is the one-step prediction of the matrix of error covariance at moment t. H(t) is the matrix of observation at moment t. \(\bar R\left (t \right) = \bar B_{t}^{- 1}\) is the covariance matrix of observation noise, in which \({\bar B_{t}}\) is the equivalent weight matrix.
3.5 Height estimation
3.6 Pressure measure
where P _{ s } is the pressure value.
where R=6356766 m is the radius of earth. Since R≫H, we can obtain \(\frac {R}{{R - H}} \approx 1\) and h≈H.
3.7 Algorithm design
Pseudo-code of height estimation
Algorithm: Height estimation |
---|
Data: P o s i t i o n i n g_r e s u l t(x,y); |
Result: Height of the target; |
1 if (x,y) is located in the staircase |
2 if Pressure value rises |
3 h _{ t }=h _{ t−1}+h _{ stair }; |
4 else if Pressure value drops |
5 h _{ t }=h _{ t−1}−h _{ stair }; |
6 else |
7 h _{ t }=h _{ t−1}; |
8 end |
9 else |
10 h _{ t }=h _{ t−1}; |
11end |
3.8 System implementation
3.8.1 Hardware platform
3.8.2 Software platform
4 Experimental results
As can be seen from Fig. 26, the large accumulative error of MEMS positioning exists, while the BLE positioning is suffered by the irregular jump error and cannot be accurate enough when the target is located in the staircase. In contrast, the proposed fusion positioning algorithm can restrain the accumulative error of MEMS positioning as well as eliminate the irregular jump error of BLE positioning. In addition, it is also verified that the proposed algorithm is featured with good height resolution, which makes the system more robust to the actual indoor multi-floor positioning.
Different percentile values with respect to the positioning error
Percentile | Fusion | BLE | MEMS |
---|---|---|---|
values (%) | positioning (m) | positioning (m) | positioning (m) |
50 | <0.60 | <3.80 | <2.29 |
70 | <0.77 | <4.48 | <3.37 |
90 | <1.03 | <5.26 | <8.41 |
5 Conclusions
In this paper, both the hardware and software of BMW system are designed and implemented for indoor multi-floor positioning. Based on the extensive experimental results, it is demonstrated that the proposed system is capable of solving the problems of accumulative error constraint and irregular jump error elimination in MEMS and BLE positioning respectively. In general, our system is featured with high positioning accuracy, good height resolution, and strong long-term stability.
Declarations
Acknowledgements
This work was supported in part by the Program for Changjiang Scholars and Innovative Research Team in University (IRT1299), Special Fund of Chongqing Key Laboratory (CSTC), Fundamental and Frontier Research Project of Chongqing (cstc2017jcyjAX0380, cstc2015jcyjBX0065, cstc2015jcyjBX0085), and University Outstanding Achievement Transformation Project of Chongqing (KJZH17117).
Authors’ contributions
The authors have contributed jointly to all parts on the preparation of this manuscript, and all authors read and approved the final manuscript.
Competing interests
The authors declare that they have no competing interests.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Authors’ Affiliations
References
- M Jia, L Wang, Q Guo, X Gu, W Xiang, A low complexity detection algorithm for fixed up-link scma system in mission critical scenario. IEEE Internet Things J. PP(99), 1–1 (2017).Google Scholar
- M Jia, X Gu, Q Guo, W Xiang, N Zhang, Broadband hybrid satellite-terrestrial communication systems based on cognitive radio towards 5g. IEEE Wirel. Commun. 23(6), 96–106 (2016).View ArticleGoogle Scholar
- M Jia, L Wang, Z Yin, Q Guo, X Gu, A novel spread slotted aloha based on cognitive radio for satellite communications system. EURASIP J. Wirel. Commun. Netw. 2016(1), 232 (2016).View ArticleGoogle Scholar
- ES Lohan, J Talvitie, PFE Silva, in International Conference on Localization and GNSS. Received signal strength models for WLAN and BLE-based indoor positioning in multi-floor buildings (IEEEGothenburg, 2015), pp. 1–6.Google Scholar
- K Yu, JP Montillet, A Rabbachin, UWB location and tracking for wireless embedded networks. Signal Process. 86(9), 2153–2171 (2006).View ArticleMATHGoogle Scholar
- YL Lai, J Cheng, A cloud-storage RFID location tracking system. IEEE Trans. Magn. 50(7), 1–4 (2014).Google Scholar
- L Zhang, J Liu, H Jiang, Senstrack: energy-efficient location tracking with smartphone sensors. IEEE Sensors J. 13(10), 3775–3784 (2013).View ArticleGoogle Scholar
- N Alsindi, Z Chaloupka, N Alkhanbashi, An empirical evaluation of a probabilistic RF signature for WLAN location fingerprinting. IEEE Trans. Wirel. Commun. 13(6), 3257–3268 (2014).View ArticleGoogle Scholar
- M Zhou, F Qiu, K Xu, Z Tian, Semi-supervised learning for indoor hybrid fingerprint database calibration with low effort. Comput. Commun. 86(C), 57–74 (2016).View ArticleGoogle Scholar
- B Koo, S Lee, M Lee, in International Conference on Indoor Positioning and Indoor Navigation. PDR/fingerprinting fusion indoor location tracking using RSS recovery and clustering (IEEEBusan, 2015), pp. 699–704.Google Scholar
- M Tanenhaus, D Carhoun, T Geis, in IEEE/ION Position Location and Navigation Symposium. Miniature IMU/INS with optimally fused low drift MEMS gyro and accelerometers for applications in GPS-denied environments (IEEESouth Carolina, 2012), pp. 259–264.Google Scholar
- K Kloch, P Lukowicz, C Fischer, in International Symposium on Wearable Computers. Collaborative PDR localisation with mobile phones (IEEESan Francisco, 2011), pp. 37–40.Google Scholar
- W Xiao, W Ni, KT Yue, in International Conference on Indoor Positioning and Indoor Navigation. Integrated Wi-Fi fingerprinting and inertial sensing for indoor positioning (IEEEGuimaraes, 2011), pp. 1–6.Google Scholar
- H Wang, H Lenz, A Szabo, in Positioning, Navigation and Communication, 2007. Wpnc ’07. Workshop On. WLAN-based pedestrian tracking using particle filters and low-cost MEMS sensors (IEEEHannover, 2007), pp. 1–7.Google Scholar
- M Atia, M Korenberg, A Noureldin, in IEEE International Symposium on Mechatronics and Its Applications. A WiFi-aided reduced inertial sensors-based navigation system with fast embedded implementation of particle filtering (IEEESharjah, 2012), pp. 1–5.Google Scholar
- V Malyavej, W Kumkeaw, M Aorpimai, in International Conference on Electrical Engineering/Electrinics, Computer, Telecommunications and Information Technology. Indoor robot localization by RSSI/IMU sensor fusion (IEEEKrabi, 2013), pp. 1–6.Google Scholar
- X Liu, Z Wu, X Lin, WLAN/MARG/GPS integrated positioning system based on a self-adaptive weighted algorithm. Qinghua Daxue Xuebao/J. Tsinghua University. 53(7), 955–960 (2013).Google Scholar
- M Zhou, Q Zhang, Y Wang, Hotspot ranking based indoor mapping and mobility analysis using crowdsourced Wi-Fi signal. IEEE Access. 5:, 3594–3602 (2017).View ArticleGoogle Scholar
- M Zhou, Y Tang, Z Tian, Semi-supervised learning for indoor hybrid fingerprint database calibration with low effort. IEEE Access. 5:, 4388–4400 (2017).View ArticleGoogle Scholar
- J Talvitie, M Renfors, ES Lohan, Distance-based interpolation and extrapolation methods for RSS-based localization with indoor wireless signals. IEEE Trans. Veh. Technol. 64(4), 1340–1353 (2015).View ArticleGoogle Scholar
- SS Jan, SJ Yeh, YW Liu, Received signal strength database interpolation by kriging for a Wi-Fi indoor positioning system. Sensors. 15(9), 21377–21393 (2015).View ArticleGoogle Scholar
- J Laneurit, C Blanc, R Chapuis, IEEE, Columbus, in Intelligent Vehicles Symposium. Multisensorial data fusion for global vehicle and obstacles absolute positioning, (2003), pp. 138–143.Google Scholar
- M Mcguire, KN Plataniotis, AN Venetsanopoulos, Data fusion of power and time measurements for mobile terminal location. IEEE Trans. Mob. Comput. 4(2), 142–153 (2005).View ArticleGoogle Scholar
- HH Liu, YN Yang, in TENCON 2011 - 2011 IEEE Region 10 Conference. WiFi-based indoor positioning for multi-floor environment (IEEECebu, 2012), pp. 597–601.Google Scholar
- IH Alshami, NA Ahmad, S Sahibuddin, in Malaysian Software Engineering Conference. A light WLAN radio map for floor detection in multi-floor environment localization (IEEESeri Kembangan, 2015), pp. 135–139.Google Scholar
- X Wang, L Yang, L Wang, in IEEE International Conference on Mobile Ad-Hoc and Sensor Networks. Improved AP clustering algorithm based on target segmentation (IEEEDalian, 2013), pp. 539–541.Google Scholar
- IR King, Density data and emission measure for a model of the coma cluster. Astrophys. J. 174(174), 123 (1972).View ArticleGoogle Scholar
- K Zhao, B Li, A Dempster, in IEEE International Conference on Wireless Communication and Sensor Network. A new approach of real time step length estimation for waist mounted PDR system (IEEEChangsha, 2015), pp. 400–406.Google Scholar
- Y Zhang, L Wu, Stock market prediction of S&P 500 via combination of improved BCO approach and BP neural network. Expert Syst. Appl. 36(5), 8849–8854 (2009).Google Scholar
- S Jia, Q Qiu, J Li, in IEEE International Conference on Information and Automation. BP neural network based localization for a front-wheel drive and differential steering mobile robot (IEEELijiang, 2015), pp. 2270–2274.Google Scholar
- Y Yang, W Gao, X Zhang, Robust Kalman filtering with constraints: a case study for integrated navigation. J. Geodesy. 84(6), 373–381 (2010).View ArticleGoogle Scholar
- S Zhao, S Chan, in IEEE International Symposium on Circuits and Systems. A novel algorithm for mobile station location estimation with none line of sight error using robust least m-estimation (IEEESeattle, 2008), pp. 1176–1179.Google Scholar
- Y Kim, Y Hwang, S Choi, in IEEE/ACM International Conference on Advanced Intelligent Mechatronics. Height estimation scheme of low-cost pedestrian dead-reckoning system using Kalman filter and walk condition estimation algorithm (IEEEWollongong, 2013), pp. 1492–1497.Google Scholar