Conference PaperPDF Available

A transfer kernel learning based strategy for adaptive localization in dynamic indoor environments: poster

Authors:

Abstract and Figures

Existing WiFi fingerprinting-based Indoor Positioning System (IPS) suffers from the vulnerability of environmental dynamics. To address this issue, we propose TKL-WinSMS as a systematic strategy, which is able to realize robust and adaptive localization in dynamic indoor environments. We developed a WiFi-based Non-intrusive Sensing and Monitoring System (WinSMS) that enables COTS WiFi routers as online reference points by extracting real-time RSS readings among them. With these online data and labeled source data from the offline calibrated radio map, we further combine the RSS readings from target mobile devices as unlabeled target data, to design a robust localization model using an emerging transfer learning algorithm, namely transfer kernel learning (TKL). It can learn a domain-invariant kernel by directly matching the source and target distributions in the reproducing kernel Hilbert space instead of the raw noisy signal space. By leveraging the resultant kernel as input for the SVR training, the trained localization model can inherit the information from online phase to adaptively enhance the offline calibrated radio map. Extensive experimental results verify the superiority of TKL-WinSMS in terms of localization accuracy compared with existing solutions in dynamic indoor environments.
Content may be subject to copyright.
Poster: A Transfer Kernel Learning based Strategy for
Adaptive Localization in Dynamic Indoor Environments
Han Zou1, Yuxun Zhou2, Hao Jiang3, Baoqi Huang4, Lihua Xie1and Costas Spanos2
1Nanyang Technological University, Singapore 2University of California, Berkeley, USA
3Fuzhou University, Fuzhou, China 4Inner Mongolia University, Hohhot, China
{zouh0005, elhxie}@ntu.edu.sg, {yxzhou, spanos}@berkeley.edu, jiangh@fzu.edu.cn,
cshbq@imu.edu.cn
ABSTRACT
Existing WiFi fingerprinting-based Indoor Positioning Sys-
tem (IPS) suffers from the vulnerability of environmental
dynamics. To address this issue, we propose TKL-WinSMS
as a systematic strategy, which is able to realize robust and
adaptive localization in dynamic indoor environments. We
developed a WiFi-based Non-intrusive Sensing and Monitor-
ing System (WinSMS) that enables COTS WiFi routers as
online reference points by extracting real-time RSS readings
among them. With these online data and labeled source da-
ta from the offline calibrated radio map, we further combine
the RSS readings from target mobile devices as unlabeled
target data, to design a robust localization model using an
emerging transfer learning algorithm, namely transfer kernel
learning (TKL). It can learn a domain-invariant kernel by
directly matching the source and target distributions in the
reproducing kernel Hilbert space instead of the raw noisy
signal space. By leveraging the resultant kernel as input for
the SVR training, the trained localization model can inherit
the information from online phase to adaptively enhance the
offline calibrated radio map. Extensive experimental results
verify the superiority of TKL-WinSMS in terms of localiza-
tion accuracy compared with existing solutions in dynamic
indoor environments.
Categories and Subject Descriptors
C.2.1 [Network Architecture and Design]: Wireless Com-
munication
Keywords
IEEE 802.11 WLAN, Localization, Transfer learning
1. INTRODUCTION
WiFi has been recognized as the most promising tech-
nique for indoor positioning services, due to the widely in-
stalled network infrastructures and pervasive WiFi-enabled
COTS mobile devices (MDs) [1, 4]. Existing WiFi-based
Permission to make digital or hard copies of part or all of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for third-party components of this work must be honored.
For all other uses, contact the owner/author(s).
MobiCom’16 October 03-07, 2016, New York City, NY, USA
c
2016 Copyright held by the owner/author(s).
ACM ISBN 978-1-4503-4226-1/16/10.
DOI: http://dx.doi.org/10.1145/2973750.2985278
IPSs for practical large-scale implementation usually adop-
t fingerprinting-based localization algorithm [3]. It localizes
an MD by comparing the real-time RSS readings with a pre-
established RSS fingerprint database, (a.k.a. radio map).
The major bottleneck of WiFi fingerprinting-based IPS is
the vulnerability to environmental dynamics. The real-time
RSS readings may deviate significantly from the fingerprints
stored in the static radio map due to severe multi-path and
shadow fading effects caused by various interferences, which
leads to huge degradation on localization accuracy. To ac-
commodate environmental dynamics, radio map recalibra-
tion is extremely time-consuming and labor-intensive and
deploying fixed reference anchors to obtain fresh RSS read-
ings introducing extra hardware cost.
To overcome this problem, we propose TKL-WinSMS as a
systematic strategy, which is able to construct a robust mod-
el for adaptive localization in dynamic indoor environments.
We developed WinSMS that enables COTS WiFi routers as
online reference points by extracting real-time RSS readings
among them. With these online data and the offline cali-
brated radio map as labeled source data, we further com-
bine the RSS readings from target MDs as unlabeled target
data, to develop a robust localization model using an emerg-
ing transfer learning algorithm TKL [2]. It is able to learn
a domain-invariant kernel by directly matching the source
and target distributions in the reproducing kernel Hilbert
space. By leveraging the resultant kernel as the input for
SVR training, the trained localization model can inherit the
information from online phase to adaptively enhance the of-
fline calibrated radio map. Extensive experiments have been
conducted and demonstrated that TKL-WinSMS can pro-
vide high localization accuracy under various environmental
dynamics consistently.
2. SYSTEM DESIGN
Suppose pWiFi access points (APs) are installed in an
indoor environment, and the signals received by MD from
these APs at an arbitrary two-dimensional location `can be
represented by a signal vector RSS. The data are collected
at totally lCP offline calibration points (CPs) as DSCP =
{(RSSi,`i)}lCP
i=1 . The proposed WinSMS enables APs as
online reference points, and the data collected at online ref-
erence points lAP are denoted as DSAP ={(RSSi,`i)}lAP
i=1 .
Since the locations of CPs and APs are known, we combine
DSCP and DSAP as labeled source data DS={(RSSi,`i)}lS
i=1
in source domain with lS=lCP +lAP . We use DT=
{RSSi}lT
i=1 to represent the RSS data of target MDs in tar-
get domain. With the labeled source data DSand unla-
462
Figure 1: System Architecture of WinSMS.
beled target data DT, our goal is to estimate the locations
of MDs by constructing a robust localization model using
TKL-WinSMS.
2.1 WinSMS
We develop WinSMS, which enables COTS WiFi router-
s as online reference points by overhearing the data pack-
ets transmitted between each MD and other routers, and
precisely retrieve the RSS values and corresponding MAC
addresses as identifiers without introducing any extra hard-
ware infrastructure. Fig. 1 presents the system architecture
of WinSMS, which includes a main AP, remote APs, a back-
end server, users and their MDs. The main AP provides the
basic WLAN Internet services, receives UDP packets sen-
t by remote APs, and forwards the data to a server. The
server is responsible to store and parse the data. We up-
grade the firmware of remote APs with OpenWrt, and use
Libpcap to capture and analyze RSS packets in the exist-
ing WiFi traffic, extract relevant data and forward them to
the main AP. Since these remote APs can overhear packet-
s of other remote APs as well, all of them become natural
online reference points with their physical coordinates and
real-time RSS readings. In this manner, WinSMS is able to
collect the data among the APs as the online labeled data
DSAP and the RSS data associated with MDs as the unla-
beled data DTwithout introducing extra infrastructure or
any intrusiveness on user side.
2.2 TKL-WinSMS
With DSand DT, TKL aims to reveal the shared knowl-
edge in signal space across different domains, select a new
domain-invariant kernel as input data for SVR training, and
then construct an adaptive localization model to precise-
ly estimate the locations of MDs in dynamic indoor envi-
ronments. The methodology of TKL-WinSMS consists of 6
steps. Step 1 is to compute the source kernel KS, the target
kernel KTand the cross-domain kernel matrix KST using a
predefined input kernel function k. The next step is to eval-
uate the distribution differences between DSand DTin the
kernel Hilbert space. However, KSand KThave distinc-
t dimensions. Thus, we construct an extrapolated source
kernel KSusing the eignesystem of KT. The eignesys-
tem of KTcan be easily obtained by eigendecomposition
KTΦT=ΦTΛTas Step 2. Then, Step 3 is to compute
the eigenvector matrix of the extrapolated source kernel KS
by ΦS'KST ΦTΛ1
T. We formalize the distribution dis-
Figure 2: Layout of the testbed.
crepancy between KSand the ground truth source kernel
KSin terms of the Nystr¨
om approximation error. As such,
to minimize the distribution divergence is equal to minimize
the approximation error, which can be achieved by using
squared loss as follows:
min
ΛkKSKSk2=kΦSΛΦ>
SKSk2,
λiζλi+1 , i = 1, . . . ,lT1, λi0, i = 1,...,lT,
(1)
where Λ= diag{λ1,...,λlT}are the lTnonnegative eigen-
spectrum parameters and ζis the eigenspectrum damping
factor. Step 4 is to estimate Λby translating the opti-
mization problem (1) into a standard convex quadratic pro-
gramming (QP) with linear constraints, which can be easily
solved by well-established convex optimization algorithms,
such as the interior-point algorithm. Step 5 is to build up
the domain-invariant kernel KAon both the source and tar-
get data using KA=ΦAΛΦ1
A, where ΦA.
= [ΦS;ΦT].
Then, the estimated domain-invariant kernel KAcan be di-
rectly adopted as the input for SVR training to construct the
adaptive localization model via LIBSVM package as Step 6.
The output of the model are the estimated locations of MDs.
3. IMPLEMENTATION AND EVALUATION
Extensive experiments were conducted in a 600m2real
multi-functional lab across 6 months to validate the perfor-
mance of TKL-WinSMS. The experiments include two phas-
es: the initial phase (T1) and a phase conducted 6 months
later (T2). During T1, RSS samples of a Nexus 6 were col-
lected at 47 offline CPs using WinSMS. At each point, we
collected 100 RSS samples and denoted all data by DSCP .
During T2, we collected RSS data (100 RSS samples per lo-
cation) at 15 testing points (TPs) as well as the 14 APs using
WinSMS. Denote the data at 14 APs as DSAP and the RSS
samples at 15 TPs as DT. The locations of the CPs, TPs
and APs are shown in Fig. 2. The localization model built
up by TKL-WinSMS is trained on all labeled data DSCP
and DSAP , as well as unlabeled target data DTto learn the
domain-invariant feature representation. The performance
of TKL-WinSMS is compared with basic SVR (use DSCP
only for SVR training) and SVR-WinSMS (use both DSCP
and DSAP for SVR training). The RBF kernel is adopted
in all the three approaches. The overall performance com-
parison is presented in Fig. 3, with the detailed location
error on each TP using the three approaches demonstrated
463
0246810
Location Error [m]
0
0.2
0.4
0.6
0.8
1
CDF
SVR
SVR-WinSMS
TKL-WinSMS
Figure 3: CDF of location error.
in Fig. 4. The mean localization accuracy of TKL-WinSMS
is 1.860 m, which obtains a tremendous performance im-
provement of 38.85% and 26.66% over SVR (3.041 m) and
SVR-WinSMS (2.536 m), respectively. TKL-WinSMS also
outperforms others at nearly every TP as presented in Fig. 4.
These consistent performance gains verify that constructing
the robust domain-invariant kernel KAby TKL is much
more advantageous than SVR and SVR-WinSMS that on-
ly leverage raw RSS data. Furthermore, by fully exploring
the eigenspaces of both the source and target domain da-
ta to learn a domain-invariant kernel, TKL-WinSMS is able
to correctly revealing the related knowledge, and kernelizing
the original RSS data across different domains (time period-
s) for adaptive indoor localization. Thus, by leveraging the
online labeled data DSAP obtained by WinSMS as well as
the domain-invariant kernel KAlearned by TKL, an adap-
tive localization model can be constructed to consistently
provide reliable positioning service with various indoor en-
vironment dynamics.
4. CONCLUSION
In this paper, we proposed TKL-WinSMS, which is able
to construct a robust localization model for adaptive lo-
calization in dynamic indoor environments. We develope-
d WinSMS to enable COTS WiFi routers as online refer-
ence points by extracting the real-time RSS readings among
them. With these online data and the offline calibrated
radio map as labeled source data, we further combine the
RSS readings from the target MDs as unlabeled target da-
ta, to design a robust localization model using TKL. It can
construct a domain-invariant kernel that minimizes the d-
ifference between the source and target distributions in the
reproducing kernel Hilbert space instead of the raw noisy
signal space. By leveraging the resultant kernel as input for
the SVR training, the trained localization model can inher-
it the information from online phase to adaptively enhance
the offline calibrated radio map. Extensive experimental re-
sults verify the superiority of TKL-WinSMS. We envision
TKL-WinSMS as a fundamental and indispensable part for
WiFi-based IPS to cope with various environmental dynam-
ics and achieve a robust localization service consistently.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
0
2
4
6
8
m
SVR
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
0
2
4
m
SVR-WinSMS
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Testing Point Index
0
2
4
m
TKL-WinSMS
Figure 4: Localization accuracy at each testing
point.
Acknowledgement
This research is funded by the Republic of Singapore Nation-
al Research Foundation (NRF) through a grant to the Berke-
ley Education Alliance for Research in Singapore (BEARS)
for the Singapore-Berkeley Building Efficiency and Sustain-
ability in the Tropics (SinBerBEST) Program. BEARS has
been established by the University of California, Berkeley as
a center for intellectual excellence in research and education
in Singapore.
5. REFERENCES
[1] S. Kumar, S. Gil, D. Katabi, and D. Rus. Accurate
indoor localization with zero start-up cost. In
Proceedings of the 20th annual international conference
on Mobile computing and networking, pages 483–494.
ACM, 2014.
[2] M. Long, J. Wang, J. Sun, and P. S. Yu. Domain
invariant transfer kernel learning. Knowledge and Data
Engineering, IEEE Transactions on, 27(6):1519–1532,
2015.
[3] D. Lymberopoulos, J. Liu, X. Yang, R. R. Choudhury,
V. Handziski, and S. Sen. A realistic evaluation and
comparison of indoor location technologies: experiences
and lessons learned. In Proceedings of the 14th
International Conference on Information Processing in
Sensor Networks, pages 178–189. ACM, 2015.
[4] H. Zou, B. Huang, X. Lu, H. Jiang, and L. Xie. A
robust indoor positioning system based on the
procrustes analysis and weighted extreme learning
machine. Wireless Communications, IEEE
Transactions on, 15(2):1252–1266, 2016.
464
... A dynamic localization system is proposed for adapting the mapping of signal space and physical location space ( [19]- [20]) and a resampling method is proposed for certain locations chosen from a set of remeasurement fingerprints ( [21]- [22]), both of which consider multiple sequential measurements over different time periods. An adaptive indoor localization using transfer kernel is proposed in [23], where the radio map and the RSSI readings are regarded as labeled and unlabeled data, respectively, and transfer learning is then applied. A wireless positioning method based on Deep Learning is prosed in [24], the positioning is casted in a four-layer Deep Neural Network(DNN) structure pre-trained by Stacked Denoising Autoencoder(SDA) that is capable of learning reliable features from a large set of noisy samples and avoids handengineering. ...
... By incorporating and extending the idea from object class recognition in [23], simultaneous AP selection and localization is proposed in this paper based on Bayesian framework, category model and EM algorithm on a shifting joint probability. The block diagram is given in Fig. 5. ...
Article
Taking advantage of widely deployed access points (AP), WiFi fingerprint based localization is of importance in indoor internet-of-things (IOT) environments. Nevertheless, spatio-temporal variation is one of its intractable problems, indicating severely environmental dynamics and uncertainty of decision. In this case, the localization accuracy drops significantly. In this paper, we attempt to overcome effects of spatio-temporal variations from two aspects: filtering of the training data and selection of partially valuable APs for matching in test phase. The key idea is to match partial unaffected measurements with ’clean’ unaffected fingerprints. Bayesian framework and category model are presented for WiFi fingerprints. Two binary hidden variables with different dimensions are introduced to identify singular fingerprints and affected measurements respectively by employing expectation-maximization (EM) algorithms. EM based filter and simultaneous AP selection and localization are then proposed to obtain an optimal matching. Experimental results show that our proposed scheme greatly improves the localization accuracy in severely dynamic indoor environments.
... Thus, by leveraging TL, the energy efficiency, which is one of requirements of 6G mRAN, can be improved. Added to the aforementioned examples, TL has been widely applied to optimize communication systems including resource management [96][97][98][99][100], channel estimation [95,100,101], caching [102][103][104], and localization [105][106][107] for improving the efficiency of learning. ...
Article
Full-text available
Predictably, the upcoming six generation (6G) networks demand ultra-massive interconnectivity comprising densely congested sustainable small-to-tiny networks. The conventional radio access network (RAN) will be redesigned to provide the necessary intelligence in all areas to meet required network flexibility, full coverage, and massive access. In this respect, this paper focuses on intelligent massive RAN (mRAN) architecture and key technologies fulfilling the requirements. Particularly, we investigate potential AI algorithms for network and resource management issues in 6G mRAN. Furthermore, we summarize the research issues in edge technologies and physical layer intelligence on 6G network architecture.
... They either require occupants to carry or wear the devices or need extra infrastructure for activity monitoring. With the pervasive and wide availability of WiFi infrastructure, leveraging WiFi signals to estimate the location of occupants and distinguish human activities becomes feasible [4,5]. Channel State Information (CSI), a fine-grained channel measurement from the physical layer that describes how WiFi signals propagate from the transmitter (TX) to receiver (RX) through multiple paths at the granularity of OFDM subcarriers, has become available from commodity WiFi NIC [2]. ...
Conference Paper
Full-text available
Existing human activity recognition approaches require either the deployment of extra infrastructure or the cooperation of occupants to carry dedicated devices, which are expensive, intrusive and inconvenient for pervasive implementation. In this paper, we propose SmartSense, a device-free human activity recognition system based on a novel machine learning algorithm with existing commercial off-the-shelf (COTS) WiFi routers. By exploiting the prevalence of existing WiFi infrastructure in buildings, we developed a novel OpenWrt based firmware for COTS WiFi routers to collect the CSI measurements from regular data frames. To identify different human activities, an automatic kernel representation learning method, namely auto-HSRL, is established to selection informative Hilbert space patterns from time, frequency, wavelet, and shape domains. A new information fusion tool based on multi-view kernel learning is proposed to combine the representations extracted from diverse perspectives and build up a robust and comprehensive activity classifier. Extensive experiments were conducted in an office and the experimental results demonstrate that SmartSense outperforms existing methods and achieves a 98% activity recognition accuracy.
... Nevertheless, they require occupants to carry a tag at all times, which introduces additional infrastructure cost. On the other hand, WiFi has certain unique merits for occupancy sensing, such as the wide availability of WiFi infrastructures in commercial buildings, and the prevalence of WiFi-enabled MDs [58][59][60]. Several WiFi-based Indoor Positioning Systems (IPSs), which can be used for occupancy sensing, have been proposed in recent decades [61][62][63][64]. The pioneering work of employing WiFi for indoor localization is RADAR [61]. ...
Article
Artificial lighting accounts for a significant proportion (19%) of energy consumption in building environments. This large contribution calls for the creation of energy-efficient lighting control schemes. In this article, we present WinLight, a novel occupancy-driven lighting control system that aims to reduce energy consumption while simultaneously preserving the lighting comfort of occupants. By leveraging the fine-grained occupancy information estimated by existing WiFi infrastructure in a non-intrusive manner, WinLight computes an appropriate dimming command for each lamp based on a novel lighting control algorithm. A centralized lighting control system assigns these commands to a zonal gateway, and occupancy-driven lighting control is achieved by actuating the brightness adjustment with a local controller integrated within each lamp. Moreover, a WinLight App is designed to enable occupants to customize their luminance preferences and to control nearby lamps using their mobile devices. We implemented WinLight in a 1500 m2 multi-functional office in Singapore and conducted an experiment during 24 weeks. The experimental results demonstrate that WinLight achieves 93.09% and 80.27% energy savings compared to static scheduling lighting control scheme and PIR sensor based lighting control scheme while guaranteeing the personalized lighting comfort of each occupant.
... RFID and Bluetooth require the occupant to carry an additional tag and install extra sensors for positioning purpose. On the other hand, WiFi has been acknowledged as the most promising alternative to GPS for indoor localization because COTS WiFi devices and infrastructures are widely available in indoor environments and most of the MDs are equipped with WiFi modules [37][38][39][40]. However, existing WiFi-based systems require a dedicated App to be installed on the occupant's MD to perform active AP scanning constantly for RSS data acquisition, which induces additional efforts on the user side as well as high battery consumption. ...
Article
Buildings accounted for half of global electricity consumption in recent years. Accurate occupancy information could improve the energy efficiency and reduce the energy consumption in built environments. Although prior studies have explored various sensing techniques for occupancy sensing, these solutions still suffer from serious drawbacks, e.g. their estimated occupancy information is coarse, extra infrastructure is required, and the privacy of occupants is exposed. In this paper, we present the design and implementation of a novel and practical occupancy sensing system, WinOSS, which is able to provide fine-grained occupancy information thoroughly by leveraging existing commodity WiFi infrastructure along with the WiFi-enabled mobile devices carried by occupants. We have implemented WinOSS in a 1500 m² built environment for four weeks to validate its performance. Extensive experimental results demonstrate that WinOSS outperforms existing occupancy sensing techniques, and provides comprehensive fine-grained occupancy information (including occupancy detection, counting and tracking) in an accurate, reliable, cost-effective and non-intrusive manner.
Article
With the rapidly growing demand for Location-Based Services in indoor environments, fingerprint-based indoor positioning has caused great interest due to its high positioning accuracy and low equipment cost. However, the standard signal radio map cannot provide consistent high positioning accuracy under environmental changes and new scenarios. To address this problem, we present a novel indoor positioning Transfer Learning(TL) system based on improved TrAdaBoost. First perform phase correction on the raw CSI(Channel State Information) phase, then use One-vs-Rest algorithm and One-Hot coding, which can realize the multi-classification ability of the TrAdaBoost algorithm. Meanwhile, optimize the iterative process of the algorithm by using a correction factor. In addition, Confidence regression is used to obtain the final estimated position. Experimental results show that the positioning accuracy can be improved by 35% in dynamic environment conditions. The proposed method can improve the positioning accuracy by an average of 30% in new scenes and the Site Survey Overhead(SSO) is reduced by 40%. Compared to other positionb algorithm, our proposed method has more robustness in time and space.
Preprint
Full-text available
With outstanding features, Machine Learning (ML) has been the backbone of numerous applications in wireless networks. However, the conventional ML approaches have been facing many challenges in practical implementation, such as the lack of labeled data, the constantly changing wireless environments , the long training process, and the limited capacity of wireless devices. These challenges, if not addressed, will impede the effectiveness and applicability of ML in future wireless networks. To address these problems, Transfer Learning (TL) has recently emerged to be a very promising solution. The core idea of TL is to leverage and synthesize distilled knowledge from similar tasks as well as from valuable experiences accumulated from the past to facilitate the learning of new problems. Doing so, TL techniques can reduce the dependence on labeled data, improve the learning speed, and enhance the ML methods' robustness to different wireless environments. This article aims to provide a comprehensive survey on applications of TL in wireless networks. Particularly, we first provide an overview of TL including formal definitions, classification, and various types of TL techniques. We then discuss diverse TL approaches proposed to address emerging issues in wireless networks. The issues include spectrum management, localization, signal recognition, security, human activity recognition and caching, which are all important to next-generation networks such as 5G and beyond. Finally, we highlight important challenges, open issues, and future research directions of TL in future wireless networks.
Chapter
With the rapid development of Mobile Internet and the popularity of mobile terminal equipment, demands on the Location-Based Services (LBS), especially the indoor localization are growing. The Access Point (AP) Filtering work is a very critical part in the indoor localization technology. Aiming at the problems which exist in the present AP filtering algorithms which include the low positioning accuracy, the long computation time and the high computational complexity, in this paper, we proposed an AP filtering algorithm based on the Cramer-Rao Lower Bound (CRLB) and Gradient standard. The AP filtering algorithm can effectively remove the APs with large noise to avoid them from negative effect to the results of localization, then to achieve the goals that reducing the calculation complexity in the positioning phase and improving the positioning accuracy. The experiment results show that the AP filtering algorithm proposed in this paper is superior to the traditional AP filtering algorithms in positioning performance, especially, it is applicable to the case of AP number limited and practical public places with multiple APs at the same time.
Article
Full-text available
Indoor Positioning System (IPS) has become one of the most attractive research fields due to the increasing demands on Location Based Services (LBSs) in indoor environments. Various IPSs have been developed under different circumstances, and most of them adopt the fingerprinting technique to mitigate pervasive indoor multipath effects. However, the performance of the fingerprinting technique severely suffers from device heterogeneity existing across commercial off-the-shelf mobile devices (e.g. smart phones, tablet computers, etc.) and indoor environmental changes (e.g. the number, distribution and activities of people, the placement of furniture, etc.). In this paper, we transform the Received Signal Strength (RSS) to a standardized location fingerprint based on the Procrustes analysis, and introduce a similarity metric, termed Signal Tendency Index (STI), for matching standardized fingerprints. An analysis on the capability of the proposed STI in handling device heterogeneity and environmental changes is presented. We further develop a robust and precise IPS by integrating the merits of both the STI and Weighted Extreme Learning Machine (WELM). Finally, extensive experiments are carried out and a performance comparison with existing solutions verifies the superiority of the proposed IPS in terms of robustness to device heterogeneity.
Conference Paper
Full-text available
We present the results, experiences and lessons learned from comparing a diverse set of technical approaches to indoor localization during the 2014 Microsoft Indoor Localization Competition. 22 different solutions to indoor localization from different teams around the world were put to test in the same unfamiliar space over the course of 2 days, allowing us to directly compare the accuracy and overhead of various technologies. In this paper, we provide a detailed analysis of the evaluation study's results, discuss the current state-of-the-art in indoor localization, and highlight the areas that, based on our experience from organizing this event, need to be improved to enable the adoption of indoor location services.
Article
Domain transfer learning generalizes a learning model across training data and testing data with different distributions. A general principle to tackle this problem is reducing the distribution difference between training data and testing data such that the generalization error can be bounded. Current methods typically model the sample distributions in input feature space, which depends on nonlinear feature mapping to embody the distribution discrepancy. However, this nonlinear feature space may not be optimal for the kernel-based learning machines. To this end, we propose a transfer kernel learning (TKL) approach to learn a domain-invariant kernel by directly matching source and target distributions in the reproducing kernel Hilbert space (RKHS). Specifically, we design a family of spectral kernels by extrapolating target eigensystem on source samples with Mercer’s theorem. The spectral kernel minimizing the approximation error to the ground truth kernel is selected to construct domain-invariant kernel machines. Comprehensive experimental evidence on a large number of text categorization, image classification, and video event recognition datasets verifies the effectiveness and efficiency of the proposed TKL approach over several state-of-the-art methods.
Article
Recent years have seen the advent of new RF-localization systems that demonstrate tens of centimeters of accuracy. However, such systems require either deployment of new infrastructure, or extensive fingerprinting of the environment through training or crowdsourcing, impeding their wide-scale adoption. We present Ubicarse, an accurate indoor localization system for commodity mobile devices, with no specialized infrastructure or fingerprinting. Ubicarse enables handheld devices to emulate large antenna arrays using a new formulation of Synthetic Aperture Radar (SAR). Past work on SAR requires measuring mechanically controlled device movement with millimeter precision, far beyond what commercial accelerometers can provide. In contrast, Ubicarse's core contribution is the ability to perform SAR on handheld devices twisted by their users along unknown paths. Ubicarse is not limited to localizing RF devices; it combines RF localization with stereo-vision algorithms to localize common objects with no RF source attached to them. We implement Ubicarse on a HP SplitX2 tablet and empirically demonstrate a median error of 39 cm in 3-D device localization and 17 cm in object geotagging in complex indoor settings.