590 IEEE INTERNET OF THINGS JOURNAL, VOL. 3, NO. 4, AUGUST 2016
A Feature-Scaling-Based k-Nearest Neighbor
Algorithm for Indoor Positioning Systems
Dong Li, Baoxian Zhang, Senior Member, IEEE,andChengLi,Senior Member, IEEE
Abstract—With the increasing popularity of WLAN infras-
tructure, WiFi fingerprint-based indoor positioning systems have
received considerable attention recently. Much existing work in
this aspect adopts classification techniques that match a vector
of radio signal strengths (RSSs) reported by a mobile station
(MS) to pretrained reference fingerprints sampled from differ-
ent access points ( APs) at different reference points (RPs) with
known positions. However, in the calculation of signal distances
between different RSS vectors, existing techniques fail to consider
the fact that equal RSS differences at different RSS levels may not
mean equal differences in geometrical distances in complex indoor
environment. To address this issue, in this paper, we propose
a feature-scaling-based k-nearest neighbor (FS-kNN) algorithm
for achieving improved localization accuracy. In FS-kNN, we
build a novel RSS-level-based FS model, which introduces RSS-
level-based scaling weights in the computation of effective signal
distances between signal vector reported by a MS and reference
fingerprints in a radio map. Experimental results show that FS-
kNN can achieve an average location error as low as 1.70 m, which
is superior to existing work.
Index Terms—Feature scaling (FS), fingerprint-based
localization, indoor positioning system, k-nearest neighbor
(kNN).
I. INTRODUCTION
I
NDOOR positioning has received great attention recently
because position information is essential for providing
location-based services (LBSs) [1], which offers intelligent ser-
vices in various fields in the context of Internet of Things (IoT)
[2]–[4]. For example, position-based navigation has been used
inside Copenhagen Airport [5]. Passengers there can use it to
plan their paths inside the airport and to get expected informa-
tion in an interactive way. Moreover, an indoor tracking system
was deployed in Hartford hospital, which helps tracking expen-
sive equipment and also assisting patients there to efficiently
Manuscript received August 07, 2015; accepted October 06, 2015. Date
of publication October 27, 2015; date of current version July 27, 2016. This
work was supported in part by the National Science Foundation (NSF) of
China under Grant 61531006, Grant 61471339, and Grant 61173158, in part
by the National Science and Technology R&D Program of China under Grant
2014BAK06B01, and in part by the Natural Sciences and Engineering Research
Council (NSERC) of Canada under Discovery Grant 293264-12 and Strategic
Project Grant STPGP 397491-10. An earlier version of this paper was presented
at the IEEE GLOBECOM 2014—Ad Hoc and Sensor Networks Symposium,
Austin, TX, USA, Dec. 8–12, 2014.
D. Li is with the Research Center of Ubiquitous Sensor Networks,
University of Chinese Academy of Sciences, Beijing 100049, China (e-mail:
lidong10b@mails.ucas.ac.cn).
B. Zhang is with the Research Center of Ubiquitous Sensor Networks,
University of Chinese Academy of Sciences, Beijing 100049, China, and also
with the Jiangsu Internet-of-Things Research and Development Center, Wuxi
214135, China (e-mail: bxzhang@ucas.ac.cn).
C. Li is with the Faculty of Engineering and Applied Science, Memorial
University, St. John’s, NL A1B 3X5, Canada (e-mail: licheng@mun.ca).
Digital Object Identifier 10.1109/JIOT.2015.2495229
use medical resources in the hospital [6]. In addition, location-
aware advertising usually delivers location-specific coupons or
discount information to customers based on their locations and
interests [7]. However, indoor environments are very compli-
cated such that there usually exist many obstacles, such as
walls, furniture, human beings, and consequently fluctuations
of wireless signals because of multipath effects. These obsta-
cles and the signal fluctuations at different scales can cause
significant degradation in the accuracy of indoor positioning,
which limits the usefulness and degree of comfortableness for
providing practical LBS services.
One popular way to estimate a mobile user’s position in
indoor environment is to use reference fingerprints based on
WiFi signals (e.g., RADAR [8]). In general, there are two
phases for a fingerprint-based positioning algorithm: 1) offline
phase and 2) online phase. During offline phase, a target envi-
ronment is calibrated by performing a site-survey with a mobile
station (MS). The whole area is typically overlaid with a set
of predetermined grid points known as reference points (RPs)
where sample data should be acquired. At each RP, many sam-
ples of radio signal strengths (RSSs) from different access
points (APs) are repeatedly collected. Each RP has a position
known apriori. Then, a vector of mean RSS values associated
with the position of the RP is recorded into a database known
as a radio map. During online phase, an MS at unknown posi-
tion reports its instantly sampled vector of RSS values from
surrounding APs to a remote server. The server uses a pattern-
matching algorithm to estimate the current position of the MS
and then returns the estimated position back to the MS. A com-
monly used pattern-matching algorithm to estimate the position
is the k-nearest neighbor (kNN) algorithm, which finds the k
fingerprints with the minimum signal distances to the instantly
reported RSS vector among different fingerprints in the radio
map that the server locally stores. The first k RPs resulting in
the minimum signal distances will be used to derive the esti-
mated location. One big issue in kNN and also its variants (e.g.,
[9], [10]) is that, i n their calculation of signal distance, equal
RSS distances are always equally treated. That is to say, equal
RSS differences are assumed to account for equal geometrical
distances. However, this may not be true in indoor environment.
In reality, equal RSS differences at different RSS levels may
often be caused by different geometrical distances, especially,
in complex indoor environment. The performance of an indoor
localization algorithm can be largely degraded if the impacts of
RSS differences at different levels on positioning accuracy are
not fully considered.
To address the above issue, in this paper, we first conduct
two motivating experiments, which demonstrate that equal RSS
2327-4662 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.