Evaluation of Low-Cost/High-Accuracy Indoor Positioning Systems

Abstract

Indoor positioning is a challenging research topic. Over the years, many different measurement principles and algorithms have been proposed. Each system has its own advantages and drawbacks, therefore trade-offs have to be made. For example, one generally needs to make a trade-off between cost and accuracy. However, recent developments in sensing technology have led to commercial systems that advertise sub-decimeter positioning accuracy for less than €1k. In this paper, we benchmark the accuracy of indoor positioning systems by Pozyx labs and Marvelmind robotics, as well as the VIVE tracker by HTC and Aruco Marker tracking in OpenCV. Results show that these systems achieve an average dynamic positioning accuracy of approximately 150 mm, 20 mm, 8 mm and 100 mm, respectively.

Full text

Evaluation of Low-Cost/High-Accuracy Indoor Positioning Systems
Robin Amsters,
Eric Demeester
and Peter Slaets
KU Leuven,
Department of Mechanical Engineering,
3000 Leuven, Belgium
Email: {firstname.lastname}@kuleuven.be
Nobby Stevens
KU Leuven,
Department of Electrical Engineering,
3000 Leuven, Belgium
Email:nobby.stevens@kuleuven.be
Quinten Lauwers
KU Leuven,
3000 Leuven, Belgium
Email: quinten.lauwers@student.kuleuven.be
Abstract—Indoor positioning is a challenging research topic. Over
the years, many different measurement principles and algorithms
have been proposed. Each system has its own advantages and
drawbacks, therefore trade-offs have to be made. For example,
one generally needs to make a trade-off between cost and
accuracy. However, recent developments in sensing technology
have led to commercial systems that advertise sub-decimeter
positioning accuracy for less than e1k. In this paper, we
benchmark the accuracy of indoor positioning systems by Pozyx
labs and Marvelmind robotics, as well as the VIVE tracker by
HTC and Aruco Marker tracking in OpenCV. Results show that
these systems achieve an average dynamic positioning accuracy of
approximately 150 mm, 20 mm, 8 mm and 100 mm, respectively.
Keywords–Indoor positioning; benchmarking; accuracy
I. INTRODUCTION
Positioning is not a particularly new problem. Mankind
has attempted to determine his position for centuries, using
instruments such as sextants, clocks, almanacs, maps, etc.,
One of the largest revolutions in this field is probably the
advent of the Global Positioning System (GPS), which can
provide position information almost anywhere on earth to
anyone with a receiver. However, GPS often does not work in
indoor environments, as the signals of the satellites lose much
strength when penetrating the walls of buildings [1], making
it difficult to receive them with traditional, low-cost sensors.
Moreover, the accuracy (that is, the Euclidean distance between
the true and the reported position) of such sensors is limited to
a couple of meters, which is insufficient for many indoor appli-
cations. A study performed by the National Exposure Research
Laboratory indicates that most people spend about 90% of
their time indoors [2]. Therefore, indoor location information
provides many business opportunities. This is illustrated by
the fact that the indoor mapping market is rapidly increasing
in size, and is estimated to be worth about $10 billion by 2020
[3]. As GPS cannot be used for indoor environments, different
technologies are required to obtain this position information.
While GPS has become the de facto standard for outdoor
environments, no such standard exists for indoor spaces [4].
The wide variety of indoor environments has prompted an
equally wide variety of Indoor Positioning Systems (IPS), each
with their own advantages and drawbacks. Usually, a trade-
off has to be made between accuracy and cost. In the past,
this trade-off was quite significant. However, in recent years,
a number of positioning systems have come to market that
should provide high accuracy (≤10 cm) at relatively low-
cost (≤e1k), potentially bridging the gap that existed before.
In this paper, we evaluate several commercially available
high-accuracy/low-cost indoor positioning systems. Different
measurement principles are represented, namely ultrasound,
ultrawideband radio, infrared light and computer vision. A
highly accurate (sub-millimeter) infrared measurement system
is used as a ground truth reference. The positioning systems
are benchmarked in the same environment, thus enabling an
objective comparison.
The rest of the paper is structured as follows; Section II
presents related work and the main contributions of this paper.
Section III elaborates on the different positioning systems
that were considered. The experimental setup is explained in
Section IV, and results are presented in Section V. Finally, a
conclusion is drawn in Section VI.
II. RE LATE D WO RK
Van Haute et al. [5] benchmarked several indoor position-
ing systems in a healthcare environment. Low-cost technolo-
gies like Wi-Fi, ZigBee and bluetooth low-energy (BLE) were
used. Accuracy of static measurements in the order of 1 to
4 meters was obtained. These kinds of radio-frequency (RF)
based systems are relatively popular for indoor localization,
due to the widespread availability of the hardware. However,
accuracy is rarely below 1 meter [1].
Ultrawideband (UWB) has been an increasingly important
topic in indoor positioning research in recent years. Typically,
UWB positioning systems determine the distance between
static anchors and a mobile node based on signal travel time.
A position estimate is then obtained via triangulation. Ruiz
and Seco [6] compared the commercial UWB systems sold
by DecaWave (the Pozyx system uses DecaWave transceivers)
and BeSpoon. Manual measurements with a ruler were used
as ground truth. Ruiz and Granja [7] added the Ubisense
system to this comparison, and extended the testing space
to a larger industrial warehouse as well (rather than the lab
environment as in [6]). A comparison was made based on
the ranging accuracy and positioning accuracy with a particle
filter. An overview of the positioning results of these papers
can be found in Table I. The Ubisense system was evaluated
separately as well by Maleek and Sadeghpour [8]. The focus of
their work was dynamic positioning and localization of factory
workers in order to increase safety. A number of experiments
were performed by placing a tag on a Lego track, as well
as experiments where the tag was used to localize a person.
Ground truth locations are calculated based on the starting
time and the known layout of the lego track for the first set
15Copyright (c) IARIA, 2019. ISBN: 978-1-61208-691-0
ALLSENSORS 2019 : The Fourth International Conference on Advances in Sensors, Actuators, Metering and Sensing

of experiments. For the experiments on worker localization,
the real position was determined with a robotics total station
(Leica iCON Robot 50). Their findings indicate 2D accuracy
of 15-31 cm depending on the experiment. Dabove et al.
evaluated the Pozyx system in [9]. In an office environment,
the average 3D positioning accuracy was 100 mm, and the
accuracy of the range measurements was found to be 320 mm.
In a narrow corridor, the horizontal accuracy and range error
were determined to be 87.4 mm and 225 mm, respectively.
Surveying equipment was used as ground truth reference for
static range measurements, and a grid pattern was used for
static positioning ground truth measurements. Finally, Ridolfi
et al. [10] evaluated the Pozyx kit as a positioning technology
for sports postures. Average positioning errors of 200 mm were
recorded (depending on tag placement and activity) and the
authors propose several implementations of filtering algorithms
to reduce this error. A motion capture system (MOCAP) was
used as ground truth.
TABLE I. OVERVIEW OF PREVIOUS UWB BENCHMARKING
RESULTS REPORTED IN [6] AND [7]. P90 REPRESENTS THE 90%
INTERVAL OF THE CUMULATIVE ERROR DISTRIBUTION
DecaWave BeSpoon Ubisense
Mean accuracy (office) [m] 0.24 0.51 /
P90 (office) [m] 0.51 0.99 /
Mean accuracy (factory) [m] 0.49 0.71 1.1
P90 (factory) [m] 1.09 1.16 2.39
The ranging of the Marvelmind system was, to the best of
our knowledge, only benchmarked by Cernohorsk and Novk
[11]. The error of the range measurements was found to be
in the order of a few centimeters, though some outliers exist
in absence of a direct line of sight. The absolute positioning
error was not evaluated.
The HTC VIVE is a relatively new system, and thus little
research about it is available. Chang et al. [12] compared sev-
eral head-mounted virtual reality systems, namely the 3Glasses
D2, Oculus Rift DK2, Google Cardboard and Samsung Gear
VR based on metrics such as positioning precision and sen-
sitivity. One degree of freedom motion was considered by
mounting the headsets on a servo motor. However, the HTC
VIVE was not considered. Niehorster et al. [13] did evaluate
the precision and latency of the HTC VIVE specifically.
Latency is conservatively estimated to be 22 ms, significantly
less than the latency of the 3Glasses D2 (44 ms) and Oculus
Rift DK2 (48 ms) measured by Chang et al. [12]. Niehorster
et al. concluded that the VIVE measures at a tilted reference
plane relative to the ground plane. However, the angles and
positions reported by the VIVE are consistent as long as the
system does not lose tracking. Positioning accuracy was not
reported.
It is clear that while some of these systems have been
evaluated in previous publications, further work is still
required. For example, position accuracy is not always
specified and even when reported, this is often the average
value. In the context of, for example, autonomous operation
of mobile robots, one is often also interested in the P95,
that is, the 95% interval of cumulative error distribution.
These results are usually not included ([6] and [7] do report
90% intervals, as shown in Table I). Additionally, static grid
measurements or measurements with a ruler are often used
as a ground truth reference. A survey of papers published in
the proceedings of the International Conference on Indoor
Positioning and Indoor Navigation (IPIN) concludes that this
method is used quite often in indoor positioning research [14].
In this work, we perform ground truth measurements with
a highly (sub-millimeter) accurate positioning system with
a relatively high update rate (50 Hz, see Section IV). This
allows better characterization of the dynamic performance of
the considered indoor positioning systems.
In summary, our main contributions are:
•Evaluation of several measurement principles in the
same environment.
•Measurements of moving receivers are compared with
a highly accurate, fast-measuring ground truth refer-
ence.
•Evaluation of positioning accuracy, rather than ranging
distance.
•Reporting of mean accuracy and P95 values, to pro-
vide both an overview of normal performance and
worst case scenarios.
III. SYS TE MS U ND ER EVALUATI ON
Table II provides an overview of the main specifications
of the IPS considered in this paper. The system cost includes
value added tax, and the specifications for the Aruco system are
based on the camera used in this work (Logitech HD webcam).
Scalability refers to whether a particular system can easily be
extended up to larger environments.
The indoor positioning system by Marvelmind robotics
uses ultrasound ranging to determine the position of one
or more mobile sensor modules (referred to as hedgehogs).
Ultrasound ranging is also used by the transmitters (referred
to as beacons) to determine their relative position. Therefore,
the Marvelmind system is self-calibrating. The sensor modules
have built-in rechargeable batteries, and whether a module is
a beacon or hedgehog can be selected in the software and
changed at will. The maximum update rate for tracking a
single hedgehog is 16 Hz. The system uses time division
multiplexing, so if multiple hedgehogs are tracked, the update
rate becomes:
Fupdate =16
nhedgehog
(1)
With Fupdate the update rate of every hedgehog and nhedgehog
the number of tracked hedgehogs.
The indoor positioning system by Pozyx labs uses ultraw-
ideband radio as a distance estimation principle. Additionally,
data from a 9-axis Inertial Measurement Unit (IMU) is fused
in order to improve the position estimate. The advantage of
ultrawideband over other RF technologies is that the increased
bandwidth makes it more likely that at least some of the
transmitted frequencies will go through or around obstacles
[15]. Therefore, accuracy can be significantly higher than, for
example, Bluetooth or Wi-Fi based positioning [4]. The system
has the option to self-calibrate, but we performed a manual
calibration to improve accuracy [16] (see Section V-B).
The HTC VIVE is sold as a virtual reality headset, and
ships with a Head-Mounted Display (HMD), two controllers
with infrared receivers and two infrared transmitters (called
lighthouses). Recently, a standalone tracker module has also
16Copyright (c) IARIA, 2019. ISBN: 978-1-61208-691-0
ALLSENSORS 2019 : The Fourth International Conference on Advances in Sensors, Actuators, Metering and Sensing

been released to enable simpler tracking of objects [17]. It
is the positioning of the tracker that was evaluated in this
paper. Each lighthouse is equipped with two lasers, which
sweep across its horizontal and vertical axes. The infrared
laser sweeps are detected by photodiodes which are mounted
on the controllers, headset or tracker modules. The difference
between arrival times of the laser at the photodiodes is used
to determine the position and orientation of the modules [13].
These laser measurements function mostly as drift correction.
In between sweeps, positions are estimated with IMU-based
dead-reckoning [18]. Contrary to the other systems considered
in this paper, the VIVE was not originally designed to be a
standalone positioning system. However, accurate position and
orientation tracking is required to provide a good virtual reality
user experience. As the OpenVR Software Development Kit
(SDK) has a published driver for the tracking hardware [19],
it is possible to access all the tracking information outside a
gaming environment, thus opening the door for a wide range
of other applications. As a positioning system, however, the
user experience is not as smooth as the Pozyx or Marvelmind
systems. For example, steamVR needs to be continuously
running in the background and the controllers need to be
connected even if one only wants to know the position of the
tracker. At the time of writing, it is also not possible to utilize
more than 2 lighthouses, thus limiting the operating space.
The final positioning system considered in this work is
based on Aruco marker detection [20] with a webcam and
OpenCV. The field of computer vision has many examples
of marker tracking, the implementation in this paper is likely
not the most accurate or user friendly system available. For
example, a calibration procedure is required to compensate
for the effects of lens distortion and to convert the measured
coordinates from pixels to meters [21]. The system is nonethe-
less included in this comparison as a representative example
of what a novice in the field could reasonably implement
themselves, and represents one of the lowest cost IPS that
can achieve sub-decimeter accuracy. The system returns z-
coordinates, but these should not be used as it is challenging
to estimate depth with a monocular camera.
TABLE II. SPECIFICATIONS OF THE CONSIDERED IPS.
Positioning system Marvelmind HTC VIVE Pozyx Aruco
Update rate (max) [Hz] 16 120 138 30
Approximate system cost
[e]
400 700 600 70
Measurement range [m] 50 5 30
Scalable ? Yes No Yes No
IV. EXP ER IM EN TAL SETUP
Experiments were conducted in a lab environment of ap-
proximately 5 meters by 5 meters. All beacons were mounted
at the edges of the test space. The Pozyx beacons are mounted
vertically at approximately the height of the receiver (see
Section V-B). The lighthouses for the VIVE are attached
to metal poles and pointed slightly downwards. Both the
Marvelmind beacons and the webcam are attached to the
ceiling at a height of approximately 2.8 meters. Figure 1
shows the experimental setup. The receivers of the various
positioning systems are mounted on top of a mobile robot
with a custom sensor platform (see Figure 2). The Pozyx tag
is mounted vertically to improve accuracy [16]. The tag is
connected to a raspberry pi 3 that also controls the robot.
Measurements for the HTC VIVE, Marvelmind and camera
system are received on a laptop. The robot moves at varying
speeds during the experiments, occasionally stopping to turn.
The maximum speed of the platform is about 0.2 m/s.
As a ground truth reference, the Krypton K600 coordi-
nate measurement machine (CMM) was used. This system is
equipped with 3 infrared cameras, which track the positions of
infrared LEDs that can be attached to objects. The accuracy of
the system is between 60 µm and 190 µm, depending on the
distance to the camera [22]. The krypton CMM is controlled
with and measurements are stored on a separate computer.
Figure 1. Experimental setup. One more Pozyx beacon is present but cannot
be seen on this perspective.
Figure 2. Robot platform used in the experiments
A. Data processing
Data from the systems under testing (SUT) are returned in
different formats (e.g., a .txt file for the HTC VIVE and as a
rosbag for the Marvelmind system). In order to compare data
from different systems, all the output is first converted to a
CSV file containing the timestamped positions. It is assumed
that timestamps recorded by different computers only have an
offset difference.
Following conversion, both the CSV data from the Krypton
CMM and the SUT are loaded into memory. The positions
of the infrared LEDs relative to the robot center are used
to determine the robot pose via Procrustes analysis [23] and
Kabsch algorithm [24], which determines the least-squares
solution for the pose matrix. Next, these pose matrices and the
position of the evaluation system relative to the robot center
are used to calculate the equivalent trajectory of the SUT (that
is, the trajectory that the SUT would report if it was placed at
the location of the krypton markers). However, this equivalent
17Copyright (c) IARIA, 2019. ISBN: 978-1-61208-691-0
ALLSENSORS 2019 : The Fourth International Conference on Advances in Sensors, Actuators, Metering and Sensing

trajectory can still be rotated or translated in space, and have
an offset time difference relative to the SUT. Additionally, the
sampling frequency of the krypton measurement system is not
necessarily the same as the SUT. To determine the points that
can be compared, virtual timestamps of the Krypton CMM
that provide the best match for the SUT are selected, based
on the assumption of a constant 50 Hz sampling rate and
the starting time of the experiment. Out of these timestamps,
those that overlap are selected for evaluation (see Figure 3).
At this stage in the post-processing, we have two datasets of
equal length (one equivalent dataset for the Krypton CMM and
one for the SUT). The position data can still be rotated and
translated relative to each other, and the time vectors can have
an offset difference. We therefore calculate the transformation
matrix that provides the best fit of the position data. The SUT
data is then transformed to the Krypton coordinate frame with
this matrix. Next, we shift the timestamps of the evaluation
samples with a period of the reference system and calculate the
transformed dataset. The time shift that provides the smallest
error is assumed to be the offset difference between the
clock. The result of this process is a reference dataset and an
evaluation dataset that is aligned in space and in time, from
which the accuracy can now be computed.
One might argue that the method described above provides
the smallest possible positioning errors as the data is trans-
formed to provide the best fit. Therefore, the entire length
of the datasets are not fitted to each other. Rather, for each
experiment approximately half of the data is used for fitting,
and the rest is used for evaluation.
Figure 3. Procedure for determining overlapping samples in the krypton and
SUT datasets. Samples in the same color are assumed to represent the same
timestep
V. EX PE RI ME NTAL RESULTS
Table III provides an overview of the measured accuracies
of the different positioning systems. The following Sections
will elaborate on these results. Accuracy is defined as follows:
ε=p(xref −xSU T )2+ (yref −ySU T )2+ (zref −zSUT )2(2)
Where x,yand zare used to denote the different coordinate
axes, and the subscripts ref and SU T indicate the ground
truth and the system under testing, respectively.
TABLE III. ACCURACY OF THE CONSIDERED POSITIONING
SYSTEMS
Marvelmind HTC VIVE Pozyx Aruco
Accuracy (mean) [mm] 19,62 8,05 150,73 99,15
Accuracy (P95) [mm] 33,28 12,62 283,21 177,09
A. Marvelmind
Figure 4 shows the Marvelmind positioning results together
with the ground truth reference. It is clear that both trajectories
are a close match. At certain sections it can be challenging to
distinguish the two from one another. The mean accuracy with
respect to the ground truth reference is 19,62 mm. The 95%
interval of the cumulative error distribution was determined to
be 33,28 mm. Therefore, it appears that the advertised accuracy
of 2 cm is consistent with the average observed accuracy.
However, it should be noted that the CMM has a relatively
limited measurement range (as can be seen in Figure 4). When
performing experiments in larger space, there inexplicably
exists a region where the Marvelmind system does not measure
at all. We were unable to determine the cause of this signal
loss, as the beacons were not obstructed in this space nor were
there any apparent sources of interference present.
Figure 4. Marvelmind positioning results alongside Krypton CMM
measurements.
B. Pozyx
The Pozyx beacons were placed at an equal height in
an approximate square. In order to improve accuracy, the
distances between the beacons were determined based on
manual measurements rather than using the self-calibration
function (this is also recommended in the documentation [16]).
The relative distances can be used to calculate the angles of the
approximate square (see Figure 5), these angles should then
sum to 360 degrees. If this is the case, the beacon angles and
distances can be used to determine the beacon locations in any
coordinate system. In this paper, one beacon is selected as the
origin, and the other beacons are defined relative to it.
Positioning results were not as expected. The large degree
of noise in positioning data means determining the best fit
is challenging. When a fit is possible, positioning errors
are on average around 150 mm, with a P95 value of 233
mm (see Figure 6). These results can likely be attributed to
the large amounts of metal present in the lab environment,
which reduces the accuracy of UWB ranging. Therefore, a
measurement was also performed with a static receiver in an
open outdoor area. The results of this experiment are shown
in Figure 7. The measured positions are spread over an area of
approximately 20 cm, thus implying that the maximum (static)
accuracy of the system is 10 cm.
Our results are slightly better than the evaluation in [10],
where a mean accuracy of 20 cm was obtained. However, a
P95 value was not specified. Our analysis reaches significantly
different results than those in [9]. However, we suspect the
authors may have used a different definition for accuracy.
Negative values for 3D accuracy are present in some of the
figures, which is impossible in the definition in (2). The
definition that the authors did use is not specified. The obtained
mean accuracy is significantly better than that of the DecaWave
18Copyright (c) IARIA, 2019. ISBN: 978-1-61208-691-0
ALLSENSORS 2019 : The Fourth International Conference on Advances in Sensors, Actuators, Metering and Sensing

kit benchmarking in [6]. Additionally, our obtained P95 value
is much better than even the 90% interval measured by the
authors. Since the Pozyx developers kit used in this paper
makes use of DecaWave transceivers, we can therefore con-
clude that their additions such as machine learning and sensor
fusion improve performance, particularly at high intervals of
the cumulative probability function.
Figure 5. Calibration of Pozyx beacon positions
Figure 6. Pozyx positioning results (red) alongside Krypton CMM
measurements (blue).
Figure 7. Reported positions by pozyx when the receiver remains static in an
outdoor environment. Beacons were placed approximately 5 meters apart.
C. Aruco marker tracking
Positioning results from the computer vision system show
significantly less noise than the Pozyx system. Accuracy is
relatively high in the center of the test space. However, this
accuracy decreases towards the edge of the image (see Figure
8), possibly due to a small degree of image distortion that
remains even after calibration. Accuracy relative to the krypton
ground truth is around 100 mm on average, with a P95 value
of 177 mm.
Figure 8. Camera positioning results (red) alongside Krypton CMM
measurements (blue). Data was fitted to the left side of the figure, which is
why deviation is smallest in this region. Evaluation is performed on the rest
of the dataset which was not fitted.
D. HTC VIVE tracker
The HTC VIVE proves to be the most accurate system out
of all the experiments. As can be seen on Figure 9, the two
datasets match very closely, in fact it is challenging to observe
any difference. Therefore, these results are also shown in a
different perspective in Figure 10. Small variations in the z-
coordinate of the reference system are present. These variations
can likely be attributed to a non-perfect smoothness of the floor
or roundness of the robot wheels. The average positioning error
relative to the ground truth is 8 mm, with a P95 value of 12
mm.
Figure 9. HTC VIVE positioning results (red) alongside Krypton CMM
measurements (blue).
Figure 10. Side view of the HTC VIVE positioning results (red) alongside
the Krypton CMM measurements (blue)
19Copyright (c) IARIA, 2019. ISBN: 978-1-61208-691-0
ALLSENSORS 2019 : The Fourth International Conference on Advances in Sensors, Actuators, Metering and Sensing

VI. CONCLUSION
In this paper, we benchmarked a number of low-cost indoor
positioning systems. We used a highly accurate reference
system rather than conventional grid measurements or mea-
surements with a ruler as a ground truth. Additionally, we
measure the ground truth position at a high rate to better
characterize dynamic performance. While there likely will
never be a single ’best’ indoor positioning system, it is clear
from our results that significant progress has been made in
recent years. For just a few hundred euros, it is now possible
to purchase a positioning system that delivers accuracy of a few
centimeters out of the box. If one has more technical expertise,
then the HTC VIVE can provide even higher accuracy. It is
even possible to use the VIVE as a ground truth reference for
positioning systems that have an accuracy that is one order of
magnitude less than 1 cm. In this case, the VIVE can be a
more interesting option than the krypton CMM due the drastic
reduction in cost and the larger measurement space. When a
very low-cost positioning solution is required, it is possible to
achieve, on average, sub-decimeter accuracy for the price of
an webcam. However, both the HTC VIVE and the camera
based solution do not scale to larger environments, unlike for
the Marvelmind or Pozyx systems.
Future work can extend this analysis to include more posi-
tioning systems. For example by including commercially avail-
able optical tracking systems. Additional performance metrics
such as precision and power use could also be evaluated, to
provide a more complete assessment of the IPS. Finally, the
experiments in this paper were performed at approximately the
same movement speed. An evaluation at a range of velocities
could be useful for highly dynamic applications (e.g., drones).
ACKNOWLEDGMENT
Robin Amsters is an SB fellow of the Research Foundation
Flanders (FWO) under grant agreement 1S57718N.
REFERENCES
[1] L. Mainetti, L. Patrono, and I. Sergi, “A survey on indoor
positioning systems,” in 2014 22nd International Conference on
Software, Telecommunications and Computer Networks, SoftCOM
2014. IEEE, sep 2014, pp. 111–120. [Online]. Available: http:
//ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=7039067
[2] N. E. Klepeis et al., “National Human Activity Pattern Survey ( NHAPS
): Use of nationwide activity data for human exposure assessment The
National Human Activity Pattern Survey ( NHAPS ): a resource for
assessing exposure to environmental pollutants,” Journal of exposure
analysis and environmental epidemiology, 2001, pp. 231–252.
[3] P. Connolly and D. Bonte, “Indoor Location in Retail:
Where is the Money?” ABI Research, Tech. Rep., 2016,
URL: https://www.abiresearch.com/market-research/product/
1025355-indoor-location-in-retail-where-is-the-mon/ [accessed:
2018-12-21].
[4] R. Mautz, “Indoor Positioning Technologies,” Habilitation thesis, ETH
Zurich, 2012, URL: http://e-collection.library.ethz.ch/eserv/eth:5659/
eth-5659- 01.pdf [accessed: 2018-12-21].
[5] T. Van Haute et al., “Performance analysis of multiple Indoor Posi-
tioning Systems in a healthcare environment,” International Journal of
Health Geographics, vol. 15, no. 1, 2016, p. 7.
[6] A. Jimenez and F. Seco, “Comparing Decawave and Bespoon UWB
location systems: Indoor/outdoor performance analysis,” in 2016 In-
ternational Conference on Indoor Positioning and Indoor Navigation
(IPIN), vol. 66, no. 8. IEEE, oct 2016, pp. 1–8, URL: http://ieeexplore.
ieee.org/document/7743686/ [accessed: 2018-12-21].
[7] A. R. J. Ruiz and F. S. Granja, “Comparing Ubisense, BeSpoon, and
DecaWave UWB Location Systems: Indoor Performance Analysis,”
IEEE Transactions on Instrumentation and Measurement, vol. 66, no. 8,
2017, pp. 2106–2117.
[8] R. Maalek and F. Sadeghpour, “Accuracy assessment of ultra-wide
band technology in locating dynamic resources in indoor scenarios,”
Automation in Construction, vol. 63, 2016, pp. 12–26. [Online].
Available: http://dx.doi.org/10.1016/j.autcon.2015.11.009
[9] P. Dabove, V. Di Pietra, M. Piras, A. A. Jabbar, and S. A. Kazim,
“Indoor positioning using Ultra-wide band (UWB) technologies: Posi-
tioning accuracies and sensors’ performances,” 2018 IEEE/ION Posi-
tion, Location and Navigation Symposium, PLANS 2018 - Proceedings,
2018, pp. 175–184.
[10] M. Ridolfi et al., “Experimental evaluation of uwb indoor positioning
for sport postures,” Sensors (Switzerland), vol. 18, no. 1, 2018, pp.
1–20.
[11] J. Cernohorsk´
y and M. Nov´
ak, “Mobile robot indoor navigation,” 2016,
pp. 151–155.
[12] C.-M. Chang, C.-H. Hsu, C.-F. Hsu, and K.-T. Chen, “Performance
Measurements of Virtual Reality Systems: Quantifying the Timing and
Positioning Accuracy,” Proceedings of the 2016 ACM on Multimedia
Conference - MM ’16, 2016, pp. 655–659. [Online]. Available:
http://dl.acm.org/citation.cfm?doid=2964284.2967303
[13] D. C. Niehorster, L. Li, and M. Lappe, “The accuracy and precision of
position and orientation tracking in the HTC vive virtual reality system
for scientific research,” i-Perception, vol. 8, no. 3, 2017, pp. 1–23.
[14] S. Adler, S. Schmitt, K. Wolter, and M. Kyas, “A survey of experi-
mental evaluation in indoor localization research,” Indoor Positioning
and Indoor Navigation (IPIN), 2015 International Conference on, no.
October, 2015, pp. 1–10.
[15] S. Gezici et al., “Localization via ultra-wideband radios a look at
positioning aspects for future sensor networks,” IEEE Signal Processing
Magazine, vol. 22, no. 4, 2005, pp. 70–84.
[16] Pozyx Labs, “Pozyx - centimeter positioning for arduino,” 2015,
URL: https://www.pozyx.io/Documentation/Tutorials/ready to localize
[accessed: 2018-09-11].
[17] Valve, “VIVE - Vive Tracker,” 2017, URL: https://www.vive.com/us/
vive-tracker/ [accessed: 2018-09-11].
[18] O. Kreylos, “Lighthouse tracking examined — Doc-Ok.org,” 2016,
URL: http://doc-ok.org/?p=1478 [accessed: 2018-09-10].
[19] J. Van den Heuvel, “Driver Documentation,” 2018, URL:
https://github.com/ValveSoftware/openvr/wiki/Driver-Documentation
[accessed: 2018-12-21].
[20] S. Garrido-Jurado, R. Mu˜
noz-Salinas, F. J. Madrid-Cuevas, and
M. J. Mar´
ın-Jim´
enez, “Automatic generation and detection of highly
reliable fiducial markers under occlusion,” Pattern Recognition,
vol. 47, no. 6, jun 2014, pp. 2280–2292. [Online]. Available:
http://linkinghub.elsevier.com/retrieve/pii/S0031320314000235
[21] OpenCV, “Camera calibration With OpenCV OpenCV 2.4.13.7
documentation,” 2014, URL: https://docs.opencv.org/2.4/doc/tutorials/
calib3d/camera{ }calibration/camera{ }calibration.html [accessed:
2018-09-11].
[22] Krypton Electronic Engineering n.v., “Krypton Help Pages on
K400/K600 Hardware & Software Guide,” Leuven, Tech. Rep.,
2003. [Online]. Available: http://nees.buffalo.edu/training/krypton/
kx00 manual EN.PDF
[23] J. C. Gower, “Generalized procrustes analysis,” Psychometrika,
vol. 40, no. 1, 1975, pp. 33–51. [Online]. Available: https:
//doi.org/10.1007/BF02291478
[24] W. Kabsch, “A solution for the best rotation to relate two sets
of vectors,” Acta Crystallographica Section A, vol. 32, no. 5,
sep 1976, pp. 922–923. [Online]. Available: https://doi.org/10.1107/
S0567739476001873
20Copyright (c) IARIA, 2019. ISBN: 978-1-61208-691-0
ALLSENSORS 2019 : The Fourth International Conference on Advances in Sensors, Actuators, Metering and Sensing