Passive Localization with Visible Light

Passive Localization with Visible Light
Delft University of Technology
Master’s Thesis in Embedded Systems
Passive Localization with Visible Light
Junwei Zhang
Passive Localization with Visible Light
Master’s Thesis in Embedded Systems
Embedded Software Section
Faculty of Electrical Engineering, Mathematics and Computer Science
Delft University of Technology
Mekelweg 4, 2628 CD Delft, The Netherlands
Junwei Zhang
J.Zhang-7@student.tudelft.nl
30th August 2016
Author
Junwei Zhang (J.Zhang-7@student.tudelft.nl)
Title
Passive Localization with Visible Light
MSc presentation
8th September 2016
Graduation Committee
Prof. dr. K.G. Langendoen (Chair)
dr. Marco Zuniga (Daily Supervisor)
dr. Christoph Lofi
Delft University of Technology
Delft University of Technology
Delft University of Technology
Abstract
Indoor localization is a topic of significant interest for academic and industrial. For outdoors, we have Global Positioning System (GPS) working
very well, but GPS does not work indoors and we spend more than 90% of
the time indoors. Thus, there is a great need to find solution for this problem. Many technologies have been proposed for indoor localization based on
radio frequency, infrared and ultrasonic sensors. Recently, researchers have
been using light as a means to achieve indoor localization. LED lights inside
our buildings can transmit information and they can be seen as “internal
GPS satellites” playing the role of localization anchors. But most of these
works require the user to have an electronic device.
In this thesis, we would like to achieve indoor localization with visible light
but in a passive manner. By passive, we mean that the object of interest
would have no electronic device or tag. This thesis takes the first step in
developing such a system and our contributions are three-fold: we develop
a system and a mathematical model to analyze it, we show a scaled-down
application in the order of decimeters and we also show a more realistic
application in the order of meters. Our work shows that this is a promising
way to achieve accurate indoor localization in a passive manner.
iv
Preface
Visible Light Communication (VLC) is a recent and hot research topic in
wireless communication. Prior to this thesis, I had no knowledge on it.
But when my advisor Prof. Marco Zuniga introduced VLC to me at the
beginning of this thesis project, I was immediately fascinated by it and its
potential applications. Using visible light for indoor localization in a passive
manner has not been explored yet. In this thesis, I take the first step towards
this direction, and have achieved some theoretical and experimental results.
Some open problems are also left for future work. I believe the achievements
in this thesis could provide inspirations to other researchers who want to
work on localization with visible light.
I would like to thank my advisor Prof. Marco Zuniga. It has been a great
honor to work with him during this nine-month thesis project. I appreciate
his great help on guiding me patiently to the correct research directions
and advising me on solving research problems, writing the thesis and giving
presentations. I also want to thank Prof. Koen Langendeon for admitting
me to this wonderful group of Embedded Software, and thanks to Prof.
Christoph Lofi for being a member of my thesis’s defense committee. I would
also like to thank Dr. Qing Wang for his valuable discussions and help on
this thesis. Furthermore, I thank Lennart Klaver, Aniruddha Deshpande
and Michail Vasilakis. They gave me many suggestions and feedback. Last,
and most important, I will be forever thankful to my beloved wife and
daughter for their strongest support.
Now that I am reaching the end of my thesis work as well as my Master’s
study, I realize that, leaving my established life in China to study aboard
for two years was not an easy decision after working for almost ten years in
industry. But I will never regret it. Although my work might need continual
improvements, I worked hard and learned a lot. This two-year studying is a
valuable experience and I am sure I will benefit greatly from it for the rest
of my life.
Junwei Zhang
v
Delft, The Netherlands
30th August 2016
vi
Contents
Preface
v
1 Introduction
1.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Organization . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
4
4
5
2 Background and Related Work
2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
7
10
3 System Model
15
3.1 System Composition . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Channel Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4 System Analysis
4.1 The Simplest System . . . . . . .
4.2 Coverage Analysis . . . . . . . .
4.3 Fingerprint Analysis . . . . . . .
4.4 Localization Algorithm . . . . . .
4.5 Impact of Object’s Surface . . . .
4.6 Further Exploitation of the RSS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5 Model Validation
5.1 Scaled-Down System Implementation . . . .
5.1.1 Starting point – the Shine platform .
5.1.2 Redesign of the Shine platform . . .
5.1.3 Object design . . . . . . . . . . . . .
5.2 Validation . . . . . . . . . . . . . . . . . . .
5.2.1 Experiment setup . . . . . . . . . . .
5.2.2 Results . . . . . . . . . . . . . . . .
vii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
21
21
21
23
25
26
29
.
.
.
.
.
.
.
33
33
33
34
36
37
37
37
6 Applicaton Test
6.1 Large-scale System Implementation
6.1.1 Platform Transformation .
6.1.2 Object Design . . . . . . .
6.2 Application Test . . . . . . . . . .
6.2.1 Experiment Setup . . . . .
6.2.2 Results . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
41
41
41
43
45
45
45
7 Conclusions and Future Work
49
7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
viii
Chapter 1
Introduction
Indoor localization is increasingly attracting people’s interest from both academic and industrial . The motivation behind this trend is the huge demand
of accurate indoor localization services. These services range from people’s
indoor navigation in buildings such as shoprrping malls, to location-based
advertising and promotions.
It is well-known that the Global Positioning System (GPS) performs well
outdoors and plays a very important role in our daily life. However, its
indoor performance is far from satisfactory due to the fact that construction
materials block the signals coming from satellites, attenuate them quickly
and make them highly stochastic. To tackle this problem, many other techniques are proposed for indoor scenarios, such as localization based on ultrasound [15, 30, 31], infrared [45, 14], and most recently, localization based on
Radio Frequency (RF) signals such as Wi-Fi [22, 6, 4], Bluetooth [34, 1, 5]
and RFID [47, 23].
However, current RF-based techniques cannot fully solve the indoor localization problem. Most of them can only achieve meter-level localization
accuracy [4, 6, 22]. Some of them might be able to achieve decimeter-level accuracy [46, 35], but with the extra cost of complicated and densely-deployed
infrastructures or complex algorithms, and can only perform well under
certain circumstances. RF-based localization also has some natural disadvantages which obstruct their deployment. For instance, RF signals have
irregular attenuation, which can affect dramatically localization accuracy,
(by up to several tens of meters). Some environments are also sensitive to
electromagnetic interference such as hospitals or airplanes, where RF-based
indoor localization systems are not allowed to be deployed.
On the contrary, Visible Light Communication (VLC) and sensing are
providing a new promising alternative for this problem and gathering an increasing attention from researchers. In VLC, Light Emitting Diodes (LEDs)
are turned on and off so rapidly that they can transmit information but this
will have no impact on the illumination perceived by the human eye like
1
shown in Fig. 1.1. Given that we have LEDs deployed everywhere, they
could be used as “indoor GPS satellites” to provide localization. In Fig. 1.2,
we show how “Indoor VLC” looks like, where at least three LEDs are transmitting the location information like satellites and a mobile phone is acting
as the receiver.
Figure 1.1: The principle behind VLC.
Figure 1.2: The illumination of indoor VLC idea.
Some ongoing active localization solutions have already been proposed to
provide precise localization results based on visible light, such as the Epsilon [21], Luxapose [17], etc., and can beat mainstream RF-based solutions.
Some visible light positioning systems are already deployed in reality as well,
for example, Philips deployed an LED indoor positioning system in a Carrefour hypermarket in France [36], ByteLight deployed their systems in several
bookstores [33], and the Lumicast indoor localization services were employed
in more than 100 stores in North America [19]. Within these active localization systems with visible light, smart-phones are normally used as the
receivers with their cameras or other additional customized boards acting
as photo-sensors. Therefore, people are normally involved in the localization
process by holding or even rotating their smart-phones.
2
In some scenarios, however, we might want to monitor or localize objects
passively. For example, cars in an indoor parking lot may need to be tracked
for management purposes. A control system in a warehouse may also need
to monitor moving objects within its space, or miners in a tunnel may need
to be tracked for safety reasons. In all these scenarios, we could be able
to exploit the visible light reflected by the exterior surfaces of the objects to
localize them. These objects can be cars or miners wearing safety helmets, as
illustrated in Fig. 1.3. In Fig. 1.4, every lamp has an LED and a photo-diode
so then lamp A sends information that is received by the photo-diode in lamp
B and can detect the presence of mobile objects. There are no electronic
devices needed on the object of interest. In this thesis, the objective is to
localize these types of objects passively based on the visible light they reflect.
(a) Car with shiny surface
(b) Miner wearing a safety helmet
Figure 1.3: Illustrated objects to be localized passively in this thesis
Figure 1.4: Examples of passive localization.
3
1.1
Problem Statement
Achieving the above objective has several challenges. First, since there is no
existing work on passive localization with reflective light, this thesis needs to
define the structure of such a system and its functional principles. Second,
each element in the system (transmitter and receiver) must be designed and
modeled properly. To achieve this, a theoretical analysis of the proposed
system must be carried out. Third, to understand the system deeply, theoretical analysis alone is far from enough. We need to develop a real testbed
to evaluate the system’s performance and compare it with analytical results.
We will tackle these challenges in this thesis.
1.2
Contributions
The contributions of this thesis are summarized as follows:
• We propose a novel passive localization system that is based on reflected visible light. The system only consists of simple transmitters (off-the-shelf LEDs), receivers (off-the-shelf photo-diodes), and
objects. The objects do not need to be equipped with any electronic
devices. The localization is mainly based on the object’s exterior properties, i.e., their surface material and shapes that are able to reflect
surrounding visible light.
• We model the proposed system, analyze it and describe how different
settings, e.g., the placement of the transmitters and receivers, affect
the performance of localization. We provide numerical evaluation results of the proposed system under various settings.
• We evaluate the performance of the proposed system preliminarily
using the Shine platform [16] in different settings. Evaluation results
validate our proposed model and analysis.
• Going further, we design and implement a system that includes offthe-shelf high-power LEDs (commonly used for illumination in daily
life). We adopt scaled-down cars and real construction hats as objects,
and locate them in a more realistic setup. Our results show that our
system can provide the accuracy with a maximal localization error of
5.3 cm at various points along the trajectory.
To the best of our knowledge, this thesis is the first to locate objects based
on reflected light emitted from off-the-shelf illumination LEDs. This work
can provide researchers with an initial framework to further exploit passive
localization with visible light .
4
1.3
Organization
The rest of this thesis is organized as follows: the necessary background
information and related work are provided in Chapter 2, followed by the
system overview and model presented in Chapter 3. The proposed system is analyzed theoretically in Chapter 4 and we present some results. In
Chapter 5, a scaled-down system is developed and evaluated. More realistic
application tests are designed, implemented, and demonstrated in Chapter 6.
Finally, the conclusions and future work are presented in Chapter 7.
5
6
Chapter 2
Background and Related
Work
This chapter presents some background information and research work related to the topic of this thesis. In Section 2.1, the difference of active and
passive localizations is given first. Then different types of light reflections
and their roles in passive localization are described. Furthermore, a brief
introduction on visible light communication is given, which is a key technology in our light-based localization system. Related works and the novelty
of tracking objects passively with visible light are described in Section 2.2.
2.1
Background
Localization systems can be categorized into two different types, active localization and passive localization [7]. Active localization systems demand
an active collaboration of the moving objects in the localization process,
which means that an object should have an electronic device or tag on it
and can receive/transmit data. When the object has no device or tag, the
localization is passive. A typical example of passive localization is the Passive Infrared Sensor (PIR) [45, 14], which can detect the presence of a person
in its Field of View (FOV) without any devices or tags carried by the person. Mainstream research in indoor localization using visible light normally
involves an electronic device, which is usually a smartphone. Our focus is
passive localization, where the tracked object is device-free.
As shown in Fig. 1.4, the proposed passive localization system in this
thesis relies on the reflected light by object’s surfaces. To understand the
proposed system, first it is very important to have a good view of different types of reflections. Light reflection can be classified into three types:
specular reflection, spread reflection, and diffuse reflection, as illustrated in
Fig. 2.1. Different reflection types heavily depend on the roughness of an
object’s surface. Reflection from smooth surfaces such as mirrors is known
7
as specular reflection, where the law of reflection1 is applicable. Rough surfaces like asphalt in roadways lead to diffuse reflection. Spread reflection
usually occurs on surfaces such as corrugated or etched metal where the
light is reflected into a cone of light rays.
Figure 2.1: Three types of light reflection [2]
In the real world, the majority of reflections are not specular but more
likely to be a mixture of these three types. Fig. 2.2 shows that different
reflection types can affect the detection results on the receiver. The artificial
light coming from light bulbs on the ceiling can hardly be detected by a
photo-detector mounted together with the bulbs, after it hits the rough
ground and then bounces back with diffuse reflection. Specular reflection
can make light reach longer ranges and keep the light direction controllable
and predictable. Spread reflection has the middle range between specular
reflection and diffuse reflection. Whether the light after spread reflection is
detected or not depends on the traversed distance.
In this thesis, we exploit the presence of smooth planes which can bring
strong reflection and enable visible light communication between the transmitters and receivers. In other words, we mainly investigate the effects of
specular and spread reflections in our research of passive light-based localization.
Visible Light Communication
One of the key technologies in the proposed passive localization is Visible
Light Communication (VLC). VLC relies on the very simple fact that it can
modulate signals to transmit information by turning the light on and off. A
symbol ONE can be represented by turning on the light, and a symbol ZERO
can be represented by turning the light off. Lights from LEDs nowadays can
be turned on/off at a very high speed such that the information can be
transmitted at speeds of Megabit or Gigabit per second. Based on this and
the fact that LED lights are pervasive in our environment e.g. car, indoor
lights, screens, VLC is growing rapidly.
There are many different techniques to modulate visible light, from OnOff Keying (OOK), Pulse-Position Modulation (PPM), Pulse-Amplitude
1
The incident light, the reflected light and the normal line are on the same plane and
the angle of incidence equals the angle of reflection.
8
Figure 2.2: The comparison of different type of reflections
Modulation (PAM), to Binary Frequency Shift Keying (BFSK) [29], ColorShift Keying (CSK) [26], and Orthogonal Frequency Division Multiplexing
(OFDM) [11]. In this thesis, we use Differential Pulse Position Modulation
(DPPM), a variation of PPM. An illustration of DPPM is shown in Fig. 2.3.
Data are transmitted with short pulses and all pulses have both the same
width and amplitude. The duration between the falling edge of the previous pulse and the next rising edge will represent a digital 0 or 1. A small
duration represents digital 1, and a large duration represents digital 0.
Figure 2.3: Differential Pulse Position Modulation.
With this prior knowledge, we now continue to present some research work
related to our topic.
9
2.2
Related Work
We classify the related work into four categories: active RF-based localization, active visible light-based localization, passive RF-based localization,
and passive visible light-based localization.
Active RF-based localization
The majority of indoor localization systems are active RF-based, where
the target objects are heavily involved in the localization process. To perform active localization, objects are normally equipped with electronic devices,
e.g., widely used smartphones.
There are many sensors available in smartphones, e.g., accelerometer,
gyroscope, etc. Based on the information collected by these sensors, various
types of active indoor localization techniques are proposed, such as exploiting Time of Arrival (TOA) [9], Time Difference of Arrival (TDOA) [10,
3], Angle of Arrival (AOA) [27], and Received Signal Strength Indicator
(RSSI) [25]. With TOA, the distance between the transmitter and the receiver is computed based on the speed of the signal and the time spent by
the signal traveling from the transmitter to the receiver. However, TOA
has a strict requirement for synchronization. To deal with this restriction,
TDOA is proposed [3], which uses two different kinds of transmitting signals. The time difference between these two types of signals at the receiver
is used to calculate the distance between the transmitter and the receiver.
AOA technique requires the receiver to have the capability of measuring the
angle of arrival of the signal. RSSI usually links the received signal strength
(RSS) to the distance from the transmitter to the receiver based on signal
propagation models.
Based on these aforementioned techniques, positioning algorithms could
take the next step to calculate the physical position of the target, for instance
trilateration/triangulation [24, 43], fingerprinting [12], and so on. Fig. 2.4
illustrates how trilateration based positioning works in two dimensions. The
distances from the target node T to three fixed non-collinear reference nodes
A, B, and C are R1, R2, and R3, and the intersection point of three circles
is declared as the physical position of the target. Triangulation in 2D can
use just two reference nodes A and B to determine the position of the target
node T . As shown in Fig. 2.5, the intersection of two angle direction lines is
the position of the target node. As for the fingerprinting positioning, RSS
is compared against the pre-stored records of geographic-signal map. As
long as a match to any records in the database is found, the corresponding
geographic position is where the target is.
In the literature we can find some research works about active wireless
localization with advanced features that can help improve the system performance. Ning Chang [4] proposes and designs an indoor positioning system
with a median positioning error of around 3 m. The wireless area local net10
Figure 2.4: Trilateration positioning.
Figure 2.5: Triangulation positioning.
work (WLAN) infrastructure is used for reference points and the distance is
calculated using RSS. A method based on the differential operation access
points is proposed to deal with the adverse effects of environmental factors.
Yih-Shyh Chiou [6] uses the Kalman filtering (KF), and radio-frequency
identification (RFID) to assist building an adaptive location estimation system achieving the accuracy of 2.1 m, which also takes advantage of WLAN
infrastructure and the radio propagation model. Hyuk Lim [22] establishes
the theoretical base and then develops a zero-configuration and robust indoor localization and tracking system taking the on-line measurements of
RSSs between a client and its neighboring 802.11 APs as input to estimate
the location of the client. A large improvement is the realization of zeroconfiguration and no need of offline training or calibration and its median
localization error does not exceed 3 m.
In order to achieve the localization accuracy of centimeter level, we can
use Ultra Wide Band (UWB) signal because UWB signal has high-temporal
resolution, multipath immunity, and simultaneous ranging and communication capability. Yuan Zhou [46] presents an asynchronous position meas11
urement system based on UWB signal, which requires the followed target
to carry a UWB transmitter to actively transmit UWB pulses to several
energy-detection receivers whose positions are known. TDOA of UWB signals captured by the receivers can be used to compute the target location.
The achieved location accuracy is less than 10 cm.
Active visible light-based localization
Considering that smart-phones have built-in light sensors (camera and
light sensors), VLC is also exploiting the user’s smart-phone for indoor localization.
A positioning system using visible LED lights and image sensor is proposed and analyzed by Masaki Yoshino [41] and he uses simulation to prove
its availability with the position of accuracy of within 1.5 m. With Epsilon [21],
a Visible Light Based Positioning System is proposed based on trilateration,
requiring at least three LED lamps as anchors. Smart-phones are enhanced
with an additional photo-sensor plugged to the audio-jack port to detect the
incident light from each transmitter and then uses a trilateration algorithm
to calculate its position. The localization accuracy by Epsilon is 0.4 m,
0.7 m, 0.8 m in different office environments.
After that, a large improvement is made in Luxapose [17]. A smart-phone
and LED luminaries are still used in this paper. The LED luminaries transmit its identifiers and/or location through OOK modulation. The built-in
camera on the smart-phone uses the rolling shutter effect to demodulate the
light and estimate the location of LED lights in the captured image. Instead
of multilateration, an AOA localization algorithm is applied in Luxapose
and a better performance of decimeter-level accuracy is acquired.
Zhice Yang [40] proposes a Light-weight Indoor Positioning with Visible
Light which makes use of polarization-based modulation instead of intensitybased modulation so that resource-constrained wearable devices, e.g. smart
glasses, can be capable of processing images. This polarization-based VLC
can be applied to all kinds of artificial luminaries by adding the liquid crystal
display (LCD) to the lighting sources. With the resource-constrained devices
as receivers, the localization error is still lower than 3 m in 90% test cases.
As a good supplement to current research, LiP ro [37] provides a solution
for insufficient reference points in some scenarios. When less than three
reference light sources are detected at the receiver, using multilateration is
impossible. So Lipro is a solution when only one transmitter is available and
the smart-phone is the receiver. The smart-phone will be rotated around
three orthogonal axes while RSS and magnetic field strengths are recorded
continuously at the same time, which are used to solve the equations for the
receiver’s position. LiP ro achieves a median error of 0.59 m in a corridor
with linearly deployed LEDs, and 0.45 m in an office.
SpinLight [38] is an indoor positioning system which realizes the localiza12
tion from a different angle. Not exactly visible light but infrared Lamps are
used as signal transmitters even though the applied communication technology is the same. For 2D localization, one LED lamp is the transmitter and
a customized hemispherical shade is created to cover and rotate around the
LED. The rings on the shade and hollow or closed cells in the ring manage
to divide the light of the transmitter into spatial beams. When a receiver,
equipped with a light sensor, detects the light signals to determine its spatial
beam, its location inside that beam can be computed. A high accuracy is
achieved by SpinLight, a median location error of 3.8 cm.
Passive RF-based localization
RF-based localization systems can also be used to track objects in a passive manner.
Moustafa Youssef [42] introduces the concept of Device-free Passive (DfP)
localization. His work describes the DfP system’s architecture and the challenges that need to be addressed to materialize a DfP system with normal
WIFI equipment.
Joey Wilson [35] proposes a localization system which can track a moving
object, a person in most cases, without the need for any electronic device or
tag on the object. In a dense peer-to-peer wireless network, RSS will change
when a person moves. All the variances of RSS can make up a motion image
with a pre-known deployment model. From the motion image, a Kalman
filter can be applied to recursively track the coordinates of a moving target.
In a 34-node experiment, the moving object in place can be located with
approximately 45 cm average accuracy.
Apart from the above RF-based passive localization, the most common
technology used in practice is passive infrared sensor (PIR). Zhiqiang Zhang [45]
and Jurgen Kemper [14] separately work out their passive localization systems based on PIR networks. The thermal radiation of humans has the
special wavelength and hence can be precisely detected by PIR sensors.
Based on this fact, the work by Zhiqiang Zhang [45] can localize a target
into a right cell of 30 cm wide with 90% accuracy and a simple triangulation
based localization algorithm is applied by Jurgen Kemper [14] to locate the
person with the maximal measured error of 80 cm.
Passive visible light-based localization
We cannot find any publications on passive localization with visible light.
Okuli [44] is however a quite-relevant system that uses visible light passively for near field sensing. The paper proposes a virtual track-pad design
using one LED and two light sensors. The goal of Okuli is to track a finger
in a small pad. Fig. 2.6 shows the structure of Okuli. One LED is in the
middle of two photo-diodes (P D1 and P D2) with a distance ds . The finger
is the target with the coordinates (x, y) in the work space and its distances
13
Figure 2.6: The Structure of Okuli.
to LED, P D1, P D2 are d1, d2, d3 respectively.
OOK is used to identify the light transmitted by the LED and the user’s
finger is the localized object on the virtual work space. Its design takes
advantage of the deterministic property of light propagation/reflection to
develop a model-driven solution. The finger is assumed to be relatively
round and be a very good diffuser of light, thus the light can be estimated
quite precisely after the emission, propagation, reflection, reception in the
model. Every pair of coordinates of the finger has the unique pair of RSSs
on P D1 and P D2. The prototype of Okuli can detect and localize random
finger positions with a median error of 0.7 cm, and 90-percentile error of
1.43 cm.
As far as we know, Okuli is the most relevant work to ours. There are no
other studies on passive localization/tracking with visible light.
14
Chapter 3
System Model
This chapter presents the theoretical foundation of a passive localization
system with visible light. We first describe the composition of such a system
by defining every necessary element and their functions. Considering that
light propagation is a very important part in the localization process and will
affect RSS on the receiver’s side, we then introduce the Lambertian law for
the channel gain in the Line-Of-Sight (LOS) scenario. Besides, we add two
extra components to make the equation suitable for the Non-Line-Of-Sight
(NLOS) case.
3.1
System Composition
The system consists of three elements: transmitters, receivers, and the object,
as shown in Fig. 3.1. Depending on the object’s position, the light emitted
by the transmitters can be reflected by the object towards the receivers.
Based on the detected light at the receivers, the system can infer the current
position of the object, localizing the object passively.
Figure 3.1: System architecture: transmitters, receivers, and the object.
15
Figure 3.2: A transmitter with m beams.
Transmitters: The transmitters are normal LED bulbs. Let N denote
the number of transmitters. They are placed at a height of h from the ground
where the object moves on. The transmitters are evenly distributed along a
line and the distance between two neighboring transmitters is denoted as d.
Each transmitter has m beams that emit lights towards different directions. Fig. 3.2 shows how m beams make up the total FOV of a transmitter.
Assume each beam has an identical field-of-view φ. All the beams of the
transmitters are assigned unique IDs containing information about which
beam and which transmitter the light is coming from. We use tij to denote
the jth beam of the ith transmitter.
Receivers: The receivers are photo-diodes. Assume there are N receivers, and each of them is placed together with a transmitter, vertically facing
down towards the ground. The receiver and the transmitter together will
be denoted as a single unit in our geometric model. All the photo-diodes
have the same FOV, that is denoted by ϕ. Let us assume the photo-diodes
can not receive line-of-sight lights from neighboring transmitters (which can
be achieved by appropriately deploying the transmitters and receivers), but
can only detect the light reflected by the object.
Object: The object moves under the transmitters and receivers, and is
assumed to have one or more flat surfaces with inclined angle α.
In our system, we put a transmitter (LED light) and a receiver (photodiode) together to create a node that has the capability of transmitting and
receiving. From this sense, it is better to call the node Transceiver and we
will use this term in the remaining part of the paper.
Fig. 3.3 shows a simple example of passive localization with visible light.
When a light beam tAj from transceiver A reaches the object’s surface, it
will be reflected by the object and then be detected by the receiving part of
16
Figure 3.3: The illustration of localizing the object passively.
transceiver B. The system can judge that the object is currently within the
region covered by the beam tAj .
Our target in this paper is to investigate how different parameters, such
as the transceiver’s height, distance between transceivers, object’s inclined
angle, objects’ surface material, etc., affect the localization accuracy.
3.2
Channel Gain
After presenting the system composition, we now introduce the light propagation model. The major concern in our proposed localization system is that
whether the receiver can detect the light emitted by the transmitter after
the light travels a given distance, and how different factors impact RSS at
the receiver.
For scenarios under LOS, an LED is usually considered as a lambertian
source and the received light intensity can be modeled by the lambertian
cosine law [13].
As shown in Fig. 3.4, the light emitted by the LED is directly received by
the photo-sensor. The received energy Pr can be expressed as the product
of the transmission power of that LED, denoted as Pt , and the channel gain
H(0)LOS,Gen.Lamb. :
Pr = Pt · H(0)LOS,Gen.Lamb.
(3.1)
where
(
H(0)LOS,Gen.Lamb. =
(m+1)A
2πd2
cosm φTs (ψ)g(ψ) cos(ψ),
0,
0 ≤ ψ ≤ Ψc
(3.2)
ψ > Ψc
where m is the lambertian order that is mainly determined by the LED’s
optical enclosure, A represents the detecting area of the photo-sensor, and d
17
Figure 3.4: LOS light propagation
is the distance between the LED and the photo-sensor. Besides, in Eq. 3.2,
ψ is the incidence angle, φ is the irradiation angle, and Ψc is the semi-angle
of the receiver. Ts (ψ) and g(ψ) are the concentrator gain and filter gain,
respectively.
However, since we insert reflections into the light propagation path in our
proposed system, there is no longer a LOS link. A light path with NLOS
is shown in Fig. 3.5. Two extra parameters are now affecting the RSS in
NLOS scenarios, namely, the active area f (Ao ) and the reflection coefficient
ρ of the object’s surface.
Figure 3.5: Light propagation in our proposed system (with light reflection)
When other parameters except for the object’s detecting area are fixed,
the RSS grows linearly with the active area of the object’s surface, until
the area reaches a certain level, after which it delivers a constant RSS. RSS
and the reflection coefficient have a monotonic increasing relationship. For
instance, a material with higher reflection coefficient like a mirror will lead to
higher RSS at the receiver. After adding these two parameters, the modified
18
version of Eq. 3.3 can be written as follows:
(
H(0)LOS,Gen.Lamb. =
(m+1)A
2πd2
0,
cosm φTs (ψ)g(ψ) cos(ψ)f (Ao )ρ, 0 ≤ ψ ≤ Ψc
θ > Ψc
(3.3)
where d = d1 + d2 .
In this chapter, we have described the equation for the lambertian model.
In Chapter 4, we will see that this equation is important to help us determine
the performance of our system in terms of RSS and potential accuracy.
19
20
Chapter 4
System Analysis
We can analyze the system from a pure geometric perspective where only the
arrangement of transceivers and objects affect the localization. In a geometric model, we assume that the light reflected by the object is always strong
enough to be detected at the receivers and focus on the influence of other
parameters. After that, we discuss the impact of RSS in the localization
process. From the analytical results, we can have a better understanding of
how all the parameters affect the performance of the localization system.
4.1
The Simplest System
For a system of N transceivers where each transceiver only has one beam
and the inclined angle of the object is α = 0, it is rather obvious that the
object can be detected when it stays at the following 2N − 1 positions: the
N positions directly under the transceivers, and the N − 1 positions in the
middle of any two neighboring transceivers. The simplest system leaves a
large blind area apart from the 2N − 1 positions, even though these 2N − 1
positions can be localized quite precisely.
4.2
Coverage Analysis
To make every point detectable between the transceivers, a possible method
is to adjust the inclined angle α. There are two types of α: as shown in
Fig. 4.1, αij enables the communication between two transceivers i and j as
shown in Fig. 4.1 , and αii or αjj reflects the light to where it comes from.
We have the following theorem:
Theorem 1. Given the distance d between the transceivers and the height
h of transceivers from the ground, for any given position between the transceivers, there always exists an inclined angle of the object that makes the
21
object reflect the light towards other transceiver(s):
x
arctan( d−x
h ) − arctan( h )
2
where x is the object’s position between the two neighboring transceivers and
we define a clockwise turns as the positive direction of angles.
In addition, for a given position x, there always exists an inclined angle
that can make the object reflect the light towards the same transceiver where
the light is coming from:
x
αii = − arctan( )
h
or
d−x
αjj = arctan(
)
h
αij =
Proof. For αij , define the incidence angle of the light on the object as θ, the
inclined angle of the object
αij = arctan(
d−x
)−θ
h
and we can see that
θ=
x
arctan( d−x
h ) − arctan( h )
2
Hence,
x
arctan( d−x
h ) − arctan( h )
2
For αii or αjj , the incidence angle of the light on the object is 0. αii or αjj
equals the irradiation angle of the light, which is arctan( hx ) or arctan( d−x
h ).
Considering the positive direction of angles, we get that
x
αii = − arctan( )
h
or
d−x
αjj = arctan(
)
h
αij =
If we fix the ratio of the height and the distance of the transceivers for
instance hd = 1, we can calculate αij , αii or αjj for every position x under
two transceivers using Theorem 1. We present the results in Fig. 4.2. From
Fig. 4.2, there are three angles for a given position x between the transceivers
i and j which can make that position detected.
Insight 1: Assuming a high luminance power and values of α between
arctan
d
arctan
d
h
] or [0, arctan( hd )], [− arctan( hd ), 0], we can have full cov[− 2 h ,
2
erage of the moving target.
22
Figure 4.1: The illustration of the existence of α .
4.3
Fingerprint Analysis
Even if we assume we can detect the moving target at every point, we will
now see that we cannot provide an unique fingerprint at each point. Here
let us assume we can have all these angles, that means we have full coverage.
We want to investigate αij mainly. We have a curved surface with the angle
arctan
d
arctan
d
h
from − 2 h to
as shown in Fig. 4.3.
2
Let us assume that each transceiver has five beams. For example, as
shown in Fig. 4.4, the curved surface is in the region covered by beam tA4 .
The black curve represents the current position and the grey ones are some
of its historical positions. As soon as the curved surface is in the region tA4 ,
we cannot exactly tell which position it is.
Let us look in detail at Fig. 4.5, when the curved surface is in the region
tA4 , the viewing angle of each beam is ω and the direction angle of beam
tA4 is µ. Then the location set of the region tA4 is [x1 , x2 ] where
ω
x1 = h · tan(µ − )
2
and
ω
x2 = h · tan(µ + )
2
1 , α2 ]:
Furthermore, we can get the range of αij for beam tA4 , which is [αij
ij
1
αij
=
x1
arctan( d−x1
arctan( hd − tan(µ − ω2 )) − (µ − ω2 )
h ) − arctan( h )
=
2
2
2
αij
=
x2
arctan( hd − tan(µ + ω2 )) − (µ + ω2 )
arctan( d−x2
h ) − arctan( h )
=
2
2
23
50
40
30
20
10
0
-10
-20
-30
-40
-50
0
0.2
0.4
0.6
0.8
1
Relative postion of object x/d
Figure 4.2: Object’s position vs. α (Theorem 1).
Figure 4.3: A curved surface with continuous angles of α.
If we use the five-beam setup and make the regions tA5 and tB1 overlap
each other as shown in Fig. 4.4. This configuration divides this whole angle
arctan
d
arctan
d
h
] into five regions, so all angles in between are going
range [− 2 h ,
2
to be detectable in the same region. In this example, we can only detect five
regions and in principle, 3 angles are sufficient to give unique information
to identify all possible regions.
Insight 2: Based on the information from tij , d and h, we can only give
coarse-grained location, which are regions.
24
Figure 4.4: The curved surface in the region tA4 .
Figure 4.5: The details about the region tA4 .
4.4
Localization Algorithm
We make the following assumption: an object has k surfaces where the
inclined angles of the surfaces are represented by a vector {α1 , α2 , ..., αk }.
The height of transceivers is h and the distance between two neighboring
transceivers is d. The object is in the region covered by beam tA4 as shown
in Fig. 4.5.
Let us assume that transceiver B receives information from beam tA4 . We
have calculated that the location range of beam tA4 is [x1 , x2 ] = [h · tan(µ −
ω
ω
2 ), h · tan(µ + 2 )].
Using Theorem 1, we compute the corresponding positions x̂1 , x̂2 , ..., x̂k
for each α by solving the following equations:
25
α1 =
α2 =
x̂1
1
arctan( d−x̂
h ) − arctan( h )
2
x̂2
2
arctan( d−x̂
h ) − arctan( h )
2
...
x̂k
k
arctan( d−x̂
h ) − arctan( h )
2
If x̂2 for instance is in the set [h · tan(µ − ω2 ), h · tan(µ + ω2 )] and is the
only one among all x̂s, x̂2 is the correct location of the object.
αk =
Another possibility is that transceiver A receives information from beam
tA4 . We compute positions for each α by solving the equations:
α1 = − arctan(
x̂1
)
h
α2 = − arctan(
x̂2
)
h
...
x̂k
)
h
This time, if x̂1 , for example, is in the location range of beam tA4 , then x̂1
is a possible location of the object.
αk = − arctan(
The possible positions can be one or more. If we have only one position,
that is the correct location of the object. If we have more than one in the
same region, we could take the average of them as the position of the target
or choose one with higher probability when we know the moving direction
and velocity of the target. This is the localization algorithm we apply when
we know the values of αs of the object.
4.5
Impact of Object’s Surface
Apart from the inclined angle α, the object surface has two other factors
which affect the localization: the active area f (Ao ) and the reflection coefficient ρ of the object’s surface. Even though we assume a sufficient RSS in
the geometric model, it might not be the case in practice.
Before we use these two factors in the localization analysis, we do simple
experiments to verify the roles of them in Eq. 3.3.
First, we put a mirror as the reflectivity surface directly under a LED
(Avago HLMP-CM1A-450DD, 5 mm, green color) and a photo-diode (Osram
SFH203P) as shown in Fig. 4.6. Their distances to the table surface are
26
Figure 4.6: The test to verify the impact of f (Ao ).
360
Output of PD / mV
340
320
300
280
260
240
0
20
40
60
80
100
Surface Area / mm 2
Figure 4.7: Output of PD vs Area of the Surface.
20 cm and 23 cm respectively. We change the area of the mirror on the
table, f (Ao ). The results are shown in Fig. 4.7. We can see the linear
relation between the output of the photo-diode and the surface area of the
mirror until a point (25 mm2 ) where the area is large enough to keep the
output of PD constant.
Second, we use one LED lamp (YPHIX LED spot Avior Plus 5W 370
Lumen) as the transmitter and a light meter (MASTECH MS8229 digital
multimeter) as the receiver. We put them in positions like in Fig. 4.8 and the
distances from the LED lamp to the reflective surface and from the reflective
surface to the light meter are both 1 m. We make the surface area of the
reflective materials large enough (10 cm × 10 cm) to focus on the impact
of ρ. A mirror, an aluminum plate and a white cardboard surface are three
kinds of reflective materials we used in the test. From the results shown
27
Figure 4.8: The test to verify the impact of ρ.
Figure 4.9: Output of the light meter vs different ρ.
in Fig. 4.9, we can see that RSS (represented by measurements of the light
meter) changes accordingly with different ρ.
The ρ includes all factors that can affect the reflectivity coefficient for us:
type of reflection (specular, spread, diffuse) and reflection coefficient itself.
Therefore the localization performance can be affected by the reflection
coefficient ρ. It is possible that RSS is too weak to detect when ρ is low. In
such a case, we will lose some detectable points in comparison to the ideal
situation in our geometric analysis, where we assume that there is always a
high enough RSS. For example, if we consider these three materials: mirror,
smooth aluminum plate, white cardboard which have ρ from high to low as
the moving object, they have different communication ranges. In a geometric
model with five detectable points, it is possible that the mirror can guarantee
detecting all five points and the aluminum plate misses the middle point. As
for the white cardboard, only two positions directly under the transceivers
are still detected. Within such a case, the localization result is shown in
Fig. 4.10.
28
Figure 4.10: Impact of ρ
4.6
Further Exploitation of the RSS
So far we are assuming that the localization is only based on detecting the
beams, but if we also consider RSS, we may be able to further distinguish
the positions in the region. Considering the reflecting point in the region,
the traveling distance d of light will be different as well as the irradiation
angle φ and the incidence angle ψ. Thus RSS might be different for different
points within the same region.
In Eq. 3.3, the parameters m, A, Ts (ψ), g(ψ), f (Ao ), ρ are constant in a
determined system. Let us assume that
T =
(m + 1)A
Ts (ψ)g(ψ)f (Ao )ρ ,
2π
where T is a constant. Then Eq. 3.3 can be rewritten as:
H(0)LOS,Gen.Lamb. = T
cosm φ cos(ψ)
d2
The lambertian order m is related to the semi-angle Φ1/2 of the LED [13].
m=−
ln 2
ln(cos Φ1/2 )
For example, m is 1 when the viewing angle of the LED is 120◦ and m is
about 45.2 when the viewing angle is 20◦ . Assume that we use five LEDs
with viewing angles of 20◦ to make up a transceiver. The height of transceivers is 1 and the distance between two neighbouring transceivers is 1.77.
Fig. 4.11 shows this configuration of two transceivers A and B. The regions
defined by beams tA5 and tB1 are the same.
29
Figure 4.11: The setup of 5 beams with FOV of 20◦
0.11
Light From A to B
Light From B to A
0.1
RSS without PtT
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0
0.5
1
1.5
2
Position of Object in [0, 1.77]
Figure 4.12: RSS vs Position of Object
For a position xi in the region tA5 , we can calculate its d, φ and ψ.
q
p
d = d1 + d2 = 1 + x2i + 1 + (1.77 − xi )2
φ = |40◦ − arctan(xi )|
1.77 − xi
ψ = arctan(
) = arctan(1.77 − xi )
1
Thus RSS (represented by Pr ) at transceiver B can be calculated by:
Pr = Pt · H(0)LOS,Gen.Lamb. = Pt T
= Pt T
cosm φ cos(ψ)
d2
cos45.2 (|40◦ − arctan(xi )|)cos(arctan(1.77 − xi ))
q
p
( 1 + x2i + 1 + (1.77 − xi )2 )2
30
where Pt T is a constant so Pr is decided by xi .
Using the same method, we can get a Pr for every position between x = 0
and x = 1.77. Fig. 4.12 shows the relationship between RSS and each
position. The y axis in Fig. 4.12 needs to multiply the constant coefficient
Pt T to get the complete value. We can observe that the fluctuations of RSS
can be mapped to the regions. In each region, no more than two positions
have the same RSS and when we have two positions with the same RSS, they
are symmetrical. If we know which region the object is in, it is technically
possible to do further localization using RSS.
31
32
Chapter 5
Model Validation
5.1
Scaled-Down System Implementation
To validate the model experimentally, we design and implement a scaleddown system based on the Shine platform [16].
5.1.1
Starting point – the Shine platform
The Shine platform was originally designed and developed for omni-directional
multi-hop visible light communication. The photo of a shine node is given in
Fig. 5.1. Each Shine node is equipped with 20 LEDs (Avago HLMP-CM1A450DD, 5mm, green color), where each LED has a Field-of-View (FoV) of
18◦ . Four PDs (Osram SFH203P) are placed in the Shine platform, and
each of them is expected to cover 90◦ . The micro-controller used in the
Shine is Atmega328P, which is low-cost and powerful. It also provides serial
communication capability that can be used to interface with a PC/laptop.
Figure 5.1: The Shine platform
33
GUI
Central Server
Localization Algorithm
UART
UART
Coding / Decoding
(Micro-­processor)
Coding / Decoding
(Micro-­processor)
…
Transmitter (LEDs)
Receiver (PD)
Transmitter (LEDs)
Shine Board #1
Receiver (PD)
Shine Board #x
Figure 5.2: Function blocks of the system.
5.1.2
Redesign of the Shine platform
The Shine platform is a good stepping start but it does not fully satisfy our
requirements. We redesign it in both the hardware and software parts, in
order to meet our system’s requirements for passive localization. A diagram
block of the redesigned Shine platform in our system is given in Fig. 5.2.
Next, we present our designs in details.
Hardware
The main change in the hardware part of the Shine platform is the PDs.
In our system, we need a very sensitive PD in each Shine node to sense
the reflected visible light. The original Osram SFH203P does not meet this
requirement. Therefore, after careful consideration, we decide to use the
Osram SFH206K to replace the original PD. This new PD is much more
sensitive with the active area increasing largely from 1 mm2 to 7.02 mm2 .
According to Eq. 3.3, RSS is seven times bigger than the original because
the photo-detector area A is seven times bigger. If we assume that the
minimum RSS for reliable communication is the same, we can increase the
range (represented by d) by about 165% after this improvement.
Another change is that the amplifying circuitry of the photo-diode signal
is improved to cope with interference. In the practical test, we found that
the original Shine board was extremely vulnerable to electromagnetic interference (EMI) and a quite obvious noise was observed on the output of the
receiving circuitry. The reason, as shown in Fig. 5.3, is that the pin of the
photo-diode is behaving like an antenna and there is no filtering in the amplifying circuit as well. Therefore we change the way how the photo-diode
34
Figure 5.3: The reason why the original Shine is vulnerable to EMI.
Figure 5.4: Comparison of the signal before and after the improvement.
is mounted and add a low-pass filter to the amplifier. Fig. 5.4 shows the
signal before and after the change. We can see that the noise is compressed
greatly now.
Software
We largely redesign the software part of the Shine platform to satisfy our
purpose.
(1) The transmitter. In our system, we hang the Shine platforms in the
air and the object to be tracked moves under them. Therefore, not all 20
lights in the Shine platform need to be used. We only need to use the
LEDs ‘facing down’. To save resources, we select seven neighboring LEDs in
each Shine as the transmitting LEDs, and assign them unique IDs. In the
software stack, these seven LEDs are controlled by the micro-controller to
send frames (modulated lights) including their own IDs in a TDMA manner.
(2) The receiver. As shown in Fig. 5.1, there are four PDs in each Shine
platform. In our system, we only adopt seven LEDs as the transmitters.
Thus not all four PDs could sense the reflected light from the object. In
the software, the micro-controller will only read light signals from the PD
‘facing down’, saving time and resources.
35
Figure 5.5: A snapshot of the GUI
(3) Medium access. During the transmissions, a Time Division Multiple
Access (TDMA) Medium Access Control (MAC) scheme is applied to all
LEDs to access the shared medium. This can avoid interference from other
LEDs when an LED is transmitting data. The Shine platform is connected
to a central server (PC/laptop), where the latter runs the MAC scheme and
schedules the LEDs ‘remotely’ through the micro-controller. Similarly, the
data received from the PDs will be decoded at the micro-controller and then
sent to the central server for follow-up process.
To increase the reliability of the communication channel, every time a
LED has a time slot in the TDMA schedule, it transmits the information
three times.
(4) Localization algorithm. We implement our localization algorithm 4.4
in the central server. The inputs of this algorithm are the decoded frames
at the Shine platforms. These frames are sent to the central server through
the UART, as shown in Fig. 5.2. Based on the information contained in
these received frames, the localization algorithm as described in Section 4.4
calculates the position of the object on-the-fly.
(5) GUI. The GUI is quite simple. As shown in Fig. 5.5, the GUI just
shows the locations of transceivers with letters and in this case we have two
transceivers A and B. Since the object is moving along the line under the
transceivers, the red knob is used to show the calculated position of the
object.
5.1.3
Object design
Regarding the moving object, we use flat mirrors as the material. The
reason behind this is that the mirror has very high reflection coefficient,
which makes sure that the reflected light can easily be detected by the
receiver. Besides this point, a flat mirror also makes the direction of the
reflected light predictable and controllable.
As analyzed in Chapter 4, to make the object detectable in more positions,
we construct it with multiple mirrors, each placed at a different angle α. We
use seven LEDs on each Shine board so each transceiver has seven beams.
The height of transceivers and the distance between transceiver can be used
36
Figure 5.6: The front view of an object made of five mirrors
to compute the regions covered by beams. We choose positions from each
region and calculate the corresponding α, as decribed in Section 4.2.
For example, Fig. 5.6 gives a front view of the object which consists of
five mirrors with different inclined angles -17.74◦ , -8.83◦ , 0◦ , 8.83◦ , 17.74◦ .
Each mirror has the same width of 5 mm. Given that we have these angles,
we only have eleven number of points that would be observable, because
for all the other ones, just having more angles will always pinpoint to the
same region. Because we do not have continual of αs, we will not have full
coverage. With these number of angles, we should expect eleven calculated
positions as shown in Table 5.1. In this example, all the inclined angles are
calculated for the purpose of detecting all possible regions, as described in
Section 4.3.
5.2
5.2.1
Validation
Experiment setup
The Shine platforms can achieve reliable visible light communications at a
maximum distance of 50 cm. In our experiments, we place two Shine nodes
at a distance of 20 cm, and at a height of 15 cm from the desktop. By
deploying the nodes like this, we can make sure that, even with the longest
path, the system can have reliable communication based on lights reflected
from the object moving on the desktop. The overall experiment setup of our
system is presented in Fig. 5.7.
5.2.2
Results
Table 5.1 illustrates the calculated positions based on the model presented
in Chapter 4 and the measured positions from the experiments (ground
truth). We repeated the testing twenty times to check the repeatability of
the results. The results are just the same when the geometric parameters
such as the distance between transceivers, the height of transceivers, and
the object do not change.
It can be observed that our analytical model is accurate, which is validated
by the results from experiments. The errors mainly come from the coarse
hand-work which prevented us to obtain accurate inclined angle, and the
37
Figure 5.7: Deployment of the Shine platforms in the validation experiment
Table 5.1: Comparison of calculated positions and measured positions
Calculated
position
Measured
position
1st
2nd
3rd
4th
5th
6th
7th
8th
9th
10th
11th
-4.7
0
4.7
3.3
6.7
10
13.3
16.7
15.3
20
24.7
-3.4
0
3.4
4.5
7.9
10
12.1
15.9
16.1
20
23.4
errors can be up to 3◦ . These comparisons are also presented in Fig. 5.8,
where we have a more pictorial representation of the results.
We can observe that there are only several points on that line detected
and accurately localized, which is the limitation of using flat mirrors as the
reflector. Adding more light beams to one transceiver and more mirrors can
make more points available in our localization algorithm.
Although the flat mirror has this disadvantage of leaving blind areas, it
can make those points localized quite accurately, which is useful in certain
applications. Back to our flat-mirror setup, the historical information about
the moving object can be used to provide localization at all non-covered
points, such as, the latest position the system knows, the moving direction
and velocity of the object. Many technologies focus on providing solutions
for such cases, for example, particle filters [8, 20, 28] or Bayesian filters [18,
32].
38
Figure 5.8: Results from scaled-down demo built of the Shine.
39
40
Chapter 6
Applicaton Test
6.1
Large-scale System Implementation
After testing our passive localization idea in the scaled-down scenario, it is
better to check if it works well in a more realistic testbed with standard
LED lights.
6.1.1
Platform Transformation
Regarding the LED lamps, we decided to find some off-the-shell LED bulbs
to use in the application test instead of creating a totally-new LED lamp
which might be good for just lab test but unrealistic in real life.
Some commercial LED bulbs were tried and tested to find appropriate
ones to guarantee detectable light intensity at long distance. In the end,
the LEDARE GU5.3 MR16 200 lumen Led-lamp (like in Fig. 6.1) is chosen
as the transmitter to build the experiment hardware. This LED lamp has
200 lumen light power and a viewing angle of 36◦ .
We mount three LED lamps together to provide three different beams
Figure 6.1: The selected LED lamp
41
Figure 6.2: The transmitter consists of three LED lamps.
Figure 6.3: The driver circuit for the high-power LED lamp
in one transceiver node. As shown in Fig. 6.2, the lamp in the middle is
horizontal and two others have adjustable inclination angles so that the three
lamps together create an appropriate wider FOV.
Naturally, the constant-current driver circuit for the high power LED lamp
needs to be redesigned and built because the original driving circuit on the
Shine cannot support such high-current output and the detailed circuitry is
depicted in Fig. 6.3. The pulse signals on the Shine used to control twenty
on-board LEDs (Avago HLMP-CM1A-450DD, 5 mm, green color) and we
now output three of those pulses to the constant-current drivers of LED
lamps. In this way, Shine is still used to control the transmission of packets.
The PD (Osram SFH206K) on the Shine is sufficient as the receiver for
our application test, which can save us a lot of trouble. So we keep the PD
and use Shine to process the received information.
Eventually, the whole transceiver which contains the Shine platform, the
new LED driver and the three LED lamps are displayed in Fig. 6.4. After
the transceiver is hung in the air with the height and distance changeable,
the Shine boards will take care of the transmission control, demodulate the
42
reflected light and send the received message to the laptop through the serial
communication.
Figure 6.4: The modified transceiver for the application test
When it comes to the software part, there is no difference from the scaleddown validation. The exact same software can work well since the control
scheme and functional principle have been verified in the previous Chapter 5.
6.1.2
Object Design
A very important aspect of the application test is that we put a real object
in our experiment or build the object to imitate the practical target quite
well. For example, since we cannot put a car in our lab room, we study the
reflection properties and angles of a car in order to recreate them.
Usually a car has several flat reflective surfaces (like in Fig. 6.5). Mainly
three planes are considered in this research, which are front wind shield, car
roof, back windshield. After going through a great number of car graphs,
we decide that the angles of three planes are -30◦ , 0◦ , 25◦ and the final ‘car’
shape is shown in Fig. 6.6. In Fig. 6.6, the surfaces are aluminum plates
and we also use mirrors as the surfaces to check the difference. The angles
in every model of car are different but they are roughly around the values
we choose. The reflection coefficient of these surfaces might be weaker than
in mirrors but still shiny enough to reflect the light to reach a long distance.
We can find more inclined planes in a car which can bring more localizable
43
Figure 6.5: Side view of a shiny car
Figure 6.6: The final shape of the ‘car’
positions, but our simple choice is already enough to prove the idea and give
some useful insights about the application.
Another example is a construction safety helmet which miners wear in
tunnels. From Fig 6.7, we can observe that the helmet is shiny enough
to reflect light and has the curved shape which can cause some diffused
reflection. Different colors might have different reflection coefficients, which
will affect the light range. We use some helmets (as in Fig. 6.7) and test
them to check the results.
Figure 6.7: The construction safety helmets
44
6.2
Application Test
6.2.1
Experiment Setup
Considering the fact that the communication range of the chosen LED lamps
is 5 m, we manage to change the height of the luminaries to various levels:
2 m, 1.5 m and 1 m to verify the influence of height in localization accuracy.
At the same time, the inter-node distance is adjusted from 2.5 m to 2 m.
We use the mirrors and then the smooth aluminum plates to construct the
moving object as discussed in Section 6.1.2 and then, test them in different
setups of height and distance. In the end, we move the safety helmets under
the platform to see if the localization works or not.
Figure 6.8: The settings of the application test.
6.2.2
Results
Five points between two transceiver nodes are detected and localized directly, which is not sufficient even though they have high accuracy. We put
the results from the setup of height 1.5 m and distance 2.5 m in Fig. 6.9 in
order to give a pictorial view. Mirrors and aluminum plates as the object
gives no difference in the result shown in Fig. 6.9. In Fig. 6.9, red bars (instead of dots as in Fig. 5.8) are used to represent the ground-truth positions
because the plates are wide enough to give a short location range, while in
the experiments of Chapter 5 the mirror is so narrow that the position is
a single value. The same testing under each configuration is repeated ten
times and the results are quite stable. Tables 6.1, 6.2, 6.3 illustrate the results we get with three configurations of distance and height, using mirrors
45
Figure 6.9: Results when Height is 1.5 m, Distance is 2.5 m.
Table 6.1: Height is 2 m and Distance is 2.5 m
Position from Model
Mirrors
Aluminum Plates
1st
0m
-0.02∼0.02
-0.02∼0.02
2nd
1.155 m
1.15∼1.17
1.13∼1.15
3rd
1.25 m
1.25∼1.27
······
4th
1.567 m
1.5∼1.52
1.51∼1.52
5th
2.5 m
2.48∼2.52
2.48∼2.52
or aluminum plates. The maximal error of these points is 5.3 cm and the
average error is 0.97 cm.
The simplest way to analyze the remaining undetected positions is that,
before the object reaches the next detectable position, we mark the object
staying at its latest position. This gives us the worst performance evaluation
of the system. Then we calculate ECDF (empirical cumulative density function) plot in Fig. 6.10. The difference between mirrors and aluminum plates
is circled in Fig. 6.10 and is caused by the fact that the third position in the
setup of height 2 m and distance 2.5 m can be detected with mirrors and
not by aluminum plates.
When it comes to the test using helmets (Fig. 6.11), the results are worse
than expected. The available range is under 0.5 m, which means that the
helmets can only be detected when they are quite near to the transceivers,
closer than 0.5 m. It is impossible for the light to travel from one transceiver
to another after reflecting on the helmet.
The similar semi-sphere shape of the helmets makes light reflect diffusely
when a beam of light hits its surface. Eq. 3.3 is not applicable any more.
Considering that the helmet is a lambertian source to the receiver, we have
to divide the light propagation into two phases. First, the received energy by
the helmet Ph is inversely proportional to the square of distance d1 between
the transmitter and the helmet. Second, the helmet is the lambertian source
46
Table 6.2: Height is 1.5 m and Distance is 2.5 m
1st
0m
-0.02∼0.02
-0.02∼0.02
Position from Model
Mirrors
Aluminum Plates
2nd
0.866 m
0.85∼0.88
0.85∼0.88
3rd
1.25 m
1.25∼1.26
1.25∼1.26
4th
1.801 m
1.76∼1.78
1.77∼1.79
5th
2.5 m
2.48∼2.52
2.48∼2.52
Table 6.3: Height is 1 m and Distance is 2 m
1st
0m
-0.02∼0.02
-0.02∼0.02
Position from Model
Mirrors
Aluminum Plates
2nd
0.577 m
0.63∼0.66
0.62∼0.65
3rd
1m
0.99∼1.01
0.99∼1.01
4th
1.534 m
1.48∼1.51
1.47∼1.5
5th
2m
1.98∼2.02
1.98∼2.02
and Ph is the transmitting energy. Then the receiver energy by the receiver
Pr can be calculated by Eq. 3.2 and is inversely proportional to the square
of distance d2 from the helmet to the receiver. From this sense, RSS at the
receiver is no longer inversely proportional to the square of the total distance
as in Eq. 3.3 but more likely to the fourth magnitude of it [39].
Meanwhile, we notice that the detected positions are not fixed like the
flat plates any more. We tested this setup ten different times, and all led
to different (more random) locations. Sometimes the helmet is even not
detected when passing through. The positions are not continuous and can
change greatly with small differences of holding the helmet. There are certainly some small variations of the height and posture of the helmet when
a miner wears a helmet and walks. The continual αs prevent us apply the
localization algorithm discussed in Section 4.4. RSS should be different at
each position theoretically but we failed to extract useful information from
RSS in our test.
mirrors
aluminum plates
1
0.8
0.8
0.6
0.6
ECDF
ECDF
1
0.4
0.4
0.2
0.2
height=1m
height=1.5m
height=2m
0
height=1m
height=1.5m
height=2m
0
0
0.2
0.4
0.6
0.8
1
1.2
0
localization error (m)
0.2
0.4
0.6
0.8
1
1.2
localization error (m)
Figure 6.10: ECDF of the worst performance from the results.
47
Figure 6.11: Results when using helmets as the object.
Therefore when we get information on the receiver, we can just tell that
the object is in a small region determined by the beam identifier.
If we would like to localize a miner who wears a helmet in practice, some
modifications to the helmet will be needed. One possible way is to enhance
its reflection property by changing the surface material to something more
shiny like metal. Another thought is to micro-adjust the shape of helmets,
by building it with small strips of metal, instead of a ‘continuous’ curve.
Besides these, we can also increase the output power of LEDs as long as
it does not jeopardize its main illumination purpose. All these actions can
work alone or add up to get a better localization performance.
48
Chapter 7
Conclusions and Future
Work
7.1
Conclusion
In this thesis, we take the first step to achieve passive localization with
visible light, where the tracked object has no electronic device or tag and
can only be involved in the localization process by the reflections on its
surface. Since no prior work on this topic could be found, we propose a
passive localization system model and define every element in the system.
In the theoretical analysis of the system, we mainly discuss the role of every
part and how they would influence the localization accuracy. Then this
passive localization idea and the developed theory are validated by a scaleddown demo which uses the Shine board. More importantly, a more realistic
experiment is further carried out to dig more insights about this idea and
verify the findings in the theoretical analysis. During the test, we exploit the
effects of the shape and surface material of the object on the localization
performance as well as the influence of the setup parameters such as the
height, the inter-transceiver distance and the number of light beams.
From the experimental results, we can find that our proposed passive
localization system can work to a certain degree in suitable setups. Several
detectable and distinguishable positions can be identified with quite high
accuracy, within the range of several centimeters.
7.2
Future Work
Although we have done some proof-of-concept work about our passive localization idea, some open questions are still waiting the further exploration.
At present, all theoretical and experimental research is still limited to one
dimension, and we can expand the system to localize an object/person in
two dimensions. In addition, we notice that our design has high accuracy,
49
but leaves a lot of space undetectable. We need to think how we could get
more locations detectable, or more effectively use the direct information we
can have, for example, by applying particle filters.
50
Bibliography
[1] M. Altini, D. Brunelli, E. Farella, and L. Benini. Bluetooth indoor localization
with multiple neural networks. In Wireless Pervasive Computing (ISWPC),
2010 5th IEEE International Symposium on, pages 295–300, May 2010.
[2] Marilyne Andersen. MIT OpenCourseWare: Types of Reflection. https://
www.flickr.com/photos/mitopencourseware/4815499473/, 2010. [Online;
accessed 21-July-2010].
[3] Y. T. Chan and K. C. Ho. A simple and efficient estimator for hyperbolic
location. IEEE Transactions on Signal Processing, 42(8):1905–1915, Aug 1994.
[4] N. Chang, R. Rashidzadeh, and M. Ahmadi. Robust indoor positioning using
differential wi-fi access points. IEEE Transactions on Consumer Electronics,
56(3):1860–1867, Aug 2010.
[5] S. S. Chawathe. Low-latency indoor localization using bluetooth beacons.
In 2009 12th International IEEE Conference on Intelligent Transportation
Systems, pages 1–7, Oct 2009.
[6] Yih-Shyh Chiou, Chin-Liang Wang, and Sheng-Cheng Yeh. An adaptive location estimator using tracking algorithms for indoor wlans. Wireless Networks,
16(7):1987–2012, 2010.
[7] Gabriel Deak, Kevin Curran, and Joan Condell. A survey of active and passive
indoor localisation systems. Computer Communications, 35(16):1939 – 1954,
2012.
[8] Frédéric Evennou and François Marx. Advanced integration of wifi and inertial
navigation systems for indoor mobile positioning. EURASIP J. Appl. Signal
Process., 2006:164–164, Jan. 2006.
[9] Stuart A. Golden and Steve S. Bateman. Sensor measurements for wi-fi location with emphasis on time-of-arrival ranging. IEEE Transactions on Mobile
Computing, 6(10):1185–1198, Oct. 2007.
[10] F. Gustafsson and F. Gunnarsson. Positioning using time-difference of arrival
measurements. In Acoustics, Speech, and Signal Processing, 2003. Proceedings.
(ICASSP ’03). 2003 IEEE International Conference on, volume 6, pages VI–
553–6 vol.6, April 2003.
[11] S. K. Hashemi, Z. Ghassemlooy, L. Chao, and D. Benhaddou. Orthogonal
frequency division multiplexing for indoor optical wireless communications
using visible light leds. In Communication Systems, Networks and Digital Signal Processing, 2008. CNSDSP 2008. 6th International Symposium on, pages
174–178, July 2008.
[12] K. Kaemarungsi and P. Krishnamurthy. Modeling of indoor positioning systems based on location fingerprinting. In INFOCOM 2004. Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies,
volume 2, pages 1012–1022 vol.2, March 2004.
51
[13] J. M. Kahn and J. R. Barry. Wireless infrared communications. Proceedings
of the IEEE, 85(2):265–298, Feb 1997.
[14] J. Kemper and H. Linde. Challenges of passive infrared indoor localization. In
Positioning, Navigation and Communication, 2008. WPNC 2008. 5th Workshop on, pages 63–70, March 2008.
[15] Hong-Shik Kim and Jong-Suk Choi. Advanced indoor localization using ultrasonic sensor and digital compass. In Control, Automation and Systems, 2008.
ICCAS 2008. International Conference on, pages 223–226, Oct 2008.
[16] L. Klaver and M. Zuniga. Shine: A step towards distributed multi-hop visible
light communication. In Mobile Ad Hoc and Sensor Systems (MASS), 2015
IEEE 12th International Conference on, pages 235–243, Oct 2015.
[17] Ye-Sheng Kuo, Pat Pannuto, Ko-Jen Hsiao, and Prabal Dutta. Luxapose:
Indoor positioning with mobile phones and visible light. In Proceedings of the
20th Annual International Conference on Mobile Computing and Networking,
MobiCom ’14, pages 447–458, New York, NY, USA, 2014. ACM.
[18] A. M. Ladd, K. E. Bekris, A. P. Rudys, D. S. Wallach, and L. E. Kavraki.
On the feasibility of using wireless ethernet for indoor localization. IEEE
Transactions on Robotics and Automation, 20(3):555–559, June 2004.
[19] Pete Lancia. Qualcomm and Acuity Brands Collaborate to Commercially
Deploy Qualcomm Lumicast Technology in More Than 100 Retail Locations. https://www.qualcomm.com/news/releases/2016/03/14/qualcommand-acuity-brands-collaborate-commercially-deploy-qualcomm, 2016.
[Online; accessed 14-March-2016].
[20] Fan Li, Chunshui Zhao, Guanzhong Ding, Jian Gong, Chenxing Liu, and
Feng Zhao. A reliable and accurate indoor localization method using phone
inertial sensors. In Proceedings of the 2012 ACM Conference on Ubiquitous
Computing, UbiComp ’12, pages 421–430, New York, NY, USA, 2012. ACM.
[21] Liqun Li, Pan Hu, Chunyi Peng, Guobin Shen, and Feng Zhao. Epsilon:
A visible light based positioning system. In 11th USENIX Symposium on
Networked Systems Design and Implementation (NSDI 14), pages 331–343,
Seattle, WA, Apr. 2014. USENIX Association.
[22] Hyuk Lim, Lu-Chuan Kung, Jennifer C. Hou, and Haiyun Luo. Zeroconfiguration indoor localization over ieee 802.11 wireless infrastructure. Wireless Networks, 16(2):405–420, 2010.
[23] Y. Ma, L. Zhou, K. Liu, and J. Wang. Iterative phase reconstruction and
weighted localization algorithm for indoor rfid-based localization in nlos environment. IEEE Sensors Journal, 14(2):597–611, Feb 2014.
[24] D. E. Manolakis. Efficient solution and performance analysis of 3-d position
estimation by trilateration. IEEE Transactions on Aerospace and Electronic
Systems, 32(4):1239–1248, Oct 1996.
[25] S. Mazuelas, A. Bahillo, R. M. Lorenzo, P. Fernandez, F. A. Lago, E. Garcia,
J. Blas, and E. J. Abril. Robust indoor positioning provided by real-time
rssi values in unmodified wlan networks. IEEE Journal of Selected Topics in
Signal Processing, 3(5):821–831, Oct 2009.
[26] E. Monteiro and S. Hranilovic. Design and implementation of color-shift
keying for visible light communications. Journal of Lightwave Technology,
32(10):2053–2060, May 2014.
[27] D. Niculescu and Badri Nath. Ad hoc positioning system (aps) using aoa.
In INFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE
52
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
Computer and Communications. IEEE Societies, volume 3, pages 1734–1743
vol.3, March 2003.
Anshul Rai, Krishna Kant Chintalapudi, Venkata N. Padmanabhan, and
Rijurekha Sen. Zee: Zero-effort crowdsourcing for indoor localization. In
Proceedings of the 18th Annual International Conference on Mobile Computing and Networking, Mobicom ’12, pages 293–304, New York, NY, USA, 2012.
ACM.
Niranjini Rajagopal, Patrick Lazik, and Anthony Rowe. Visual light landmarks for mobile devices. In Proceedings of the 13th international symposium
on Information processing in sensor networks, pages 249–260. IEEE Press,
2014.
M. Scherhufl, R. Pfeil, M. Pichler, and A. Berger. A novel ultrasonic indoor
localization system with simultaneous estimation of position and velocity. In
Wireless Sensors and Sensor Networks (WiSNet), 2012 IEEE Topical Conference on, pages 21–24, Jan 2012.
Herbert Schweinzer and Georg Kaniak. Ultrasonic device localization and its
potential for wireless sensor network security. Control Engineering Practice,
18(8):852 – 862, 2010.
J. Wang, Q. Gao, Y. Yu, H. Wang, and M. Jin. Toward robust indoor localization based on bayesian filter using chirp-spread-spectrum ranging. IEEE
Transactions on Industrial Electronics, 59(3):1622–1629, March 2012.
Ucilia Wang. The lighting of the future is here, and it’s designed to make
you spend. https://www.theguardian.com/sustainable-business/2015/
feb/12/lighting-future-sale-customers-shop-energy, 2015. [Online;
accessed 21-Feb-2015].
Y. Wang, Q. Ye, J. Cheng, and L. Wang. Rssi-based bluetooth indoor localization. In 2015 11th International Conference on Mobile Ad-hoc and Sensor
Networks (MSN), pages 165–171, Dec 2015.
Joey Wilson and Neal Patwari. Through-wall tracking using variance-based
radio tomography networks. CoRR, abs/0909.5417, 2009.
Maury Wright. Philips Lighting deploys LED-based indoor positioning in
Carrefour hypermarket. http://www.ledsmagazine.com/articles/2015/
05/philips-lighting-deploys-led-based-indoor-positioning-incarrefour-hypermarket.html, 2015. [Online; accessed 21-May-2015].
Bo Xie, Shimin Gong, and Guang Tan. Lipro: light-based indoor positioning
with rotating handheld devices. Wireless Networks, pages 1–11, 2016.
Bo Xie, Guang Tan, and Tian He. Spinlight: A high accuracy and robust
light positioning system for indoor applications. In Proceedings of the 13th
ACM Conference on Embedded Networked Sensor Systems, SenSys ’15, pages
211–223, New York, NY, USA, 2015. ACM.
H. Yang and A. Pandharipande. Full-duplex relay vlc in led lighting linear
system topology. In Industrial Electronics Society, IECON 2013 - 39th Annual
Conference of the IEEE, pages 6075–6080, Nov 2013.
Zhice Yang, Zeyu Wang, Jiansong Zhang, Chenyu Huang, and Qian Zhang.
Wearables can afford: Light-weight indoor positioning with visible light. In
Proceedings of the 13th Annual International Conference on Mobile Systems,
Applications, and Services, MobiSys ’15, pages 317–330, New York, NY, USA,
2015. ACM.
Masaki Yoshino, Shinichiro Haruyama, and Masao Nakagawa. High-accuracy
positioning system using visible led lights and image sensor. In 2008 IEEE
Radio and Wireless Symposium, pages 439–442, Jan 2008.
53
[42] Moustafa Youssef, Matthew Mah, and Ashok Agrawala. Challenges: Devicefree passive localization for wireless environments. In Proceedings of the 13th
Annual ACM International Conference on Mobile Computing and Networking,
MobiCom ’07, pages 222–229, New York, NY, USA, 2007. ACM.
[43] D. C. K. Yuen and B. A. MacDonald. Vision-based localization algorithm
based on landmark matching, triangulation, reconstruction, and comparison.
IEEE Transactions on Robotics, 21(2):217–226, April 2005.
[44] Chi Zhang, Josh Tabor, Jialiang Zhang, and Xinyu Zhang. Extending mobile interaction through near-field visible light sensing. In Proceedings of the
21st Annual International Conference on Mobile Computing and Networking,
MobiCom ’15, pages 345–357, New York, NY, USA, 2015. ACM.
[45] Zhiqiang Zhang, Xuebin Gao, Jit Biswas, and Jian Kang Wu. Moving targets
detection and localization in passive infrared sensor networks. In Information
Fusion, 2007 10th International Conference on, pages 1–6, July 2007.
[46] Y. Zhou, C. L. Law, Y. L. Guan, and F. Chin. Indoor elliptical localization based on asynchronous uwb range measurement. IEEE Transactions on
Instrumentation and Measurement, 60(1):248–257, Jan 2011.
[47] Y. Zhou, W. Liu, and P. Huang. Laser-activated rfid-based indoor localization
system for mobile robots. In Proceedings 2007 IEEE International Conference
on Robotics and Automation, pages 4600–4605, April 2007.
54
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising