INOM
EXAMENSARBETE TEKNIK, GRUNDNIVÅ, 15 HP
STOCKHOLM SVERIGE 2018,
Indoor positioning aided survey and map generation
FREDRIK BJÖRKMAN GUSTAF ERIKSSON
KTH
SKOLAN FÖR ELEKTROTEKNIK OCH DATAVETENSKAP
i
Abstract
Indoor positioning systems can be of a great help when trying to local- ize in an indoor environment. To be able to navigate, there is a need for an indoor map that to some degree represent the reality. This thesis aims to come up with a proof of concept of a system that maps rectan- gular objects onto a two dimensional map in an indoor environment.
The maps will be drawn with the help of the position of a hand-held device that marks the corners of the objects. A problem that arises is that the error in the positioning can contribute to objects that have the same size and are aligned in reality can vary in size and be misaligned on the map. Another problem is that with the error in the position- ing, the position might end up within an existing object, which leads to a bad user experience. The problems were solved by developing a system, that given a map with objects mapped from the help of posi- tioning, can adjust the shape and positions of the objects based on the mean-values of multiple identical objects. To avoid positions within objects, a function was written that always checks if the positions is within an existing object and if that is the case, then move the correct the position to the closest position outside of the object. The final re- sult was a proof of concept of a system that can generate maps with mapped objects and users that can traverse the map with a good user experience.
Keywords
Real-time locating systems; Indoor positioning; Embedded systems;
User experience
iii
Sammanfattning
Inomhuspositioneringssystem kan vara till stor hjälp när man vill lo- kalisera sig i en inomhusmiljö. För att kunna navigera sig så krävs det en inomhuskarta som i viss mån representerar verkligheten. Den- na avhandling ämnar att komma fram till en konceptvalidering för ett system som ska mappa upp rektangulära objekt på tvådimensio- nella kartor i en inomhusmiljö. Kartorna ritas upp med hjälp av po- sitionen från en handhållen enhet som markerar hörnen på objekten.
Problem som uppstår då är att felet från positionen kan få objekt som är av samma storlek och uppställda på rad i verkligheten, att se oli- ka stora ut och vara placerade lite huller om buller på kartan. Ett an- nat problem är att positionen kan med det här felet i positioneringen, komma att befinna sig i ett existerande objekt, vilket leder till en dålig användarupplevelse. Problemen löstes genom att utveckla ett system, som utifrån en karta med objekt mappade med hjälp av positionering, kan justera objektens form och position baserat på medelvärdena av flertalet likadana objekt. För att undvika positioner innanför objekt så skrevs en funktion som hela tiden kollar om positionen befinner sig i något befintligt objekt och om så skulle vara fallet så korrigeras po- sitionen till närmsta position utanför objektet. Slutresultatet blev en konceptvalidering av ett system som kan generera kartor med mappa- de objekt och användare som kan traversera dessa kartor med en god användarupplevelse.
Keywords
Realtidslokaliseringssystem; Inomhus positionering; Inbyggda system;
Användarupplevelse
Contents
1 Introduction 1
1.1 Background . . . 1
1.2 Problem . . . 2
1.3 Purpose . . . 2
1.4 Goal . . . 3
1.4.1 Goals . . . 3
1.4.2 Deliverables . . . 3
1.4.3 Result . . . 3
1.4.4 Benefits, Ethics and Sustainability . . . 3
1.5 Methodology . . . 4
1.6 Delimitations . . . 5
1.7 Outline (Disposition) . . . 5
2 Theory 7 2.1 Positioning . . . 7
2.1.1 Filter . . . 7
2.1.2 Inertial Measurement Unit . . . 8
2.1.3 Unreachable areas . . . 9
2.2 Map generation . . . 9
2.2.1 Floor plans or Blueprints . . . 9
2.2.2 Simultaneous Localization And Mapping . . . 10
2.2.3 Mapping with positioning . . . 10
3 Method 13 3.1 Pilot study . . . 13
3.2 Practical work . . . 13
3.2.1 Map Generation . . . 14
3.2.2 Improving the position . . . 15
3.3 Thesis . . . 15
v
vi CONTENTS
4 Result 17
4.1 Map generation . . . 17
4.2 Positioning . . . 18
5 Discussion 21 5.1 Map generation . . . 21
5.1.1 Mapping from raw data . . . 21
5.1.2 Improved user experience . . . 22
5.1.3 Non-implemented methods . . . 22
5.2 Positioning . . . 22
5.2.1 Unreachable areas . . . 23
5.2.2 Non-implemented methods . . . 23
6 Conclusion and future work 25 6.1 Conclusion . . . 25
6.2 Future work . . . 25
Bibliography 27
List of Figures 31
Chapter 1 Introduction
This section describes the aim of the project and is divided in differ- ent sections. Section 1.1 describes the background to the problem area.
Section 1.2 describes the problem. In section 1.3 the reader is intro- duced to the purpose of the project and the project goal in section 1.4.
Section 1.5 discloses what methods and methodologies that is used in the project. Section 1.6 presents the delimitations. In 1.7 the reader is presented with the disposition of the project thesis.
1.1 Background
A regular GPS can guide you almost anywhere in the world. It can guide you to the hospital, to the grocery store or in a foreign country.
It can guide you anywhere as long as you travel outside. By using indoor positioning you could extend this to also guide you inside the hospital to a specific department, help you find inside a grocery store or even guide you inside a museum. The possibilities are endless.
When implementing indoor positioning and mapping inside a build- ing the normal practice is to take an existing map, a floor plan or blueprints, and then choose a suitable location system [12].
Real-time locating systems (RTLS) are widely used today in programs and systems. One problem that occurs when using locating systems is that we always have an error margin between the measured location
1
2 CHAPTER 1. INTRODUCTION
and the true location. When analyzing different RTLS, accuracy is the most important feature in a working system. The Accuracy measures the average Euclidean distance between the estimated location and the true location. Even though the accuracy is the most important feature of a RTLS, with a high accuracy there is often a trade off between ac- curacy and other characteristic features such as precision, complexity, robustness, scalability and cost [9].
The RTLS that this thesis and project will address is a system that are connected to multiple hand-held devices with a screen, that shows that device’s current location on a two dimensional map.
1.2 Problem
The problem with indoor positioning is that it is not always accurate enough. There are multiple techniques to improve the accuracy of in- door positioning but they are not always applicable due to restriction of hardware and too complicated implementations.
A problem occurs when mapping objects with the help of the posi- tion provided from an indoor positioning system (IPS). Objects that actually have the same size and are aligned with each other, might be different sizes and not aligned because of the error between the true location and the measured location. When an object on the map is drawn, the location shown on the device should not be inside any ob- ject.
What methods are suitable to use to make the map correspond to real- ity and how can a working collision function be implemented?
1.3 Purpose
The purpose of the thesis is to present the work of our project, what decisions we took and why we took them. It also presents the theoretic background needed for the project as well as what methods we used to conduct the project.
The Purpose of the project is to come up with a proof of concept of
CHAPTER 1. INTRODUCTION 3
a system for a company. The system is supposed to map larger objects, like shelves, benches, tables, in an indoor environment onto a two di- mensional map, with only the use of the hand-held device’s position as input data.
1.4 Goal
This section is divided into four different subsections. Subsection 1.4.1 contains the goals of this thesis. Subsection 1.4.2 explains the deliv- erables. In subsection 1.4.3 the results will be introduced. Subsection 1.4.4 describes who will benefit from this project, if there are any ethi- cal issues that can arise and the sustainability aspects of the project.
1.4.1 Goals
The goals of this project are to improve the current system the com- pany is using, by first develop a system that easily can map objects onto a two dimensional map and the drawn map should visually cor- respond to reality. The users should then be able to walk around with a seemingly correct positioning and without any inconveniences, like standing within an object or jumping around all over the place. In other words, the system should have a good user experience.
1.4.2 Deliverables
This project is divided in two deliverables:
• To be able to easily generate a map that corresponds to reality.
• Improve the user experience by improving the position.
1.4.3 Result
The result of the project is to deliver a working proof of concept of a system that can be analyzed and evaluated for its usability.
1.4.4 Benefits, Ethics and Sustainability
This project is an assignation from a company called Virtual Stores.
The company will benefit from this project by having this proof of con- cept explored.
4 CHAPTER 1. INTRODUCTION
This system will only be a tool for administrator users to make it eas- ier to map objects onto a two dimensional map and improve the user experience, therefor no ethical issues should arise from this project.
There are no direct sustainability impacts from this system. The sys- tem, when in use, will make it more efficient to draw and generate maps. This results in less time needed to be spend on generating maps which might be able to contribute to a more economical sustainability for the company.
1.5 Methodology
The first couple of weeks of this project will be spend on literature studies, researching different strategies to approach this problem and getting familiar with the current system in the company. There are dif- ferent methods of reasoning when approaching a problem. One can either use a deductive or inductive approach. A deductive approach consists of first develop a theory. This theory is then narrowed down into specific hypothesis that is testable. The hypothesis are further narrowed down when observations are collected that addresses the hypothesis. This ultimately leads to being able to test the hypothesis with specific data that can lead to a confirmation or annulment of the original theory [16]. The inductive approach is the other way around.
Here observations are made first in order to try to discover patterns.
From the patterns a broader generalization and theory is formed.
There is also a third approach to a problem, an abductive approach.
The abductive approach is given an observation or a set of observa- tions one seeks to find the simplest and most likely explanation [15].
In this project the abductive approach will be used, along with the con- cept of trial and error to try to find a solution that fulfills the goals of this project. The abductive approach fits our needs because we do not have any specific data to compare with a hypothesis. Since the prob- lem is mostly based on user experience the trial and error approach suits the project.
CHAPTER 1. INTRODUCTION 5
1.6 Delimitations
These delimitations was taken in action to fit the scope of the project:
• Software development only on android.
• No hardware change can be made in the handheld device, e.g.
include Inertial Measurement units (IMU) .
• Limit the indoor mapping to simple rectangular objects.
• Indoor mapping is done on one device only.
1.7 Outline (Disposition)
Chapter 2 will be include the theory needed for the project. Chapter 3 will be about what methodologies and what was done during the project. Chapter 4 will present the project results. Chapter 5 will dis- cuss the results and Chapter 6 will conclude the results and present future work.
Chapter 2 Theory
This chapter contains theory about improving indoor positioning and indoor topographic map generation relevant for this thesis.
2.1 Positioning
This section will go over some techniques used to estimate or improve indoor positioning.
2.1.1 Filter
Filters can be used in many ways to improve a position or a signal. It can for instance try to predict where the position should be and com- pare it with the raw data and with the combined data make a more accurate position. Filtering can also filter out noise that can interfere with the signals that provides the raw data, to keep the position as accurate as possible [8]. In this thesis two filters will be explained, Kalman filter and particle filter.
Kalman filter
A Kalman filter is an recursive algorithm that estimates the state of a linear system with Gaussian noise (normal distributed signal noise).
When using a Kalman filter, one must first recognize what the filter should measure. It’s not always suitable to measure and estimate ev- ery variable of the system[6].
The filter takes the state vector from each new measurement and cal-
7
8 CHAPTER 2. THEORY
culates an estimation based on the previous state and the probability of the estimated state. In terms of indoor positioning the state vector could include the position and velocity in the x- and y-axis [11]. The Kalman filter will then estimate the next position and velocity based on all previous measurements.
Particle filter
A particle filter is the implementation of the formal recursive Bayesian Filter with Sequential Monte Carlo methods [4][17]. The idea with particle filters is to estimate a position, based on a set of samples of particles and each particle’s respective weight. The weight is based on the probability that the sample is the actual location. This requires a measurable reference of some kind, for example the distance from the position to a reference object. When the weight is calculated for each sample, the samples with low weight (low probability) will be removed and a resample is made around the high probability samples where all samples are equally weighted particles. These three steps (sample, compute weight and resample) will be repeated every itera- tion and each time the sample will be spread around the highest prob- ability particles and the lowest probability particles will be removed from the set [4].
2.1.2 Inertial Measurement Unit
Inertial Measurement Unit or IMU for short, is an electronic device that measures rotational changes, acceleration and magnetism in x-, y- and z-directions. What the IMU measures depends on what com- ponents the IMU contains. These components is often a gyroscope, an accelerometer and a magnetometer. When it comes to indoor posi- tioning, an IMU can be used as a complement to the main positioning system to better estimate the position [7].
Dead Reckoning is the method to constantly measure the changes in speed, acceleration and direction with an IMU to predict where a ob- ject moved based on the previous location. Dead reckoning for indoor environments have become common enough that it gained its own name: Pedestrian Dead Reckoning (PDR) [1]. The problem with PDR is that humans have a tendency to often change speed and direction which results in an inaccurate position. To compensate for this error,
CHAPTER 2. THEORY 9
usage of a positioning system together with PDR is necessary [1].
PDR is a effective way to improve an indoor positioning but since we don’t have access to any IMU inside the handheld device and cannot change the hardware, the PDR method ends outside of our delimita- tions.
2.1.3 Unreachable areas
When making indoor maps and positioning systems, it’s important to separate areas where a user can or cannot walk on. In e.g. a user should not be able to visually stand/walk inside a wall, shelf or other objects [14]. If the positioning system cannot provide an exact position at all times, this could happen and it would affect the user experience in a negative way.
In our project, this could be a good way to improve the user experi- ence by not letting a user walk inside an object or visually step out of the map due to the error margin in the positioning system.
2.2 Map generation
This section is about different ways to create maps for indoor environ- ments used by positioning systems.
2.2.1 Floor plans or Blueprints
Indoor floor plans or blueprints is often used when setting up indoor positioning in a large building like airports, stadiums and even uni- versities [18]. A floor plan is a 2D topographical map that shows the layout of an entire floor. It only shows rooms and where they are, not details like tables, obstacles and other objects that can be moved.
Therefore it’s suitable for large and open buildings. By using already existing floor plans or blueprints, the time for setting up indoor posi- tioning in a building is reduced [18].
This could be used on the improving position part of the project but it defeats the purpose of mapping an environment based on the position system, thus this technique is not suitable for this project.
10 CHAPTER 2. THEORY
2.2.2 Simultaneous Localization And Mapping
Simultaneous Localization And Mapping, also known as SLAM, is a combination of a localization and mapping system used in an at first unknown environment. A SLAM device uses a combination of sen- sors to scan the environment. The sensors can be cameras, ultrasonic sensors, radar, etc. depending on what environment SLAM is used in [3] [5].
When a SLAM device scans an environment it pinpoints different land- marks to create reference points that are used in the localization. In an indoor environment, these landmarks can be walls, corners, door openings or any kind of unmovable object [13]. These landmarks are then evaluated and an topographical map is created.
To calculate the location of the SLAM device, some kind of position estimation is needed (e.g. kalman- and particle filter, section 2.1.1). To- gether with the position estimation and the sensors locating the land- marks, the SLAM device have a fairly accurate position. Because the SLAM technique is both used for both localization and mapping at the same time, it’s dynamic and generates a better result over time [13].
SLAM is a good technique for robots that need to navigate and cre- ate map in open buildings like warehouses [2]. This does not fit our project due to the complicity of the SLAM technique and we must work with the hardware accessible on the handheld device.
2.2.3 Mapping with positioning
Mapping from coordinates based from a positioning system is not the standard practice but it is a good way to create maps in environments that changes a lot or where no blueprints/floor plans exists. Examples of these environments can be supermarkets or warehouses. The tech- nique is to use a coordinate like a landmark in SLAM(section 2.2.2), and with several coordinates map an entire environment [14]. It’s easy to generate maps if the positioning system is fairly accurate but if it’s not, then the map will not correspond to the real environment.
Due to an already existing positioning system in the environment this thesis will be conducted in, this will likely be the mapping method we
CHAPTER 2. THEORY 11
will use. The position system that generates our coordinates have an error margin that makes this method usable for us.
Chapter 3 Method
In this chapter, the research methodology and the methods used in the degree project will be described. Section 3.1 describes the pilot study. In section 3.2, the work that was made during the project will be presented. Section 3.3 explains the writing of the thesis.
3.1 Pilot study
In the beginning of the project, a theoretic pilot study was done to find out what methods are used for generating maps and improving positions for indoor navigation. The study was done by reading other thesis reports on similar subjects and analyzing the methods they used to see if it will fit inside the scope of this project. The pilot study set the base for all the theory in the project and is concluded in chapter 2.
While the pilot study of this project was made, no thesis were found that had done research on the same subject within map generation.
Therefore most of the techniques used for the map generation were constructed during the project.
3.2 Practical work
The practical work done for the project is presented in the following two subsections. 3.2.1 will describe the work done on the map gener- ation application and the subsection 3.2.2 will describe the work done on improving the position.
13
14 CHAPTER 3. METHOD
3.2.1 Map Generation
The first thing to be implemented was the selection of what kind of rectangle one would like to draw. For that two modes were created:
Mode 1 This mode draws straight rectangles that adjusts the sides to the mean value of the x- and y-coordinates.
Mode 2 This mode draws rotated rectangles with the line between the first two corners as a reference and the length to the third mark will be drawn at a 90 degrees angle from the first side.
Both modes will sort the corners, so that the first corner is always at the top left and then continue clockwise.
The next thing to be implemented was a way to make the map visu- ally look better. Therefore a number of functions were implemented.
The first function to be implemented was a function that made it possi- ble to select objects. This was made by registering screen touches and check if the position was inside any of the existing objects. When ob- jects could be selected, eight other functions were implemented. Two of these function were made to change the size of the objects, one for the width and one for the height. The functions iterated through all se- lected objects and calculated the mean width or mean hight (based on which of the functions were selected) and then it adjusted all objects to the mean length.
The other six function that were implemented were functions that changed the position of the objects without changing its size. Two of these func- tions adjusts the gaps between the objects (one for the horizontal gap and one for the vertical gap). These functions iterates through all the selected objects and calculate the mean gap and then sets the mean gap between every object. Worth noting is that the total gap (between the first and last object) will still be the same, which mean that those objects will not move.
The last four functions are alignment functions. The first object that is selected will be a reference object. The four functions are as follows:
• Align right
• Align left
CHAPTER 3. METHOD 15
• Align top
• Align bottom
These will align all the selected object along with the reference object.
3.2.2 Improving the position
In order to improve the position the concept of filters and its use were looked into. There was already an existing particle filter at the com- pany, but Kalman filter was researched to investigate if that filter would be better. The Kalman filter was only partly implemented when it was decided that another approach would be made.
Instead of the Kalman filter to improve the position, it was decided to focus purely on the user experience view and solve the problem with the ability to stand within an object. A function that detected if a position was within an object was created with the help of the same function that was used to select object. If a position was going to land within an object, the position was moved to the closest position out- side the object.
3.3 Thesis
Most of the theoretic part in the thesis was written throughout the pilot study and in the beginning of the practical work. The rest was written as the project proceeded.
Chapter 4 Result
This chapter includes the result divided into two sections. 4.1 Map generation describes the software created to generate maps and 4.2 describes what was done in order to improve the position.
4.1 Map generation
The final map generating software is able to create topographical 2D maps of the environment inside a grocery store. This map generat- ing tool can create simple rectangular objects, both horizontal/vertical rectangles as well as rotated ones.
The theory behind the map generating software is "Mapping with po- sitioning" (section 2.2.3). Three raw coordinates from the IPS was used to create each objects, this made it possible to create rotated rectangles and improve the error margin for the horizontal/vertical rectangles.
Due to the irregular error margin in the IPS, several functions was cre- ated to visually improve the map. Both the raw data map and the visually improved map can be seen in figure 1. These functions was made only to improve the User experience through aligning of objects, change them to the same width / height and also set an equal distance between selected objects.
17
18 CHAPTER 4. RESULT
Figure 1: Two topographic maps of a grocery store.
4.2 Positioning
Several attempts to improve the positioning was made. A Kalman fil- ter was partly implemented only to discover that it didn’t fit the project due to the projects delimitations. After discarding the Kalman filter an successful attempt to implement the unreachable areas was made.
Every time the user position is updated, a function checks if the user coordinates is inside an object on the map and then moves the user to the nearest edge if a collision occurred. This can be seen in figure 2.
CHAPTER 4. RESULT 19
Figure 2: Map illustrating the unreachable areas method.
Chapter 5 Discussion
This chapter will discuss the results of this degree project. It will also reflect upon the positive effects and the drawbacks in each method implemented.
5.1 Map generation
Here will we present our general thoughts on the map generating soft- ware and the positive and negative aspects on the theory behind it.
Due to limited hardware and no access to a floor plan our options for the map generating software was strictly limited. We created a good map generating tool that we further developed to improve the user experience.
5.1.1 Mapping from raw data
The result of the basic map generating tool was fairly good. The gen- erated map based on raw data only, see figure 1, looked similar to the real environment. Every object had a slight error due to the irregular error in the IPS and therefor identical objects wasn’t identical at all on the map. The only solutions we came up with that could eliminate this problem (using a floor plan excluded) is to either have an IPS that can pinpoint an exact location at all times, such system doesn’t exists[10], or create functions that could visually fix these errors in the raw data.
We choose to visually improve the data and base the reconstruction of the shelves on average measurements on similar shelves. This cre-
21
22 CHAPTER 5. DISCUSSION
ated a more appealing view and because the similarities in the shelves and multiple data points, it actually improved the true location of the shelves.
5.1.2 Improved user experience
We think user experience is the most important when it comes to in- door maps. The user needs to be able to look at the map and recog- nize the surrounding area, therefor we thought that implementing the align, same width/height and equal space functions was absolute nec- essary to create realistic indoor maps. The drawback with this method is that when changing the appearance of the objects we also changes the coordinates of each objects. When doing this we cannot guarantee that an object on the map corresponds to where the IPS thinks it is. On the other hand, due to the error margin of IPS, we couldn’t guarantee this before either so it isn’t that big of a problem if we keep the changes to a minimum. If we look at figure 1 we can see that the final map isn’t that off from the raw data. That tells us that the final map will work well together with the raw data position.
5.1.3 Non-implemented methods
Floor plan or blueprints
The usage of a floor plan as the indoor map is not to recommend in an environment that changes on a regular basis. We think it’s an good idea to use the floor plan method in environments similar to airports or shopping malls where the structure isn’t going to change regularly.
Unfortunately the environment changes quite often in a grocery store.
Simultaneous Localization and Mapping
We think SLAM is a perfect method for robotics to generate maps and localize itself in an unknown environment. Sadly SLAM isn’t appro- priate for us to use due to the hardware requirements.
5.2 Positioning
In this section we will discuss the positive and negative things about our attempts to improve the positioning. This part of the project was
CHAPTER 5. DISCUSSION 23
the most difficult one. Improving the positioning was the most chal- lenging part of this project. We think unreachable areas was the best way to go because we didn’t have an IMU to analyze how the user moves.
5.2.1 Unreachable areas
Unreachable areas is straight forward, lightweight and was fairly easy to implement because the map generation had already created all co- ordinates for every object. This method worked really well together with the improved user experience method in the map generating part of the project. The modified map could, as said before, be a little off from the IPS point of view and it could appear that a user was standing inside an object. Unreachable areas method took care of this problem as well as the same problem happening from the error created by the IPS. With this implemented the user experience was greatly increased because the user would never appear to stand inside of an object.
5.2.2 Non-implemented methods
Kalman filter
We went in with our heads high and thought Kalman filter was the way to go, except it wasn’t. It could have been if we had an IMU to work with and therefor introduce more variables. Only working with the coordinates in an Kalman filter isn’t going to work well for this kind of project.
Inertial Measurement Unit
Including an IMU would have been a great way to improve the posi- tioning through the Pedestrian Dead Reckoning method. Every part of this project could benefit from the data the IMU provides.
Chapter 6
Conclusion and future work
This last chapter concludes the results and presents potential future work.
6.1 Conclusion
This project resulted in a new way to create indoor maps, creating them based on the coordinates generated by the indoor positioning system.
It also resulted in a position improvement by not letting a position visually be inside a object. These implementations, together with the functions that improves the user experience when traversing the map, making it even and neat, creates a strong map generating software with an visually improved position.
6.2 Future work
There are a numerous things that could be done to improve this project even further.
• An IMU could help improve the positioning with an PDR imple- mentation.
• The method for the unreachable areas could be developed fur- ther and take the adjustments made on the map in consideration.
• If an IMU were included and a PDR implemented, a Kalman fil- ter could help the positioning ever further.
25
Bibliography
[1] Kenneth C Cheung, Stephen S Intille, and Kent Larson. “An in- expensive bluetooth-based indoor positioning hack”. In: Proceed- ings of UbiComp. Vol. 6. 2006.
[2] A.J. Davison. “Real-time simultaneous localisation and mapping with a single camera”. In: vol. 2. 2003, pp. 1403–1410.URL: https:
//www.scopus.com/inward/record.uri?eid=2-s2.0-
0345414201&partnerID=40&md5=4bee7b580662769f1c1cab7a8c912af3.
[3] M. W. M. G. Dissanayake et al. “A solution to the simultaneous localization and map building (SLAM) problem”. In: IEEE Trans- actions on Robotics and Automation 17.3 (June 2001), pp. 229–241.
ISSN: 1042-296X.DOI: 10.1109/70.938381.
[4] Arnaud Doucet and Adam M Johansen. “A tutorial on particle filtering and smoothing: Fifteen years later”. In: Handbook of non- linear filtering 12.656-704 (2009), p. 3.
[5] H. Durrant-Whyte and T. Bailey. “Simultaneous localization and mapping: Part I”. In: IEEE Robotics and Automation Magazine 13.2 (2006), pp. 99–108. DOI: 10.1109/MRA.2006.1638022. URL: https://www.scopus.com/inward/record.uri?eid=
2 - s2 . 0 - 33745178432 & doi = 10 . 1109 % 2fMRA . 2006 .
1638022&partnerID=40&md5=53372260cf7799b3db416d34b82ed0fc.
[6] Mohinder S. Grewal. “Kalman Filtering”. In: International Ency- clopedia of Statistical Science. Ed. by Miodrag Lovric. Berlin, Hei- delberg: Springer Berlin Heidelberg, 2011, pp. 705–708.ISBN: 978- 3-642-04898-2. DOI: 10.1007/978- 3- 642- 04898- 2_321.
URL: https://doi.org/10.1007/978- 3- 642- 04898- 2_321.
27
28 BIBLIOGRAPHY
[7] Magdy Ibrahim and Osama Moselhi. “Inertial measurement unit based indoor localization for construction applications”. In: Au- tomation in Construction 71 (2016). The Special Issue of 32nd In- ternational Symposium on Automation and Robotics in Con- struction, pp. 13–20.ISSN: 0926-5805.DOI: https://doi.org/
10 . 1016 / j . autcon . 2016 . 05 . 006. URL: http://www.
sciencedirect.com/science/article/pii/S0926580516300863.
[8] Rudolph Emil Kalman. “A new approach to linear filtering and prediction problems”. In: Journal of basic Engineering 82.1 (1960), pp. 35–45.DOI: 10.1115/1.3662552.
[9] H. Liu et al. “Survey of Wireless Indoor Positioning Techniques and Systems”. In: IEEE Transactions on Systems, Man, and Cyber- netics, Part C (Applications and Reviews) 37.6 (Nov. 2007), pp. 1067–
1080. ISSN: 1094-6977. DOI: 10.1109/TSMCC.2007.905750.
[10] H. Liu et al. “Survey of Wireless Indoor Positioning Techniques and Systems”. In: IEEE Transactions on Systems, Man, and Cyber- netics, Part C (Applications and Reviews) 37.6 (Nov. 2007), pp. 1067–
1080. ISSN: 1094-6977. DOI: 10.1109/TSMCC.2007.905750.
[11] M. Nabil, M. B. Abdelhalim, and A. AbdelRaouf. “A new Kalman filter-based algorithm to improve the indoor positioning”. In:
2016 5th International Conference on Multimedia Computing and Sys- tems (ICMCS). Sept. 2016, pp. 236–242. DOI: 10.1109/ICMCS.
2016.7905588.
[12] EE Publishers Pierre Potgieter. Indoor mapping: beyond the front door. July 2014. URL: http : / / www . ee . co . za / article / indoor-mapping-beyond-front-door.html.
[13] Morten Rufus Blas Søren Riisgaard. SLAM for dummies. 2014.
URL: https://www.aci-na.org/sites/default/files/
session-4.pdf.
[14] M. S. Svalastog. “Indoor Positioning - Technologies, Services and Architectures”. Master thesis. University of Oslo, May 2007.
[15] Jan Svennevig. “Abduction as a methodological approach to the study of spoken interaction”. In: 103 (Jan. 2001).
[16] William M.K. Trochim. Deduction & Induction. 2006.URL: https:
//socialresearchmethods.net/kb/dedind.php.
BIBLIOGRAPHY 29
[17] A. F. M. S. Uddin et al. “Particle filter based moving object track- ing with adaptive observation model”. In: 2017 6th International Conference on Informatics, Electronics and Vision 2017 7th Interna- tional Symposium in Computational Medical and Health Technology (ICIEV-ISCMHT). Sept. 2017, pp. 1–6. DOI: 10.1109/ICIEV.
2017.8338569.
[18] Shaun Vann. Indoor Google Maps. 2014. URL: https : / / www . aci-na.org/sites/default/files/session-4.pdf.
List of Figures
1 Two topographic maps of a grocery store. . . 18 2 Map illustrating the unreachable areas method. . . 19
31
TRITA-EECS-EX-2018:158
www.kth.se
