Indoor positioning system for human and slam for robot in disaster relief

TÊN ĐỀ TÀI: Hệ thống định vị trong nhà cho con người và SLAM cho robot trong cứu hộ Indoor positioning system for human and SLAM for robot in disaster relief II.. This work presents tw

ĐẠI HỌC QUỐC GIA TP HCM

TRƯỜNG ĐẠI HỌC BÁCH KHOA



HỒ SỸ THÔNG

HỆ THỐNG ĐỊNH VỊ TRONG NHÀ CHO CON NGƯỜI VÀ SLAM CHO ROBOT

VIETNAM NATIONAL UNIVERSITY OF HO CHI MINH CITY

HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY

CÔNG TRÌNH ĐƯỢC HOÀN THÀNH TẠI TRƯỜNG ĐẠI HỌC BÁCH KHOA –ĐHQG -HCM Cán bộ hướng dẫn khoa học : Tiến sĩ Trương Quang Vinh

Thành phần Hội đồng đánh giá luận văn thạc sĩ gồm:

(Ghi rõ họ, tên, học hàm, học vị của Hội đồng chấm bảo vệ luận văn thạc sĩ)

1 PGS.TS Hoàng Trang

2 TS Bùi Trọng Tú

3 TS Nguyễn Minh Sơn

4 TS Trần Hoàng Linh

5 TS Nguyễn Lý Thiên Trường

Xác nhận của Chủ tịch Hội đồng đánh giá LV và Trưởng Khoa quản lý chuyên ngành sau khi luận văn đã được sửa chữa (nếu có)

CHỦ TỊCH HỘI ĐỒNG TRƯỞNG KHOA

ĐẠI HỌC QUỐC GIA TP.HCM

TRƯỜNG ĐẠI HỌC BÁCH KHOA

CỘNG HÒA XÃ HỘI CHỦ NGHĨA VIỆT NAM

Độc lập – Tự do – Hạnh phúc

NHIỆM VỤ LUẬN VĂN THẠC SĨ

I TÊN ĐỀ TÀI:

Hệ thống định vị trong nhà cho con người và SLAM cho robot trong cứu hộ (Indoor positioning system for human and SLAM for robot in disaster relief)

II NHIỆM VỤ VÀ NỘI DUNG:

Khảo sát, tìm hiểu các kỹ thuật định vị trong nhà và SLAM

Nghiên cứu và thiết kế hệ thống định vị trong nhà ứng dụng cho cứu hộ

Nghiên cứu và thiết kế robot sử dụng SLAM ứng dụng cho cứu hộ

III NGÀY GIAO NHIỆM VỤ : / /2018

IV NGÀY HOÀN THÀNH NHIỆM VỤ: 28/06/2019

First and foremost, I would also like to thank my first supervisor Dr Vinh Truong Quang

of the Faculty of Electrical and Electronics Engineering at the Ho Chi Minh City University ofTechnology for great deal of technical support and guidance He has motivated me to chooseand study further in this research

I would like to thank my second supervisor Professor Filippo Sanfilippo of the Faculty ofTechnology, Natural Sciences and Maritime Sciences at the University of South-Eastern Norwayfor his valuable suggestions and support during the planning and development of this researchwork He consistently allowed this thesis to be my own work but steered me in the right directionwhenever he thought I needed it

I would also like to thank other professors of Ho Chi Minh City University of Technologyand USN, my colleagues, and my friends with their supports during my efforts for this study.Finally, I must express my very profound gratitude to my wife and my lovely daughter forproviding me with unfailing support and daily encouragement throughout my years of study andthrough the process of researching and writing this thesis This accomplishment would not havebeen possible without them

Indoor positioning has attracted a great deal of attention these days due to the rising number

of location-based services It is not only a useful system for daily human activities but also

a key component of emergency rescue evacuation support system to minimize the damages

of disaster However, the most challenging conditions such as darkness, power outages, hightemperatures, flames can prevent indoor positioning working The technologies based on theWireless Fidelity (WiFi) protocol and the Pedestrian Dead Reckoning (PDR) approach havewidely exploited because of the existing WiFi infrastructure in buildings and the advancement

of integrated sensors on smart-phones However, these techniques have several limitations, such

as the fluctuation of the WiFi signal and the accumulation of positioning errors due to the driftcaused by noise in the sensors for the PDR technology The Bluetooth Low Energy (BLE) isanother potential solution for indoor applications because of the low energy consumption andfast response RSS measurements This work presents two main parts: Indoor positioning systemfor human and Simultaneous Mapping and localization (SLAM) for Robot in disaster relief

In the first part, a hybrid algorithm that combines WiFi fingerprinting, BLE, and PDR toboth exploit their advantages as well as limiting the impact of their disadvantages is proposed forlocalization human in the building Specifically, to build a probability map from noisy ReceivedSignal Strength (RSS), a Gaussian Process (GP) regression is deployed to estimate and constructthe RSS fingerprints with incomplete data Mean and variance of generated points are used

to estimate WiFi fingerprinting position by K-nearest weights from the probability of visibleRSS measurements of the online phase In addition, a particle filter is applied to fuse WiFifingerprinting, BLE, and PDR by using the information from RSS, inertial sensors and features

of indoor maps The proposed framework can be low cost, easy integration in all building, highaccuracy and efficiency as the infrastructure becomes collapsed gradually

In the second part, a robot is designed using SLAM to build a dynamic map for indoorpositioning system and search and rescue victims in disaster scenarios The robot focus on theapproach that consumes low computational resource and thus can be used on low-weight, low-power and low-cost processor working int small-scales autonomous system Furthermore, therobot can solve the challenging that when it moves in different infrastructures, the odometryinformation may be not available

Tóm tắt

Hệ thống định vị trong nhà đang được chú trọng gần đây bởi sự gia tăng các dịch vụ về định

vị Hệ thống này không chỉ ứng dụng hữu ích trong các hoạt động hàng ngày mà còn là nhân

tố quan trọng trong sơ tán để giảm thiểu tác động của thiên tai Tuy nhiên, trong điều kiện môitrường như tối, cắt điện lưới cung cấp, nhiệt độ cao, lửa cháy là những thách thức lớn tác độngtới sự làm việc của hệ thống định vị trong nhà Các công nghệ như WiFi hay PDR được khaithác rộng rãi do tận dụng được các kiến trúc mạng WiFi hiện có trong các tòa nhà và các cảmbiến được tích hợp bên trong của các điện thoại thông minh Tuy vậy, các kỹ thuật này có nhữnggiới hạn nhất định, ví dụ sự giao động của tín hiệu WiFi hay sự lũy tiến lỗi vị trí sự trôi các cảmbiến trong công nghệ PDR Bluetooth năng lượng thấp (BLE) là một giải pháp tiềm năng chocác ứng dụng định vị trong nhà do năng lượng tiêu thụ thấp và tốc độ đáp ứng của đo lương tínhiệu sóng (RSS) cao Trong Luận văn này trình bày 2 phần chính: Hệ thống định vị trong nhàcho con người và SLAM cho robot trong cứu hộ

Trong phần đầu, giải thuật lai được đề xuất trong ứng dụng định vị trong nhà kết hợp kỹthuật dấu vân tay WiFi, BLE, và PDR để khai thác được những ưu điểm của các kỹ thuật nàycũng như giảm thiểu ảnh hưởng của các hạn chế của chúng Đặc biệt, để xây dựng bản đồ xácxuất từ nhiễu của tín hiệu cường độ sóng (RSS) , Giải thuật hồi quy Gaussian process được khaithác để ước lượng và xây dựng các vị trí dấu vân tay của RSS từ tập dữ liệu không đầy đủ Giátrị trung bình và phương sai của các điểm tạo ra được sử dụng để xác định vị trí của phươngpháp dấu vân tay WiFi bằng cách tính toán trọng số xác xuất RSS của các điểm hàng xóm gầnnhất trong cơ sở dữ liệu với các giá trị đo tức thời Hơn nữa, Bộ lọc hạt được áp dụng để kết hợpdấu vân tay WiFi, BLE và PDR bằng cách sử dụng thông tin từ RSS, các cảm biến trong điệnthoại và đặc điểm của bản đồ tòa nhà Phương pháp đề xuất có chi phí thấp, dễ dàng tích hợptrong các tòa nhà, độ chính xác cao và hiệu quả khi các cơ sở hạ tầng dần bị sụp đổ do thiên tai.Trong phần còn lại, Một Robot được thiết kế sử dụng SLAm để xây dựng bản đồ động cho

hệ thống định vị trong nhà, hỗ trợ tìm kiếm và cứu hộ các nạn nhân trong các trường hợp thiêntai Robot này tập trung vào phương pháp sử dụng chi phí tính toán thấp và do đó có thể sử dụngcác bộ xử lý nhẹ, năng lượng tiêu thụ và giá thành thấp ứng dụng trong các hệ thống tự độngquy mô nhỏ Hơn nữa, robot có thể giải quyết được khó khăn trong di chuyển ở các địa hìnhkhác nhau khi các thông tin về động lực của robot không sẵn có

Mục lục

1 Overview and Motivation 1

2 Research Questions 4

3 Approach 4

3.1 Indoor Positioning System 4

3.2 Simultaneous Localization and Mapping for Robot 6

4 Outline 6

2 Backgrounds and Related Works 8 1 Indoor Positioning System 8

1.1 General Indoor Positioning System 10

1.2 Indoor Positioning in Disaster relief 12

2 SLAM for Robot 13

2.1 Simultaneous Localization and Mapping Problem 13

2.2 SLAM for Disaster Relief 16

3 Methods for Indoor Positioning of Human 17 1 RSS-Based and Path-loss Model 17

1.1 WiFi RSS based 17

1.2 iBeacon based Localization 18

2 WiFi Fingerprinting using Gaussian Process Regression 20

2.1 Offline phase: Building WiFi Fingerprinting Maps Using a Gaussian Process Regression 20

2.2 Online phase- WiFi position estimation 22

3 Pedestrian Dead Reckoning 23

3.1 Step Detection using Accelerometer 23

3.2 Stride Length Estimation 25

3.3 Heading Estimation 26

4 Fusing Algorithm with Particle Filter 26

4 Methods for ROBOT using SLAM 30

1 Localization and Mapping 31

1.1 Hector SLAM 31

1.2 Pose Estimation 31

2 Kinematics for Two Wheels Robot 31

2.1 Differential Controller 31

2.2 Close Loop Control using PID 32

2.3 Odometry Estimation 33

3 Sensors 35

3.1 Laser Range Finder 35

3.2 Inertial Measurement Unit (IMU) 36

3.3 Camera 36

4 Navigation 37

5 Clients Nodes 37

5 Experiment and Result of Indoor Positioning System 38 1 Indoor Positioning System Architecture 38

1.1 Android Application 38

1.2 Web Server and Database 40

2 Result and Evaluation 40

2.1 Experimental Setup 40

2.2 Evaluation Methodology 42

2.3 Results and Evaluation 42

3 Discussion 48

6 Experiments and Results of SLAM 50 1 Experimental Setup 50

1.1 Hardware on Robot 50

1.2 Test Map 50

1.3 Remote Controller using Smart-phone 50

1.4 ROS Nodes for Hector Slam 51

2 The Results and Evaluation 51

7 Conclusion and Future Work 54 1 Conclusion 54

2 Future Works 54

A Appendix 56 1 Android Application 56

2 Web Server 56

Danh sách hình vẽ

1.1 The survey in Fires from U.S 2

1.2 The general scheme of Indoor Positioning System 4

1.3 The hybrid framework for IPS 5

1.4 The architecture of the robot using SLAM for disaster relief 6

2.1 A functional block diagram of positioning system 8

2.2 Literature review of indoor positioning system 9

2.3 A basic indoor positioning system using fingerprinting technique 10

2.4 Indoor positioning approaches using WiFi Fingerprinting 11

2.5 General SLAM techinique 14

2.6 A literature review for SLAM 15

3.1 The relationship between RSS and distance using path-loss model 18

3.2 Histogram of RSS 19

3.3 An experiment of Kalman filter for BLE 19

3.4 A Gaussian process (GP) regression for an indoor positioning system 20

3.5 This is the caption in list of figures 22

3.6 An experiment of GP for hyper-parameter estimation 22

3.7 Step detection using the accelerometer 25

3.8 The proposed hybrid framework for indoor positioning 27

4.1 Overview schematic diagram of the ROBOT 30

4.2 The close loop control 32

4.3 PID velocity control 34

4.4 The PID experiment 35

4.5 Two instantaneous velocity scenarios 36

4.6 RPLidar A1 36

5.1 The experimental architecture of Indoor Positioning System 39

5.2 The wireless sensors modules are used for IPS 40

5.3 The summary architecture of Android Application for IPS 41

5.4 The sequence diagram of map view 41

5.5 The sequence diagram of training data 42

5.6 MVC model for Sailjs 43

5.7 Training positions of test-bed 1 44

5.8 Training positions of test-bed 2 45

5.9 CDF results of test-bed 1 46

5.11 Localization performance in different scenarios 46

5.10 The experiments of test-bed 2 49

6.1 The designed robot 50

6.2 The real map of the fifth floor of Krona building 51

6.3 The Android application on smart-phone to control robot 52

6.4 ROS nodes network of Hector slam running from bag file 52

6.5 The maps are built from Hector slam and gmapping 53

A.1 Some user interface of Android experiment for IPS 58

A.2 The database of web server 59

Danh sách bảng

1.1 The taxonomy of disaster 1

2.1 The Taxonomy and References for Literature review in Figure 2.2 9

5.1 Mean errors of GP and KNN 43

5.2 Mean errors of different experiments 46

5.3 The localization performance with the partially available infrastructure 47

5.4 Computational time 47

5.5 Mean error of different approaches 47

A.1 The web services API are desined 57

List of Acronyms

SLAM Simultaneous Localization and Mapping WiFi Wireless Fidelity

PDR Pedestrian Dead Reckoning

RSS Received Signal Strength

GP Gaussian Process

KNN K-nearest N

BLE Bluetooth Low Energy

RFID Radio Frequency Identification

UWB Ultra Wide Band

AOA Angle of Arrival

TOA Time of Arrival

TDOA Time of Different Arrival

POA Phase of Arrival

KF Kalman Filter

AP Access Point

RP Reference Point

LGD Logarithmic Gaussian Distance

SAR Search and Rescue

Chương 1

Introduction

1 Overview and Motivation

According to the World Health Organization (WHO), "Disaster is an occurrence disruptingthe normal conditions of existence and causing a level of suffering that exceeds the capacity ofadjustment of the affected community" [1] Disasters around the world including natural andman-made hazard have been damaging to communities and countries They can be classifiedinto two main types in Table 1.1 In the case of natural disasters, such as the well-known Hur-

Type DisastersNatural Earthquakes, typhoons, hurricanes (storm), cyclone,

volcanic eruptionMan-made Fire, Explosion, collision, shipwreck, structural col-

lapse, environmental pollution

Bảng 1.1: The taxonomy of disaster

ricane Katrina in 2005 and the Haiti earthquake in 2010, the affected victims and areas weresizeable so that the rescue required worldwide and long-term action However, the minor tomoderate natural disasters may bring slight damage to a building and hurt people inside In thecase of man-made disasters, fire in a building is the most frequent disaster Figure 1.1 showsthe trends in fires, deaths, injuries and dollar loss from U.S of fire statistic from 2008 to 2017.About 1.3 million fires are reported and they cause more than 18000 casualties every year [2].Though large disasters impact society more, small to moderate disasters are more frequent andlikely to occur in our lives

In the case of small-scale disasters, people are apt to get trapped inside buildings For ple, a moderate earthquake may cause a wall to collapse or a column of the structure or knockover furniture like bookcases Consequently, victims might be trapped in the building, and getinjured or even die if they do not get help to escape In the case of fires in a building, most vic-tims die from smoke or toxic gases and not from burns [3] Therefore, the rapid rescue of peopleinside the building is the key to reducing casualties in the case of small to moderate disasters.Considering the effects of the disaster in residential buildings in urban areas An IndoorPositioning System (IPS) can greatly assist the victims and rescuers in case of disasters:

exam-1 In Emergency Rescue Evacuation Support System by locating individuals inside the ing and guiding them to a safer place [4]

build-2 Localizing the position of the victims, it may be include estimating emergency situations

Hình 1.1: Trends in fires, deaths, injuries and dollar loss from U.S fire statistic from 2008 to

2017 [2]

3 Supporting for the rescuers to identify the location of the victim and reach their position

However, the use of IPS in disaster-relief is known to have challenges According to [5],indoor environment can be classified as structured or known, semi-structured and unstructured

or unknown depending on the control that the IPS possesses over them as follows:

• Structured or Known Environment: these environments do not change their shape over

time or the IPS is updated when a change occurs [5] That means the configuration ofthe Wireless systems (e.g WiFi access points, Bluetooth Beacons) are not changed overtime In addition, the positions of the landmarks (such as the map of the building, doors,elevators) are known beforehand by the system

• Semi-structured Environment: in this type, it is assumed that parts of the building are

changed by disasters Some of the wireless networks are not worked

• Unstructured or Unknown Environment: in this type of scenario, the IPS does not have

control over the environment The electrical system may not work The structure of thebuilding can change and the wave propagation conditions are affected This is the mostchallenging environment for the localization process, since the IPS has to deal with dy-namic changes of the environment

Recently, different technologies are applied for these location services Nevertheless, to have

an IPS that can assist in finding and rescuing victims is still a big challenge for the ments mentioned above Since the GPS cannot guarantee location service because a line-of-sight transmission between receivers and satellites is not possible in an indoor environment,many approaches in current positioning technologies have been proposed: including Ultra WideBand (UWB) [6], Ultrasound [7], Radio Frequency Identification tags (RFID) [8], BluetoothLow Energy (BLE) [9], WiFi [10], PDR using inertial sensors [11, 12, 13],and Camera [14].Among them, WiFi and BLE technology based on the Received Signal Strength (RSS) havebecome the common solutions for indoor positioning because of the convenience of measuringthis value directly in most devices such as smart-phones, laptops

environ-In order to estimate a position from RSS, path-loss model-based or fingerprinting basedapproaches can be used Firstly, a “path-loss” based approach is a technique that converts thevalues of RSS from the access points to the mobile receiver into distances based on a signalpropagation model [15] However, the position relationship is highly complex due to multi-path,

metal reflection and interference noise [16] Thus, the path-loss model may not be adequatelycaptured by an invariant model Secondly, fingerprinting/sense analysis is a technique that esti-mates the position based on a scene analysis This technique estimates the user’s position withregards to the similarities between the RSS measurements of online phase and RSS of the offlinephase training [17, 18] The main advantages of WiFi fingerprinting are that it takes advantage

of current WiFi infrastructures, and the location of the access points can be unknown On theother hand, the disadvantages of the fingerprinting method include the need for dense trainingcoverage and the poor extrapolation of areas not covered during the training phase During theoffline phase, it can be extremely time-consuming and labour-intensive to build substantiallylarge fingerprinting databases [18]

Similar to WiFi, RSS of BLE technology can be used to locate an object It is also affected

by indoor environments However, it has advantages of bi-transmission range, low energy sumption and getting information about the location in a few milliseconds [19] Hence, the IPSusing BLE can use path-loss model-based to avoid time-consuming of the training phase, but itcan get high accuracy by different filters for fast response RSS measurements

con-Another widely adopted localization approach is PDR [11, 13], which leverages inertialsensors to estimate the displacement of pedestrians relatively to their previous position Themain challenge in this approach is that the inertial sensors in commercial smart-phones oftensuffer from imperfect calibration and noisy measurements [18] In addition, step counting iscurrently a major method to capture the walking path and the movement of pedestrians [11].The estimated location of PDR is often drifted when travelling a long distance due to inaccuratemeasurement of step detection, step length and heading

In the urban areas, the most challenging conditions such as darkness, power outages, hightemperatures, smoke, flames, and noise can prevent an IPS from working There are many dif-ferent ways to support rescuers and communication between rescuers and victims However,taking into account the human in the building, an Indoor positioning system plays a critical role

to minimize the damage of disasters, especially for evacuation by locating individuals insidethe building and guiding them to safety The IPS should be convenient for everyone to use andhighly accurate in structured environments

Another important mission in disaster relief is Search and Rescue (SAR) Rescue teamshave to explore a large terrain within a short amount of time in order to locate survivors after

a disaster Simultaneous Localization and Mapping (SLAM) for robotics is very important toexplore the environment autonomously or partially guided by the incident commander Theirtasks are to jointly create a map of the terrain and to register victim locations, which can further

be utilized by human task forces for rescue In unstructured environments, a dynamic map fromthe robot can be useful for the indoor positioning system

It is clear that the IPS cannot work when all infrastructures of the building were collapsed Inthis thesis, an indoor position framework is proposed with a partial infrastructure, which refers

to the case where only a part of the infrastructure provides its functionality like electricity Inaddition, a ROBOT using SLAM techniques is also presented in this thesis to draw the buildingmaps for both indoor positioning system and SAR moreover, with the rescue robot, it providespromising solution to assist people in urban areas in term of: 1) reducing personal risk to peopleand rescue dogs by entering unstable structures, 2) increasing speed of response by penetratingordinarily inaccessible voids, and 3) through the information from cameras and sensor fusion inorder to extend the reach of rescuers to regions that are otherwise inaccessible [20]

Hình 1.2: The general scheme of Indoor Positioning System

3.1 Indoor Positioning System

The general scheme of the proposed framework shows in Figure 1.2 The IPS deploys ent sensors on smart-phone to build an effective indoor positioning system that can work in dif-ferent previous mentioned indoor environments Especially, the proposed framework focuses onfusing different indoor techniques (Fingerprinting, path-loss, PDR) and different technologies(WiFi, BLE, inertial sensors) and building map to improve the position accuracy and efficiency

differ-as the infrdiffer-astructure becomes collapsed gradually This IPS is low-cost, ediffer-asy to integrate into

Hình 1.3: The hybrid framework of indoor positioning system in the building for disaster relief

the building and convenient for people inside the building It also can combine with robot to usedynamic map (robot map) when the environments change Furthermore, In the offline phase ofWiFi fingerprinting techniques, robot can be used to collected WiFi RSS at the reference points

to reduce training time The general scheme of IPS in the building for disaster relief as shown

in Figure 1.3 The proposed framework aims at achieving location robustness and accuracy fordisaster relief by combining a number of techniques:

• Rss based path-loss for BLE iBeacon: the BLE beacons are deployed by RSS based using

path-loss model to measure the distance between mobile users and these beacons Thenoise of RSS measurement can be reduced by Kalman filter (KF) The benefit of BLEbeacon is low energy consumption So, it can work in blackout condition

• Constructing a "WiFi map" With the aim of reducing the time needed for data training

during the offline phase and for improving the accuracy of WiFi fingerprinting, a sian process (GP) regression is employed This makes it possible to obtain the mean andvariance of the considered WiFi map based on the correlation between RSS of sparsetraining points Moreover, An efficient method is proposed to evaluate the user’s position

Gaus-by real-time RSS measurement and the WiFi map

• Motion estimation of PDR To detect motion and calculate the movement of a pedestrian

using smart-phones, we aim at improving the step detection and stride length algorithm

by using only the accelerometer Besides, instead of using the absolute heading from thecompass, a Magdwick filters [21] is applied by combining values from the accelerometerand the gyroscope to avoid the effect of magnetic fields on the magnetometer and toestimate the relative heading

• Location hybrid method a hybrid method is applied by using a particle filter for

combin-ing the WiFi, BLE estimation with the PDR and the features of the buildcombin-ing map (robot

Hình 1.4: The architecture of the robot using SLAM for disaster relief

map) This hybrid makes the indoor positioning system possible to achieve high accuracyand robustness

3.2 Simultaneous Localization and Mapping for Robot

It is clear that a robot using wheels is difficult to operate in disaster in some situations.However, in this thesis, taking into account the SLAM technique and low-cost robot, a twowheels robot is designed to implement SLAM techniques The Robot uses Hector slam tech-niques for fast online learning of occupancy grid maps requiring low computational resources.The architecture of the robot shows in Figure 1.4 The robot uses Lidar, Inertial MeasurementUnits (IMU), camera and odometry of two wheels to draw 2D map The main center processing

is used Raspberry pi 3B+ running the Robot Operating System (ROS) which is a frameworkfor Robot to implement different algorithms, for instance, differential controller, hectorSLAM,Navigation, sensor fusing using Extended Kalman filter, Filter for IMU in addition, a micro-controller is applied to implement PID velocity control for Robot as well as communicate withROS A key challenge for this scenario is that when the robot moves in different infrastructures,the odometry information may be not available Therefore, in this work, the robot is applied

hector_slamthat is an open source package using EFK SLAM in ROS to generate a two sional map with LIDAR without odometry information

dimen-4 Outline

The rest of the thesis is organized as follows:

• Chapter 2 provides the overview of related works on indoor positioning and SLAM

• Chapter 3 describes the methodology of the indoor positioning system

• Chapter 4 describes the methodology for ROBOT using SLAM

• Chapter 5 presents experiments and results of IPS

• Chapter 6 presents experiments and result of SLAM for ROBOT

• Chapter 7 concludes the thesis with a summary and future work

Chương 2

Backgrounds and Related Works

This chapter presents the general indoor positioning approaches, related works of IPS fordisaster relief and the SLAM techniques for robot

1 Indoor Positioning System

According to the suggestion by Pahlavan et al [22], a basic diagram of the indoor ing system which describes these components and their relationships in Figure 2.1 It includes anumber of location sensing devices, a positioning algorithm, a display system First, a number

position-of location sensing devices measure metrics related to the relative position position-of the transmitter orreceiver which respect to a known reference point (RP) It can use a different type of sensingtechnologies such as infrared (IR), ultrasound, WiFi, Bluetooth, Zigbee, RFID, UWB, Visiblelight The location metrics show the relationships of the direction (angle) or distance with time,phase, or received signal strength level between reference points and mobile target (MT) Thosemetrics are the angle of arrival (AOA), time of arrival (TOA), time of different arrival (TDOA),carrier signal phase of arrival (POA), or received signal strength (RSS) [16] Then, the posi-tioning algorithm process the location metrics and estimates the coordinate of MT Finally, thedisplay system converts such coordinate into the suitable format for the end user The litera-ture review diagram of IPS and corresponding related works show in Figure 2.2 and Table 2.1,respectively

Hình 2.1: A functional block diagram of positioning system

Hình 2.2: Literature review of indoor positioning system

Fingerprinting Deterministic [39, 4, 34, 40, 16, 10, 41, 42],

Prob-ability [43, 25, 44, 45, 16, 46], Pattern tion [47, 48, 49, 50, 51], Clustering [10, 52]

Recogni-Bảng 2.1: The Taxonomy and References for Literature review in Figure 2.2

Hình 2.3: A basic indoor positioning system using fingerprinting technique

1.1 General Indoor Positioning System

Wireless localization can be categorized into two main approaches: Geometrical-calculationbased and fingerprinting based (or scene-analysis based) [53] The former approach relies onmeasurement of geometrical parameters (i.e distance, angle) using various physical properties

of radio signal, such as Time of Arrival (TOA) [26], Time Difference of Arrival (TDOA) [26],Angle of Arrival (AoA) [27], Received Signal Strength (RSS) Systems that incorporate AOA,TOA or TDOA usually achieve high localization accuracy with errors lower than 1m, but theyare complicated to synchronize between transmitters and receivers [44], hence unavailable ingeneral RSS is the parameter used in most of these studies, however, achieving robustnessand accuracy is challenging for most of these studies because the radio signal is affected byreflection, refraction, shadowing and scattering in the indoor environment

Fingerprinting/Sense Analysis is a technique that estimate the position base on scene sis [25] This method can directly exploit existing infrastructures to estimate the user’s positionwith a lower cost WiFi Fingerprinting technique takes advantages of the similarities betweenthe RSSs It is usually conducted in two phases [25], [40]: offline phase or training or calibra-tion or training phase and online phase or tracking phase In Figure 2.3 the basic operation offingerprinting [25] are described In the offline phase, the ”radio map” is built with observedRSS of all the detected WiFi Signals from different access points (APs) at many reference points(RPs) of known locations and it is saved into the database In the online phase, a mobile device(or target) measures the vector RSS from available APs and estimate its position by using thefingerprinting database and positioning algorithms Based on the information contained intothe database regarding the "radio map", several research works have been proposed to solvethe indoor positioning A diagram summarizing some of these approaches is shown in Fig 2.4.Firstly, the deterministic approaches usually estimate the position based on the closest RSSs

analy-in the pre-stored radio map with real-time measurements The most typical algorithms for thisapproach are K-nearest neighborhood (KNN) [39, 4, 34], nearest-neighbour [40, 16], weight k-nearest neighborhood [10, 41], and Median Filtering [42] Euclidean distance is normally used

to measure the similarities between the observed RSSs and the mean of the fingerprints lected at each training point in deterministic approaches There are also some distances, for ex-ample, Logarithmic Gaussian Distance (LGD) [23], Penalized Logarithmic Gaussian Distance

col-Hình 2.4: Indoor positioning approaches using WiFi Fingerprinting

(PLGD) [10], Cosine distance [52], that can be used for these methods

Secondly, probability approaches have concentrated on a more precise distance measurethat can take into account the variability of the RSS training vectors These methods estimateprobability density for the training of RSS and then calculate likelihood or a posteriori esti-mates during the online phase using the observed RSSs and the estimated densities User po-sition is performed by a maximum-likelihood [43, 25, 44] or maximum a posteriori (MAP)estimation [45, 16], or histogram matching by generating fingerprinting distributions rely onradio-map fingerprints [46] Compared to deterministic approaches, they often require largercomputational resources and training sets

Thirdly, pattern recognition techniques are based on classifiers, that estimate the most likelylocation of the user’s position by using discriminating observed RSS during online phase withsurveyed fingerprinting data Support vector machine [47, 48], neural network [49, 48], deep

learning with CSI in [50], DeepFi in [51] are examples of pattern recognition schemes.

Lastly, clustering approaches are the basic idea to reduce high computation when the ber of RPs increases with the area size These algorithms reduce the search space of the userlocation to the smaller number of RPs based on the dependence of characteristics of RSS fin-gerprinting on environment features and available RPs In [10], the authors applied a K-Meanclustering approach by using Euclidean distance to find the centroid of each cluster To enhanceclassification between each cluster, He, Suining et al in [52] used cosine metric for K-MeanClustering by evaluating the similarity between two signal vectors and Cramariuc et al in [10],the authors used a penalized logarithmic Gaussian Distance approach

num-The major disadvantages of the fingerprinting method include the need for dense trainingcoverage and the poor extrapolation to areas not covered during training [32] During the of-fline phase it can be extremely time-consuming and labor-intensive to build substantially largefingerprinting databases [54, 15] In contrast to fingerprinting, "path loss" model RF signal can

be applied to compute the relationship between RSS and distance of APs and user’s position

Shchekotov, Maxim in [54] using a simple signal propagation model as equation follows:

P= P0− 10γlog(d

where P is the signal power at an RF distance d, and P0 is the know signal power at the ence distance d0 and γ is the path-loss exponent P0, d0 and γ can calculated from experimentresults Then, trilateration technique can be applied to estimate the user’s position Dao, Trung-

refer-Kien et al in [15] added some more parameters that related effects of walls and floors and used a genetic algorithm to search these parameters in the experiment Yang et al in [55] used

Linear regression and correlation constraint-based method to improve the location accuracy ing lateration methods However, the position relationship is highly complex due to multi-path,mental reflection, and interference noise [56] Thus, the path-loss model may not be adequatelycaptured by a fixed invariant model

us-Another widely-adopted localization approach is pedestrian dead reckoning (PDR) [12, 13,

28, 29], which leverages inertial sensors to measure the pedestrian displacement relative tothe previous position The main challenge in these approaches is that the inertial sensors incommercial smart-phones often suffer from imperfect calibration and noisy measurement [18].Step counting is currently a major method to capture the walking path and the movement ofpedestrians [12, 11] A number of variants on probabilistic Bayesian inference approaches haveappeared in the literature [30]-[13] which sequentially estimates the unknown state from noisyobservations using a dynamic predictive model based on pedestrian’s stride and direction andthe observation likelihood It can also provide an uncertainty measure of the estimates Kalmanfilter [30], Extended Kalman filter [31], sigma-point Kalman smoother [32, 33], Particle fil-ter [34, 13] have been applied to improve the accuracy in indoor localization systems Kalmanfiltering and its variants are the most efficient in terms of memory and computation while parti-cle filters can converge to the true posterior state distribution for non-Gaussian and multimodelcases [32]

1.2 Indoor Positioning in Disaster relief

PierfrancescoBellini et al [57] present a solution for the guiding personnel during nance and/or emergency conditions They propose integrating indoor/outdoor position and nav-igation in their system developing for hospital emergency management It’s purposes to providesupport to teams to get detail about reaching the event location; involved personnel in getting theclosest and updated exit, and registered users in reaching points of interest This indoor naviga-tion base on low-cost mobile sensor and Adaptive Extended Kalman filter In order to estimatethe current position, the mobile using the sensors to perform adjustments with respect to theposition set using a QR code then, the movements taking into account of the device’s sensorssuch as gyroscopes, magnetic compass, and accelerometers as an inertial navigation system.This system can be low cost and get a better result that compares to classical Kalman and deadreckoning The final error is lower than 20cm at the end of the path with 40m of length

mainte-Yoon, Hyungchul, et al [58] present in-building emergency response assistance system thatfocuses on getting information on the location and physical statuses of trapped victims inside abuilding during a disaster It comprises two subsystems: Victim positioning system (VPS) and

a Victim Assessment system (VAS) The VPS is developed for smart-phones using RSS of Wifisignal and Fingerprinting technique with referencing a pre-established Wifi-fingerprinting map

of the building The VAS uses patterns obtained from measured 3D acceleration changes bythe status of a victim The VPS find the locations of smart-phones inside a building following

a disaster that relies on Wifi signals from wireless access points It is further assumed thatmany wireless access points will survive a minor to moderate disaster and continuous working.The location information can be displayed on the smart-phone locally as part of a system toguide the victim to the nearest safe exit and can be transmitted for use by the on-site emergencyresponders.

The victim assessment system is designed to assess and inform the emergency responders ofthe status of the victims by collecting real-time data from sensors, such as accelerometer, gyro-scope, and a magnetic field sensor embedded in smart-phone A passive VAS recognizes eightdifferent types of activities: walking, running, standing, sitting/lying, rolling fainting, stepping

up stairs, and stepping down stairs The Naive Bayes classifier can estimate activities linked intothe four physical statues of victims (highly ambulatory, ambulatory, nonambulatory, and uncon-scious) in order to aid emergency responders in coordinating evaluation and rescue efforts.Son, Donghyun, et [4] present the indoor positioning system for emergency rescue evacua-tion support system with partial infrastructure It deploys PDR and WiFi fingerprinting based

on RSS which combine Particle filter algorithm to improve the accuracy the mobile users usesmart-phone to measure heading and distance from inertial sensors and collect RSS values fromWiFi access points Then combine the wall filter and RSS particle filter to estimate user posi-tion The simulation result shows the average error distance from 0.94 m to 2.81 m depending

on numbers of the available infrastructure of Wifi system The benefits off this system are lowcost, working in partial infrastructure

Lee, Hyo Won et al [59] propose the indoor localization scheme for disaster relief tion like rescuing people in the building It comprises two phases: In the first phase, using aprobability base on the path loss model to estimate the subarea where the victim nodes located

applica-In the second phase (or online phase), mobile nodes corresponding to rescuers go around thebuilding utilizing pedestrian dead reckoning technique (PDR) using internal sensor measure-ments of smart-phone to track its position and send a signal to stationary nodes (victims) nearits node When station nodes receive a signal from mobile nodes, they report RSS to the mo-bile node Then, the rescuers estimate the location of the stationary node base on a path lossmodel and a subarea database constructed in the offline phase The stationary node on eachsubarea is determined by localization server estimating the probabilities based on the path lossmodel This scheme can reduce the cost of the off-line phase compared with fingerprinting-based method because it does not measure RSS in that phase another advantage of this is that

it use communication between devices instead of pre-installed facilities

2 SLAM for Robot

2.1 Simultaneous Localization and Mapping Problem

SLAM is an abbreviation for simultaneous localization and mapping also know as rent Mapping and Localization (CML), which is a technique for estimating sensor motion andreconstruction structure in an unknown environment [24] Slam dealing with the necessity ofbuilding a map of the environment while simultaneously determining the location of the robotwithin this map [60] The accuracy of the map depends on the accuracy of localization andvice versa This technique was originally proposed to achieve autonomous control of robots inrobotics [24]

Concur-Consider a robot moving in an unknown environments as depicted in Figure 2.5 with:

• The robot is given: the robot’s controls ukto drive the robot at time k, and the Observation

Hình 2.5: A Robot in an unknown environment in [61]: Landmarks being observed at differentposition along the robot’s trajectory

Hình 2.6: A literature review for SLAM

zik taken from the robot of the location of the ith landmark at time k

• It wants to have: map of the environment mi describing the location of the ith landmarkwhich is time invariant, and the state vector xk describing the location and orientation ofthe vehicle (the path of the robot)

SLAM consists of multiple parts: Landmark extraction, data association, state estimation,state update and landmark update [62] The literature review for SLAM shows in Figure 2.6.The most common sensors using for SLAM can be categorized into laser-based, sonar-based,vision-based systems Besides that, some RF technologies are also used for SLAM such asWiFi [63], UWB [64], RFID [65] The Solution to the SLAM problem can be fallen into threemain categories: Kalman Filters (KF) based, Particle based and graph-based SLAM Kalmanfilters are Bayes filters that represent posteriors using Gaussians [60], i.e unimodal, multivari-ate distributions KF SLAM relied on the assumption that the observations and the state tran-sition functions are linear with added Gaussian noise, the initial posteriors are also Gaussian.There are some main variations of KF that commonly use for SLAM: the Extended KalmanFilter (EKF) [66], Extended Information Filter (EIF) [67], Sparse Extended Information Filter(SEIF) [68], Unscented Kalman Filter (UKF) [69] The main drawbacks of EKF and KF im-plementation are that high computational complexity and large linearization error [70] ParticleFilter (PF) is another implementation of a Bayes filter In contrast to the KF based, it does not use

a parametric model for the probability distributions which makes it capable of handling highlynonlinear sensors and non-Gaussian noise The PF can combine with other techniques to deal

with SLAM problem, for example, FastSLAM [71] and the fastSLAM2.0 [72], unscented ticle filter (UFastSLAM) [73] FastSLAM relaxes the limitation of Gaussian distribution noiseand divides the SLAM problem into two parts: robot path and landmark estimation A maincharacteristic of fastSLAM is that each particle makes its own local data association so thecomputational complexity and memory usage are improved Another wide Solution for SLAM

par-is graph-based SLAM [74] which represents robot positions and observations as nodes and surement constraints as edges [61] This algorithm solves the full SLAM problem that the entiremap and path are recovered, instead of just recent pose and map in online SLAM [75] Thisdifferent allows considering dependencies between previous and current poses Graph-basedSLAM can work with all of the data at once to find the optimal SLAM solution

mea-In ROS, there are some packages that are implemented 2D mapping using LIDAR such

as hector_slam (2011) use EFK SLAM, Gmapping(2007) using fastSLAM, and Karto SLAM (2010) using Graph-based SLAM [76], ETHZASL-ICP-Mapper(2013) hector_slam does not

detect loop closures but it can work without odometry information In disaster scenarios, theloop-closures is not really important during the short period of SAR Therefore, it can be apromising solution for this work

2.2 SLAM for Disaster Relief

Alexander Kleiner in [65] propose the novel method for realtime exploration and SLAMbased on RFID tags for Robot search and rescue The tags are autonomously distributed in theenvironments Their approach allows the computationally efficient construction of a map withinharsh environments, and coordinated exploration of a team of robots [65]

In [77], the authors use hector_slam that is an Open Source Modules for Autonomous

Map-ping and Navigation with Rescue Robots This robot using ROS solves SLAM problem to ate sufficiently accurate metric maps useful for navigation of first responders or a robot system.This system also solve unreliable odometry information in SAR by using purely relying on fastLIDAR data scan matching at full LIDAR update rate

gener-In Uban Search and Rescue (USAR), SLAM using different perception techniques can beapplied for robots to generate 3D maps and localize themselves In [67], an extended informationfilter (EIF) based SLAM algorithm was proposed for building dense 3D maps in indoor USARenvironments via the use of mobile rescue robots The data association is performed using acombination of scale invariant feature transformation (SIFT) feature detection and matching.The SLAM techniques using a camera to for robot move in 6D and no odometry information

In [66], The EKF SLAM techniques using laser range finders is proposed for constructing a3D map of rubble by teleoperated mobile robots In [78] propose an online multi-robot SLAMsystem for 3D LIDARs for disaster scenario

Chương 3

Methods for Indoor Positioning of Human

In this thesis, an indoor positioning system is considered to work in different environments.With a structured environment, a combining WiFi fingerprinting, BLE based path-loss and PDRtechniques can easy to deploy in all type of building and smart-phone Furthermore, this fusingcan get high accuracy in normal conditions In the case of disaster, partially access points ofthe WiFi network may not work or electrical system even does not work The IPS can worklocally based on PDR techniques using inertial sensors of smart-phone and the measurementsfrom low-power BLE iBeacons This chapter describes the detail of the different localizationtechniques for this framework and the proposed framework for indoor positioning system fordisaster relief

1 RSS-Based and Path-loss Model

This technique deploys the relationship between RSS and distance of transmitters and ceivers In real environments, however, the wireless radio channel is affected by noise, interfer-ence, and other channel impediments The basic of Attenuation model of RSS has been mention

re-in most of research [15, 54, 9] The relationship between RSS and transmission distance based

on commonly used logarithmic attenuation model is shows:

P(dt) = P(d0) − 10γlog10(dt

d0) + Xσ (3.1)

where P(d) is the received signal strength of receiving, P(d0) is the free space propagationloss at reference distance d0 (typically assumed to be 1 m), and γ is the path loss exponent (orattenuation factor for RSS)

The Figure 3.2 describes the histogram of RSS from two different training data at 1 meterand 2 meters of three WiFi access points The values of the different beacon in light of sightcondition are heterogeneous Therefore in this thesis, we using fingerprinting method for WiFiusing pre-training data

Hình 3.1: The relationship between RSS and distance using path-loss model

1.2 iBeacon based Localization

The iBeacon technology is established upon BLE4.0, thus, it is very energy efficient and can

be utilized for localization based on RSS for BLE devices, i.e a smart-phone [38] An iBeaconwill periodically broadcast an advertisement packet containing a unique ID and a calibratedRSS value at a one-meter distance This value allows us to determine the distance between aniBeacon and a device Note that no paired connections are required for receiving these packets.More details can be found in [79] The main advantage of the iBeacon is that it has very longbattery life, for example, Estimote iBeacon uses four CR2477 batteries and Estimote Locationenabled by default This allows for up to 5 years of battery life [80]

The iBeacon based localization leverages the RSS of BLE The signal propagation of aniBeacon can be formulated using equation 3.1 Therefore, we can obtain distance of dt as

dt = 10

P (d0)−P(dt )

Based on the estimated distances between iBeacon and a device, the lateration techniques can

be applied for localization In this work, these distances are used to fusing with PDR techniques

By this fusing, the number of iBeacon can be reduced However, RSS values are affected by theenvironment and have, consequently, high levels of noise In order to deal with these noises AKalman filter can be applied to filter RSS measurements [81, 82] RSS values change randomly,therefore, the transition matrix F and the measurement matrix H are set to one Moreover, there

is no external control input Donate rBLEt and zBLEt are estimated RSS and raw RSS at time t,respectively; Q is noise measurement, R is noise process With these assumptions, the predictionand update phase can be shown as:

(a) 1 meter (b) 2 meter

Hình 3.2: Histogram of RSS from two different training data of 3 beacons (access points)

• Prediction Phase:

ˆrtBLE= rt−1BLE,ˆ

Pt= Pt+ Rt (3.3)

• Update Phase:

Kt= Pˆtˆ

Hình 3.3: Raw, and filtered RSS at distance 1m from an experiment of two BLE beacons (R =0.001and Q = 1)

Hình 3.4: A Gaussian process (GP) regression for an indoor positioning system.

2 WiFi Fingerprinting using Gaussian Process Regression

WiFi fingerprinting includes two phases: In the offline phase, WiFi RSS are collected tothe database by smartphones or robot at specific points to build "WiFi map" On the onlinephase, the real-time RSS measurements are compared with the "WiFi map" to estimate theuser position This section presents two phases of WiFi techniques to build "WiFi map" andAlgorithm to estimate position from this map

2.1 Offline phase : Building WiFi Fingerprinting Maps Using a Gaussian Process Regression

Fingerprinting techniques require high-density training data to get high accuracy However,the data collection is labour intensive In this proposed framework, a GP regression is used

to minimise the training time as well as for improving the effectiveness on WiFi ing [17] GP has many advantages that make it applicable for indoor positioning systems usingWiFi RSS [17, 83, 84] It is non-parametric, continuous and correctly handling uncertainty inboth process and estimation [85] GP is especially useful because of the noisy RSS WiFi mea-surements due to various phenomena such as reflection, scatting and diffraction

fingerprint-To generate a WiFi map using GP regression for indoor positioning system from the trainingdata, the GP relies on a covariance function kernel that establishes the correlation of values atdifferent points, as shown in Figure 3.4 The conjugate gradient descent method is utilised tooptimise the hyperparameters of the function kernel Finally, the GP generates prediction points

by estimating the posterior distribution of the WiFi map in an interested space

Assuming that r = {ri, i = 1, n} is the observed RSS vector that includes n received accesspoints (AP) at corresponding coordinate points in d dimension x = {xi, i = 1, n}, xi∈ Rd,

so that the pair (xi, ri) represents the training data Each observation ri can be related to a

transformation f (xi) through a Gaussian noise model from a noisy process as:

ri= f (xi) + ε, (3.5)where {ε} is the generated measurement noise from a Gaussian distribution with zero meanand variance σi2 Any two output values, rpand rqare assumed to be correlated by a covariancefunction based on their input values xpand xq:

cov(rp, rq) = k(xp, xq) + σn2δpq, (3.6)where k(xp, xq) is a kernel, σn2 is the variance , δpq is 1 if p = q and 0 otherwise The kernelfunction considered in this work is is squared exponential kernel as equation:

p(r∗|x∗, x, r) ∼N (µr ∗, σr2∗), (3.9)

µz∗ = kT∗(K + σn2In)−1z,

σz2∗ = k∗∗− kT∗(K + σn2In)−1k∗ (3.10)where k∗∗= cov(x∗, x∗) is the vector variance of generated points x∗and k∗= cov(x∗, x) isthe vector of covariance between x∗and training points x In this work, the WiFi map is built up

by predicted points spaced 1m×1m apart

An example for generated WiFi map for one access point is show in the Figure 3.5a andFigure 3.5b, the mean generate for signal strength and the variance is higher in areas far fromtraining points

∂

∂ θjlog p(y|x, θ ) = 1

2tr

(K−1y)(K−1y)T ∂ K

∂ θj



(3.12)

(a) Predicted mean (b) Predicted variance

Hình 3.5: An example of generated WiFi map of one access point

An experiment of GP for hyper-parameter estimation shows in Figure 3.6b The length scale

lcharacterizes how smooth of the predicted mean is The σ2f represent the WiFi signal variance.Smaller the σ2f, slower the variation of the function

2.2 Online phase - WiFi position estimation

On the online phase, the WiFi fingerprinting position is estimated by measuring the bilities of new visible RSS in generated training points (WiFi map) All the steps of this algo-rithm are as follows:

proba-• Step 1: for each access point l, the likelihood is computed as:

ψm=

L

∑

l=1log(p(rl|x∗)) (3.15)

• Step 3: Sort the weights, get K nearest weights (ψk, k = [1, K], K ≤ M), then they arenormalised by:

where xmis the K nearest predicted training points

3 Pedestrian Dead Reckoning

PDR is the technique that uses Inertial Measuring Units (IMU) to estimate the movement

of a person by detecting steps, estimating stride lengths and the directions of motions An IMU

is integrated into most of the smart-phone and provides triaxial orthogonal accelerometers,gyroscopes, magnetometers, and even pressure sensors PDR determines the next position usingthe previous position, step length and walking direction, which is expressed as follows:

3.1 Step Detection using Accelerometer

Steps can be detected by measurements from accelerometers [30] or gyroscopes [12] In[36], an accelerometer is used by Normalised Auto-correlation based Step Counting In thispaper, we use an accelerometer based on the technique in [36, 11] The algorithm for stepdetection consists of the following steps:

• Step 1 Calculate the magnitude for the normalisation factor of every sample i:

i+w

∑

j=i(aj− ¯aj), (3.20)

where ¯ajis a local mean acceleration value, computed by this expression: ¯aj= w1∑iq=iaq,and w defines the size of the averaging window (w = 50 samples)

• Step 3 To discriminate between the state of walking and the state of standing, the

devi-ation is gained by 10 times using the accumulated window (wg= 10), according to thefollowing equation:

σd=

i+w g

∑

j=i(σaj), (3.21)

where σdis the deviation for the step detection

• Step 4 Thresholding: a first threshold is applied to detect the step with high accelerations,

the value can be calibrated by the user at T (m/s2) If σdi > T and previous value σdi−1 <

T, the steps will increase by one

Figure 3.7 shows the magnitude of the accelerator deviation before and after using the mulated window of this algorithm from an experiment The threshold can be calibrated by theusers on Android application to fit with these smart-phone In our experiments, we set a thresh-old equal to 4

accu-(a) Raw data of the accelerometer.

(b) The step detection from the accelerometer with the proposed

algorithm

Hình 3.7: Step detection using the accelerometer

3.2 Stride Length Estimation

Stride Length (SL) at every detected step is necessary in order to calculate the travelleddistance by the person while walking The SL can be approximated as a constant value [34],however, it varies significantly depending on the person and according to different parameters,such as the length of the legs, walking speed and frequency In this work, we use the algorithmproposed by Weinberg that archive high accuracy by using an accelerometer with the PDRtechnique [11] for dynamic walking The Weinberg algorithm is as follows:

• Compute the magnitude of accelerations, ai, as in eq 3.19

• Low-Pass filter this signal (aei= LP(a)i) We use a filter of order 4 and cut-off frequency

at 3 Hz

• Estimate the SL using the Weiberg expression:

L= Kq4

e

ajmax−aejmin, (3.22)whereaejmax andaejmin are the maximum and minimum acceleration values after the low-passfilter, respectively K is the coefficient that needs to be selected experimentally or calibrated.This approach takes the dynamics of step length during walking into consideration

3.3 Heading Estimation

To track the user’s path in PDR, the most crucial factor is the pedestrian walking direction

It can be estimated from orientation sensors, such as magnetometers and gyroscopes The pass calculates the phone orientation relative to the perceived magnetic north [36] However,the magnetometer will be affected by noise Therefore, an accelerometer is also used as an alter-native or in combination with the magnetometer to improve accuracy [11] By fusing the outputfrom these sensors, the heading can be accurately estimated by different algorithms, such as anKalman filter [30], Complimentary filter, Madgwick filter, or Mahony filter [86] It is important

com-to have a reference for the heading direction for initialisation Since it would be inconvenientfor a user to start with a specific direction [30], this in work, the relative direction is calculatedusing an accelerometer and a gyroscope with a Madgwick filter [86, 21] It is not necessary tomeasure the absolute heading since an accurate heading-change estimate can be determined byusing the particle filter This is enough to guide the particles to propagate in the right direction.Madgwick filter uses a quaternion representation, allowing accelerometer and magnetome-ter data to be used in an analytically derived and optimised gradient-descent algorithm to com-pute the direction of the gyroscope measurement error as a quaternion derivative The benefit offilter include: (1) computational inexpensive, (2) effective at low sample rate and, (3) contains

1 (IMU) or 2 (MARG) adjustable parameters defined by observable system characteristics [87].The details of this filter is presented in []

4 Fusing Algorithm with Particle Filter

Only WiFi fingerprinting cannot achieve high accuracy due to the effects discussed earlier

in indoor environments Moreover, the speed rate to get RSS measurements have significantlatency In our experiments, the RSS is approximately updated every two seconds This meansthat the indoor positioning system using only WiFi can experience a low response in real-timenavigation if pedestrians are moving fastly On the contrary, PDR can provide a high positionaccuracy in a short range, then slowly drifts walking distance In this work, a particle filter

is utilised to combine PDR with WiFi fingerprinting and BLE as shown in Figure 3.8 Theparticle filter is based on a set of randomly weighted samples (i.e., the particles) representingthe density function of the user’s position Each particle explores the environment according

to the motion model of the PDR These weights are updated at each step once a new positionfrom WiFi fingerprinting is estimated or when BLE measurements are updated It is possible toconstrain some moves like crossing the walls from a building map by forcing the weight at 0 forthe particles having such a behaviour

Assume that the state of particle ithat step k as xik= [xik, yik, θki] has weight wik The Particlefilter algorithm to combine WiFi fingerprinting and PDR is as follows:

• Step 1 Initialisation : set k=0, generate randomly N position and heading particles xi

0and an equal weight wi0