eileen su lee ming - eprints.utm.myeprints.utm.my/id/eprint/5145/1/eileensuleemingmfke2006.pdf ·...

DEVELOPMENT OF A ROOM RECOGNITION SYSTEM USING A

CATADIOPTRIC SENSOR AND ARTIFICIAL NEURAL NETWORK

EILEEN SU LEE MING

UNIVERSITI TEKNOLOGI MALAYSIA

DEVELOPMENT OF A ROOM RECOGNITION SYSTEM USING A

CATADIOPTRIC SENSOR AND ARTIFICIAL NEURAL NETWORK

EILEEN SU LEE MING

A thesis submitted in fulfilment of the

requirements for the award of the degree of

Master of Engineering (Electrical)

Faculty of Electrical Engineering

Universiti Teknologi Malaysia

JANUARY 2006

iii

DEDICATION

Dedicated to my beloved family and my dearest, Che Fai.

iv

ACKNOWLEDGEMENT

First and foremost, my most sincere appreciation to my project supervisor,

Professor Dr. Shamsudin Hj. Mohd. Amin, who sets a noble example of a devoted

educator. His encouragement and constant support will always be treasured.

A big thank you to my co-supervisor, Associate Professor Dr. Rosbi Mamat, for

his kind guidance and invaluable advice.

My deepest gratitude to Yeong Che Fai, for his continuous motivation, technical

assistance and most importantly, his insightful counsel throughout the entire period of

this project. I can never thank him enough for the inspiration, for his patience and for

always being there as my pillar of strength.

To my dearest family members for their love, care and prayers. Thank you for

persevering with me.

To everyone who assisted me in one way or another, be it through advice or

words of encouragement - my dedicated friends, thanks for making this possible.

A special thanks to Universiti Teknologi Malaysia, especially the Department of

Mechatronics and Robotics, for providing financial support and excellent facilities to

complete my research.

Above all, I thank God for everything that He has given and planned for me.

v

ABSTRACT

Robot localization has been a challenging issue in robot navigation. In recent

years, there has been increasing interest in topological localization. One popular

approach for vision based topological localization is the appearance based method,

where the image can be used for recognition in its basic form without extracting local

features. The general aim of this work is to develop a room recognition system using

appearance-based method for topological localization. In this work, the room

recognition is achieved by matching color histogram of image using the Artificial

Neural Network. A hardware module and a software module have been developed

for this project. The hardware module consists of a catadioptric sensor system

implemented on a mobile platform. The software module encompasses several sub

modules namely image acquisition; image pre-processing; histogram plotting;

histogram filtering, sampling and normalization; neural network for offline training

and testing, and finally real time room recognition. A few experiments have been

conducted to evaluate the performance of the system and the results have been

favorable. Testing for suitable network setting was also carried out and a

recommendable setting was proposed.

vi

ABSTRAK

Lokalisasi robot merupakan satu isu yang mencabar dalam navigasi robot.

Sejak kebelakangan ini, terdapat perhatian yang meningkat terhadap lokalisasi secara

topologikal. Salah satu kaedah yang popular bagi lokalisasi topologikal

menggunakan ‘penglihatan’ adalah melalui pendekatan berasaskan penampilan, di

mana imej dapat digunakan untuk pengenalan dalam bentuk asas tanpa mengekstrak

ciri-ciri tempatan. Objektif keseluruhan projek ini adalah untuk menghasilkan satu

sistem pengenalan bilik yang menggunakan kaedah berasaskan penampilan bagi

lokalisasi topologikal. Dalam projek ini, pengenalan bilik dicapai dengan

memadankan histogram warna menggunakan Jaringan Neural Buatan (Artificial

Neural Network). Satu modul perkakasan dan satu modul perisian telah dihasilkan

bagi projek ini. Modul perkakasan merangkumi sistem penderia catadioptric yang

diimplementasikan atas platform bergerak. Modul perisian pula merangkumi

beberapa sub-modul iaitu pengambilan imej; pra-pemprosesan imej; pemplotan

histogram; penapisan histogram; pensampelan; normalisasi; jaringan neural untuk

latihan dan pengujian secara offline; dan akhir sekali, pengenalan bilik secara nyata.

Beberapa eksperimen telah dijalankan untuk mengevaluasi pencapaian sistem ini dan

keputusan yang diperoleh adalah memuaskan. Pengujian turut dilaksanakan untuk

mendapatkan aturan yang bersesuaian dan satu aturan saranan telah dicadangkan.

vii

TABLE OF CONTENT

CHAPTER TITLE PAGE

DECLARATION

DEDICATION

ACKNOWLEDGEMENT

ABSTRACT

ABSTRAK

CONTENTS

LIST OF TABLES

LIST OF FIGURES

LIST OF ABBREVIATIONS

ii

iii

iv

v

vi

vii

xiii

xiv

xviii

1 INTRODUCTION 1

1.1 Challenges in Mobile Robot Navigation

1.2 An Overview of Robot Localization

1.3 Problem Background

1.4 Project Objectives

1.5 Outline of the Proposed Approach

1.6 Outline of Thesis

1

2

3

5

5

7

viii

2 LITERATURE REVIEW ON ROBOT

LOCALIZATION 9

2.1 Types of localization

2.1.1 Position tracking or local techniques

2.1.2 Global localization

2.1.3 Place Recognition

2.2 Reviews on Sensors and Techniques

2.2.1 Odometry

2.2.2 Inertial Navigation

2.2.3 Magnetic Compasses

2.2.4 Active Beacons

2.2.5 Global Positioning System

2.2.6 Landmarks

2.2.7 Map Based Positioning

2.3 Vision Based Localization

2.3.1 Appearance Based Methods

2.3.2 Related Works

2.4 Features of the Proposed Technique

2.5 Summary

9

10

12

12

13

14

14

14

15

15

16

17

18

19

20

22

22

3 SYSTEM DEVELOPMENT METHODOLOGY AND

DESIGN

24

3.1 Design Methodology

3.2 Hardware requirement and specifications

3.3 Software Architecture

3.3.1 Image Acquisition

3.3.2 Pre-processing

3.3.3 Image Classification

25

27

28

29

30

30

ix

3.3.4 Data Retrieval and Matching

3.4 System Integration

3.5 Performance Analysis

3.6 Summary

30

31

31

32

4 APPEARANCE BASED ROOM RECOGNITION

SYSTEM DEVELOPMENT

33

4.1 Input Images

4.2 Color Extraction for Appearance Based

Room Recognition

4.2.1 RGB color space

4.2.2 RGB Components

4.3 Plotting Histogram

4.4 Histogram for Single Color Component

Image

4.5 Filtering

4.6 Rotation Invariant Property of Color

Histogram

4.7 Histogram Patterns To Represent Rooms

4.8 Data Sampling

4.9 Normalization

4.10 Artificial Neural Network for Room

Recognition

4.10.1 Multi Layer Perceptron

4.10.2 MLP Network Architecture

4.10.3 Activation Function

4.10.4 Learning Algorithm

4.11 Summary

34

35

35

36

43

45

50

52

54

55

56

57

58

58

60

61

67

x

5 SYSTEM IMPLEMENTATION

68

5.1 Hardware Implementation

5.1.1 The Mobile Robot

5.1.2 Main Controller

5.1.3 Catadioptric Sensor

5.1.4 Design Specifications

5.2 Software Implementation

5.2.1 Image Acquisition

5.2.2 Image Conversion To Single R,G, or B Color

Component

5.2.3 Color Histogram

5.2.4 Single Color Component Histogram

5.2.5 Histogram Filtering, Sampling and

Normalization

5.2.6 Artificial Neural Network

5.3 Summary

68

69

70

71

76

80

82

84

85

86

87

90

99

6 EXPERIMENTAL RESULTS AND DISCUSSION 100

6.1 Objectives of Experiment

6.1.1 Performance Comparison Based On

Different Color Component For Appearance

Based Room Recognition


Different Sampling Rates For Appearance



Different Normalization Denominators For

101

101

101

xi

Appearance Based Room Recognition


Different Network Settings For Appearance


6.1.5 Real Time Recognition Performance

6.2 Experimental Setup

6.2.1 Experimental field

6.2.2 The Mobile Platform

6.2.3 Datasets

6.3 Experiment Procedure

6.3.1 Experiment 1: Performance Comparison

Based On Different Color Component For



Based On Different Sampling Rates For



Based On Different Normalization

Denominators For Appearance Based Room

Recognition


Based On Different Network Settings For


6.3.5 Experiment 5: Real Time Recognition

Performance

6.4 Experimental Results


Based On Different Color Component For


102

102

102

102

103

104

104

106

107

108

110

113

114

116

117

xii


Based On Different Sampling Rates For



Based On Different Normalization

Denominators For Appearance Based Room

Recognition


Based On Different Network Settings For


6.4.5 Experiment 5: Real Time Recognition

Performance

6.5 Summary

119

121

124

126

130

7 CONCLUSION 132

7.1 Research Summary

7.2 Contributions

7.3 Limitations and Future Work

132

135

136

REFERENCES

138

xiii

LIST OF TABLES

TABLE NO. TITLE PAGE

5.1

6.1

6.2

6.3

6.4

6.5

6.6

6.7

Software environments for the system

Performance comparison based on different color

components

Performance comparison based on different sampling rates

Performance comparison of different normalization

denominators

Performance comparison based on different network

settings

Results for random real time testing

Detailed result for continuous real time testing

Result summary for continuous real time testing

81

117

119

122

124

127

129

130

xiv

LIST OF FIGURES

FIGURE NO. TITLE PAGE

1.1

1.2

3.1

3.2

3.3

Topological map of the indoor environment

Sample image of fish-eye field of view

Design Process of a Room Recognition System

Hardware requirements for the room recognition system

Software architecture for overall system

6

7

26

28

29

4.1

4.2

4.3

4.4

4.5

4.6

4.7

4.8

4.9

4.10

4.11

4.12

4.13

Overview of the room recognition development process

A Catadioptric Image of Room A

Converting from RGB color values to single color

component values

Flowchart showing how image in RGB color space is

converted into single color component image

Sample image of (a) original RGB color space (b) Red

color component (c) Blue color component (d) Green color

component

A 100-pixel image with different colors

The color histogram for image in Figure 4.6

The image in Figure 4.6 with only the R color component

Color intensity histogram for image in Figure 4.8

(a) A room image (b) The image with only its Red

component and (c) its corresponding color histogram

Flowchart showing the process of plotting the color

intensity histogram after the image has been reduced to an

8-bit color space.

A typical room image

A spike at the left side in all the histograms plotted for the

34

35

41

42

43

44

45

46

47

48

49

50

xv

4.14

4.15

4.16

4.17

4.18

4.19

4.20

4.21

4.22

4.23

4.24

room images

Thresholding histogram at a certain color intensity

(a) Before Thresholding (b) After thresholding

The rotation invariant property of a color histogram

Similar histogram pattern for Room A is obtained for

images taken in the same room irrespective of camera

location or image rotation.

Histogram for three different room images

Histogram patterns that represent different rooms

Histogram patterns sampled at different rates (a) at 200

bins per image (b) 100 bins per image (c) 50 bins per

image (d) 25 bins per image

Color histogram (a) before normalization and (b) after

normalization

A two-layer Multi Layer Perceptron

Binary sigmoid graph, range (0,1)

Local minimum and global minimum

Flowchart of the Backpropagation learning algorithm

50

51

52

53

54

54

55

56

59

60

62

66

5.1

5.2

5.3

5.4

5.5

5.6

5.7

5.8

5.9

5.10

5.11

BeMR : The robot used as test platform for the vision

system

Different catadioptric sensor designs

Catadioptric sensor design

Image taken with the catadioptric sensor

The Logitech Quick Cam Pro after modifications

Shiny ornament as reflector ball

The black cardboard used as shade

The sequence of mounting the reflector ball, webcam and

shade onto the aluminium rod and then mounting the rod

onto the BeMR.

The catadioptric sensor mounted onto the BeMR

Recommended distance between reflector ball and floor

(a) Camera fixed more than 3cm away from the reflector

(b) Image taken when the camera is more than 3cm away.

70

71

72

72

73

74

74

75

76

77

78

xvi

5.12

5.13

5.14

5.15

5.16

5.17

5.18

5.19

5.20

5.21

5.22

5.23

5.24

5.25

5.26

5.27

5.28

5.29

5.30

5.31

5.32

5.33

5.34

5.35

5.36

(a) Camera fixed less than 3cm away from the reflector

(b) Image taken when the camera is less than 3cm away.

(a) Camera fixed at 3cm away from the reflector (b) Image

taken when the camera is at the 3cm position.

Images taken without shade

Images taken with shade

Sub modules in the appearance based recognition software

GUI for the Save Picture algorithm

Original images are saved into folders

The last image saved can also be displayed

Original image conversion to single color component

image

Room A data for three different color components

Histogram of different color components for a room

Histograms plotted for different rooms

A histogram (a) before filtering and (b) after filtering

GUI for filtering, sampling and normalization

Histogram values saved as text files after filtering,

sampling and normalization

Histogram values (a) original (b) after filtering (c) after

sampling and normalization

An example shape of a network using Method A

An example shape of a network using Method B

An example shape of a network using Method C

GUI with function for network training, save net and load

net

Network is saved into their respective folders

Network can be retrieved using the ‘Load Net’ function

The network information available from the files

GUI with function for single image testing and result

display

GUI with automatic testing to recognize Training and Test

Set images

78

79

80

80

81

82

83

83

84

85

86

86

87

88

89

89

90

91

91

92

93

93

94

95

96

xvii

5.37 GUI for Real Time Room Recognition 98

6.1

6.2

6.3

6.4

6.5

6.6

6.7

6.8

6.9

6.10

Layout of the experiment field

Random path taken when capturing training images

Random path taken when capturing testing images

Illustration of the random locations for real time

recognition

Illustration of the locations for robot continuous real time

recognition

Number of correctly recognized images using inputs of

different color components.

Number of correctly recognized images using inputs

sampled with different sampling rates

Number of correctly recognized images using inputs

normalized with different denominator values

Number of correctly recognized images using different

network settings

Number of correct recognitions during random real time

testing

103

105

106

115

116

118

120

123

125

128

xviii

LIST OF ABBREVIATIONS

2D - Two dimensional

3D - Three dimensional

ANN - Artificial Neural Network

B - Blue

BeMR - Bluetooth enabled Mobile Robot

BP - Back Propagation

CAD - Computer Aided Design

CCD - Charge-coupled Device

DGPS - Differential Global Positioning System

G - Green

GPS - Global Positioning System

GUI - Graphical User Interface

MLP - Multi Layer Perceptron

MSE - Mean Square Error

NASA - National Aeronautics and Space Administrations

PCA - Principal Components Analysis

PDA - Personal Digital Assistant

R - Red

RGB - Red, Green, Blue

SA - Selective Availability

ZPR - Zero Phase Representation

1

CHAPTER 1

INTRODUCTION

Autonomous mobile robots designed to move freely in the world have the

same problems as humans when navigating. The world is a complex environment

and if robots only move around without ‘looking’ at where their action takes them,

they might get lost due to the imperfections in their moving mechanisms and the

environment. The research here therefore, focuses on a topological localization

strategy that employs vision and neural network for the robot to ‘see’ its

surroundings and then estimate its own location.

1.1 Challenges in Mobile Robot Navigation

In order for a mobile robot to perform its assigned tasks, it often requires a

representation of its environment, a knowledge of how to navigate in its

environment, and a method for determining its position in the environment. These

problems have been characterized by the three fundamental questions of mobile

robotics, which are “Where am I?”, “Where am I going?” and “How can I get there?’

(Leonard and Durrant-Whyte, 1991).

The first question is one of localization. The robot has to know where it is in

a given environment based on what it sees and what it was previously told. The

second and third questions are essentially those of specifying a goal and being able to

plan a path to achieve that goal. Therefore, finding a robust and reliable solution to

2

the first question of localization is a precursor to answering the remaining two

questions. This can be related with the example of going into a bookshop. You will

firstly need to know where you are standing (localization) and where is your

destination relative to your current position (identifying goal) so that you can plan

which direction you should take in order to reach your destination (path planning).

This is why many researchers (Duckett and Nehmzow , 2001; Cox and Wilfong,

1990; Fox, 1998; Borenstein et al.,1996) maintained that localization is one of the

most fundamental problems in mobile robotics.

1.2 An Overview of Robot Localization

Robot localization is the problem of estimating a robot’s pose relative to a

map of its environment (Fox et al., 1999). A wide variety of localization methods

have been proposed and a number of successful laboratory prototypes have been

developed. Some of these systems have been validated in larger environments,

generally consisting of enclosed areas within public buildings (Yamauchi and

Langley, 1997; Weiss and von Puttkamer, 1995; Burgard et al., 1998; Thrun et al.,

2000; Duckett and Nehmzow, 2000) and some attempts have been made for

localization in outdoor environments (Kweon and Kanade, 1991; Takeuchi and

Herbert 1998).

The localization issue can usually be categorized as being geometric or

topological (Andreasson and Duckett, 2004, Ulrich and Nourbakhsh, 2000).

Geometric approaches attempt to estimate the position of the robot as accurately as

possible (x, y, θ) with respect to the map’s coordinate system. Topological

localization gives a more abstract position estimate, for example “This is the coffee

room”.

The task of localization can again be divided into two sub-problems: position

tracking and global localization. In position tracking, a robot knows its initial

position and only has to accommodate small errors in its odometry as it moves (Fox

3

et al., 1999). The global localization is the ability to estimate the position of the robot

without knowledge of its initial location and the ability to relocalize if its position is

lost (Fox, 1998). Global localization hence, has to solve a much more difficult

localization problem, that of estimating its position from scratch. This includes the

kidnapped robot problem where the robot is all of a sudden transferred or

‘kidnapped’ to another location without the robot being aware of this.

Some researchers, however, categorize the localization tasks using different

terms. Yamauchi et al. (1998) divided the localization tasks to that of continuous

localization and place recognition. Continuous localization is like driving downtown

without getting lost which is similar to position tracking. In contrast, place

recognition is like waking up in a hotel room and trying to determine which city one

is in. In place recognition approaches, accurate coordinates are not needed and thus,

place recognition approaches are normally used in topological localization.

In determining its location, a robot needs access to two kinds of information.

First is the information or map, either gathered by the robot itself or supplied by an

external source during the training or initialization phase. This map specifies certain

features of the environment that are time-invariant and thus can be used to determine

a location. The second kind of information is the navigational information which the

robot gathers from its sensors during navigation. Generally, when a map or

information of the environment is available, the robot position is computed thanks to

a matching technique applied between currently observed part of the environment

and the global map.

1.3 Problem Background

Over the past few years, there has been tremendous scientific interest in

algorithms for estimating a robot’s location from sensor data. Many different

approaches have been introduced to handle various challenges in the robot

4

localization problem. A few challenges have been identified as dynamism of

environment, noise and errors, computational cost and ease of use.

A robot may be confronted with the problem of dynamic environment. Due to

the changes of furniture arrangement, the environment may look different from the

representation of a robot’s initial map. This raises the question of how to make a

robot localization method robust against such dynamic effects.

A general factor that complicates the robot localization is the existence of

noise and errors, particularly in the sensor readings. To perceive changes in the

environment, the robot has to sense repeatedly and often. However, knowledge

gained via sensing is incomplete, inaccurate and uncertain. One example is the use of

odometry sensors which count the revolutions that the wheels make while moving

and turning. The readings can be used to help in estimating the displacement over the

floor to give an indication of the location of the robot. Due to wheel slippage and

irregularities of the floor texture, the odometer readings may give inaccurate results.

With more revolutions, the cumulative error increases.

Another issue in localization system for robot is computational power. For a

robot to accurately determine its location, a detailed metric map will be a good input.

However, this type of maps requires extremely large memory and for the matching

algorithms to quickly determine a robot’s location in such a detailed map, a very fast

processing is needed. The computational cost is even higher if the robot is to do real

time image processing. Often, a two-dimensional map is used to avoid the

computation and space explosion that a three-dimensional representation may entail

(Kaelbling et al., 1998; Nourbakhsh et al., 1995; Schultz et al., 1999; Simmons and

Koenig, 1995). Statistical sampling is also used to alleviate this computational

burden at the cost of completeness (Dellaert et al., 1999).

Some approaches for robot localization face the problems in terms of ease of

use when a robot is transferred to a new domain. It will require a period of

initialization or retraining. Methods of localization which involve artificial

landmarks or beacons need a lot of additional engineering effort and modification of

5

the environment when a robot has to change or expand its environment. Localization

method which employs metric mapping will even need a new set of representation.

1.4 Project Objectives

The objective of this work is therefore to design a topological localization

strategy which can potentially help a robot recognize its surroundings despite the

dynamism, errors and computational limitations. Strategically, the system should

require no or little alteration when used in a new domain.

The objectives of this project are:

1. To develop a room recognition system using vision sensor and artificial

neural network

a. To develop hardware combination for color based sensing.

b. To develop an algorithm for room recognition using color histogram.

c. To implement Multi Layer Perceptron (MLP) network as recognition

engine.

2. To evaluate the performance of the room recognition system.

3. To propose a suitable setting for the room recognition system.

1.5 Outline of the Proposed Approach

The interest of this work is primarily in the place recognition aspect of the

topological localization problem. Putting together the issues to be solved, a room

recognition method employing vision and neural network for robot topological

localization is introduced. The proposed approach can be outlined as follows:

6

1. Adopting topological map as the world representation.

2. Room recognition (global localization) using visual information obtained

from camera.

3. Image matching using Artificial Neural Network.

4. Testing and evaluating this combination of sensor and techniques in an

unmodified indoor environment.

In this approach, the topological representation of the world is chosen

because creating such map takes little effort as there is no need to measure the

dimensions of the environment. By nature of their compactness, it has the potential

for representing environments which are several magnitudes larger than those which

can be tractably navigated using metric maps. Topological localization uses a graph

representation that captures the connectivity of a set of features in the environment.

Nodes of the graph represent locations while arcs represent the connectivity between

the locations. The topological map of a section of the P08 Robotics Building is

shown in Figure 1. This map and section of the building were used during testing.

Figure 1.1 Topological map of the indoor environment

A robust localization system requires a sensor that provides rich information

in order to allow the system to reliably distinguish between adjacent locations. For

this reason, a color web camera is used as the sensor. The images captured during the

training stage need to be representative of the environment. In order to keep the

number of necessary reference images low, the ‘fish-eye’ camera set-up which

provides a 360° circular image of its surroundings is implemented (Greiner and

Isukapalli, 1994; Betke and Gurvits, 1997). An example of fish-eye field of view

image is shown in Figure 1.2.

Research Room Pantry Robotic Lab

7

Figure 1.2 Sample image of fish-eye field of view

This vision-based localization system must be trained before it can be used in

an environment. During training, representative images are captured from the

environment and associated with corresponding locations. Classification of images is

done through the Multi Layer back propagation technique. At runtime, the acquired

image is compared to the map’s reference images. The location whose reference

image best matches the input image is then considered to be currently visible

location. Image matching is achieved through the same neural network.

For validation, this localization system is tested in an indoor environment. All

tests were performed in unmodified environments.

1.6 Outline of Thesis

The remainder of this thesis is organized in four main chapters. Chapter 2

reviews related works on robot localization and existing methods for vision-based

system focusing on appearance based method. Chapter 3 describes the development

methodology and guidelines in designing an appearance based room recognition

system for robot localization. Chapter 4 explains the theory and concept of the

8

appearance based room recognition system design. Chapter 5 describes the hardware

and software implementation of the system. Chapter 6 presents the experimental

results of the system performance analysis. Finally, Chapter 7 concludes the thesis

with summary of contributions and suggestions for future development.

Designing a robot that can navigate and operate in a real world environment

is a challenging task. The most fundamental competence it should have is the

localization capability. To localize, a robot needs to have external sensor information

and be able to give an estimate of its location. Localization is made complicated due

to a few factors such dynamism of environment, noise and errors, limitations in

computational resources and the ease of use of the system. Various techniques and

sensors have been introduced to tackle these issues. In this work, a vision-based

topological localization system is developed and investigated for an unmodified,

indoor environment.

137

enhancement, the system can be made to return several fuzzy states of

answers for Room A – for example, CONFIDENT RIGHT for (0.9, 0.01, 0.0)

; NOT CONFIDENT RIGHT for (0.4, 0.0,0.0) ; CONFUSED for (0.4, 0.4,

0.0) or WRONG for (0.0,0.9,0.0). When in NOT CONFIDENT or

CONFUSED states, a dynamic voting system can be designed to vote for a

confident answer to hopefully increase confidence and recognition rate.

4. The aluminium rod which serves as a mounting platform for the webcam,

reflector ball and shade is generating some occlusion factor in the system.

The rod can be replaced with a transparent glass cylinder. Ideally, the items

can be mounted within the transparent glass cylinder to avoid any occlusion

in the view. Then, the system can be fused with orientation tracking sensors

such as magnetic compass to determine a robot’s current heading position.

5. The software module can be modified to incorporate multiple color spaces for

future investigations in color and recognition.

6. Currently, the laptop used for this room recognition work is different from the

laptop used to control the BeMR in previous work by Yeong (2005). The

integration of both systems on the same laptop will be excellent to present a

multi purpose service robot with global localization capability that can be

controlled with a portable device.

The appearance based room recognition system successfully developed in this

work offers a partial solution to the problem of mobile robot localization in an

unmodified indoor environment. Many improvements can still be carried out to

improve the system with more capabilities and higher accuracy.

138

REFERENCES

Abe, Y., Shikano, M., Fukuda, T., Arai, F. and Tanaka, Y. (1999). Vision Based

Navigation System for Autonomous Mobile Robot with Global Matching.

Proceedings of the 1999 IEEE International Conference on Robotics and

Automation. May 10-15. Detroit, Michigan : IEEE, 1299-1304.

Andreasson, H. and Duckett, T. (2004). Topological Localization for Mobile Robots

Using Omni-directional Vision and Local Features. Proceedings of IAV 2004, the

5th IFAC Symposium on Intelligent Autonomous Vehicles. July 5-7. Lisbon,

Portugal.

Ascensio, J. R., Montiel, J. M. M. and Montano, L. (1999). Goal Directed Reactive

Robot Navigation with Relocation Using Laser and Vision. Proceedings of the

1999 IEEE International Conference on Robotics and Automation. May 10-15.

Detroit, Michigan : IEEE, 2905-2910.

Betke, M. and Gurvits, L. (1997). Mobile Robot Localization Using Landmarks.

IEEE Transactions on Robotics and Automation. 13(2): 251-263.

Borenstein, J., Everett, H.R. and Feng, L. (1996 ). “Where am I?" Systems and

Methods for Mobile Robot Positioning, University of Michigan : Technical

Report UM-MEAM-94-21.

Borenstein, J., Everett, H.R. , Feng, L., and Wehe, D. (1997). Mobile Robot

Positioning: Sensors and Techniques. Journal of Robotic Systems, Special Issue

on Mobile Robots. 14(4) : 231-249.

139

Brown, R. G. and Donald, B. R. (2000). Mobile Robot Self Localization Without

Explicit Landmarks. Algorithmica. 26 (3-4): 515-559.

Burgard, W., Fox, D., Hennig, D. and Schmidt, T. (1996). Estimating the Absolute

Position of a Mobile Robot Using Position Probability Grids. Proceedings of the

13th National Conference on Artificial Intelligence (AAAI ’96). August 4-8.

Portland, Oregon: AAAI, 896-901.

Burgard, W., Derr, A., Fox, D. and Cremers, A. B. (1998). Integrating Global

Position Estimation and Position Tracking for Mobile Robots: The Dynamic

Markov Localization Approach. Proceedings of the 1998 IEEE/RSJ International

Conference on Intelligent Robots and Systems (IROS ’98). October 13-17.

Victoria BC : IEEE, 730-735.

Byrne, R. H., Klarer, P. R. and Pletta, J. B. (1992). Techniques for Autonomous

Navigation. Sandia National Laboratories (Albuquerque, N.M.): Sandia Report

SAND92-0457.

Cassinis, R., Grana, D. and Rizzi, A. (1996). Proceedings of the 1st Euromicro

Workshop on Advanced Mobile Robots (Eurobot'96). October 9-11.

Kaiserslautern, Germany : IEEE, 172-176.

Chan, W. K. (2003). Detection of Arching Fault In Underground Distribution Cable

Using Artificial Neural Network. Universiti Teknologi Malaysia : M.Eng Thesis.

Cox, I. J. (1989). Blanche: Position Estimation For An Autonomous Robot Vehicle.

Proceedings of the 1989 IEEE/RSJ International Workshop on Intelligent Robots

and Systems (IROS ’89). September 4-6. Tsukuba, Japan : IEEE, 432 – 439.

Cox, I. J. and Wilfong, G. ed. (1990). Autonomous Robot Vehicles. New York :

Springer-Verlag.

140

Dellaert, F., Fox, D., Burgard, W. and Thrun, S. (1999). Monte Carlo Localization

for Mobile Robots. Proceedings of the 1999 IEEE Conference on Robotics and

Automation. May 10-15. Detroit, Michigan : IEEE, 1322-1328.

Dellaert, F., Burgard, W., Fox, D. and Thrun, S. (2001). Using the

CONDENSATION Algorithm for Robust, Vision-Based Robot Localization.

State-of-the-Art in Content-Based Image and Video Retrieval, Kluwer Academic

Publishers.

Duckett, T. and Nehmzow, U. (2000). Performance Comparison of Landmark

Recognition Systems for Navigating Mobile Robots. Proceedings of the 17th

National Conference on Artificial Intelligence (AAAI 2000). July 30- August 3.

Austin, TX: AAAI.

Duckett, T. and Nehmzow, U. (2001). Mobile Robot Self Localization using

Occupancy Histograms and A Mixture of Gaussian Location Hyphotheses.

Robotics and Autonomous Systems. 34: 117-129.

Fausett, L. (1994). Fundamentals of Neural Networks : Architechture, Algorithm and

Applications. New Jersey: Prentice Hall, Inc.

Fox, D. (1998). Markov Localization: A Probabilistic Framework for Mobile Robot

Localization and Navigation. University of Bonn : Ph.D. Dissertation.

Fox, D., Burgard, W., Dellaert, F. and Thrun, S. (1999). Monte Carlo Localization:

Efficient Position Estimation for Mobile Robots. Proceedings of the 16th

National Conference on Artificial Intelligence. July 18-22. Orlando, Florida :

AAAI Press / The MIT Press, 343 - 349.

Gothard, B. M., Etersky, R. D. and Ewing, R.E. (1993). Lessons Learned on a Low-

Cost Global Navigation System for the Surrogate Semi-Autonomous Vehicle.

Proceedings of the SPIE Conference on Mobile Robots VIII. September 9-10.

Boston MA : SPIE, 2058: 258-269.

141

Greiner, R. and Isukapalli, R. (1994). Learning to Select Useful Landmarks.

Proceedings of the 12th National Conference on Artificial Intelligence (AAA1

’94). August 1-4. Seattle, Washington: AAAI, 1251-1256.

Gutmann, J.-S. and Schlegel, C. (1996). AMOS: Comparison of Scan Matching

Approaches for Self-Localization in Indoor Environments. Proceedings of the 1st

Euromicro Workshop on Advanced Mobile Robots (Eurobot'96). October 9-11.

Kaiserslautern, Germany : IEEE, 61-67.

Haykin, S. (1994). Neural Networks – A Comprehensive Foundation. New York :

Macmillan College Publishing Company, Inc.

Howard, D. and Beale, M. (2002).Neural Network Toolbox User’s Guide. Version 4.

Natick, M. A. : The Mathworks, Inc.

Ishiguro, H. and Tsuji, S. (1996). Image-based Memory of Environment.

Proceedings of the 1996 IEEE/RSJ International Conference on Intelligent

Robots and Systems. November 4-8. Osaka, Japan: IEEE, 634–639.

Jennings, C., Murray, D. and Little, J. J. (1999). Cooperative Robot Localization

with Vision-based Mapping. Proceedings of the 1999 IEEE International

Conference on Robotics and Automation. May 10-15. Detroit, Michigan : IEEE,

2659-2665.

Jogan, M. and Leonardis, A. (2003). Robust localization using an omnidirectional

appearance-based subspace model of environment. Robotics and Autonomous

Systems. 45 : 51–72.

Kaelbling, L., Cassandra, A. and Kurien, J. (1996). Acting Under Uncertainty:

Discrete Bayesian Models for Mobile Robot Navigation. Proceedings of the 1996

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS

’96). November 4-8. Osaka, Japan : IEEE, 963-972.

Kaelbling, L., Littman, M. and Cassandra, A. (1998). Planning and Acting in

Partially Stochastic Domains. Artificial Intelligence. 101(1-2): 99-134.

142

Kak, A., Andress, K., Lopez-Abadia and Carroll, M. (1990). Hierarchical Evidence

Accumulation in the PSEIKI System and Experiments in Model-driven Mobile

Robot Navigation. Uncertainty in Artificial Intelligence. 5 : 353-369.

Kalman, R. E. (1960). A new approach to linear filtering and prediction problems.

Transaction of the ASME – Journal of Basic Engineering. 82 : 35-45.

Kröse, B. J. A., Vlassis, N, Bunschoten, R. and Motomura, Y. (2001). A probabilistic

model for appearance-based robot localization. Image and Vision Computing.

6(19): 381–391.

Kweon, I. S. and Kanade, T. (1991). Extracting Topographic Features for Outdoor

Mobile Robotics. Proceedings of the 1991 IEEE International Conference on

Robotics and Automation. April 9-11. Sacramento CA : IEEE, pp 1992-1997.

Langley, P., Pfleger, K. and Sahami, M. (1997). Lazy Acquisition of Place

Knowledge. Artificial Intelligence Review on Lazy Learning. 11(1-5) : 315-342.

Leonard, J. and Durrant-Whyte, H. (1991). Mobile Robot Localization by Tracking

Geometric Beacons. IEEE Transactions on Robotics and Automation. 7(3) : 376-

382.

Leonardis, A. and Bischof, H. (2000). Robust recognition using eigenimages.

Computer Vision and Image Understanding— Special Issue on Robust Statistical

Techniques in Image Understanding. 78 (1): 99–118.

Lippmann, R. P. (1987). An introduction to computing with neural nets. IEEE

Acoustics,

Speech and Signal Processing Magazine. 4(2): 4-22.

Masters, T. (1993). Practical Neural Network Recipes in C++. San Diego, C.A.:

Academic Press, Inc.

143

Moravec, H. P. (1988). Sensor Fusion in certainty grids for mobile robots. AI

Magazine. 9 (2) : 61-74.

Muller, B. and Reindhart, J. (1990). Neural Networks – An Introduction. Berlin :

Springer-Verlag.

Murase, H. and Nayar, S. K. (1997). Detection of 3D objects in cluttered scenes

using hierarchical eigenspace. Pattern Recognition Letters. 18 (4): 375–384.

Nayar, S. K., Nene, S. A. and Murase, H. (1996). Subspace methods for robot vision.

IEEE Transactions on Robotics and Automation. 12(5) : 750–758.

Negenborn, R. (2003). Robot Localization and Kalman Filters On finding your

position in a noisy world. Utrecht University : M.Sc Thesis.

Nourbakhsh, I., Powers, R. and Birchfield, S. (1995). Dervish An Office Navigating

Robot. AI Magazine. 16(2) : 53-60.

Ramos, F. T., Upcroft, B., Kumar, S. and Durrant-Whyte, H., F. (2005). A Bayesian

Approach for Place Recognition. Proceedings of the IJCAI Workshop on

Reasoning with Uncertainty in Robotics. July 30. Edinburgh, Scotland.

Schiele, B. and Crowley, J. L. (1994). A Comparison of Position Estimation

Techniques Using Occupancy Grids. Proceedings of the 1994 IEEE International

Conference on Robotics and Automation. May 8-13. San Diego, USA : IEEE,

1628-1634.

Schultz, A., Adams, W. and Yamauchi, B. (1999). Integrating Exploration,

Localization, Navigation and Planning with a Common Representation.

Autonomous Robots. 6(3): 293 - 308.

Shaffer, G., Gonzalez, J. and Stentz, A. (1992). Comparison of Two Range-based

Estimators for a Mobile Robot. Proceedings of the SPIE Conference on Mobile

Robots VII. November 18. Boston, USA : SPIE, 1831: 661-667.

144

Simmons, R. and Koenig, S. (1995). Probabilistic Robot Navigation in Partially

Observable Environments. Proceedings of the 14th International Joint

Conference on Artificial Intelligence. July . San Mateo, California : Morgan

Kaufmann, 1080 -1087.

Takeuchi, Y. and Hebert, M. (1998). Evaluation of Image Based Landmark

Recognition Techniques. Carnegie Mellon University (Pittsburgh, P.A.): CMU

Technical Report CMU-RI-TR-98-20.

Thrun, S. (1998). Finding Landmarks for Mobile Robot Navigation. Proceedings of

the 1998 IEEE International Conference on Robotics and Automation. May 16-

20. Leuven, Belgium: IEEE, 958-963.

Thrun, S., Beetz, M., Bennewitz, M., Burgard, W., Cremers, A., Dellaert, F., Fox, D.,

Hahnel, D., Rosenberg, C., Roy, N., Schulte, J. and Schulz, D. (2000).

Probabilistic Algorithms and the interactive Musuem Tour Guide Robot Minerva.

International Journal of Robotics Research. 19 : 972-999.

Ting, H. N. (2002). Speech Analysis and Classification Using Neural-Networks for

Computer-Based Malay Speech Theraphy. Universiti Teknologi Malaysia :

M.Eng Thesis.

Ulrich, I. and Nourbakhsh, I. (2000). Appearance-Based Place Recognition for

Topological Localization. Proceedings of the 2000 IEEE International

Conference on Robotics and Automation. April 24-28. , San Francisco, CA :

IEEE, 1023-1029.

Weiss, G., Wetzler, C. and von Puttkamer, E. (1994). Keeping Track of Position and

Orientation of Moving Indoor Systems by Correlation of Rangefinder Scans.

Proceedings of the 1994 IEEE/RSJ International Conference on Intelligent

Robots and Systems (IROS ’94). September 12-16. Munich, Germany: IEEE,

595 – 601.

145

Weiss, G. and von Puttkamer, E. (1995). A Map Based On Laser Scans Without

Geometric Interpretation. Proceedings of the 4th International Conference on

Intelligent Autonomous Systems. March 27 - 30. Karlsruhe, Germany : IOS, 4:

403-407.

Yamauchi, B. (1996). Mobile Robot Localization in Dynamic Environments Using

Dead Reckoning and Evidence Grids. Proceedings of the 1996 IEEE

International Conference on Robotics and Automation (ICRA ’96). April 22-28.

Minneapolis, MN : IEEE, 1401-1406.

Yamauchi, B. and Langley, P. (1997). Place Recognition in Dynamic Environments.

Journal of Robotic Systems, Special Issue on Mobile Robots Special Issue. 14 (2):

107-120.

Yamauchi, B., Langley, P., Schultz, A. C., Grefenstette, J., and Adams, W. (1998).

Magellan: An integrated adaptive architecture for mobile robotics. Institute for

the Study of Learning and Expertise (Palo Alto, CA) : Technical Report 98-2.

Yeong, C. F. (2005). Development and Evaluation of Various Modes of Human

Robot Interface for Mobile Robot. . Universiti Teknologi Malaysia : M.Eng

Thesis.

eileen su lee ming - eprints.utm.myeprints.utm.my/id/eprint/5145/1/eileensuleemingmfke2006.pdf ·...

Documents