Reconstruction and Analysis of 3D Scenes (eBook)

Reconstruction and Analysis of 3D Scenes 5
Foreword 7
Preface 9
Target Audience 9
Difficulty 10
Organization of the Content 10
Abstract 12
Kurzfassung 12
Note 14
Acknowledgments 15
Contents 17
1 Introduction 21
1.1 Goals 23
1.2 Challenges and Main Contributions 25
1.3 Publications 27
1.4 The Proposed Framework 31
1.5 Book Outline 33
References 34
2 Preliminaries of 3D Point Cloud Processing 36
2.1 From the Real World to a Scene and Its Representation 36
2.2 On Points and Clouds, and Point Clouds 38
2.3 Point Cloud Acquisition 39
2.3.1 Passive Techniques 40
2.3.2 Active Techniques 41
2.4 Generation of 2D Image Representations for 3D Point Clouds 43
2.5 Point Quality Assessment 44
2.5.1 Influencing Factors and Related Work 45
2.5.2 Filtering Based on Intensity Information 46
2.5.3 Filtering Based on Range Reliability 47
2.5.4 Filtering Based on Local Planarity 48
2.5.5 A Qualitative Comparison of Different Measures 50
2.5.6 A Quantitative Comparison of Different Measures 52
2.6 Conclusions 54
References 55
3 A Brief Survey on 2D and 3D Feature Extraction 58
3.1 What Is a Feature? 58
3.2 2D Feature Extraction 59
3.2.1 Overall Appearance 60
3.2.2 Pixel Attributes 61
3.2.3 Texture 61
3.2.4 Shape 62
3.2.5 Local Features 63
3.3 3D Feature Extraction 64
3.3.1 Point Attributes 64
3.3.2 Shape 65
3.3.3 Local Features 66
3.4 Discussion 67
3.5 Conclusions 69
References 70
4 Point Cloud Registration 73
4.1 Motivation and Contributions 74
4.2 Related Work 77
4.2.1 Feature Extraction 77
4.2.2 Keypoint-Based Point Cloud Registration 81
4.3 A Novel Framework for Keypoint-Based Point Cloud Registration 82
4.3.1 2D Image Representations 82
4.3.2 Point Quality Assessment 84
4.3.3 Feature Extraction and Matching 84
4.3.4 Forward Projection of 2D Keypoints to 3D Space 90
4.3.5 Correspondence Weighting 91
4.3.6 Point Cloud Registration 95
4.4 Experimental Results 103
4.4.1 Dataset 103
4.4.2 Experiments 104
4.4.3 Results 105
4.5 Discussion 116
4.6 Conclusions 122
References 123
5 Co-Registration of 2D Imagery and 3D Point Cloud Data 129
5.1 Motivation and Contributions 130
5.2 Related Work 132
5.2.1 Indirect Co-Registration of 3D Point Clouds and Thermal Infrared Images 133
5.2.2 Direct Co-Registration of 3D Point Clouds and Thermal Infrared Images 134
5.2.3 Direct Generation of 3D Point Clouds from Thermal Infrared Images 134
5.3 A Novel Framework for Keypoint-Based 3D Mapping of Thermal Information 135
5.3.1 Radiometric Correction 135
5.3.2 Geometric Calibration 136
5.3.3 Feature Extraction and Matching 139
5.3.4 Keypoint-Based Co-Registration of 3D and 2D Information 141
5.4 Experimental Results 146
5.4.1 Data Acquisition 146
5.4.2 Experiments 148
5.4.3 Results 148
5.5 Discussion 149
5.6 Conclusions 155
References 156
6 3D Scene Analysis 159
6.1 Motivation and Contributions 160
6.2 Related Work 163
6.2.1 Neighborhood Selection: Single-Scale Representation Versus Multi-Scale Representation 163
6.2.2 Feature Extraction: Sampled Features Versus Interpretable Features 166
6.2.3 Feature Selection: All Features Versus Relevant Features 167
6.2.4 Classification: Individual Feature Vectors Versus Contextual Information 169
6.3 A Novel Framework for 3D Scene Analysis 171
6.3.1 Neighborhood Selection 172
6.3.2 Feature Extraction 177
6.3.3 Feature Selection 183
6.3.4 Classification 188
6.4 Extension Toward Large-Scale 3D Scene Analysis 193
6.5 Extension by Involving Contextual Information 195
6.5.1 Association Potentials 197
6.5.2 Interaction Potentials 197
6.5.3 Definition of the Neighborhood 198
6.5.4 Training and Inference 199
6.6 Experimental Results 200
6.6.1 Datasets 200
6.6.2 Experiments 203
6.6.3 Results 206
6.7 Discussion 226
6.7.1 Our Main Framework 227
6.7.2 Extension Toward Data-Intensive Processing 231
6.7.3 Extension by Involving Contextual Information 233
6.8 Conclusions 234
References 235
7 Conclusions and Future Work 243
Index 248