Robust Range Image Registration Using Genetic Algorithms And The Surface Interpenetration Measure

✍ Scribed by Luciano Silva, Olga R. P. Bellon, Kim L. Boyer

• Publisher: World Scientific Publishing Company
• Year: 2004
• Tongue: English
• Leaves: 175
• Series: Series in Machine Perception and Artificial Intelligence
• Category: Library

✦ Synopsis

This book addresses the range image registration problem for automatic 3D model construction. The focus is on obtaining highly precise alignments between different view pairs of the same object to avoid 3D model distortions; in contrast to most prior work, the view pairs may exhibit relatively little overlap and need not be prealigned. To this end, a novel effective evaluation metric for registration, the Surface Interpenetration Measure (SIM) is defined. This measure quantifies the interleaving of two surfaces as their alignment is refined, putting the qualitative evaluation of "splotchiness," often used in reference to renderings of the aligned surfaces, onto a solid mathematical footing. The SIM is shown to be superior to mean squared error (i.e. more sensitive to fine scale changes) in controlling the final stages of the alignment process. The authors go on to combine the SIM with Genetic Algorithms (GAs) to develop a robust approach for range image registration. The results confirm that this technique achieves precise surface registration with no need for prealignment, as opposed to methods based on the Iterative Closest Point (ICP) algorithm, the most popular to date. Thorough experimental results including an extensive comparative study are presented and enhanced GA-based approaches to improve the registration still further are proposed. The authors also develop a global multiview registration technique using the GA-based approach. The results show considerable promise in terms of accuracy for 3D modeling.

✦ Table of Contents

Team DDU......Page 1
Preface......Page 8
Contents......Page 10
1 Introduction......Page 12
1.1 Range images......Page 14
1.2 Applications......Page 18
1.3 Book outline......Page 19
2.1 Definition......Page 20
2.2 Registration approaches......Page 21
2.3 Outlier rejection rules......Page 25
2.4 Registration quality measures......Page 27
2.5 Summary......Page 29
3.1 Definition......Page 30
3.2 Obtaining precise alignments......Page 36
3.3 Parameters and constraints on the SIM......Page 46
3.4 Stability against noise......Page 52
3.5 Discussion......Page 54
4.1 Concepts......Page 56
4.2 Chromosome encoding......Page 60
4.3 Robust fitness function......Page 61
4.4 GA parameter settings......Page 64
4.5 Enhanced GAs......Page 74
4.6 Results for other range image databases......Page 86
4.7 GAs and SA......Page 89
4.8 Low-overlap registration......Page 94
4.9 Evaluation time......Page 96
4.10 Discussion......Page 98
5.1 SIM as fitness function......Page 100
5.2 Experimental results......Page 102
5.3 Multiobjective Evolutionary Algorithms......Page 109
5.4 Discussion......Page 114
6.1 Aligning common overlapping areas......Page 116
6.2 Global multiview registration......Page 121
6.3 Experimental results......Page 124
6.4 Alignment consistency......Page 127
6.5 Discussion......Page 129
7.1 Contributions......Page 132
7.2 Future work......Page 134
Appendix Experimental Results......Page 136
Bibliography......Page 168
Index......Page 174