On averaging multiview relations for 3D scan registration.

Area of Science:

• Computer Vision
• Geometric Computing
• 3D Data Processing

Background:

• Iterative Closest Point (ICP) algorithm is a standard for 3D point cloud registration.
• Existing ICP methods often fail to leverage multiview constraints effectively.
• Simultaneous registration of multiple 3D scans presents challenges in accuracy and efficiency.

Purpose of the Study:

• To develop an extended ICP algorithm for simultaneous registration of multiple 3D scans.
• To exploit information redundancy in multiview 3D scans for improved registration.
• To enhance the efficiency and accuracy of 3D registration using multiview data.

Main Methods:

• Extension of the iterative closest point (ICP) algorithm.
• Utilizing averaging of relative motions based on Lie group structure.
• Introducing causality-obeying and transitive correspondence variants for multiview registration.

Main Results:

• Demonstrated superior accuracy compared to existing multiview registration methods.
• Experimental validation on real-world 3D scan datasets.
• Characterization of the method's behavior and performance.

Conclusions:

• The proposed multiview 3D registration method offers significant improvements in accuracy and efficiency.
• Exploiting multiview constraints through motion averaging enhances 3D registration performance.
• The developed variants provide robust solutions for complex multiview registration problems.