Guaranteed Outlier Removal for Point Cloud Registration with Correspondences.

Area of Science:

• Computer Vision
• Geometric Algorithms
• 3D Data Processing

Background:

• 3D point cloud registration commonly uses 3D keypoint correspondences, but high outlier rates from inaccurate keypoints necessitate robust estimation.
• Existing robust methods face high computational costs due to large data sizes and numerous outliers, hindering efficient registration.

Purpose of the Study:

• To present a novel preprocessing method, guaranteed outlier removal, for accelerating 3D point cloud registration.
• To significantly reduce the number of outliers while ensuring the globally optimal solution is preserved.

Main Methods:

• A preprocessing technique called guaranteed outlier removal is introduced.
• This method employs deterministic and fast purely geometric operations to reduce the set of correspondences.
• Any rejected correspondence is guaranteed not to be part of the globally optimal solution.

Main Results:

• The method effectively reduces the outlier population in 3D point cloud registration.
• This reduction enables significantly faster subsequent optimization processes.
• Experiments on synthetic and real data confirm the method's effectiveness.

Conclusions:

• Guaranteed outlier removal is an effective preprocessing step for 3D point cloud registration.
• The method offers a significant speedup in computation by efficiently handling outliers.
• It preserves the globally optimal solution, making it suitable for various applications.