SliceLRF: A Local Reference Frame Sliced along the Height on the 3D Surface

Area of Science:

• Computer Vision
• 3D Geometry Processing
• Machine Learning

Background:

• Local Reference Frames (LRFs) are crucial for 3D shape analysis and matching.
• Existing LRF methods often lack robustness to noise, varying resolutions, clutter, and occlusion.
• Many LRFs are heuristic, limiting their generalizability across applications and data types.

Purpose of the Study:

• To define and analyze the 'degree of distinction' for 2D point cloud distributions.
• To identify factors influencing LRF error based on distinction degree deviation.
• To develop a novel, robust, and generalizable LRF method.

Main Methods:

• Defined 'degree of distinction' to characterize point cloud distributions.
• Analyzed the relationship between distinction degree deviation and LRF error under noise and sampling variations.
• Developed a scoring criterion to assess point cloud distribution robustness.
• Proposed SliceLRF, a Z-axis slicing method using robust slice selection for X-axis estimation.

Main Results:

• Identified key factors affecting distinction degree deviation and LRF error.
• SliceLRF demonstrated superior repeatability and robustness compared to state-of-the-art LRFs.
• Performance was validated across diverse benchmark datasets and data modalities.

Conclusions:

• SliceLRF offers a significant advancement in robust local reference frame estimation.
• The proposed method provides improved generalizability for 3D shape description and matching tasks.
• The 'degree of distinction' metric offers new insights into point cloud distribution properties.