A Performance Evaluation of Correspondence Grouping Methods for 3D Rigid Data Matching

Area of Science:

• Computer Vision
• 3D Data Analysis

Background:

• Establishing point-to-point correspondences in 3D rigid data is crucial for many computer vision tasks.
• Existing correspondence selection methods lack clear comparative analysis for diverse applications and perturbations.

Purpose of the Study:

• To comprehensively evaluate and compare nine state-of-the-art 3D correspondence grouping methods.
• To clarify the advantages and disadvantages of these methods under various conditions.

Main Methods:

• Experimental evaluation of nine 3D correspondence grouping algorithms.
• Testing across three benchmarks: shape retrieval, 3D object recognition, and point cloud registration.
• Investigating performance under perturbations: noise, density variation, clutter, occlusion, partial overlap, and varying initial correspondences.

Main Results:

• Performance variations observed across different application contexts and perturbation types.
• No single method demonstrated superiority across all tested scenarios.
• Insights gained into the robustness and limitations of each evaluated approach.

Conclusions:

• The study provides a detailed comparison of 3D correspondence grouping methods.
• Identifies strengths and weaknesses of current techniques for practical applications.
• Highlights areas for future research in robust 3D correspondence matching.