A Hough Voting-Based 2-Point RANSAC Solution to the Perspective-n-Point Problem.

Area of Science:

• Computer Vision
• Multi-view Geometry

Background:

• Perspective-n-Point (PnP) is crucial in multi-view geometry.
• Existing PnP methods struggle with high outlier rates and degenerate cases, leading to errors or high computational costs.
• The assumption that the maximum inlier set yields the correct pose is questionable at ultra-high outlier rates.

Purpose of the Study:

• To address the robustness limitations of current PnP methods under challenging conditions.
• To investigate the maximum outlier tolerance of PnP algorithms.
• To propose a novel, efficient, and robust PnP solution.

Main Methods:

• A novel Hough voting based 2-point RANSAC algorithm is proposed.
• The method is evaluated on synthetic and real-world datasets.
• Systematic examination under ultra-high outlier rates was performed.

Main Results:

• The proposed PnP solution demonstrates robustness in near-degenerate cases with high outlier rates.
• The method successfully handles up to 99% outliers on random synthetic data.
• The traditional maximum inlier-set assumption fails for real-world data when outlier rates exceed 97%.

Conclusions:

• The novel Hough voting based 2-point RANSAC offers accurate and efficient PnP solutions for high outlier rates and degenerate cases.
• The findings challenge the universal applicability of the maximum inlier-set assumption in extreme outlier scenarios.
• The study provides a more reliable PnP method for challenging computer vision applications.