A New Approach to Robust Estimation of Parametric Structures.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Robust estimators often require complex parameter tuning, limiting their practical use.
• Existing methods like RANSAC struggle with datasets containing diverse inlier noise levels.

Purpose of the Study:

• Introduce the Multiple Input Structures with Robust Estimator (MISRE) algorithm.
• Develop a robust estimation method that minimizes parameter tuning and handles varied noise.

Main Methods:

• MISRE processes inlier and outlier structures independently using shared scale estimation constants.
• Data is ordered to facilitate straightforward inlier/outlier classification.
• The algorithm's performance is evaluated on 2D image and 3D point cloud datasets.

Main Results:

• MISRE achieves performance comparable to RANSAC when inlier noises are similar.
• MISRE provides accurate inlier estimates even with highly variable inlier noises, surpassing RANSAC.
• The algorithm exhibits gradual failure modes in the presence of numerous outliers.

Conclusions:

• MISRE offers a robust and practical alternative to traditional robust estimators.
• The method's independent structure processing and straightforward classification enhance its applicability.
• MISRE demonstrates superior performance in challenging datasets with heterogeneous noise.