Less Biased Noise Scale Estimation for Threshold-Robust RANSAC

2025-03-17

AI Summary

This paper addresses a critical bottleneck in RANSAC-based relative pose estimation: its reliance on manually tuned inlier thresholds and the difficulty of optimization without ground-truth annotations. To overcome this, we propose an adaptive threshold determination method. Our key contributions are: (1) correcting the noise-scale underestimation bias in SIMFIT++ caused by data reuse and neglect of threshold sensitivity; (2) introducing a multi-image-pair collaborative filtering mechanism to enhance cross-scenario consistency and robustness of threshold estimation; and (3) establishing an analytical mapping between noise scale and threshold, enabling end-to-end adaptive threshold derivation. Experiments demonstrate that our method maintains high performance across a wide range of threshold values and significantly improves both robustness and accuracy of pose estimation—especially in the absence of ground-truth labels.

Technology Category

• Intelligent Robots: State Estimation
• Reasoning under Uncertainty: Relational Probabilistic Models
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

The gold-standard for robustly estimating relative pose through image matching is RANSAC. While RANSAC is powerful, it requires setting the inlier threshold that determines whether the error of a correspondence under an estimated model is sufficiently small to be included in its consensus set. Setting this threshold is typically done by hand, and is difficult to tune without a access to ground truth data. Thus, a method capable of automatically determining the optimal threshold would be desirable. In this paper we revisit inlier noise scale estimation, which is an attractive approach as the inlier noise scale is linear to the optimal threshold. We revisit the noise scale estimation method SIMFIT and find bias in the estimate of the noise scale. In particular, we fix underestimates from using the same data for fitting the model as estimating the inlier noise, and from not taking the threshold itself into account. Secondly, since the optimal threshold within a scene is approximately constant we propose a multi-pair extension of SIMFIT++, by filtering of estimates, which improves results. Our approach yields robust performance across a range of thresholds, shown in Figure 1.

Problem

• Automatically determine optimal inlier threshold for RANSAC.
• Address bias in noise scale estimation for robust pose estimation.
• Improve SIMFIT++ with multi-pair extension for consistent thresholds.

Innovation

• Automated optimal threshold determination for RANSAC
• Improved noise scale estimation with SIMFIT++
• Multi-pair extension for consistent threshold estimation