Are Minimal Radial Distortion Solvers Necessary for Relative Pose Estimation?

2024-10-08

ECCV Workshops

AI Summary

In relative pose estimation, radial distortion modeling typically relies on computationally expensive and implementationally complex minimal solvers; ignoring distortion, however, severely degrades accuracy. This paper challenges the necessity of such analytical approaches and proposes a lightweight alternative: within the RANSAC framework, coupling an efficient pinhole solver with coarse-grained sampling-based optimization of radial distortion parameters. By avoiding analytical solutions to the distortion equations, our method significantly reduces implementation complexity and computational overhead. Experiments across multiple real and synthetic datasets—and under various RANSAC variants—demonstrate that our approach matches or surpasses state-of-the-art minimal distortion solvers in accuracy, while achieving substantially higher speed and outperforming existing fast non-minimal solvers. The core contribution is the empirical demonstration that high-accuracy distortion compensation does not require intricate analytical modeling; instead, sampling-driven, lightweight co-optimization achieves superior accuracy–efficiency trade-offs.

Technology Category

• Intelligent Robots: State Estimation
• Computer Vision: Motion & Tracking
• Search and Optimization: Non-convex Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Estimating the relative pose between two cameras is a fundamental step in many applications such as Structure-from-Motion. The common approach to relative pose estimation is to apply a minimal solver inside a RANSAC loop. Highly efficient solvers exist for pinhole cameras. Yet, (nearly) all cameras exhibit radial distortion. Not modeling radial distortion leads to (significantly) worse results. However, minimal radial distortion solvers are significantly more complex than pinhole solvers, both in terms of run-time and implementation efforts. This paper compares radial distortion solvers with a simple-to-implement approach that combines an efficient pinhole solver with sampled radial distortion parameters. Extensive experiments on multiple datasets and RANSAC variants show that this simple approach performs similarly or better than the most accurate minimal distortion solvers at faster run-times while being significantly more accurate than faster non-minimal solvers. We clearly show that complex radial distortion solvers are not necessary in practice. Code and benchmark are available at https://github.com/kocurvik/rd.

Problem

• Evaluating necessity of minimal radial distortion solvers for camera pose estimation
• Comparing simple pinhole-plus-sampling approach with complex distortion solvers
• Determining optimal balance between accuracy and computational efficiency

Innovation

• Combining pinhole solver with sampled radial distortion parameters
• Achieving faster run-times than minimal distortion solvers
• Maintaining accuracy comparable to complex radial distortion solvers