On the Robustness of Multi-View Rotation Averaging

TL;DR

This paper introduces a robust rotation averaging framework that effectively handles noisy measurements in multi-view structure from motion, improving accuracy and reliability in pose estimation tasks.

Contribution

The paper proposes an innovative robust initialization scheme using epsilon-cycle consistency integrated into IRLS, avoiding costly edge removal and enhancing robustness against noise.

Findings

Outperforms state-of-the-art methods on benchmark datasets.

Effectively handles high levels of measurement noise.

Reduces failures caused by poor initialization.

Abstract

Rotation averaging is a synchronization process on single or multiple rotation groups, and is a fundamental problem in many computer vision tasks such as multi-view structure from motion (SfM). Specifically, rotation averaging involves the recovery of an underlying pose-graph consistency from pairwise relative camera poses. Specifically, given pairwise motion in rotation groups, especially 3-dimensional rotation groups (\eg, $\mathbb{SO}(3)$), one is interested in recovering the original signal of multiple rotations with respect to a fixed frame. In this paper, we propose a robust framework to solve multiple rotation averaging problem, especially in the cases that a significant amount of noisy measurements are present. By introducing the $\epsilon$-cycle consistency term into the solver, we enable the robust initialization scheme to be implemented into the IRLS solver. Instead of conducting the costly edge removal, we implicitly constrain the negative effect of erroneous measurements by weight reducing, such that IRLS failures caused by poor initialization can be effectively avoided. Experiment results demonstrate that our proposed approach outperforms state of the arts on various benchmarks.