Generalized Maximum Likelihood Estimation for Perspective-n-Point Problem

TL;DR

This paper introduces GMLPnP, a novel generalized maximum likelihood solver for the Perspective-n-Point problem that accounts for observation anisotropy, improving pose estimation accuracy especially in noisy conditions.

Contribution

The paper proposes GMLPnP, a decoupled, iterative GLS-based method that estimates pose and uncertainty simultaneously, addressing limitations of existing PnP solutions.

Findings

GMLPnP improves rotation accuracy by 4.7% and translation accuracy by 2.0% on TUM-RGBD.

GMLPnP enhances rotation and translation accuracy by 18.6% and 18.4% on KITTI-360.

In UAV localization, GMLPnP outperforms baselines by 34.4% in translation accuracy.

Abstract

The Perspective-n-Point (PnP) problem has been widely studied in the literature and applied in various vision-based pose estimation scenarios. However, existing methods ignore the anisotropy uncertainty of observations, as demonstrated in several real-world datasets in this paper. This oversight may lead to suboptimal and inaccurate estimation, particularly in the presence of noisy observations. To this end, we propose a generalized maximum likelihood PnP solver, named GMLPnP, that minimizes the determinant criterion by iterating the GLS procedure to estimate the pose and uncertainty simultaneously. Further, the proposed method is decoupled from the camera model. Results of synthetic and real experiments show that our method achieves better accuracy in common pose estimation scenarios, GMLPnP improves rotation/translation accuracy by 4.7%/2.0% on TUM-RGBD and 18.6%/18.4% on KITTI-360 dataset compared to the best baseline. It is more accurate under very noisy observations in a vision-based UAV localization task, outperforming the best baseline by 34.4% in translation estimation accuracy.