Leveraging Feature Uncertainty in the PnP Problem

Abstract

We propose a real-time and accurate solution to the Perspective-n-Point (PnP) problem -estimating the pose of a calibrated camera from n 3D-to-2D point correspondencesthat exploits the fact that in practice the 2D position of not all 2D features is estimated with the same accuracy. Assuming a model of such feature uncertainties is known in advance, we reformulate the PnP problem as a maximum likelihood minimization approximated by an unconstrained Sampson error function, which naturally penalizes the most noisy correspondences. The advantages of this approach are thoroughly demonstrated in synthetic experiments where feature uncertainties are exactly known.Pre-estimating the features uncertainties in real experiments is, though, not easy. In this paper we model feature uncertainty as 2D Gaussian distributions representing the sensitivity of the 2D feature detectors to different camera viewpoints. When using these noise models with our PnP formulation we still obtain promising pose estimation results that outperform the most recent approaches.

MLPNP – A REAL-TIME MAXIMUM LIKELIHOOD SOLUTION TO THE PERSPECTIVE-N-POINT PROBLEM

“…Matlab OPnP (Zheng et al, 2013a), EPPnP (Ferraz et al, 2014b), CEPPnP (Ferraz et al, 2014a), DLS (Hesch and Roumeliotis, 2011), LHM (Lu et al, 2000), ASPnP (Zheng et al, 2013b), SDP (Schweighofer and Pinz, 2008), RPnP (Li et al, 2012), PPnP (Garro et al, 2012) C++,mex UPnP , EPnP+GN (Lepetit et al, 2009), MLPnP (this paper)…”

“…We use the Matlab toolbox provided by the authors of (Ferraz et al, 2014a, Li et al, 2012, Zheng et al, 2013a) and compare our algorithm in terms of accuracy and speed. The simulation configurations as well as the evaluation metrics are also part of the Now, the image plane coordinates x are usually transformed to normalized image plane coordinates applying the forward projection π.…”

MLPNP – A Real-Time Maximum Likelihood Solution to the Perspective-N-Point Problem

“…Yet another extension to the aforementioned EPPnP is termed Covariant EPPnP (CEPPnP) (Ferraz et al, 2014a). It is the first algorithm to inherently incorporate observation uncertainty into the framework.…”

“…3 depict the mean accuracy for rotation and translation for an increasing number of points (I=10,..,200). Following the experiments of (Ferraz et al, 2014a), the image plane coordinates are perturbed by Gaussians with σ = [1,..,10] and 10% of the features are perturbed by each noise level, respectively.…”

“…Most algorithms focus on geometric but not statistic optimality. To the best of our knowledge, the only work, that includes observation uncertainty into their framework is the Covariant EPPnP of Ferraz et al (Ferraz et al, 2014a).…”