VKFPos: A Learning-Based Monocular Positioning with Variational Bayesian Extended Kalman Filter Integration

TL;DR

VKFPos introduces a learning-based monocular positioning method that combines absolute and relative pose regression through an Extended Kalman Filter within a variational Bayesian framework, improving accuracy in indoor and outdoor scenarios.

Contribution

It presents a novel integration of APR and RPR via EKF within a variational Bayesian inference framework, with covariance prediction enhancing uncertainty modeling.

Findings

Single-shot APR matches state-of-the-art accuracy.

VKFPos outperforms existing methods in temporal positioning.

Covariance prediction improves uncertainty handling.

Abstract

This paper addresses the challenges in learning-based monocular positioning by proposing VKFPos, a novel approach that integrates Absolute Pose Regression (APR) and Relative Pose Regression (RPR) via an Extended Kalman Filter (EKF) within a variational Bayesian inference framework. Our method shows that the essential posterior probability of the monocular positioning problem can be decomposed into APR and RPR components. This decomposition is embedded in the deep learning model by predicting covariances in both APR and RPR branches, allowing them to account for associated uncertainties. These covariances enhance the loss functions and facilitate EKF integration. Experimental evaluations on both indoor and outdoor datasets show that the single-shot APR branch achieves accuracy on par with state-of-the-art methods. Furthermore, for temporal positioning, where consecutive images allow for RPR and EKF integration, VKFPos outperforms temporal APR and model-based integration methods, achieving superior accuracy.