New Gravity-Aware Pose Estimation Method Needs Only 1-2 Feature Correspondences

zhenjun_zhao · x · 2026-08-21

This paper presents a novel approach for absolute pose and focal length estimation under partial calibration by leveraging IMU gravity vectors and local geometric information from features. The authors derive two efficient solvers: UP1PfAC, which operates with a single affine correspondence, and UP2PfORI, requiring two orientation-covariant features. Unlike traditional methods that need four point correspondences, these solvers require fewer samples and lower computational costs, significantly simplifying robust estimation within RANSAC-like frameworks.

Original post →

More from Research

Research channel →