Striking a Georeferenced Pose: RTK and ARKit Fusion in Learned 3D Map Reconstruction

Daniel Neamati, Mira Partha, Lance Legel, and Grace Gao

Abstract:	Many modern navigation systems use 3D maps to provide users with an accurate and precise navigation solution, especially for autonomous navigation systems. Recent advances in the computer vision and computer graphics communities, grouped as “radiance fields,” have enabled realistic 3D reconstructions from images and their associated poses (i.e., position and orientation). However, despite these advances, camera pose estimation and uncertainty have been a long-standing challenge in building a sufficiently accurate 3D map. We develop a novel approach to quantify the camera pose uncertainty leveraging the image realism of recent 3D maps. We consider various pose solutions, including Augmented Reality Kit (ARKit), Real-Time Kinematic (RTK), and fused solutions, from which we assess the impact of quality on the reconstructed map and evaluate the pose uncertainty. We collect a real-world dataset that includes aerial and phone camera imagery, and we find that the contributions of ARKit, RTK, and the learned 3D map, together, provide confidence in phone and drone georeferenced pose estimates at the decimeter level for most cases. For the remaining cases, we show how our approach helps explain how map errors can create a covariance matrix that is not positive definite. Overall, we demonstrate that uncertainty quantification using the full image, rather than features, shows promise in leveraging pixel-level information for a more precise understanding of the georeferenced camera poses.
Published in:	Proceedings of the 38th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2025) September 8 - 12, 2025 Hilton Baltimore Inner Harbor Baltimore, Maryland
Pages:	1083 - 1095
Cite this article:	Neamati, Daniel, Partha, Mira, Legel, Lance, Gao, Grace, "Striking a Georeferenced Pose: RTK and ARKit Fusion in Learned 3D Map Reconstruction," Proceedings of the 38th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2025), Baltimore, Maryland, September 2025, pp. 1083-1095. https://doi.org/10.33012/2025.20273
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In