GSplatLoc: Ultra-Precise Camera Localization via 3D Gaussian Splatting

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Zeller, Atticus J., Wu, Haijuan
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909613829390336
author Zeller, Atticus J.
Wu, Haijuan
author_facet Zeller, Atticus J.
Wu, Haijuan
contents We present GSplatLoc, a camera localization method that leverages the differentiable rendering capabilities of 3D Gaussian splatting for ultra-precise pose estimation. By formulating pose estimation as a gradient-based optimization problem that minimizes discrepancies between rendered depth maps from a pre-existing 3D Gaussian scene and observed depth images, GSplatLoc achieves translational errors within 0.01 cm and near-zero rotational errors on the Replica dataset - significantly outperforming existing methods. Evaluations on the Replica and TUM RGB-D datasets demonstrate the method's robustness in challenging indoor environments with complex camera motions. GSplatLoc sets a new benchmark for localization in dense mapping, with important implications for applications requiring accurate real-time localization, such as robotics and augmented reality.
format Preprint
id arxiv_https___arxiv_org_abs_2412_20056
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle GSplatLoc: Ultra-Precise Camera Localization via 3D Gaussian Splatting
Zeller, Atticus J.
Wu, Haijuan
Computer Vision and Pattern Recognition
We present GSplatLoc, a camera localization method that leverages the differentiable rendering capabilities of 3D Gaussian splatting for ultra-precise pose estimation. By formulating pose estimation as a gradient-based optimization problem that minimizes discrepancies between rendered depth maps from a pre-existing 3D Gaussian scene and observed depth images, GSplatLoc achieves translational errors within 0.01 cm and near-zero rotational errors on the Replica dataset - significantly outperforming existing methods. Evaluations on the Replica and TUM RGB-D datasets demonstrate the method's robustness in challenging indoor environments with complex camera motions. GSplatLoc sets a new benchmark for localization in dense mapping, with important implications for applications requiring accurate real-time localization, such as robotics and augmented reality.
title GSplatLoc: Ultra-Precise Camera Localization via 3D Gaussian Splatting
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2412.20056