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

TL;DR

GSplatLoc introduces a differentiable rendering-based camera localization method using 3D Gaussian splatting, achieving ultra-precise pose estimation with errors as low as 0.01 cm, outperforming existing approaches in indoor environments.

Contribution

The paper proposes GSplatLoc, a novel localization approach that leverages 3D Gaussian splatting and gradient optimization for highly accurate camera pose estimation.

Findings

Achieves translational errors within 0.01 cm on the Replica dataset.

Demonstrates robustness in complex indoor environments.

Sets new benchmarks for dense mapping localization accuracy.

Abstract

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.