SDF-based RGB-D Camera Tracking in Neural Scene Representations

TL;DR

This paper introduces a method for tracking the 6D pose of RGB-D cameras using neural scene representations, specifically signed distance fields, which improves speed and robustness over density-based methods.

Contribution

The paper proposes a novel SDF-based neural scene representation for RGB-D camera tracking, demonstrating faster and more accurate pose estimation compared to existing density-based approaches.

Findings

SDF-based tracking is faster than density-based methods.

SDF approach yields more robust pose estimates.

The method improves accuracy under limited computation time.

Abstract

We consider the problem of tracking the 6D pose of a moving RGB-D camera in a neural scene representation. Different such representations have recently emerged, and we investigate the suitability of them for the task of camera tracking. In particular, we propose to track an RGB-D camera using a signed distance field-based representation and show that compared to density-based representations, tracking can be sped up, which enables more robust and accurate pose estimates when computation time is limited.