Fine-To-Coarse Global Registration of RGB-D Scans

TL;DR

This paper introduces a novel fine-to-coarse global registration algorithm for RGB-D scans that improves long sequence registration accuracy in indoor environments, addressing issues of drift and tracking loss.

Contribution

The paper presents a new fine-to-coarse registration method and a benchmark dataset for evaluating global registration of RGB-D scans.

Findings

Our algorithm outperforms previous methods in long sequence registration.

The benchmark includes 10,401 manually annotated correspondences across 25 scenes.

The approach effectively reduces drift and improves structural consistency.

Abstract

RGB-D scanning of indoor environments is important for many applications, including real estate, interior design, and virtual reality. However, it is still challenging to register RGB-D images from a hand-held camera over a long video sequence into a globally consistent 3D model. Current methods often can lose tracking or drift and thus fail to reconstruct salient structures in large environments (e.g., parallel walls in different rooms). To address this problem, we propose a "fine-to-coarse" global registration algorithm that leverages robust registrations at finer scales to seed detection and enforcement of new correspondence and structural constraints at coarser scales. To test global registration algorithms, we provide a benchmark with 10,401 manually-clicked point correspondences in 25 scenes from the SUN3D dataset. During experiments with this benchmark, we find that our fine-to-coarse algorithm registers long RGB-D sequences better than previous methods.