RIGA: Rotation-Invariant and Globally-Aware Descriptors for Point Cloud Registration

TL;DR

RIGA introduces rotation-invariant and globally-aware descriptors for point cloud registration, significantly improving accuracy under large rotations by encoding local and global 3D structures.

Contribution

The paper proposes a novel descriptor learning method that is both rotation-invariant and globally-aware, enhancing point cloud registration performance.

Findings

Outperforms state-of-the-art methods on ModelNet40 with 8° less rotation error.

Achieves at least 5% higher feature matching recall on 3DLoMatch.

Effective in both object- and scene-level registration tasks.

Abstract

Successful point cloud registration relies on accurate correspondences established upon powerful descriptors. However, existing neural descriptors either leverage a rotation-variant backbone whose performance declines under large rotations, or encode local geometry that is less distinctive. To address this issue, we introduce RIGA to learn descriptors that are Rotation-Invariant by design and Globally-Aware. From the Point Pair Features (PPFs) of sparse local regions, rotation-invariant local geometry is encoded into geometric descriptors. Global awareness of 3D structures and geometric context is subsequently incorporated, both in a rotation-invariant fashion. More specifically, 3D structures of the whole frame are first represented by our global PPF signatures, from which structural descriptors are learned to help geometric descriptors sense the 3D world beyond local regions. Geometric context from the whole scene is then globally aggregated into descriptors. Finally, the description of sparse regions is interpolated to dense point descriptors, from which correspondences are extracted for registration. To validate our approach, we conduct extensive experiments on both object- and scene-level data. With large rotations, RIGA surpasses the state-of-the-art methods by a margin of 8\degree in terms of the Relative Rotation Error on ModelNet40 and improves the Feature Matching Recall by at least 5 percentage points on 3DLoMatch.