Geometric Transformer for Fast and Robust Point Cloud Registration

TL;DR

This paper introduces Geometric Transformer, a novel method that learns geometric features for superpoint matching in point cloud registration, achieving high accuracy and speed without RANSAC, especially effective in low-overlap scenarios.

Contribution

It proposes a geometric transformer model that encodes pair-wise distances and triplet-wise angles for robust superpoint matching, eliminating the need for RANSAC and significantly accelerating registration.

Findings

Improves inlier ratio by 17-30 percentage points.

Increases registration recall by over 7 points.

Achieves 100x faster registration without RANSAC.

Abstract

We study the problem of extracting accurate correspondences for point cloud registration. Recent keypoint-free methods bypass the detection of repeatable keypoints which is difficult in low-overlap scenarios, showing great potential in registration. They seek correspondences over downsampled superpoints, which are then propagated to dense points. Superpoints are matched based on whether their neighboring patches overlap. Such sparse and loose matching requires contextual features capturing the geometric structure of the point clouds. We propose Geometric Transformer to learn geometric feature for robust superpoint matching. It encodes pair-wise distances and triplet-wise angles, making it robust in low-overlap cases and invariant to rigid transformation. The simplistic design attains surprisingly high matching accuracy such that no RANSAC is required in the estimation of alignment transformation, leading to $100$ times acceleration. Our method improves the inlier ratio by $17{\sim}30$ percentage points and the registration recall by over $7$ points on the challenging 3DLoMatch benchmark. Our code and models are available at https://github.com/qinzheng93/GeoTransformer.