GeoTransformer: Fast and Robust Point Cloud Registration with Geometric Transformer

Zheng Qin,Hao Yu,Changjian Wang,Yulan Guo,Yuxing Peng,Slobodan Ilic,Dewen Hu,Kai Xu

2023-07-25

Abstract:We study the problem of extracting accurate correspondences for point cloud registration. Recent keypoint-free methods have shown great potential through bypassing the detection of repeatable keypoints which is difficult to do especially in low-overlap scenarios. 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, or GeoTransformer for short, to learn geometric feature for robust superpoint matching. It encodes pair-wise distances and triplet-wise angles, making it invariant to rigid transformation and robust in low-overlap cases. 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. Extensive experiments on rich benchmarks encompassing indoor, outdoor, synthetic, multiway and non-rigid demonstrate the efficacy of GeoTransformer. Notably, our method improves the inlier ratio by $18{\sim}31$ percentage points and the registration recall by over $7$ points on the challenging 3DLoMatch benchmark. Our code and models are available at \url{<a class="link-external link-https" href="https://github.com/qinzheng93/GeoTransformer" rel="external noopener nofollow">this https URL</a>}.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper attempts to address several key issues in point cloud registration:

1. **Accurate Correspondence Extraction**: In point cloud registration tasks, it is crucial to accurately find correspondences between two partially overlapping 3D point clouds. Traditional keypoint-based methods struggle to detect repeatable keypoints in low-overlap scenarios, leading to inaccurate matching results.

2. **Improving Robustness and Efficiency**: Existing methods often rely on robust estimators like RANSAC to handle a large number of outliers, which not only has high computational complexity but also performs poorly in low-overlap situations. The paper proposes a new method that improves matching robustness by learning geometric features and eliminates the need for RANSAC, thereby significantly enhancing computational efficiency.

3. **Reducing Dependence on Keypoint Detection**: The paper proposes a keypoint-free approach by downsampling the input point cloud into superpoints and then matching the local neighborhoods (patches) of these superpoints, eventually propagating the matching results to dense points. This method reduces dependence on keypoint detection and is suitable for low-overlap scenarios.

Specifically, the paper presents the following innovations:

- **GeoTransformer**: By encoding point-pair distances and triplet angles, it learns the geometric structure features of the point cloud, making it invariant to rigid transformations, thus enabling robust superpoint matching even in low-overlap situations.
- **Overlap-aware Circle Loss**: It reweights the loss of each superpoint match to focus more on highly overlapping matches, thereby improving the model's convergence and accuracy.
- **Fast and Accurate Registration Method**: Through high-quality superpoint matching, combined with the Optimal Transport Layer to extract dense point correspondences, it ultimately achieves efficient registration without the need for RANSAC.

Experimental results show that this method performs excellently on multiple benchmark datasets, significantly improving inlier rates and registration recall rates, especially in challenging scenarios with low overlap and large rotations.