PointDifformer: Robust Point Cloud Registration With Neural Diffusion and Transformer

Rui She,Qiyu Kang,Sijie Wang,Wee Peng Tay,Kai Zhao,Yang Song,Tianyu Geng,Yi Xu,Diego Navarro Navarro,Andreas Hartmannsgruber

DOI: https://doi.org/10.1109/TGRS.2024.3351286

2024-04-22

Abstract:Point cloud registration is a fundamental technique in 3-D computer vision with applications in graphics, autonomous driving, and robotics. However, registration tasks under challenging conditions, under which noise or perturbations are prevalent, can be difficult. We propose a robust point cloud registration approach that leverages graph neural partial differential equations (PDEs) and heat kernel signatures. Our method first uses graph neural PDE modules to extract high dimensional features from point clouds by aggregating information from the 3-D point neighborhood, thereby enhancing the robustness of the feature representations. Then, we incorporate heat kernel signatures into an attention mechanism to efficiently obtain corresponding keypoints. Finally, a singular value decomposition (SVD) module with learnable weights is used to predict the transformation between two point clouds. Empirical experiments on a 3-D point cloud dataset demonstrate that our approach not only achieves state-of-the-art performance for point cloud registration but also exhibits better robustness to additive noise or 3-D shape perturbations.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper primarily focuses on the fundamental technique of point cloud registration in computer vision, aiming to address the challenges of point cloud registration under challenging conditions (such as the presence of noise or perturbations).

Specifically, the paper proposes a new method called PointDifformer, which combines Graph Neural Partial Differential Equations (GnPDEs) and heat kernel signatures to enhance the robustness of point cloud registration. The main contributions include:

1. **Designing a 3D point cloud representation module based on graph neural PDE learning**: By constructing a neighborhood graph in the point cloud feature space and utilizing the graph neural PDE module to extract high-dimensional features, the robustness of feature representation is improved.

2. **Proposing a robust 3D point cloud registration method using a graph neural diffusion module and heat kernel signature attention mechanism**: By leveraging the shape isometry invariance of heat kernel signatures and incorporating them into the attention mechanism, the identification of corresponding key points is improved, resulting in more robust registration outcomes.

3. **Experimental results demonstrate that the proposed point cloud registration method performs excellently not only under normal conditions but also maintains good performance in the presence of noise and perturbations**: Through a series of experiments, comparisons with existing baseline methods show the superiority of PointDifformer in various scenarios.

In summary, this research aims to develop a method capable of effectively handling point cloud registration tasks in complex environments, especially maintaining good performance in the face of challenges such as noise and non-rigid deformations. This is of significant importance for fields such as autonomous driving, robotic navigation, and scene modeling.