Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

TL;DR

PointDifformer tackles robust 3D point cloud registration under noise and perturbations by marrying graph neural PDE-based feature learning with a heat kernel signature–augmented self-cross attention for robust correspondences. A weighted SVD with learnable weights estimates the rigid transformation from a compact set of correspondences. Extensive experiments on indoor and outdoor datasets (vReLoc, Boreas, KITTI) demonstrate state-of-the-art accuracy and superior robustness to synthetic and natural perturbations, highlighting practical applicability in autonomous driving and robotics.

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.

PointDifformer: Robust Point Cloud Registration With Neural Diffusion and Transformer

TL;DR

PointDifformer tackles robust 3D point cloud registration under noise and perturbations by marrying graph neural PDE-based feature learning with a heat kernel signature–augmented self-cross attention for robust correspondences. A weighted SVD with learnable weights estimates the rigid transformation from a compact set of correspondences. Extensive experiments on indoor and outdoor datasets (vReLoc, Boreas, KITTI) demonstrate state-of-the-art accuracy and superior robustness to synthetic and natural perturbations, highlighting practical applicability in autonomous driving and robotics.

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.
Paper Structure (29 sections, 26 equations, 10 figures, 15 tables)

This paper contains 29 sections, 26 equations, 10 figures, 15 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Point Cloud Registration Methods
  4. Point Cloud Feature Representation
  5. Neural Diffusion
  6. Heat Kernel Signature
  7. Point Cloud Registration Method Based on Graph Neural PDE
  8. Overview of PointDifformer
  9. Model Details
  10. Point Cloud Representation with Neural Diffusion
  11. Self-Cross Attention Embedding Based on Heat Kernel Signature
  12. Attention-Based Keypoint Correspondence
  13. Transformation Prediction
  14. Loss function
  15. Experiments
  16. ...and 14 more sections

Figures (10)

  • Figure 1: PointDifformer for point cloud registration. The details of the modules are provided in \ref{['subsect:model_details']}.
  • Figure 2: The Point-Diffusion Net based on the graph neural PDEs for the point cloud representation. The details are provided in \ref{['subsubsect:NDPCR']}.
  • Figure 3: The modules of the self-attention with heat kernel signature and the cross-attention.
  • Figure 4: Examples of point cloud frame pairs transformed using different prediction methods to align the second frame with the coordinate system of the first frame.
  • Figure 5: The empirical probability of relative translation (centimeter [cm]) and rotation (degree [$^{\circ}$]) errors on the vReLoc dataset.
  • ...and 5 more figures