PosDiffNet: Positional Neural Diffusion for Point Cloud Registration in a Large Field of View with Perturbations

Abstract

Point cloud registration is a crucial technique in 3D computer vision with a wide range of applications. However, this task can be challenging, particularly in large fields of view with dynamic objects, environmental noise, or other perturbations. To address this challenge, we propose a model called PosDiffNet. Our approach performs hierarchical registration based on window-level, patch-level, and point-level correspondence. We leverage a graph neural partial differential equation (PDE) based on Beltrami flow to obtain high-dimensional features and position embeddings for point clouds. We incorporate position embeddings into a Transformer module based on a neural ordinary differential equation (ODE) to efficiently represent patches within points. We employ the multi-level correspondence derived from the high feature similarity scores to facilitate alignment between point clouds. Subsequently, we use registration methods such as SVD-based algorithms to predict the transformation using corresponding point pairs. We evaluate PosDiffNet on several 3D point cloud datasets, verifying that it achieves state-of-the-art (SOTA) performance for point cloud registration in large fields of view with perturbations. The implementation code of experiments is available at https://github.com/AI-IT-AVs/PosDiffNet.

References in corpus (26)

• Adam: A Method for Stochastic Optimization
• PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation
• PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies
• Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework
• Deep Closest Point: Learning Representations for Point Cloud Registration
• LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping
• Learning general and distinctive 3D local deep descriptors for point cloud registration
• LoFTR: Detector-Free Local Feature Matching with Transformers
• D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features
• CoFiNet: Reliable Coarse-to-fine Correspondences for Robust Point Cloud Registration
• Geometric Transformer for Fast and Robust Point Cloud Registration
• PointDSC: Robust Point Cloud Registration using Deep Spatial Consistency
• Learning a Task-specific Descriptor for Robust Matching of 3D Point Clouds
• SpinNet: Learning a General Surface Descriptor for 3D Point Cloud Registration
• Beltrami Flow and Neural Diffusion on Graphs
• On the Robustness of Graph Neural Diffusion to Topology Perturbations
• RobustMat: Neural Diffusion for Street Landmark Patch Matching under Challenging Environments
• SC^2-PCR: A Second Order Spatial Compatibility for Efficient and Robust Point Cloud Registration
• Adversarial Robustness in Graph Neural Networks: A Hamiltonian Approach
• StickyPillars: Robust and Efficient Feature Matching on Point Clouds using Graph Neural Networks
• Lepard: Learning partial point cloud matching in rigid and deformable scenes
• Rotation-Invariant Transformer for Point Cloud Matching
• Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration
• HypLiLoc: Towards Effective LiDAR Pose Regression with Hyperbolic Fusion
• RobustLoc: Robust Camera Pose Regression in Challenging Driving Environments
• SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud