HyperGCT: A Dynamic Hyper-GNN-Learned Geometric Constraint for 3D Registration

2025-03-04

AI Summary

In 3D point cloud registration, explicit consistency graph modeling is highly sensitive to noise and struggles to enforce geometric consistency among correspondences. To address this, we propose Dynamic Hyper-GNN—a novel framework that abandons conventional pairwise graphs and instead directly models high-order matching relationships via dynamic hypergraphs. Our method constructs learnable dynamic hyperedges and jointly aggregates features across vertices and hyperedges to end-to-end learn robust geometric constraints, enabling noise-resilient hypothesis generation. The key contribution is the first introduction of dynamic hypergraphs to point cloud registration, unifying high-order consistency modeling with geometric constraint optimization. Extensive experiments demonstrate state-of-the-art performance on 3DMatch, 3DLoMatch, KITTI-LC, and ETH benchmarks, with significant improvements in generalization capability and robustness to noise and outliers.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Graph-based Machine Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Geometric constraints between feature matches are critical in 3D point cloud registration problems. Existing approaches typically model unordered matches as a consistency graph and sample consistent matches to generate hypotheses. However, explicit graph construction introduces noise, posing great challenges for handcrafted geometric constraints to render consistency among matches. To overcome this, we propose HyperGCT, a flexible dynamic Hyper-GNN-learned geometric constraint that leverages high-order consistency among 3D correspondences. To our knowledge, HyperGCT is the first method that mines robust geometric constraints from dynamic hypergraphs for 3D registration. By dynamically optimizing the hypergraph through vertex and edge feature aggregation, HyperGCT effectively captures the correlations among correspondences, leading to accurate hypothesis generation. Extensive experiments on 3DMatch, 3DLoMatch, KITTI-LC, and ETH show that HyperGCT achieves state-of-the-art performance. Furthermore, our method is robust to graph noise, demonstrating a significant advantage in terms of generalization. The code will be released.

Problem

• Overcomes noise in 3D point cloud registration
• Leverages high-order consistency in 3D correspondences
• Dynamically optimizes hypergraph for accurate hypothesis generation

Innovation

• Dynamic Hyper-GNN for 3D registration
• High-order consistency in geometric constraints
• Vertex and edge feature aggregation optimization