Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FF-LOGO: Cross-Modality Point Cloud Registration with Feature Filtering and Local to Global Optimization

Nan Ma, Mohan Wang, Yiheng Han, Yong-Jin Liu

TL;DR

This paper tackles cross-modality point cloud registration by integrating deep feature extraction with precise optimization. The FF-LOGO framework first filters cross-modality feature correspondences with a geometric self-attention mechanism and isopycnic downsampling, producing a coarse pose. It then performs local adaptive key-region aggregation followed by a global modality-consistency fusion optimization to refine the pose, achieving a substantial recall improvement on the 3DCSR dataset (from 40.59% to 75.74%). The approach demonstrates strong generalization and practical viability, including a real-world robotics demonstration showing robust cross-modality localization with sub-centimeter accuracy. Overall, FF-LOGO advances cross-modality registration by coupling robust feature matching with principled local-to-global pose refinement.

Abstract

Cross-modality point cloud registration is confronted with significant challenges due to inherent differences in modalities between different sensors. We propose a cross-modality point cloud registration framework FF-LOGO: a cross-modality point cloud registration method with feature filtering and local-global optimization. The cross-modality feature correlation filtering module extracts geometric transformation-invariant features from cross-modality point clouds and achieves point selection by feature matching. We also introduce a cross-modality optimization process, including a local adaptive key region aggregation module and a global modality consistency fusion optimization module. Experimental results demonstrate that our two-stage optimization significantly improves the registration accuracy of the feature association and selection module. Our method achieves a substantial increase in recall rate compared to the current state-of-the-art methods on the 3DCSR dataset, improving from 40.59% to 75.74%. Our code will be available at https://github.com/wangmohan17/FFLOGO.

FF-LOGO: Cross-Modality Point Cloud Registration with Feature Filtering and Local to Global Optimization

TL;DR

This paper tackles cross-modality point cloud registration by integrating deep feature extraction with precise optimization. The FF-LOGO framework first filters cross-modality feature correspondences with a geometric self-attention mechanism and isopycnic downsampling, producing a coarse pose. It then performs local adaptive key-region aggregation followed by a global modality-consistency fusion optimization to refine the pose, achieving a substantial recall improvement on the 3DCSR dataset (from 40.59% to 75.74%). The approach demonstrates strong generalization and practical viability, including a real-world robotics demonstration showing robust cross-modality localization with sub-centimeter accuracy. Overall, FF-LOGO advances cross-modality registration by coupling robust feature matching with principled local-to-global pose refinement.

Abstract

Cross-modality point cloud registration is confronted with significant challenges due to inherent differences in modalities between different sensors. We propose a cross-modality point cloud registration framework FF-LOGO: a cross-modality point cloud registration method with feature filtering and local-global optimization. The cross-modality feature correlation filtering module extracts geometric transformation-invariant features from cross-modality point clouds and achieves point selection by feature matching. We also introduce a cross-modality optimization process, including a local adaptive key region aggregation module and a global modality consistency fusion optimization module. Experimental results demonstrate that our two-stage optimization significantly improves the registration accuracy of the feature association and selection module. Our method achieves a substantial increase in recall rate compared to the current state-of-the-art methods on the 3DCSR dataset, improving from 40.59% to 75.74%. Our code will be available at https://github.com/wangmohan17/FFLOGO.
Paper Structure (17 sections, 12 equations, 2 figures, 2 tables, 1 algorithm)

This paper contains 17 sections, 12 equations, 2 figures, 2 tables, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORKS
  3. Conventional Optimization Methods
  4. Deep neural network methods
  5. METHODOLOGY
  6. The Cross-Modality Feature Correlation Filtering Module
  7. The Local Adaptive Key Region Aggregation Module
  8. The Global Modality Consistency Fusion Optimization Module
  9. EXPERIMENT
  10. Implementation Details
  11. Parameters
  12. Dataset
  13. Evaluation Metrics
  14. Performance
  15. Ablations
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Overview of the proposed pipeline. The Cross-Modality Feature Correlation Filtering Module extracts and filters feature-correlated points, obtaining an initial pose estimation. Key regions identified by The Local Adaptive Key Region Aggregation Module are then optimized with the Global Modality Consistency Fusion Optimization Module to achieve the final optimized registration.
  • Figure 2: Application of FF-LOGO