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

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges faced by cross - modality point cloud registration (Cross - modality Point Cloud Registration). Due to the inherent modality differences between different sensors, such as uneven point density distribution, outliers caused by different sensor precisions, and the inconsistency in pose and field of view of point clouds captured by different sensors, cross - modality point cloud registration becomes a difficult problem. Both traditional optimization methods and deep - learning - based methods have limitations when dealing with these challenges. The former is sensitive to initial pose estimation and outliers, while the latter is affected by noise and density changes in feature extraction. To overcome these problems, the paper proposes a framework named FF - LOGO, which combines feature filtering and local - global optimization techniques. Specifically: 1. **Feature Filtering Module**: Through the cross - modality feature association filtering module, geometric transformation - invariant features are extracted from cross - modality point clouds, and point selection is achieved through feature matching. 2. **Local - Global Optimization Module**: It includes a local adaptive key region aggregation module and a global modality consistency fusion optimization module. Through these two modules, local - global joint optimization is carried out, which significantly improves the pose estimation accuracy of the feature association and selection module. The experimental results show that the recall rate of FF - LOGO on the 3DCSR dataset has increased from 40.59% of the existing best method to 75.74%, significantly improving the performance of cross - modality point cloud registration. ### Main Contributions 1. **Proposed a new cross - modality point cloud registration framework**: Combined the feature - related filtering module and the local - global optimization module to solve the difficult problem of cross - modality point cloud registration. 2. **Introduced an optimization method from local to global**: Significantly improved the initial accuracy after feature filtering. 3. **Fully utilized the advantages of deep learning in fuzzy correspondence and the advantages of traditional optimization algorithms in pose fine - tuning**: Achieved a significant improvement from 40.59% to 75.74% on the 3DCSR dataset. ### Method Overview 1. **Feature - Related Filtering Module**: - Use the geometric self - attention mechanism to learn the features and global associations between point clouds. - Extract features through geometric structure embedding (including pairwise distance embedding and triple - angle embedding). - Select a high - confidence feature - related point set based on feature matching. 2. **Local Adaptive Key Region Aggregation Module**: - Select scattered and representative key points from the point set. - Use the K - nearest - neighbor algorithm to gather neighboring points around the key points to form a local adaptive key region point set. 3. **Global Modality Consistency Fusion Optimization Module**: - Match the feature - related point set with the local adaptive key region point set and calculate the point - to - plane residual. - Solve the optimal transformation pose by minimizing the residual. - Perform global key point least - squares optimization and integrate the adjustment of local modality consistency. ### Experimental Verification The paper conducted extensive experiments on the 3DCSR dataset, using rotation error (RE) and translation error (TE) as evaluation metrics. The experimental results show that FF - LOGO is significantly superior to existing methods in terms of recall rate and estimation error, especially achieving a significant improvement of 75.74% in recall rate. ### Conclusion FF - LOGO effectively solves the challenges in cross - modality point cloud registration by combining feature filtering and local - global optimization, providing new ideas and technical solutions for research in this field.

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

A Robust Multi-Local to Global with Outlier Filtering for Point Cloud Registration

CFI2P: Coarse-to-Fine Cross-Modal Correspondence Learning for Image-to-Point Cloud Registration

VRHCF: Cross-Source Point Cloud Registration via Voxel Representation and Hierarchical Correspondence Filtering

HybridFusion: LiDAR and Vision Cross-Source Point Cloud Fusion

Accurate Registration of Cross-Modality Geometry via Consistent Clustering

A Robust Multimodal Remote Sensing Image Registration Method and System Using Steerable Filters with First- and Second-order Gradients

Discriminative optimization algorithm with global–local feature for LIDAR point cloud registration

OLFF-Net: Robust Registration of 3D Point Cloud based on Overlapped Local Feature Fusion

Cross-Modal Information-Guided Network using Contrastive Learning for Point Cloud Registration

Image-to-Point Registration Via Cross-Modality Correspondence Retrieval

ModaLink: Unifying Modalities for Efficient Image-to-PointCloud Place Recognition

A Global-to-Local Algorithm for High-Resolution Optical and SAR Image Registration.

An overlap estimation guided feature metric approach for real point cloud registration

Fast Registration for cross-source point clouds by using weak regional affinity and pixel-wise refinement

MS-HLMO: Multiscale Histogram of Local Main Orientation for Remote Sensing Image Registration

IMFNet: Interpretable Multimodal Fusion for Point Cloud Registration

MS-HLMO: Multi-scale Histogram of Local Main Orientation for Remote Sensing Image Registration

A Systematic Approach for Cross-source Point Cloud Registration by Preserving Macro and Micro Structures

Optimization of the 3D Point Cloud Registration Algorithm Based on FPFH Features