A New Outlier Removal Strategy Based on Reliability of Correspondence Graph for Fast Point Cloud Registration (2205.07404v1)

Abstract: Registration is a basic yet crucial task in point cloud processing. In correspondence-based point cloud registration, matching correspondences by point feature techniques may lead to an extremely high outlier ratio. Current methods still suffer from low efficiency, accuracy, and recall rate. We use a simple and intuitive method to describe the 6-DOF (degree of freedom) curtailment process in point cloud registration and propose an outlier removal strategy based on the reliability of the correspondence graph. The method constructs the corresponding graph according to the given correspondences and designs the concept of the reliability degree of the graph node for optimal candidate selection and the reliability degree of the graph edge to obtain the global maximum consensus set. The presented method could achieve fast and accurate outliers removal along with gradual aligning parameters estimation. Extensive experiments on simulations and challenging real-world datasets demonstrate that the proposed method can still perform effective point cloud registration even the correspondence outlier ratio is over 99%, and the efficiency is better than the state-of-the-art. Code is available at https://github.com/WPC-WHU/GROR.

• The paper introduces GROR, a novel graph reliability-based method that robustly removes outliers in point cloud registration.
• The method leverages node and edge reliability measures with loose and tight constraints to achieve precise alignment even with extreme outlier ratios.
• Experimental results demonstrate significant improvements in rotation and translation accuracy while reducing computation time compared to state-of-the-art methods.

Overview: A New Strategy for Outlier Removal in Point Cloud Registration

Point cloud registration is a fundamental task in 3D data processing that involves aligning different point clouds into a common coordinate framework. A key challenge in this domain is managing the high outlier ratio produced during correspondence matching, which impacts the efficiency, accuracy, and recall rate of point cloud alignment. The paper "A New Outlier Removal Strategy Based on Reliability of Correspondence Graph for Fast Point Cloud Registration" addresses this challenge by proposing an innovative outlier removal strategy, referred to as Graph Reliability Outlier Removal (GROR).

Methodology

The core approach of GROR focuses on the reliability of correspondence graphs, which are constructed from given correspondences of point clouds. The method introduces the concept of node and edge reliability within these graphs to robustly identify and prune outliers during the alignment process:

1. Node Reliability for Optimal Selection: The method calculates the reliability degree of graph nodes by creating adjacency matrices for correspondence graphs. These matrices are based on geometric constraints between points, where the sum of elements in a row (or degree of a vertex) signifies node reliability. A high degree indicates a high-reliability node, and correspondences with the top reliability nodes are selected as optimal candidates.
2. Edge Reliability for Alignment: Two functions, termed as loose and tight constraints, are implemented. The loose constraint function evaluates the projection distance of nodes to the edge vector, facilitating quick elimination of some outliers. The tight constraint function further evaluates correspondence consistency and completes the 6-degree of freedom alignment via the maximum consensus set approach, using the edge as a reference axis for the final rotation.
3. Efficiency Enhancements: The algorithm accelerates computation by focusing on global maximum consensus parameters. It leverages a structured strategy that prioritizes calculations based on edge reliability determined under a loose constraint before validating under a tight constraint, significantly improving computational efficiency.

Experimental Evaluation

The proposed method is evaluated through extensive simulations and real-world datasets, demonstrating its robustness across various outlier ratios, including scenarios exceeding 99% outliers. The simulation experiments underscore its precision, with rotation and translation errors significantly lower than other state-of-the-art methods like RANSAC and Teaser++. Moreover, GROR consistently outperforms in terms of execution speed, crucial for processing large datasets inherent to real-world applications.

Implications and Future Directions

The method's robustness in handling high outlier ratios and its efficiency make it a valuable tool for real-time applications in fields such as robotics, computer vision, and urban mapping. Its performance on real-world datasets like ETH and WHU-TLS further confirms its practical applicability.

Future work could explore integrating GROR with advanced point feature descriptors boosted by deep learning models, potentially enhancing outlier rejection at the pre-processing stage. Additionally, adapting GROR to handle dynamic or continuously evolving point clouds could open new possibilities in real-time 3D scene reconstruction and environmental monitoring.

Conclusion

The research presents a sophisticated yet practical approach to outlier removal in point cloud registration, emphasizing reliability within the correspondence graph structure. GROR exemplifies an effective synthesis of geometric principles and robust statistical techniques, offering a compelling solution to the challenges posed by high outlier ratios in correspondence-based 3D registration tasks. Its application extends beyond traditional scenarios, indicating promising directions for future enhancements in the field of automated and scalable point cloud processing.