DualReg: Dual-Space Filtering and Reinforcement for Rigid Registration

2025-08-23

AI Summary

Addressing the challenge of simultaneously achieving noise robustness, partial overlap handling, and real-time performance in rigid point cloud registration, this paper proposes a dual-space fusion registration paradigm: robust large-displacement matching in the feature space and high-precision local optimization in a geometric proxy space. We innovatively design an anchor-guided lightweight single-point RANSAC filter and formulate a co-designed objective function with a customized solver tailored for dual-space synergy. This approach seamlessly integrates the global robustness of feature-based matching with the fine-grained geometric alignment capability of local modeling. Evaluated on the KITTI dataset, our method achieves accuracy comparable to the state-of-the-art MAC while attaining up to 32× speedup on CPU inference—demonstrating a significant advance in balancing accuracy, robustness, and efficiency.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Motion & Tracking
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Security and Privacy: Large-scale security measurements
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Rigid registration, aiming to estimate a rigid transformation to align source and target data, play a crucial role in applications such as SLAM and 3D reconstruction. However, noisy, partially overlapping data and the need for real-time processing pose major challenges for rigid registration. Considering that feature-based matching can handle large transformation differences but suffers from limited accuracy, while local geometry-based matching can achieve fine-grained local alignment but relies heavily on a good initial transformation, we propose a novel dual-space paradigm to fully leverage the strengths of both approaches. First, we introduce an efficient filtering mechanism that incorporates a computationally lightweight single-point RANSAC algorithm followed by a refinement module to eliminate unreliable feature-based correspondences. Subsequently, we treat filtered correspondences as anchor points, extract geometric proxies, and formulates an effective objective function with a tailored solver to estimate the transformation. Experiments verify our method's effectiveness, as shown by achieving up to a 32x CPU-time speedup over MAC on KITTI with comparable accuracy.

Problem

• Improving rigid registration accuracy with noisy, partially overlapping data
• Combining feature-based and geometry-based methods for better alignment
• Achieving real-time processing efficiency in SLAM applications

Innovation

• Dual-space filtering with single-point RANSAC
• Geometric proxies from filtered correspondences as anchors
• Tailored solver for efficient transformation estimation