Correspondence-Free Multiview Point Cloud Registration via Depth-Guided Joint Optimisation

2025-06-18

AI Summary

Multi-view point cloud registration often fails to converge to the global optimum in complex scenes due to its reliance on explicit feature matching and hand-crafted data association. To address this, we propose a correspondence-free, depth-map-guided joint optimization framework: the global map is parameterized as a differentiable depth map, and camera poses of all frames and the depth map structure are jointly optimized within a nonlinear least-squares formulation, enabling implicit and dynamic data association. Our method eliminates conventional feature extraction and explicit correspondence estimation, instead leveraging raw depth observations as supervision for end-to-end consistent 3D reconstruction. Experiments on real-world datasets demonstrate that our approach significantly outperforms existing state-of-the-art methods, achieving substantial improvements in both registration accuracy and robustness—particularly under challenging conditions such as textureless regions, motion blur, and large viewpoint variations.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Multiview point cloud registration is a fundamental task for constructing globally consistent 3D models. Existing approaches typically rely on feature extraction and data association across multiple point clouds; however, these processes are challenging to obtain global optimal solution in complex environments. In this paper, we introduce a novel correspondence-free multiview point cloud registration method. Specifically, we represent the global map as a depth map and leverage raw depth information to formulate a non-linear least squares optimisation that jointly estimates poses of point clouds and the global map. Unlike traditional feature-based bundle adjustment methods, which rely on explicit feature extraction and data association, our method bypasses these challenges by associating multi-frame point clouds with a global depth map through their corresponding poses. This data association is implicitly incorporated and dynamically refined during the optimisation process. Extensive evaluations on real-world datasets demonstrate that our method outperforms state-of-the-art approaches in accuracy, particularly in challenging environments where feature extraction and data association are difficult.

Problem

• Multiview point cloud registration without correspondence
• Joint optimization of poses and global depth map
• Overcoming feature extraction challenges in complex environments

Innovation

• Correspondence-free registration via depth-guided optimization
• Joint estimation of poses and global depth map
• Implicit dynamic refinement during optimization