Incremental Multiview Point Cloud Registration

Xiaoya Cheng,Yu Liu,Maojun Zhang,Shen Yan

2024-07-06

Abstract:In this paper, we present a novel approach for multiview point cloud registration. Different from previous researches that typically employ a global scheme for multiview registration, we propose to adopt an incremental pipeline to progressively align scans into a canonical coordinate system. Specifically, drawing inspiration from image-based 3D reconstruction, our approach first builds a sparse scan graph with scan retrieval and geometric verification. Then, we perform incremental registration via initialization, next scan selection and registration, Track create and continue, and Bundle Adjustment. Additionally, for detector-free matchers, we incorporate a Track refinement process. This process primarily constructs a coarse multiview registration and refines the model by adjusting the positions of the keypoints on the Track. Experiments demonstrate that the proposed framework outperforms existing multiview registration methods on three benchmark datasets. The code is available at <a class="link-external link-https" href="https://github.com/Choyaa/IncreMVR" rel="external noopener nofollow">this https URL</a>.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper primarily focuses on addressing the problem of multiview point cloud registration, which is a fundamental and crucial task in 3D vision for achieving complete scene reconstruction. Specifically, the goal of the paper is to align all these point clouds into a unified coordinate system given a series of unordered point cloud scans captured from different perspectives, thereby constructing a complete and consistent 3D scene.

Traditional methods typically adopt a global approach to handle the multiview point cloud registration problem, i.e., first computing the relative transformations between point clouds and then estimating the absolute positions of the point clouds through graph optimization. Although effective, this approach may encounter challenges when dealing with noisy or complex input data. Additionally, due to the impact of accumulated errors, existing incremental methods may also lead to suboptimal final results.

To address the above issues, the authors propose a novel incremental framework, whose core steps include:

1. **Sparse Scan Graph Construction**: Establishing a sparse scan graph through matching techniques and geometric verification, which captures reliable correspondences between point clouds and removes incorrect matches.

2. **Incremental Scan Registration**: Selecting an appropriate initial pair of point clouds to start with, and then gradually adding new scans, aligning them to the existing model. During this process, Bundle Adjustment is utilized to reduce accumulated errors.

3. **Trajectory Refinement** (optional, mainly for detector-free matchers): For detector-free matchers like GeoTransformer, a keypoint refinement module is proposed to adjust the positions of keypoints on the trajectory to improve registration accuracy.

The experimental section demonstrates the superior performance of this method on three benchmark datasets (3D(Lo)Match, ScanNet, and ETH), proving that the method outperforms existing techniques in terms of accuracy, robustness, and completeness. Particularly, the method excels in handling data with low overlap rates and full scan graphs with a large number of incorrect connections.