PARTE: Plane-Assisted Robust Transformation Estimation for Point Cloud Registration

Notation

The main notation used throughout the paper is summarized in Table I . For a finite set $\mathcal{X}$, $/\mathcal{X}/$ denotes its cardinality. For a vector $x\in\mathbb{R}^{n}$, we use $\/x\/_{1}=\sum_{i}/x_{i}/$ and $\/x\/_{2}=(\sum_{i}x_{i}^{2})^{1/2}$ to denote the $\ell_{1}$ and $\ell_{2}$ norms, respectively.

For nonnegative vectors $h,g\in\mathbb{R}^{n}$, we define their $\chi^{2}$-distance as Superscripts $P$ and $Q$ identify quantities associated with the source and target point clouds, respectively.

Proposed Approach

PARTE follows the standard feature-based registration stages of feature extraction, descriptor matching, outlier rejection, and transformation estimation, as illustrated in Fig. 2 . Its distinction lies in how planar information is represented and integrated throughout these stages.

PARTE describes each extracted planar patch through the surrounding scene geometry using PCH, while retaining conventional FPFH point features for non-planar regions.

Experimental Evaluation

In this section, we evaluate PARTE at three different levels. First, we evaluate the proposed plane descriptor and matching procedure directly by measuring the quality of the resulting plane correspondences. We then evaluate the complete registration pipeline across indoor RGB-D, outdoor LiDAR, and partial-to-full registration problems.

Finally, we study the main design choices through ablation experiments and analyze the runtime of the complete pipeline. Datasets.

Sources and Demonstrations

Figures are reproduced from Babanazari et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.