PCR-99: A Practical Method for Point Cloud Registration with 99 Percent Outliers

TL;DR

A robust method for point cloud registration that can handle both unknown scales and extreme outlier ratios, and outperforms the state of the art for both known-scale and unknown-scale problems.

Abstract

We propose a robust method for point cloud registration that can handle both unknown scales and extreme outlier ratios. Our method, dubbed PCR-99, uses a deterministic 3-point sampling approach with two novel mechanisms that significantly boost the speed: (1) an improved ordering of the samples based on pairwise scale consistency, prioritizing the point correspondences that are more likely to be inliers, and (2) an efficient outlier rejection scheme based on triplet scale consistency, prescreening bad samples and reducing the number of hypotheses to be tested. Our evaluation shows that, up to 98% outlier ratio, the proposed method achieves comparable performance to the state of the art. At 99% outlier ratio, however, it outperforms the state of the art for both known-scale and unknown-scale problems. Especially for the latter, we observe a clear superiority in terms of robustness and speed.

Figures (3)

• Figure 1: The Bunny dataset bunny used in our evaluation. The inlier correspondences are shown as the thick green lines and the outliers as the thin red lines. Here, the outlier ratio is 96%. Image credit: sra2.
• Figure 2: Evaluation results for unknown-scale problems (500 Monte Carlo runs): RANSIC and ICOS are not robust at 99% outlier ratio when the computation time is limited to 100 seconds. In contrast, PCR-99 (with or without sample ordering) is robust up to 99% outlier ratio. The sample ordering has a very small impact on the accuracy and robustness of PCR-99. Comparing the median times of the two versions of PCR-99, the random sampling approach is faster up to 98% outlier ratio, but slower at 99% outlier ratio.
• Figure 3: Evaluation results for known-scale problems (500 Monte Carlo runs): TriVoC, VODRAC, and PCR-99 (with or without sample ordering) are more robust than the other methods, producing fewer large rotation errors (e.g., larger than $10^\circ$). They exhibit similar levels of robustness across all outlier ratios. For PCR-99, the sample ordering boosts the speed significantly at high outlier ratios, without causing a noticeable change in the accuracy or robustness. Among all methods, PCR-99 is the fastest one at 99% outlier ratio.