Comparison of Point Cloud Registration Techniques on Scanned Physical Objects

This paper presents a comparative analysis of six prominent registration techniques for solving CAD model alignment problems. Unlike the typical approach of assessing registration algorithms with synthetic data sets, our study utilizes point clouds generated from the Cranfield benchmark. Point clouds are sampled from both existing CAD models and from 3D-scans of physical objects, introducing real-world complexities such as noise and outliers. The acquired point cloud scans, including ground truth transformations, are made publicly available. Comparison results show a high accuracy for GO-ICP, as well as PointNetLK, RANSAC and RPMNet combined with ICP refinement. However, except for GO-ICP, all methods show a significant number of failure cases on difficult scans, which include more noise or require larger transformations. FGR and RANSAC are among the quickest methods, while GO-ICP takes several seconds to solve. Finally, learning-based methods, although yielding good performances and being very fast, have difficulties in training and generalizing.