Benchmarking Classical and Learning-Based Multibeam Point Cloud Registration

TL;DR

We address the challenge of registering bathymetric MBES point clouds by introducing the DotsonEast Dataset, a large semi-synthetic MBES benchmark built from AUV surveys in West Antarctica, and by systematically evaluating both classical and learning-based registration methods. The study benchmarks 2 classical methods and 4 learning-based approaches, revealing that learning-based models excel at coarse alignment in high-overlap scenarios while GICP provides robust, precise refinement even at very low overlap, supporting a two-stage registration strategy. Key contributions include the semi-synthetic dataset construction with ground-truth transformations, a diverse evaluation framework, and the first cross-family benchmarking of MBES registration methods with open-source data and code. The findings have practical impact for MBES-based SLAM and localization in challenging underwater environments, guiding the design of robust, hierarchical registration pipelines.

Abstract

Deep learning has shown promising results for multiple 3D point cloud registration datasets. However, in the underwater domain, most registration of multibeam echo-sounder (MBES) point cloud data are still performed using classical methods in the iterative closest point (ICP) family. In this work, we curate and release DotsonEast Dataset, a semi-synthetic MBES registration dataset constructed from an autonomous underwater vehicle in West Antarctica. Using this dataset, we systematically benchmark the performance of 2 classical and 4 learning-based methods. The experimental results show that the learning-based methods work well for coarse alignment, and are better at recovering rough transforms consistently at high overlap (20-50%). In comparison, GICP (a variant of ICP) performs well for fine alignment and is better across all metrics at extremely low overlap (10%). To the best of our knowledge, this is the first work to benchmark both learning-based and classical registration methods on an AUV-based MBES dataset. To facilitate future research, both the code and data are made available online.

Figures (6)

• Figure 1: Example MBES submap pair from the proposed DotsonEast Dataset . Each row showcases the predicted transformations (left), the consistency error of the map (middle) and the t-SNE van2008visualizing embedding of the feature descriptors (right). For methods without feature descriptors, the right column is left blank. The ground truth and null transforms are provided for comparison. The right column on the ground truth row shows the point cloud pair colored by depth. More details on the dataset and its metrics can be found in \ref{['sec:sec3-multibeam-benchmark']}.
• Figure 2: Kongsberg's Hugin AUV (orange) during the recovery of a survey in West Antarctica.
• Figure 3: AUV trajectory throughout the mission, as reported by the onboard INS system. The color of the trajectory represents the vehicle depth. The red rectangle highlights the test data segment, which is chosen to be physically furthest away from the training data.
• Figure 4: Map consistency metrics for decreasing overlap ratios. Note that the consistency and predicted overlap (%) are only computed for successful registrations. Specifically, the success rate (%) reported here is the rate at which the method returns a transform, regardless of its correctness. This is consistent with real-life AUV missions, where ground truth transformation is unknown. Transformation accuracy is evaluated in the subsequent metrics.
• Figure 5: Registration error metrics given decreasing overlap ratios. Benefiting from the synthetic ground truth transformation, the recall (%) here accounts for the cases where the predicted transformation is has an $RRE < 5 \degree$ and $RTE < 10m$. The RRE and RTE are only computed for the correctly recalled pairs.