Point Cloud Structural Similarity-based Underwater Sonar Loop Detection

TL;DR

The paper addresses loop closure in underwater SLAM with sparse MBES sonar data by introducing rotation-invariant, point-wise feature maps based on geometry, normals, and curvature, enabling direct 3D point-cloud comparison without 2D projection or learning. It demonstrates robust, training-free loop detection across deep-sea and freshwater environments, outperforming learning-based and BoW baselines on Antarctica and Seaward datasets. Ablation studies confirm the importance of square cropping and geometry-based features, and reveal that combining mean and variance across feature maps improves reliability. The approach offers practical impact by reducing preprocessing and vocabulary requirements while maintaining high detection performance in diverse bathymetric conditions.

Abstract

In this letter, we propose a point cloud structural similarity-based loop detection method for underwater Simultaneous Localization and Mapping using sonar sensors. Existing sonar-based loop detection approaches often rely on 2D projection and keypoint extraction, which can lead to data loss and poor performance in feature-scarce environments. Additionally, methods based on neural networks or Bag-of-Words require extensive preprocessing, such as model training or vocabulary creation, reducing adaptability to new environments. To address these challenges, our method directly utilizes 3D sonar point clouds without projection and computes point-wise structural feature maps based on geometry, normals, and curvature. By leveraging rotation-invariant similarity comparisons, the proposed approach eliminates the need for keypoint detection and ensures robust loop detection across diverse underwater terrains. We validate our method using two real-world datasets: the Antarctica dataset obtained from deep underwater and the Seaward dataset collected from rivers and lakes. Experimental results show that our method achieves the highest loop detection performance compared to existing keypointbased and learning-based approaches while requiring no additional training or preprocessing. Our code is available at https://github.com/donghwijung/point_cloud_structural_similarity_based_underwater_sonar_loop_detection.

Figures (6)

• Figure 1: Process of our proposed method. The inputs are sensor data from sonar, DVL, and IMU, and the output is the detected loop pairs.
• Figure 2: Process of square cropping. (a) Accumulated point clouds based on the estimated poses, (b) Cropped point cloud centered on the pose of vehicle. The colors of the point cloud were assigned arbitrarily based on the z-values of the points to facilitate visualization.
• Figure 3: Feature maps. Geometry (left), Normal (middle), and Curvature (right). Mean (top) and Variance (bottom). Each feature map, matched to each point in the point cloud, is shown as a 2D grid image for visualization purposes.
• Figure 4: Examples of a point cloud and feature maps. From the left, a raw point cloud, a query feature map, the true positive feature map, and the true negative feature map. (a) represents the 3D point cloud, while (b)-(d) depict the feature maps projected onto 2D images.
• Figure 5: The results of sonar-based loop detection methods. From top to bottom, the trajectories and predicted loop pairs for each method are shown in the datasets of the North River, Wiggles bank, and Antarctica. Green indicates true positive pairs, and red indicates false positive pairs.