A Depth-Sensor-Free Next-Best-View Framework for Efficient ArUco Marker Recognition in Autonomous Underwater Vehicle Charging Operation

Abstract

The sustainability of the Internet of Underwater Things (IoUTs) relies on the continuous operation of autonomous underwater vehicles (AUVs), where ArUco markers serve as a fundamental interface for autonomous charging and data offloading. However, in docking scenarios involving torpedo-shaped AUVs, marker recognition frequently fails due to severe occlusion and geometric distortion, creating a bottleneck for reliable IoUT maintenance. To address this challenge, this study proposes a monocular next-best-view (NBV) planning framework that operates independently of depth information. By leveraging prior geometric knowledge of cooperative targets, the proposed method utilizes a single RGB image from a failed recognition attempt to generate a set of candidate viewpoints. A novel scoring function is introduced to quantify recognizability by comprehensively evaluating visibility, projection flatness, and self-occlusion constraints. Furthermore, an end-to-end network integrating a vision transformer (ViT) and U-Net is designed to predict scores for these candidates, effectively capturing both global context and local features. Both simulation and real-world experiments demonstrate that the proposed method significantly enhances real-time performance while maintaining high recognition accuracy, offering an efficient, low-cost active vision solution for AUV docking.