Answers are grounded in this paper and cite their sources.
This study proposes a camera-based relative position estimation technique for autonomous berthing of unmanned surface vehicles (USVs). The system estimates the relative distance and bearing angle from the USV to the berth using only a single onboard camera. Existing camera-based approaches for berth recognition have primarily relied on marker detection, which is vulnerable to light reflection and requires costly installation and maintenance. To overcome these limitations, this study adopts a 3D detection approach based on deep learning. A key challenge in applying deep learning is the need for large amounts of training data, which is often expensive and time-consuming to collect. To address this issue, we employed a 3D scene reconstruction technique to generate training data. First, images of the berth were captured using a camera mounted on a drone. Then, a 3D reconstruction of the environment was performed using Neural Radiance Fields (NeRF), enabling the generation of multi-view images of the berth as seen from various positions and angles from the perspective of the USV. These synthetic images were then used to train a Real-Time Monocular 3D Detection (RTM3D) model, which performs keypoint-based 3D object detection to estimate relative distance and bearing. To validate the proposed method, we conducted experiments using a camera mounted on the USV Aragon Cat 40-02 at the berth of Jebu Marina in Hwaseong-si, Gyeonggi-do, South Korea. The results confirm that the proposed method can successfully estimate the relative distance and bearing angle from the USV to the berth, demonstrating its potential for reliable, marker-free autonomous berthing.
No takes yet. Share an insight, caveat, or question.
Lim et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: