Automated Ship Berthing Guidance Method Based on Three-dimensional Target Measurement

Abstract

Automatic berthing guidance is an important aspect of automated ship technology to obtain the ship-shore position relationship. The current mainstream measurement methods for ship-shore position relationships are based on radar, multisensor fusion, and visual detection technologies. This paper proposes an automated ship berthing guidance method based on three-dimensional (3D) target measurement and compares it with a single-target recognition method using a binocular camera. An improved deep object pose estimation (DOPE) network is used in this method to predict the pixel coordinates of the two-dimensional (2D) keypoints of the shore target in the image. The pixel coordinates are then converted into 3D coordinates through the camera imaging principle, and an algorithm for calculating the relationship between the ship and the shore is proposed. Experiments were conducted on the improved DOPE network and the actual ship guidance performance to verify the effectiveness of the method. Results show that the proposed method with a monocular camera has high stability and accuracy and can meet the requirements of automatic berthing.

Keywords

Automated ship / Automatic berthing / Berthing guidance / 3D measurement / Neural networks / Deep learning / Position estimation