Localization for Ships during Automated Docking using a Monocular Camera

Abstract

Automating docking operations of ships requires at least two robust and precise localization systems for reaching the typically required safety and redundancy levels. Global navigation satellite systems are typically chosen as one of these, while the second needs to be independent and preferably use another measurement principle. Optical sensors are versatile, low-cost, and assumed to provide sufficient localization range. Used to simultaneously locate the vessel and map the harbor environment, this technology is believed to offer the necessary properties to complement navigation satellite systems in a resilient manner. In this paper, a monocular camera together with the state-of-the-art ORB-SLAM3 algorithm is used for localization. Umeyama's method is used to create an initialization procedure to determine the unknown scale factor encountered in monocular camera odometry and to find the transformation between the camera and a world-fixed coordinate frame. The proposed system is validated using data recorded on a commercial high-speed passenger ferry in nominal operation. The results indicate a localization range of more than 200 m. The mean absolute position error is less than 0.5 m with an estimated heading error of 0.5° in favorable weather conditions.