Visual Estimation of Relative Ship-Crane Movement From Point and Line Correspondences

Abstract

Abstract. This article presents a method where a monocular camera is used to estimate the relative movement between a stationary crane and a moving target vessel in a load transfer scenario. The proposed method is based on identifying points and lines on the deck of the target vessel with a camera mounted on the crane tip, and leverages point and line correspondences to estimate the relative position and rotation between crane and ship deck for use in the crane control task of landing a load on the moving ship deck. The estimation is done with a recently developed approach where a perspective-n-point-and-line (PnPL) technique is used for accurate pose determination, and where outlier rejection is achieved with dynamical pose estimation and graduated nonconvexity (GNC). This method for outlier rejection gives better results than the usual solution based on RANdom SAmple Consensus (RANSAC), especially for high-outlier rates. We validate our approach through extensive simulation and model tests where a model ship is moving with a realistic wave motion. The tests demonstrate high accuracy, robustness to outliers, and computational efficiency. The results show that the method can handle challenging scenarios, such as high-noise levels and up to 90% outliers, making it suitable for real-time marine applications.