Vision-based deck state estimation for autonomous ship-board landing

This paper describes a method for ship deck estimation using only sensing carried aboard an autonomous rotorcraft: specifically, sensing is limited to a vision system, an inertial measurement unit and GPS. Using bearings to features on the ship deck and knowledge of helicopter state provided by the INS/GPS, a state estimator computes estimates of deck state and covariance. This deck state estimate can then be used to compute a safe, feasible trajectory to landing. This paper presents an Unscented Kalman Filter based implementation that uses a generic second order kinematic model driven by zero mean Gaussian noise for the ship deck motion model: while this deck motion model contains significant unmodeled dynamics it is not specific to a particular ship. Results of Monte Carlo simulations illustrate the utility of the proposed approach: good estimation results are obtained for stochastic deck motion (with a Pierson-Moskowitz power spectral density) and a fast ferry ship model.