Situational Awareness for Autonomous Interception Of Piracy Vessels Using Unmanned Surface Vessel

Abstract

Globally, a growing number of kidnappings executed by pirates occur each year. The current protection consists of mercenaries which are flown in by helicopter to protect merchant ships in threat areas. This is not only expensive, sometimes innocent fishermen are seen for pirates, in some cases with a lethal ending for the fishermen. An alternative could be in the form of an autonomous unmanned vessel, which intercepts potential pirates in a non-lethal manner. This thesis presents the design of a situational awareness system of such a vessel. The autonomous interception vessel utilizes radar and vision systems to obtain information about the environment. The camera detection algorithm makes use of a modified version of boolean map saliency to segment the target from the background. The radar and camera detections are passed to a global nearest neighbor data association algorithm. The states of the interception vessel are also estimated using sensor data using an EKF. This estimation is used to compensate for the self movement of the interception vessel in the estimation of the states of the target or targets. The target states are are tracked by an interacting multiple model framework, using extended Kalman filters. The algorithm is tested on scenario-specific simulations, which include a visual feed and sensor data. The simulated scenarios are case studies with staged pursuits of the interception vessel and its targets. In the evaluation, multiple algorithms and design parameters are tested and compared. Next to the design of the algorithm, some modifications and improvements to the existing literature are proposed. A modification to the boolean map saliency algorithm is made to make it less sensitive to target disappearance from the image frame. Also, a motion model is proposed, based on ship dynamics. It is shown that for the generated data set, this model obtained better performance in estimating the velocity and heading of the target. Finally, the measurements are obtained in polar coordinates, while the motion models are defined in Cartesian coordinates. The coordinate change is done using the converted measurement Kalman filter. To convert the single bearing obtained by the camera, the predicted range is used. In order to deal with the increasing uncertainty in the range prediction, a time dependency in the range variance is proposed.