Ship Pose Estimation for Autonomous Drone Landing Using IR Cameras and Neural Networks

Maritime surveillance plays a crucial role in various domains including security, environmental monitoring, and maritime traffic management. The use of unmanned aerial vehicle (UAVs) is emerging as a promising technol- ogy, given their faster movement speed compared to traditional surveillance vessel, reduce cost in maintenance, and the improve capability for quick deployment. Specifically, these aerial platforms have demonstrated efficacy in multiple maritime applications, ranging from ship detection to border patrol. Regarding the ship detection task, traditional methods of ship detection predominantly rely on onboard radar and optical sensors, witch face limitations such as weather dependence, illumination, and susceptibility to camouflage techniques. Infrared (IR) imaging, however, has emerged as a promising alternative due to its ability to detect thermal signatures of vessels regardless of weather conditions and time of day. In light of this fact, this thesis investigates the advantages derived from the use of IR camera imagery in the ship detection task, with a particular focus on detecting ships from a distance of up to 750[m], mimicking the approach procedure that a human pilot must adhere to when landing on a ship deck. In more detail, a novel neural network approach is designed to address short-distance detection, an aspect that is less explored in the existing literature.