Unmanned Aerial Vehicle (UAV) is especially interesting among aircraft due its holonomic constraints, in other words, UAV can be capable of moving on 6-Degrees of freedom resulting in the possibility of landing and taking off vertically using their rotors to translate on Z-axis without moving on X-axis, unlike fixed-wing planes. Position and orientation estimation has been the focus of much research in recent decades. Onboard sensors widely utilized for such tasks include, but not limited, IMU, GPS, and a camera. In addition, since UAVs can operate in areas where the GPS isn't always available [ab], a vision sensor can be an important element of any autonomous system, and because of its adaptability a lot of information can be collected from only image frames to help with scene interpretation, navigation and pose estimation. Due to the incapability of monocular vision to perceive depth using a single image frame without any extra knowledge of the captured image, research has recently been done to estimate pose using multiple camera arrangements that rely on stereo disparity principle[ac][ad], which refers to the pixel location discrepancy when comparing object on overlapped images captured by different cameras, where object triangulation can be done since the distance between lenses is a well-known parameter, while other studies focus on compute disparity between image sequences using monocular vision[ae] and how disparity evolves in time. Because both techniques increase processing costs due to the complexity of image comparison, another branch of monocular vision research the use of cues on the captured image, that is, a reference point which dimensions were previously measured, as results a ratio between pixels and world units can be obtained and then target depth and distance estimation can be computed by comparison. Cues are not limited to knowing reference size; it is also possible to know cues geometry or even have numerous cues arranged on a well-known pattern or geometry, and use such fact to estimate target position and orientation. Furthermore, in environments where orientation and position play a crucial role, such as UAV landing, where it is expected that the UAV knows at all moment its position regarding the landing area, which, could or not has a constrained axis, that is, a 6 degrees of freedom landing area where translation and rotation is possible, there exist the need to develop reliable estimation methods that use sensors that are available to the UAV. For that reason, this thesis addresses the difficulty of computing the UAV relative height and absolute orientation of an Unmanned Surface Vehicle (USV) due to sea state 6 conditions, based primarily on a single image capture of n beacons, that is signal lights, arranged as a circular pattern around the center of the USV's deck, being image captured by a camera mounted on the UAV, which is hovering above the USV. The final objective is to be able to accurately estimate the dynamic state of the USV so that the UAV can land on the USV autonomously in up to sea state 6 conditions.
This scenario is similar to landing on a dynamic terrestrial target, but with the added challenge of the sea's motion causing continuous USV translation and orientation.