Refinement and Characterization of Feature-Based Vision Algorithm for Ship-Deck Landing under Degraded Visibility

The objective of this work is to refine and experimentally characterize a two-dimensional, feature-based vision algorithm for tracking a stochastically moving ship-deck under degraded visual conditions. A 2.75-kg quadrotor UAV (unmanned aerial vehicle), with only the accessories essential for vision-based navigation, is specifically designed to establish the minimal system requirements, fabricated, and tested in-house. The algorithm was integrated into this quadrotor and was performance tested in controlled, hand-held tests as well as piloted free-flight hover. All results were validated against Vicon ground-truth data. In the controlled tests, the algorithm was first used to estimate the motion of a Stewart platform simulating a stochastically moving ship-deck. Next, tests were conducted under visually degraded conditions, specifically glare, low illumination, and occlusion of the landing pad. Free-flight tests were conducted with the quadrotor hovering above the landing pad at varying levels of illumination and occlusion as well as with ship-deck motion. The algorithm could accurately estimate the pose of a ship-deck undergoing Sea-state 6 motions in visually degraded conditions in both hand-held and free-flight tests. Performance was observed to reduce slightly in free-flight compared to hand-held tests due to aircraft motion and vibration.