Addressing corner detection issues for machine vision based UAV aerial refueling.

摘要

The need for developing autonomous aerial refueling capabilities for an Unmanned Aerial Vehicle (UAV) has risen out of the growing importance of UAVs in military and non-military applications. The AAR capabilities would improve the range and the loiter time capabilities of UAVs. A number of AAR techniques have been proposed, based on GPS based measurements and Machine Vision based measurements. The GPS based measurements suffer from distorted data in the wake of the tanker. The MV based techniques proposed the use of optical markers which---when detected---were used to determine relative orientation and position of the tanker and the UAV. The drawback of the MV based techniques is the assumption that all the optical markers are always visible and functional. This research effort proposes an alternative approach where the pose estimation does not depend on optical markers but on Feature Extraction methods. The thesis describes the results of the analysis of specific 'corner detection' algorithms within a Machine Vision---based approach for the problem of Aerial Refueling for Unmanned Aerial Vehicles. Specifically, the performances of the SUSAN and the Harris corner detection algorithms have been compared. Special emphasis was placed on evaluating their accuracy, the required computational effort, and the robustness of both methods to different sources of noise. Closed loop simulations were performed using a detailed SimulinkRTM -based simulation environment to reproduce docking maneuvers, using the US Air Force refueling boom.