(The) development of vision and ultrasonic beacon fusion localization of UAV for indoor airplane inspection

Abstract

Recently, autonomous aircraft inspection in a hangar by drones has been highlighted because it can reduce inspection time and prevent accidents. Although, for indoor drone navigation, trilateration based RTLS (Real-time Localization System) is widely used, it is prone to largely drift in a hangar environment due to interrupted LOS (Line of Sight) by aircraft structure. Another candidate solution is visual localization method, but it suffers from lack of feature points in near distance in hangar environment. In this research, fusion of visual odometry and ultrasonic RTLS by means of graph optimization and outlier rejection is proposed as an alternative to either RTLS or vision only. The proposed system compensates visual odometry error with ultrasonic position and rejects outlier ultrasonic position based on fused position. This algorithm provides more accurate localization than visual odometry only as well keeps estimating position in absence of RTLS data. In this thesis, at first, camera type, visual odometry algorithm and directions of view in a hangar are evaluated and suitable system is selected. In hangar experiments, it is shown that fusion localization solution which is robust to drift of VIO and LOS interrupt of RTLS. Real-time speed is implemented and real-time performance is evaluated in a hangar experiment. From flight test in large open space similar to a hangar environment, feasibility of this algorithm in real world is verified.