Airborne vision-based attitude estimation and localisation

Abstract

Vision plays an integral part in a pilot's ability to navigate and control an aircraft. Therefore Visual Flight Rules have been developed around the pilot's ability to see the environment outside of the cockpit in order to control the attitude of the aircraft, to navigate and to avoid obstacles. The automation of these processes using a vision system could greatly increase the reliability and autonomy of unmanned aircraft and flight automation systems. This thesis investigates the development and implementation of a robust vision system which fuses inertial information with visual information in a probabilistic framework with the aim of aircraft navigation. The horizon appearance is a strong visual indicator of the attitude of the aircraft. This leads to the first research area of this thesis, visual horizon attitude determination. An image processing method was developed to provide high performance horizon detection and extraction from camera imagery. A number of horizon models were developed to link the detected horizon to the attitude of the aircraft with varying degrees of accuracy. The second area investigated in this thesis was visual localisation of the aircraft. A terrain-aided horizon model was developed to estimate the position, altitude as well as attitude of the aircraft. This gives rough positions estimates with highly accurate attitude information. The visual localisation accuracy was improved by incorporating ground feature-based map-aided navigation. Road intersections were detected using a developed image processing algorithm and then they were matched to a database to provide positional information. The developed vision system show comparable performance to other non-vision-based systems while removing the dependence on external systems for navigation. The vision system and techniques developed in this thesis helps to increase the autonomy of unmanned aircraft and flight automation systems for manned flight.