Steven Mills

Visual Guidance for Fixed-wing Unmanned Aerial Vehicles Using Feature Detection and Tracking: Application to Power Line Inspection

SteveMills 130pxSq
Queensland University of Technology
Supervisor (Academic)
Drs Luis Mejias & Jason Ford, QUT
Thesis Abstract

As the use of Unmanned Aerial Vehicles (UAVs) grows within the civilian sector, one application that is likely to attract the attention of industry is the inspection of infrastructure, in particular, those situated in rural and remote regions. Automating the process of data collection would appear to be a task well suited to the UAV and one that can draw upon years of research in areas of machine vision, guidance and control, and automated data processing. Fixed-wing UAVs can be expected to play a crucial role in this, particularly for tasks covering large areas, due to the platforms inherent efficiency and generous payload capabilities that directly contribute to long range.

Successful completion of these tasks introduces the challenge of performing guidance and control in a manner that establishes favourable conditions for data collection. While various tracking solutions exist, a common approach is to guide the vehicle directly over the feature that inevitably sees data collection controlled indirectly as a by-product of aircraft position. In particular, these solutions overlook sensor line-of-sight that is directly affected by aircraft attitude that varies as a result of rotation induced by manoeuvres used to maintain track. In the context of downward facing sensors that are likely to be fitted to fixed-wing UAVs, the impact is most evident through Bank-to-Turn manoeuvres that form the predominant means of altering heading.

Current solutions addressing these issues are limited and generally seek to address the problem through path planning and following that assumes knowledge of infrastructure location. Obtaining this information at a level of accuracy that can take advantage of these techniques however is not always possible. In this work, solutions are presented in the form of vision based control, offering realtime control capable of actively tracking infrastructure. Guidance and control is developed on the principal of providing ideal conditions for data collection from body-xed sensors, removing the need for gimballed mounts and thus alleviating payload requirements that are crucial on small UAV systems. Utilising Image Based Visual Servo (IBVS) techniques, data collection is controlled directly as viewed from an inspection sensor; a technique that is then extended to provide coverage as the UAV transitions between segments of locally linear infrastructure.

In the rst of two developments, Skid-to-Turn (STT) manoeuvres are utilised through an IBVS control design to view the feature at a Desired Line Angle, calculated as a function of Sensor Track Error, that allows recentring of the feature in one smooth motion. The second development augments the interaction matrix of a line feature with the aircraft equations of motion. This allows the design of an optimal state feedback controller that enables tracking to be performed through Forward-Slip (FS) manoeuvres. These manoeuvres are shown to improve tracking performance at reduced control effort compared to STT, while control through state feedback provides a direct means to suppress unwanted motion that could otherwise degrade data collection.

Another contribution is made to the direct management of data collection through an analysis of visual tracking in the presence of wind. To track a desired course in the presence of wind requires heading to be altered by a Wind Correction Angle. This presents an issue for visual control formed on a desired view of features that does not account for wind. The issue is investigated through the inclusion of a wind model in the interaction matrix, linking relative motion of image features with aircraft motion and wind. The effect of a steady wind disturbance is found to introduce a constant term in the interaction matrix and shown to be offset with desired line angle set to the Wind Correction Angle. 

A final contribution extends these developments to negotiating transitions between locally linear segments of infrastructure. Transitions present discrete changes in the direction of infrastructure that require a UAV performing inspection to alter course whilst ensuring continued data collection. Both the STT IBVS and FS IBVS developments are extended to this task, the rst using a smoothing feature to manage the transition, while the latter switches between features at a predetermined distance in the image frame. These provide separate solutions with variations in overshoot, time to recentre and maximum transition angle. 

Each of these developments is tested extensively through simulation, in an environment developed to generate imagery as would be captured during inspection, while allowing realistic test conditions including turbulence and wind gusts.