Visual Controllers for Indoor Navigation
This work aims at developing visual controllers that can work at moderately high speed so a UAV can navigate in a GPS-denied environment. The idea behind this approach is such that decouples the rotational from translational degrees of freedom using an intermediate decay function between the rotational and translation components of the control law.
Navigation Circuit: Rectangle
Reinforced Deep Learning Approach to Visual Servoing
Straight Line Navigation
Navigation Circuit: Triangle
The aim of this project is to investigate automated vision-based aircraft collision warning technologies. We have developed and tested algorithms that detect and avoid other aircraft using a machine vision camera. Several experiments have been conducted to replicate collision scenarios using two Cessna and a UAV vs Cessna.
More info. Here
Example detection on real images
Power Line Inspection and Navigation
This main objective of this research is to extend the aircraft path planning, data capture and flight assist capabilities of a highly automated aircraft to enable automated and optimised flight management and planning for the purposes of large-scale infrastructure aerial surveys.
The intended outcome of which is a capability that can be applied to both highly automated manned and unmanned aircraft to contribute to a direct reduction in aerial survey costs and optimise data capture and decision making under a wide range of conditions. More info
Automated Emergency Landing System
This project is developing novel visual detection, control and planning algorithms that can land an aircraft in the event of an onboard failure. We focus on technologies that can be used by manned and unmanned aircraft. The approach includes site detection algorithms to visually identify areas on the ground suitable for landing, path planning and guidance approaches that account for environmental disturbances (e.g wind), decision making algorithms to assess multiple attributes internal and external to the aircraft and choose the optimal landing area on the ground as well as the best approach trajectory and finally fault detection & identification algorithms to identify the type of onboard failure. More info.
Marine Mammal Detection
This research is investigating approaches to automatically detect dugongs in aerial images captured using a customised payload onboard an Unmanned Aircraft. Aerial surveys generate many thousands of still images that require processing post-survey to record marine mammal detections. Several methods are being investigated and developed drawing techniques from pattern recognition and machine learning fields to process images automatically. Further info:
Mejias, Luis, Duclos, Gwenael, Hodgson, Amanda, & Maire, Frederic D. (2013) Automated marine mammal detection from aerial imagery. In Proceedings of OCEANS ’13 IEEE/MTS, Town and Country Resort Hotel, San Diego, CA. pp. 1-5.
Maire, Frederic, Mejias, Luis, Hodgson, Amanda, & Duclos, Gwenael (2013) Detection of dugongs from unmanned aerial vehicles. InProceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems 2013, Tokyo Big Sight, Tokyo. pp. 2750 – 2756.
Forced Landing (Automated Emergency Landing) – 2005
Some preliminary work I did on the use of computer vision techniques for forced landing of UAVs. This application was intended for UAVs that conducting a task (in case of power line inspection) and require to perform an emergency landing avoiding the power lines.
Stereo Visual Odometry for UAVs – 2006
Traditionally, unmanned aircraft rely on their onboard GPS/INS sensor to localise themselves. A loss of GPS signal can severely impact aircraft navigation. This research deals with the use of cameras to estimate platform motion.