Autonomous robotic helicopter platform for reliable target tracking

·With the rapid advancement of computer hardware and software technology, . truly autonomous mobile robots are becoming a reality. Due to the constantly progressing miniaturisation of components, it is now possible to incorporate enough computing power into a package that can be carried by a miniature helicopter platform to perform useful autonomous missions. In contrast to full size helicopters, miniature rotorcraft are not only much cheaper, but can also be deployed in hazardous application scenarios. This thesis investigates novel techniques for Unmanned Aerial Vehicles (UAVs) in the areas of visual target tracking, three-dimensional path planning, and target reacquisition. The experimental testing platform of the Oxford Aerial Tracking System (OATS) consists of a petrol-powered helicopter model that is equipped for autonomous operation. In order to focus on high-level tasks, OATS uses an off-the-shelf automatic flight controller that stabilises the aircraft on six degrees of freedom and allows for waypoint navigation. Unlike previous devices of this type, OATS constitutes the first robotic helicopter to incorporate an active vision system that solely relies on onboard based vision processing. The key contributions of this thesis include the implementation of a visual object tracking algorithm that utilises Mean-Shift to reliably track moving objects on the ground. Building on this, a geolocalisation technique was developed that estimates the position of the tracked target within the Global Positioning System's reference frame in real-time. Furthermore, a novel path planning and collision avoidance technique is proposed that incorporates an automatic height change mechanism which enables the aerial robot to safely operate at low flight altitudes. In order to improve the system's target reacquisition capabilities during line-of-sight occlusion, a target trajectory prediction method is presented that employs artificial neural networks which adapt themselves to target movement patterns through learning. Results from simulation experiments as well as field experiments with the robotic helicopter demonstrate the functionality of the proposed system.EThOS - Electronic Theses Online ServiceGBUnited Kingdo