Visual inertial odometry for multirotors
Multirotor drones are widely used as sensor platforms. In many cases the drone itself does not need to move large distances but provide a persistent aerial platform. For these applications, one solution is a tethered drone that is connected to a power source on the ground. Such a product is developed by CAFA Tech.
The drone needs automatic position hold system to stay above the power source. However, in many applications GPS, which is the de facto standard for outdoor localisation, can not be used due to signal jamming or spoofing. Therefore, an alternative positioning method was needed.
The thesis provides a theoretical overview of the computer vision and sensor fusion methods used in state-of-the-art algorithms. Based on this knowledge, I proposed an application specific visual inertial odometry method that utilises the geometry and hardware of the given problem. This results in a method that is computationally more efficient than general state-of-the-art algorithms and more tolerant to noise due to additional physical constraints.
The thesis has a publication restriction but the abstract is published here.