Abstract
Fiducial markers provide a computationally cheap way for drones to determine their location with respect to a landing pad and execute precision landings. However, most existing work in this field uses a fixed, downward facing camera that does not leverage the common gimbal-mounted camera setup found on many drones. Such rigid systems cannot easily track detected markers, and may lose sight of the markers in non-ideal conditions (e.g. wind gusts). This paper evaluates April Tag and WhyCode fiducial systems for drone landing with a gimbal-mounted, monocular camera, with the advantage that the drone system can track the marker over time. However, since the orientation of the camera changes, we must know the orientation of the marker, which is unreliable in monocular fiducial systems. Additionally, the system must be fast. We propose 2 methods for mitigating the orientation ambiguity of WhyCode, and 1 method for increasing the runtime detection rate of April Tag. We evaluate our 3 systems against 2 default systems in terms of marker orientation ambiguity, and detection rate. We test rates of marker detection in a ROS framework on a Raspberry Pi 4, and we rank the systems in terms of their performance. Our first WhyCode variant significantly reduces orientation ambiguity with an insignificant reduction in detection rate. Our second WhyCode variant does not show significantly different orientation ambiguity from the default WhyCode system, but does provide additional functionality in terms of multi-marker WhyCode bundle arrangements. Our April Tag variant does not show performance improvements on a Raspberry Pi 4.
Abstract (translated)
URL
https://arxiv.org/abs/2203.10180
