Aerial object tracking from an airborne platform

The integration of drones into the civil airspace is still an unresolved problem. In this paper we present an experimental Sense and Avoid system integrated into an aircraft to detect and track other aerial objects with electro-optical sensors. The system is based on a custom aircraft nose-pod with two integrated cameras and several additional sensors. First test flights were successfully completed where data from artificial collision scenarios executed by two aircraft were recorded. We give an overview of the recorded dataset and show the challenges to be faced with processing videos from a mobile airborne platform in a mountainous area. The proposed tracking framework is based on measurements from multiple detectors fused onto a virtual sphere centered at the aircraft position. To reduce false tracks from ground clutter, clouds or dirt on the lens, a hierarchical multi-layer filter pipeline is applied. The aerial object tracking framework is evaluated on various scenarios from our challenging dataset. We show that aerial objects are successfully detected and tracked at large distances, even in front of terrain.