Position Correlated Vision Dataset from Multirotor and Fixed-wing sUAS of General Aviation, Fixed-wing sUAS, Multirotor sUAS, and Birds

Onboard autonomy is needed for safe Advanced Air Mobility (AAM) flight operations to keep aircraft from colliding with each other and buildings. Despite decades of Sense and Avoid research in the aviation domain, the specific requirements from an aviation authority for widespread autonomous operations for small aircraft and small Unmanned Aircraft Systems (sUAS) does not exist. AAM researchers need to provide reproducible methodologies for developing collision avoidance systems. One promising sensor for Sense and Avoid systems is the camera as it is lightweight, data rich, and can provide high angular resolution of objects. In this work, we present a vision dataset collected from multirotor sUAS and fixed-wing sUAS where the target aircraft include multirotor sUAS, fixed-wing sUAS, General Aviation, and birds captured over seven days with varying flight conditions through thirty-one videos. A novel semi-autonomous data labeling technique is presented utilizing an object detection and tracking pipeline. Dataset includes onboard flight controller logs along with the computed separation distance between the ownship and intruder aircraft per frame.