Victim Detection from a Fixed-Wing UAV: Experimental Results.

This paper outlines a method to identify humans from a low-altitude fixed-wing UAV relying on various visual and inertial sensors including an infrared camera. The work draws inspiration from the need to detect victims in disaster scenarios in real-time, providing needed aid to rescue efforts. Such work can also be easily employed for surveillance related applications. We start by pointing out various challenges from camera imperfections, viewpoint, altitude, and synchronization. We provide a pipeline to efficiently fuse thermal and visual aerial imagery for robust real-time detections. Confident detections are tracked across various frames and the real-time GPS locations of the victims are conveyed. Performance of our detection algorithm is evaluated in a real-world victim detection scenario from an autonomous fixed-wing aircaft.