Autonomous Aerial Search Using a Particle Filter and Satellite Imagery

Unmanned aerial systems (UAS) have become ubiquitous in outdoor search missions, offering expansive and largely unobstructed views of vast areas. Generating efficient paths for these systems that minimize the expected target detection time and reduce real-time computational load is of significant value in the field. This work presents an offline path generation approach that uses segmented and classified satellite terrain imagery to seed a particle filter, which tracks target belief. We propose a variant artificial potential field (APF) to steer the agent. When compared with canonical search patterns on uniform, Gaussian, and terrain-seeded distributions, the proposed approach performs best when the initial target belief is estimated from terrain classification. When applied to realistic terrain environments, the APF approach achieved search completion times 6% and 17% faster (p