A Multiple UAV Path-Planning Approach to Small Object Counting with Aerial Images
Miles Krusniak, Alec James, Angel Flores, Yi Shang · 2021
UAV and object detection technology can automate object counting tasks, but low-altitude images, providing better inferences, require greater energy expenditure. Cooperative UAV teams can overcome this shortcoming, particularly when object density information is predetermined. We consider an aerial object-counting problem for a conservation use case: multiple quadrotor UAVs operating in a continuous region with a known probability distribution of objects (waterfowl). We build upon single-drone algorithms, generating density-based subregions covered by UAVs with unique roles. This approach scales with mission parameters, resulting in an F1 score improvement of more than 0.2 under various conditions compared to naive multiple-drone approaches.