Fuzzy Community Geometry Adaptive Density Map Generation for UAVs Crowd Counting

Owing to effective and flexible data acquisition, drones, or general UAVs has been widely applied in crowd counting tasks, including crowd monitoring, traffic management, precision agriculture, aerial photography and situational awareness, etc. Density map generated by Gaussian kernel is widely used in crowd counting task. Most of the researchers focus on density map estimation, but neglect the density map generation methods. In this paper, we propose a point annotation based adaptive density map generation method and a well-performed framework which could satisfy higher-level crowd analysis. We verified our proposed method on 5 popular crowd counting datasets include DroneCrowd.