Towards a system for analyzing accidents of unmanned aerial vehicles

Our goal was to analyze accidents of unmanned aerial vehicles (UAVs) by replicating their circumstances, using data obtained from the sensors and flight recorder installed on the UAVs. In this paper, we investigated the performance of three tools for 3-D mapping to reproduce the surrounding environment along the path of the UAV, and found that LIDAR data can construct more accurate and broader maps than methods that use monocular or stereo camera images. We then applied an optical flow method to images taken by a rotating monocular camera and found that imaging at more than 120 fps is appropriate for accurate motion tracking of a spinning and falling UAV. Finally, we developed a visualization system that displays replications of UAV flights and its surrounding environment on a computer screen.