This research aims to create a 3D map with 3D boundaries of buildings within an operating range, which will be used for collision avoidance with buildings in autonomous driving of drones, UAVs (unmanned aerial vehicles), eVTOLs (electric vertical take-off and landing) aircraft, and air taxis. As a first step, a computer will be trained using deep learning to obtain 2D boundaries of target buildings corresponding to the slopes captured by cameras, etc., from a dataset of video and images captured by cameras such as those on drones. The learning results will then be evaluated by having the computer recognize the 2D boundaries in real time on test video. This research note reports on the selection of datasets performed in this first step, the deep learning methods using those datasets, and the evaluation of the learning results using test video.
