Multi-view Geometry and Deep Learning Based Drone Detection and Localization

A real-time vision based detection and 3D pose estimation of intruder Unmanned Aerial Vehicles (UAVs) is presented. Proposed method utilizes a cooperative team of UAVs mounted with monocular camera to detect the intruder UAV and simultaneously localize it. Detection is performed by using YOLOv2 deep learning network architecture while the position is obtained using multi-view geometry. General dynamics of the drone are considered to improve the target position estimate obtained from the image based estimate.