GVPS: Global Visual Position System for Drones

Drones are contributing significantly to transportation, rescue, commercial, and safety purposes through their rapid technology development. Operating in an unknown outdoor environment is a stringent requirement in these applications, while usually Global Positioning System (GPS) is exploited to determine the global position of the drone. However, relying on an external input source is not a secure solution due to the hijacking possibilities. To tackle this problem, a positioning system that is independent of any external signals is proposed in this paper. This is accomplished in two steps of relative position estimation and global position derivation. First, the relative position of a UAV is estimated by a monocular camera. Since the estimated relative position has an unknown scale, an extended Kalman filter (EKF) is employed to fuse IMU data to that of the relative position. Next, the global position of UAV is derived by using satellite images in addition to the estimated relative position. The proposed algorithm is implemented on a custom made drone equipped with a Jetson TX2 GPU. Experimental results verify the effectiveness of the online implementation of the proposed algorithm in practice.