Deep Neural Network for Precision Landing and Variable Flight Planning of Autonomous UAV

The article is focused on autonomous unmanned aerial vehicle control for precise guidance to the ground landing target with variable creation of another flight plan. Object recognition is performed in real-time by a neural network using a camera located on Unmanned Aerial Vehicle (UAV). Object recognition is performed in the ground station with which the aircraft maintains a communication channel. The ground station computer evaluates the relative position of the aircraft with the position of the monitored landing field in the field of view of the image and after successful detection sends back flight instructions to the aircraft control unit.The neural network is pre-trained on landing patterns carrying additionally encoded information with flight instructions about the next waypoints of the flight plan according to which the drone performs an autonomous flight. The created neural network thus serves not only for precise landing, but also for finding the following points of the flight plan for a given aircraft.