Vision-based Object Tracking Guidance of a Multirotor UAV using Deep Reinforcement Learning

Abstract

The recent market demands on multirotor Unmanned Aerial Vehicles (UAVs) include various industrial and military fields, such as aerial photography, asset monitoring, delivering, farming, structural inspections, surveillance and reconnaissance, search and rescue, and so forth (Canis 2015). In these demands, most UAVs are required to capture objects of interest, which are fixed or moving in various environments. In order to capture the object clearly and confidently, UAVs also need to fly near the target, which requires autonomous target following capability when the pilot's workload needs to be alleviated.