A binocular vision-based autonomous aerial docking system design for UAVs

In this paper，a binocular vision-based Unmanned Aerial Vehicle （UAV） autonomous aerial docking system is designed for the docking segment in autonomous aerial refueling mission，which requires high accuracy，efficiency and security. Firstly，a target detection module based on binocular vision and YOLOv4-Tiny network is proposed for target detection and position estimation，and data enhancement is performed on the drogue image dataset taken under various environments. Test results of the module indicate that it can perform drogue detection and positioning in multiple intricate circumstances，maintaining the depth estimation error within 2%. After that，to satisfy the control accuracy and robustness requirements of UAV controller during the docking process in the presence of strong external environment disturbance，dynamics-modelling and disturbance-observer based active anti-disturbance control law design is carried out on a quadcopter testing platform，and the designed control law has also been verified by simulation tests. Finally，flight experiments of autonomous aerial docking are conducted on the quadcopter testing platform. Experiments results show that the autonomous aerial docking system has a properly designed structure，which can maintain attitude stability under strong external wind disturbance，perform accurate drogue detection and positioning，and complete the docking mission successfully and efficiently. © 2023 AAAS Press of Chinese Society of Aeronautics and Astronautics. All rights reserved.