Monocular-Vision-Based Positioning Method for UAV Formation

Abstract

Aiming at the issue of the relative positioning of unmanned aerial vehicles (UAVs) within formation in a GPS-suppressed environment, we propose a relative positioning method based on monocular vision. Firstly, such a method is used to identify UAVs online by a detection model and calculate their actual positions by only using visual information without auxiliary tags or other sensor information. Secondly, we improve the detection speed of this model by pruning redundant channels to meet the requirements of real-time detection when it is transplanted to the onboard computer with limited computing power. To adapt to this relative positioning method, we redesign an autonomous tracking controller with visual information as inputs. Finally, simulation experiments are carried out to verify the feasibility of the proposed method.