Vision-Only Pose and Relative Distance Estimation for Leading Quadrotor UAV from Following UAV

Multi-rotor unmanned aerial vehicle (UAV) systems are widely utilized across various scenarios to enhance task flexibility and effectiveness. Common tasks include pursuing, intercepting and training. To increase efficiency and independent operation, particularly by reducing reliance on wireless communication and enhancing stability during cooperative control of systems, this paper introduces a vision-only and universal method for pose estimation and relative distance calculation. Firstly, key components of a UAV are detected through a two-step method. Secondly, with the detection of UAVs keypoints, the UAV pose and relative distance from the observer can be estimated through geometry. Flight experiments were conducted to evaluate the proposed approach in GPS-denied environments using a two-UAV leader-follower system. The results demonstrate a calculation error of less than 5%, i.e. less than a meter at relative distances of up to 20 meters, achieved at a processing speed of 30 milliseconds per frame. This signifies superior precision and processing speed compared to other methods of similar nature.