Image-based Visual-Servoing for Air-to-Air Drone Tracking & Following with Model Predictive Control

This paper proposes an algorithm for three dimensional image-based visual servo (IBVS) control of airto-air drone tracking and following applications. The control algorithm is based on Model Predictive Control (MPC). Using video feed from a monocular camera, the controller estimates the position and orientation of a target drone. The controller ensures that the target drone is always within the Field of View (FOV) of the camera by controlling the relative positions between the target drone and the observer drone. The challenge is to ensure that the velocity input to the observer drone is feasible. This objective is achieved by developing a model predictive controller, where the control inputs are bounded while keeping the target drone within the camera FOV. The model predictive control (MPC) performed better than PID controller, providing feedback to keep target drone within the field of view (FOV) by controlling the relative yaw positions between the target and observer drones. We have achieved visual tracking by MPC for the challenging curvy trajectories at 100% (0% of readings outside 80% tolerance) and 33% (67% of readings outside 20% tolerance).