On-board object tracking control of a quadcopter with monocular vision

Abstract

The development of an object tracking controller for a quadcopter using an on-board vision system is presented. Using low-cost components, a novel system is introduced that operates entirely on board the quadcopter, without external localization sensors or GPS. A low-frequency monocular computer vision algorithm is applied in closed-loop control to track an object of known color. Parallel PID controllers for aircraft bearing, relative height and range are implemented with feedback from object offset and size in the image frame. The noise exaggerated by measuring range from object pixel area is mitigated with a Kalman filter. Stable closed-loop tracking is demonstrated experimentally for all three control axes when tested individually and coupled together. Individual settling times were under 10 seconds and coupled control settling times under 25 seconds.