A follow-me algorithm for AR.Drone using MobileNet-SSD and PID control

Abstract

[en] In recent years the industry of quadcopters has experimented a boost. The appearance of inexpensive drones has led to the growth of the recreational use of this vehicles, which opens the door to the creation of new applications and technologies. This thesis presents a vision-based autonomous control system for an AR.Drone 2.0. A tracking algorithm is developed using onboard vision systems without relying on additional external inputs. In particular, the tracking algorithm is the combination of a trained MobileNet-SSD object detector and a KCF tracker. The noise induced by the tracker is decreased with a Kalman filter. Furthermore, PID controllers are implemented for the motion control of the quadcopter, which process the output of the tracking algorithm to move the drone to the desired position. The final implementation was tested indoors and the system yields acceptable results.