Fiducial Marker-Based Autonomous Landing Using Image Filter and Kalman Filter

Autonomous take-off and landing are essential to perform continuous missions of unmanned aerial vehicles (UAVs) using drone stations. Generally, this is impossible to land on the drone station only using a global positioning system (GPS) because of sensor errors. This paper deals with the vision-based autonomous precision landing of UAV using fiducial markers. To improve landing accuracy, an image sensor attached to the bottom of the UAV and a ground-based fiducial marker is commonly used. However, if the fiducial marker is not recognized in the image frame, the guidance value cannot be calculated and it is impossible to land on the fiducial marker. To overcome this, we applied an image filter and Kalman filter to the algorithm for fiducial-marker recognition, and we used the calculated position and orientation data of fiducial markers as landing guidance values. To verify the developed algorithm and landing stability, a software-in-the-loop simulation and flight test were performed, resulting in a landing position error up to a radius of 2.70 cm with heading root mean square error (RMSE) of 1.4393 & DEG; in simulation, and within a radius of 9.470 cm with a heading RMSE of 2.3152 & DEG; in flight tests with disturbances. Thus, we confirmed that a stable and accurate landing is possible using the proposed algorithm and fiducial markers.