Study on ellipse fitting problem for vision-based autonomous landing of an UAV

In this paper, we studied for the ellipse fitting problem when an unmanned aerial vehicle (UAV) reaches near a landing marker for vision-based autonomous landing. Due to limited field of view of a camera, the entire shape of the landing marker cannot be captured as the vehicle approaches it. Also, depending on the vehicle's pose, the shape of the maker is shown distortedly. In order to overcome these problems, we designed the novel landing marker using concentric circles and the letter ‘H’. By calculating the conic parameters from portions of ellipse curves, we can estimate accurate relative navigation for the precise autonomous landing. Several basic image processing algorithms are employed to find the ellipse curves before estimating the ellipse parameters and the center of the marker using the direct least square method. From indoor experiments, we analyzed the ellipse fitting performance based on well-fitted and ill-fitted ellipses. Additionally, we studied how we can improve the ill-fitted case.