Automatic landing of a low-cost quadrotor using monocular vision and Kalman filter in GPS-denied environments

Abstract

Unmanned aerial vehicles are becoming an important part of the modern life. Despite some recent advancesin GPS-aided navigation of quadrotors, the concern of crash and collision still overshadows their reliability and safety,especially in GPS-denied environments. Therefore, the necessity for developing fully automatic methods for safe, accurate,and independent landing of drones increases over time. This paper investigates the autolanding process by focusing on anaccurate and continuous position estimation of the drone using a monocular vision system and the fusion with the inertialmeasurement unit and ultrasonic sensors’ data. An ARUCO marker is used as the landing pad, and the information isprocessed in the ground station through a real-time Wi-Fi link. In order to overcome the closed loop instability caused bythe communication and localization delays, we propose a method called ”movement slicing method”. This method dividesthe moves around the marker into moving and waiting slices and makes the landing process not only more accurate butalso faster. Experimental results show a successful landing of the UAV on a predefined location, while it is accuratelyaligned with the marker using the proposed method.