Vision-based auto landing of a UAV

Unmanned aerial vehicles (UAVs) have recently become very useful in human's life. Unsuccessful landings or the danger of collision in landing is one of the problems of quadrotor UAVs. The goal of this paper is to present a precise and continues pose estimation method using monocular machine vision for a quadrotor to automatically land on a predefined place. For an accurate landing and to reduce the effects of existing delays in the drone's motion we propose an algorithm called "time slicing method", which divides the drone's moves close to the marker into smaller intervals called "movement" and "waiting". The time and the speed of the movements are proportional to the distance of the drone from the marker. The processing is parallel and of a minimum delay. Experimental results verify the success of our method and show that the drone can successfully land on the marker with an error of less than 3cm and in a time less than 15 seconds.