Vision-based qualitative path-following control of quadrotor aerial vehicle

Abstract

This paper presents a vision-based qualitative 3D navigation technique as well as first results of adapting Funnel Lane theory into path-following control of quadrotor aerial vehicle. The image's Kanade-Lucas-Tomasi (KLT) corner features are detected along the reference path in order to build a funnel lane for navigation. Then a funnel-lane navigation calculation is developed to estimate the desired yaw angle and height for the next movement. The proposed algorithm uses the front camera, heading measurement and altimeter of the Ar.Drone quadrotor for navigation. The remarkable advantage of the proposed technique is independently working in GPS-denied environments without the support of the external tracking system as well as computationally efficient. As compared to other available approaches, at-least one matched feature is required during path following. The proposed navigation technique can be implemented for visual-homing, visual-servoing and visual-teach-and-repeat (VT&R) applications. The proposed method is simulated in ROS and Gazebo simulator followed by a realtime experiment with the Ar.Drone quadrotor.