Appearance-Based Visual-Teach-And-Repeat Navigation Technique for Micro Aerial Vehicle

The objective of this paper is to develop a vision-based navigation technique for micro aerial vehicles, quadrotor type, to operate in GPS-denied environment. The navigation method has been developed while using appearance-based Visual-Teach-andRepeat (VT&R) technique. In a teaching phase, a quadrotor is manually navigated along a desired route to collect a set of reference images. In a repeating phase, the quadrotor is able to autonomously follow the desired route using these reference images. Self-localization is developed to determine the current segment of the desired route by a number of SpeededUp Robust Features (SURF), matched between the current image and the reference images. In this paper, three methods of self-localization are proposed and T. Nguyen ( ) · G. K. I. Mann · R. G. Gosine · A. Vardy Intelligent Systems Lab, Faculty of Engineering and Applied Science, Memorial University of Newfoundland, A1B 3X5, St. John’s, NL, Canada e-mail: [email protected] G. K. I. Mann e-mail: [email protected] R. G. Gosine e-mail: [email protected] A. Vardy e-mail: [email protected] compared. After performing self-localization, the quadrotor computes appearance-based motion control commands (desired yaw and height) for the next movement in order to keep track of the desired route. This computation is developed on Funnel Lane theory, which was originally proposed in Chen and Birchfield (IEEE Trans. Robot. 25(3), 749–754 (2009)) for 2D navigation of a ground vehicle. The paper extends this theory to 3D navigation of a quadrotor. The proposed self-localization methods are tested with several image databases. Finally, an online experiment of proposed VT&R technique is demonstrated using Ar.Drone quadrotor model.