Vision-Based UAV Navigation in Orchards**This work is supported in part by NSF Awards 1111638, 1525045 and the MnDrive initiative. The authors thank Professors Emily Hoover, Cindy Tong and James Luby from the Department of Horticultural Science, University of Minnesota for access to their orchards and useful discussions. We also thank Dr. Krishna Doddapaneni for his help with the experiments.

Abstract

Unmanned Aerial Vehicles (UAV) are becoming increasingly common in agricultural applications. Currently, they are primarily used to fly over fields in open space. Navigation inside orchard-like environments remains challenging. We study the problem of orchard navigation with cameras on an aerial vehicle. We study both the controller and the vision component. For the vision component, we provide two methods for detecting orchard rows with frontal facing cameras. In the monocular case, we present a pipeline to extract the geometry of tree rows when there is a well defined path structure. In the binocular case, we present a depth-based navigation algorithm to extract the rows. For the controller component, we design a controller that uses both frontal and downward facing cameras and provides reliable performance even on the presence of strong wind disturbances.