Localization Method Using Camera and LiDAR and its Application to Autonomous Mowing in Orchards
Hiroki Kurita, Masaki Oku, Takeshi Nakamura, Takeshi Yoshida, Takanori Fukao
Abstract
This paper presents a localization method using deep learning and light detection and ranging (LiDAR) for unmanned ground vehicle (UGV) in field environment. We develop a sensor fusion algorithm that UGV recognizes natural objects from RGB camera using deep learning and measures the distance to the recognized objects with LiDAR. UGV calculates its position relative to the objects, creates a reference path, and then executes path following control. The method is applied to autonomous mowing operation in orchard. A mower is tracked by UGV. UGV needs to run along a tree row keeping an appropriate distance from tree trunks, by which the mowing arm of the tracked mower properly touches the trunks. Field experiments are conducted in pear and apple orchards. UGV localizes self position relative to trees and performs autonomous mowing successfully. The results show that the presented method is effective.