Automatic Navigation of Pesticide Spraying Vehicle for Orchard Tree Trunk Detection

Traditional Japanese orchards use nets to limit the height of fruit trees to make it convenient for farmers, but the low and dense branches are not conducive to the operation of large-sized machinery. A small automatic spraying robot can alleviate the problem of labor shortage in orchard. The dense canopy of the orchard not only blocks the GNSS signals. However, the lack of light reduces the accuracy of object recognition by normal RGB camera. In order to overcome these problems, Light Detection and Ranging (LiDAR) was chosen as a single sensor to implement a prototype robot navigation system in this study. In this study, the planning path of the vehicle was obtained by sequentially using various algorithms to process the LiDAR return points. First, density-based spatial clustering of applications with noise (DBSCAN) clusters the LiDAR return points to find the targets, the cones and the tree trunks, after that K-means divided the targets into left and right groups, and finally, random sample consensus (RANSAC) was calculated to get the left and right boundary lines, and the midline of the two boundary lines was the planning path for navigation. In field tests, the results show that the position root mean square error (RMSE) of this vehicle was as 13.8 cm in right turn and 11.4 cm in left turn. The system was able to calculate the path in real time based on the position of the cones and trees and guide the vehicle operate safely.