Using Assembled 2D LiDAR Data for Single Plant Detection

Abstract

A 2D laser scanner was mounted on the front of the small 4-wheel autonomous robot with differential steering, in an angle of 30 degrees pointing downwards. The machine was able to drive between maize rows and collect timestamped data simultaneously. The position of the vehicle was tracked by a highly precise total station. The data of the total station and the laser scanner was fused to generate a 3D point cloud. This 3D representation was used to search for single plant positions, what could later be used for additional applications like single plant treatment and precision weeding. First all points belonging to the ground plane were removed. Afterwards outliers were filtered. For separating the resulting points, a k-d tree clustering was used. Of each single point cloud cluster the 3D centroid was evaluated and assumed as the resulting plant position. This was done on three different growth stages of the plants. Results showed good detection rates up to 70.7 % with a root mean square error of 3.6 cm, precise enough to allow single plant treatment.