LiDAR Odometry and Mapping Based on Semantic Information for Maize Field

Abstract

Agricultural environment mapping is the premise of the autonomous navigation of agricultural robots. Due to the undulating terrain and chaotic environment, it is challenging to accurately map the environmental maize field using existing LOAM (LiDAR odometry and mapping) methods. This paper proposes a LOAM method based on maize stalk semantic features for 6-DOF (degrees of freedom) pose estimation and field mapping with agricultural robots operating in a dynamic environment. The piecewise plane fitting method filters the ground points for the complex farmland terrain. To eliminate the unstable factors in the environment, we introduce the semantic information of maize plants into the feature extraction. The regional growth method segments the maize stalk instances, the instances are parameterized to a line model, and the optimization method calculates the pose transformation. Finally, the mapping method corrects the drift error of the odometry and outputs the maize field map. This paper compares our method with the GICP and LOAM methods. The trajectory relative errors of our method are 0.88%, 0.96%, and 2.12%, respectively, better than other methods. At the same time, the map drawn by our method has less ghosting and clearer plant edges. The results show that our method is more robust and accurate than other methods due to the introduction of semantic information in the environment. The mapping of corn fields can be further used in precision agriculture.