Laser Odometry for Agricultural Environment Based on Point Cloud Intensity and Covariance

Abstract

Laser-based localization and mapping in the agricultural environment is challenging due to the unstructured scene with unstable point cloud features, high feature repetition, bumpy roads, and dynamic environmental objects. In order to solve these challenges, we proposed a laser odometry system based on point cloud intensity and covariance with modifications on LOAM. Aiming at the characteristics of less and unstable geometric feature extraction of point cloud in agricultural scene, we extract the intensity feature of point cloud to improve the accuracy of pose calculation. While in the geometric feature extraction module, we judge the extracted plane features strictly and discard the plane feature whose fitting degree is insufficient due to the irregular distribution of point clouds in the agricultural scene and the uneven ground. In addition, we dynamically measure the accuracy of point cloud matching by calculating the intensity covariance of source point cloud and target point cloud, and the robustness of the system is improved, too. Our system has achieved the state-of-the-art results on the KITTI and the recently released AgRob Vxx. The results show that this method is superior to the existing laser SLAM method.