LiDAR mapping using point cloud segmentation by intensity calibration for localization in seasonal changing environment

Abstract

In this research, we propose a LiDAR mapping method for autonomous robot localization using point cloud segmentation by intensity calibration in agricultural environment with changing seasons. This method segments the tree trunk point cloud using calibrated intensity values correlated with the reflectance characteristics of objects in an agricultural environment. By incorporating calibrated intensity, this research improves the mapping performance of LeGO-LOAM with intensity calibration in seasonal changing agricultural environments, resulting in maps with strong robustness to environmental variations. This research also conducts localization tests on established intensity-calibrated maps, it compares the localization performance using raw LiDAR data from one season on maps generated by other different seasons. The localization accuracy in RMSE of intensity-calibrated maps for all seasons fluctuated within 0.17 m. The results demonstrate that LiDAR mapping by intensity calibration is robust to seasonal changes in agricultural environments and has lower localization errors compared to other maps, showcasing the feasibility of this method. • Invariant trunk objects were extracted using different reflectance correlated with calibrated intensities. • Developed a LiDAR mapping method using intensity calibration to create robust trunk maps to the environment. • Demonstrated that intensity-calibrated maps significantly improve localization accuracy across different seasons.