P2-LOAM: LiDAR Odometry and Mapping with Pole-Plane Landmark

For spatial perception, object-level SLAM (Simultaneous Localization and Mapping) has shown an advantage by leveraging semantic information to comprehend unknown environments. Poles are considered significant semantic landmark objects in urban roads and man-made constructions like stone pillars, street light lamps, and tree trunks. The rich pole landmarks enhance to the robustness and accuracy of SLAM. In this paper, we propose a LiDAR-based object-level SLAM named P-2-LOAM, which simultaneously estimates the pose and constructs a sparse pole landmark map. We propose a multi-RANSAC method for pole segmentation and estimate the parametric representation of pole objects in various scenes. Based on the segmented pole outcome, a coarse-to-fine data association for the pole object method is designed. Furthermore, a plane-assisted cost function for the pole landmark residual construction is developed. We demonstrate the accuracy and robustness of the proposed method in public datasets and real-world experiments.