Efficient Feature Extraction and Localizability Based Matching for Lidar SLAM

Abstract

We proposed a lidar SLAM system for mobile robot running in feature sparse and degraded environments. The existing algorithms are affected by the unstable and similar features in environment which highly motivate our work. First, we analyze the segmented point clouds based on PCA to select the stable planner and linear features. Then, localizability based on Fisher information matrix is used to estimate the localization performance of robot and dynamically adjust the matching prameters. Finally, we carried out experiments in KITTI and environment containing parking lots, street and long straight corridor, which show the robustness and accurancy of our algorithm.