EPSC-LOAM: Feature Submap Based LiDAR Odometry and Mapping with Edge-Planar Scan Context for Mobile Robots

In Simultaneous Localization and Mapping (SLAM) systems for mobile robots, the implementation of loop closure detection can help correct cumulative errors and build a globally consistent map. This paper reports on a LiDAR-based SLAM framework that combines edge, planar feature submaps and loop closure detection with global descriptor. Based on the 6D pose transformation obtained from the registration between scans, the edge-based and planar-based submaps are constructed by fusing features from keyframes, respectively. Using the feature submap which has more feature information compared with the LiDAR scan helps increase robustness and the accuracy of the complete SLAM system. From the work of Scan context (SC) and Intensity Scan Context (ISC), a global descriptor, Edge-Planar Scan Context (EPSC), is constructed using the statistics of the feature information in the submap, exhibits a superior or comparable precision and recall rate. Results of experiments over public datasets and an actual college environment demonstrate the higher robustness and global consistency performance of mapping of the proposed framework against others LiDAR-based SLAM works.