Lidar Mapping Optimization Based on Lightweight Semantic Segmentation

Lidar Simultaneous Localization and Mapping (LiDAR-SLAM) algorithm with semantic information is an open research question and it is such a time consuming task. The related work that focus on real-time LiDAR-SLAM has poor accuracy. To solve these problems, a lightweight semantic segmentation network to assist in localization and mapping is proposed in this paper. The method uses the lidar point clouds generated by the simulator and annotated manually in real world as the original input. Then, the semantic cloud is segmented by the semantic segmentation network to obtain the semantic information. Finally, the semantic information obtained by the segmentation is used to assist the localization and mapping.