SGT-LLC: LiDAR Loop Closing Based on Semantic Graph With Triangular Spatial Topology

Abstract

Inspired by how humans perceive, remember, and understand the world, semantic graphs have become an efficient solution for place representation and location. However, many current graph-based LiDAR loop closing methods focus on extracting adjacency matrices or semantic histograms to describe the scene, which ignore a lot of multifaceted topology information for efficiency. In this letter, we propose a LiDAR loop closing method based on semantic graph with triangular spatial topology (SGT-LLC), which fully considers both semantic and spatial topological information. To ensure that descriptors contain robust spatial information while maintaining good rotation invariance, a local descriptor based on semantic topological encoding and triangular spatial topology is proposed, which can effectively correlate scenes and estimate 6-DoF poses. In addition, we aggregate local descriptors from various nodes in the graph using fuzzy classification to create lightweight database and efficient global search. Extensive experiments on KITTI, KITTI360, Apollo, MulRAN and MCD datasets prove the superiority of our approach, compared with state-of-art methods.