基于语义拓扑信息的物体级SLAM回环检测算法OA

Loop closure detection is a crucial component of SLAM systems,enabling the elimination of accumulated odometry errors.Traditional appearance-based methods face challenges in handling large viewpoint changes.This paper proposes a visual loop closure detection method based on semantic topological information.By leveraging the viewpoint invariance of object landmarks and encoding scenes through an object-oriented semantic topological map,the method significantly enhances system robustness under large viewpoint changes.Specifically,the method maintains a hierarchical semantic landmark database and adopts a"coarse-to-fine"detection strategy.First,high-level macroscopic object landmarks are utilized to extract topological graphs for coarse matching via local topological descriptors;to effectively eliminate mismatches,local and global topological graphs are unified in a polar coordinate system to evaluate spatial distribution similarity.Subsequently,guided by accurate object matches,fine registration is performed using low-level microscopic point landmarks to optimize pose estimation.Experimental results on the TUM and USTC datasets demonstrate that the proposed method exhibits superior performance in both precision and recall,achieving an average precision of over 80%.Notably,in large-disparity loop closure scenarios,positioning accuracy is improved by more than 40%.