.png){kind=icon}
.png){kind=icon}
Return
LCD-SSIC: enhancing object-level loop closure detection based on spatial-semantic information consistency
.png){kind=icon}
PRE
AI
DOI：10.1088/1361-6501/ae44b5.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Loop closure detection is pivotal for correcting drift in visual simultaneous localization and mapping (SLAM) but remains challenging in complex indoor environments with lighting variations and repetitive textures. To address the limitations of traditional feature-based methods in utilizing high-level semantics, this paper proposes Loop Closure Detection method based on Spatial-Semantic Information Consistency (LCD-SSIC), a coarse-to-fine object-level loop closure detection method leveraging spatial-semantic information consistency. The proposed approach effectively fuses visual appearance and depth information through a hierarchical strategy: deep global features (NetVLAD) are first employed for efficient candidate retrieval, followed by a fine-grained consistency check. This fine stage integrates geometric-semantic view probability, depth-derived 3D relative positioning, and color-based semantic features into a unified similarity metric to strictly verify spatial and semantic alignment. Extensive experiments on Technical University of Munich (TUM) RGB-D and ScanNet datasets demonstrate that LCD-SSIC outperforms state-of-the-art techniques, including ORB-SLAM3 and SemanticTopoLoop. The results show that our method achieves 100% precision and superior LCD-scores (e.g. 0.980 on TUM fr2-desk), proving its robust capability in distinguishing geometrically similar but distinct scenes.
Keywords:
Loop closure detection
Spatial-semantic information consistency
Visual SLAM
Object-level detection
Semantic feature fusion
Journal
IF:
3.4
Papers:
2.6K
Citations:
2.3W