SG-LPR: Semantic-Guided LiDAR-Based Place Recognition

• DOI: 10.3390/electronics13224532
• Publication Date: 2024-11-19T11:06:54Z
• Authors: Weizhong Jiang; Hanzhang Xue; Shubin Si; Chen Min; Liang Xiao; Yiming Nie; Bin Dai
• Keywords: 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology

Place recognition plays a crucial role in tasks such as loop closure detection and re-localization robotic navigation. As high-level representation within scenes, semantics enables models to effectively distinguish geometrically similar places, therefore enhancing their robustness environmental changes. Unlike most existing semantic-based LiDAR place (LPR) methods that adopt multi-stage relatively segregated data-processing storage pipeline, we propose novel end-to-end LPR model guided by semantic information—SG-LPR. This introduces segmentation auxiliary task guide the autonomously capturing information from scene, implicitly integrating these features into main task, thus providing unified framework of “segmentation-while-describing” avoiding additional intermediate steps. Moreover, operates only during training, not adding any time overhead testing phase. The also combines advantages Swin Transformer U-Net address shortcomings current global contextual extracting fine-grained features. Extensive experiments conducted on multiple sequences KITTI NCLT datasets validate effectiveness, robustness, generalization ability our proposed method. Our approach achieves notable performance improvements over state-of-the-art methods.