Loop Closure Detection (LCD) is an essential component of visual simultaneous localization and mapping (SLAM) systems. It enables the recognition of previously visited scenes to eliminate pose and map estimate drifts arising from long-term exploration. However, current appearance-based LCD methods face significant challenges, including high computational costs, viewpoint variance, and dynamic objects in scenes. This paper introduced an online appearance based LCD using local superpixel grids descriptor and dynamic node, i.e, LSGDDN-LCD, to find similarities between scenes via hand-crafted features extracted from LSGD. Unlike traditional Bag-of-Words (BoW) based LCD, which requires pre-training, we proposed an adaptive mechanism to group similar images called $\textbf{\textit{dynamic}}$ $\textbf{\textit{node}}$, which incrementally adjusted the database in an online manner, allowing for efficient and online retrieval of previously viewed images without need of the pre-training. Experimental results confirmed that the LSGDDN-LCD significantly improved LCD precision-recall and efficiency, and outperformed several state-of-the-art (SOTA) approaches on multiple typical datasets, indicating its great potential as a generic LCD framework.

翻译：闭环检测是视觉同步定位与地图构建系统的核心组件，能够识别先前访问的场景，消除长期探索过程中产生的位姿与地图估计漂移。然而，现有基于外观的闭环检测方法面临计算成本高昂、视角变化显著及场景动态物体干扰等重大挑战。本文提出一种基于局部超像素网格描述符与动态节点的在线外观闭环检测方法（LSGDDN-LCD），通过提取LSGD的手工特征实现场景相似性判定。不同于需要预训练的传统词袋模型闭环检测，我们提出了一种名为“动态节点”的自适应机制对相似图像进行分组，以在线方式增量调整数据库，无需预训练即可高效检索历史场景。实验结果表明，LSGDDN-LCD显著提升了闭环检测的精度-召回率与效率，在多个典型数据集上超越多种现有最优方法，展现出作为通用闭环检测框架的巨大潜力。