Spatial Graph Attentional Network Based Place Recognition with Visual Mamba Embedding

Abstract

Visual Place Recognition (VPR) plays a vital role in mobile robotics and autonomous navigation by retrieving reference images from a pre-established database. However, VPR systems frequently encounter performance degradation due to environmental variations. To overcome these challenges, we propose a re-ranking based VPR framework incorporating two key components: (1) A Visual Mamba Embedding (VME) module that optimizes spatial-channel feature interactions to generate discriminative global descriptors; and (2) A Spatial Graph Attentional Network (SGAN) that replaces conventional RANSAC-based verification with an efficient graph attention mechanism, improving matching accuracy while reducing computation. Comprehensive evaluations across multiple benchmark datasets demonstrate that the proposed method achieves superior performance compared to existing state-of-the-art methods, while maintaining advantages in computational efficiency and storage requirements.