Loop Closure Detection With Reweighting NetVLAD and Local Motion and Structure Consensus

Dear Editor, Loop closure detection (LCD) is an important module in simultaneous localization and mapping (SLAM). In this letter, we address the LCD task from semantic aspect to geometric one. To end, a network termed as AttentionNetVLAD which can simultaneously extract global local features proposed. It leverages attentive selection for features, coupling with reweighting soft assignment NetVLAD via attention map features. Given query image, candidate frames are first identified coarsely by retrieving similar database hierarchical navigable small world (HNSW). As mainly summarize information of images lead compact representation, about spatial arrangement visual elements lost. provide fine results, further propose feature matching method motion structure consensus (LMSC) conduct verification between pairs. constructs neighborhood structures through consistency manifold formulates problem into optimization model, enabling linearithmic time complexity closed-form solution. Experiments on several public datasets demonstrate that LMSC performs well matching, proposed system yield satisfying results.