Year: 2025
Type: article
Abstract: Visual positioning is a crucial technology for localization in GNSS-denied environments, forming the basis for scene understanding and accurate interaction. This paper proposes a graph-driven visual positioning framework that combines knowledge graphs with deep neural networks to determine the position of targets in images. The framework consists of three main steps. First, we construct a multimodal knowledge graph by integrating images, text, and structured data. Second, visual features are extracted using a pre-trained convolutional neural network. Third, a graph neural network is used to model the entities and relationships in the knowledge graph, enabling reasoning from visual data to spatial information. The proposed method leverages a multimodal knowledge graph with diverse entities and relationships, employs deep learning models to extract image features for semantic representation, and utilizes graph neural networks to process structured information for image localization. Experimental results show that the framework achieves higher positioning accuracy compared to traditional methods, with reduced errors across different distance ranges. This study provides a new approach to visual positioning and demonstrates the potential of integrating knowledge graphs into visual applications.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,390
Award: G807399062
OpenAlex ID: W7117580625