Novel VSLAM positioning through synthetic data and deep learning: Applications in virtual archaeology, ArQVIA

Virtual archaeology enhances the visitor’s understanding of historical sites through 3D reconstructions of original locations. However, a major obstacle to interactive visualization is the VSLAM (Visual Simultaneous Localization and Mapping) problem, which revolves around accurately matching real-world positions with scaled virtual replicas. Current methods—such as marker recognition or image feature comparison against keyframes or point clouds—often involve high computational costs, complex setups, or software dependencies (embedded systems, GPUs, library compilations).

Our proposed method leverages Content-Based Image Retrieval (CBIR) to match RGB input images taken on-site with features learned from photogrammetric models. A single image is transmitted via a server-based system, and an easy-to-use request-response interface is guided by the gyroscope. A 360-degree panorama that is in line with the nearest camera perspective is returned after the match. This approach is demonstrated by the mobile app ArQVIA, which makes 3D visualizations accessible and interactive for any monocular device without the need for specialized hardware.