GeoLocTrack: a scalable visual tracking system based on quasi-omnidirectional fiducial markers for edge-oriented digital twins

The realization of Digital Earth requires not only macro-scale geospatial mapping but also the micro-scale synchronization of massive dynamic entities (e.g. crowds, logistics). However, current solutions often face trade-offs between precision, coverage, and deployment cost. To cope with this gap, we propose GeoLocTrack, a cost-effective, edge-native system serving as the synchronization layer for Digital Twins. Distinct from traditional fiducial markers, our proposed Color-Texture Fiducial Marker (CTFM) supports ID mapping onto arbitrary geometries without altering physical structures. To decode these patterns, Texture-Preserving YOLO (TP-YOLO) integrates a Texture-Preserving Feature Pyramid (TP-FPN) and Texture-Guided Attention (TGA) to explicitly capture high-frequency synthetic texture details, ensuring robust multi-angle recognition. Via marker multiplexing and fusing 2D-3D tracking framework, Grouping-based Trajectory Similarity Tracking (GTST) eliminates reliance on active positioning hardware and achieves near real-time tracking. Alternatively, Hybrid Active-Passive Tracking (HAPT) achieves real-time tracking utilizing sparse passive localizations to correct active localization drift. Comprehensive validations across real-world VR and large-scale simulation experiments demonstrate that GeoLocTrack delivers decimeter-level accuracy (10-20 cm) and supports over 100 concurrent targets on commercial edge devices. Crucially, the system’s robustness has been verified through long-term field deployment in multiple VR scenes, offering a proven, ‘infrastructure-light’ solution for bridging the physical-digital divide.