This work presents an indoor localization system based on Ultra-Wide Band technology, enhanced with a semantic inference layer through graph-based modeling. These systems, focused on the person using them, have high applicability in the context of the Internet of Things and the Internet of Everything, as they provide not only precise location but also contextual information through reliable room identification. The proposed methodology models the monitored space as an undirected graph, where the nodes represent rooms defined by bounding boxes, and the edges indicate physically valid transitions. This improves spatial coherence and reduces erroneous room changes. The solution was validated in the SmartLab of the University of Jaén, achieving a 94.08% accuracy in room classification compared to the classic algorithm, while maintaining real-time performance on low-cost hardware. The results confirm its potential as a robust, scalable, and context-aware localization layer for intelligent environments.