Enhancing Indoor Positioning Accuracy with a Wi-Fi Fingerprinting-Based Transformer System

Indoor positioning within a floor poses significant challenges to researchers and industrialists. In this paper, we propose an indoor positioning system based on Wi-Fi fingerprinting and a transformer model in deep learning. Initially, we convert Wi-Fi samples into AP-RSSI range pairs. Subsequently, we construct a Location Space and generate spatially correlated and combined Wi-Fi samples as the model inputs to enhance the positioning accuracy. Furthermore, we introduce a novel spatial label smoothing method to improve model training efficiency and positioning precision. Experimental results show that our approach effectively enhances indoor positioning accuracy at a lower sampling cost compared to previous methods in the literature.