Wi-Fi fingerprinting localization by graph-based similarity propagation

Abstract Wi-Fi fingerprinting localization is one of the mainstream indoor localization technologies, which uses the Wi-Fi received signal strength indicator (RSSI) as the signal feature of the reference points and compares the online RSSI with that to predict the location. To promote localization accuracy, various methods are proposed to build the fingerprint database using the spatial information of the reference points in the offline stage. However, traditional techniques usually compare the RSSI of unknown locations with the fingerprints individually, whose structure is ignored in the online stage. Inspired by the label propagation algorithm, in this paper, we propose Graph-based Similarity Propagation Localization (GSPL), which refines the distance metric (DM) iteratively with the similarities between the RSSIs at the online stage to improve localization accuracy. Based on the offline data defined as the known data, the problem is formulated as a semi-supervised learning problem treating the online RSSI as the labeled one, which could be solved by the typical information propagation method. Compared with the traditional distance metric, the proposed one considers the structures of the feature spaces benefitting the correct selection. In the experiments, the proposed algorithm is accessed on UJIIndoorLoc and museum datasets, on which the classification accuracy and positioning accuracy are significantly improved.