LAUSR

Various methods have been developed for indoor localization. In particular, localization methods based on received signal strength indication (RSSI) fingerprinting can achieve accuracies of a few meters. A drawback that RSSI fingerprinting-based methods still suffer from is, in practice, the RSSI observed by certain devices commonly shows a more complex distribution rather than a single Gaussian distribution. In this paper, we propose an adaptive model recognition and construction (AMRC) method to adjust to the RSSI distribution characteristics for fingerprint-based methods, which improves the positioning accuracy and robustness of the localization system. The different components and factors that influence the performances of the proposed AMRC method are discussed, with experiments conducted in two indoor testbeds. The test results have demonstrated that the proposed AMRC method outperforms traditional localization methods due to more accurate modelling for the non-Gaussian RSSI distribution.