Improved Gaussian mixture modeling for accurate Wi-Fi based indoor localization systems

This paper deals with the problem of device’s location using Wi-Fi signal strength in an indoor environment. In order to improve the location accuracy a new fingerprinting-probabilistic algorithm is proposed. In this algorithm, the probability distribution of the Received Signal Strength (RSS) by a Mobile User (MU) from several Access Points (AP) is approximated using the Gaussian Mixture Model (GMM) approach. This probability distribution is then exploited to improve the location estimation of the mobile user. To tackle the initialization problem of the Expectation Maximization (EM) algorithm, which is used to estimate the Gaussian mixture parameters, a deterministic initialization method is proposed. This method named Manual initialization (MI) initializes manually the mixture parameters directly from the data. To cope with the MI shortcomings, an Adaptive Initialization (AI) technique is proposed. The performance of the resulting method, named the Improved GMM (IGMM) is evaluated experimentally and compared to that of other robust methods. The obtained results illustrate the efficiency of the IGMM method. Compared to the standard initialization technique, the MI and AI initialization techniques improve the location accuracy by 14.4% and 18.4%, respectively. Compared to other location methods, the improvement brought about by the IGMM method, varies from 5.1% to 17.7%, when the MI technique is used, and from 10% to 21.5%, when the AI technique is used.