A fingerprint technique for indoor localization using autoencoder based semi-supervised deep extreme learning machine

In recent years, because of the growing demand for location-based services in indoor environments and the development of Wi-Fi, fingerprint-based indoor localization has attracted many researchers’ interest. In Wireless Sensor Networks (WSNs), fingerprint-based localization methods estimate the target location by using a pattern matching model for the measurements of the Received Signal Strength (RSS) from the
available transmitter sensors, which are collected by a smartphone with internal sensors. Due to the dynamic nature of the environment, the fingerprint database needs to be updated, periodically. Hence, it is better to add new fingerprint data to the primary database in order to update them. However, collecting the labeled data is time-consuming and labor-intensive. In this paper, we propose a novel algorithm, which uses high level extracted features by an autoencoder to improve the localization performance in the classification process. Furthermore, to update the fingerprint database, we also add crowd-sourced labeled and unlabeled data in order to improve the localization performance, gradually. Simulation results indicate that the proposed method provides a significant improvement in localization performance, using high level extracted features by the autoencoder, and by increasing the number of unlabeled training data.