A robust and accurate indoor localization system using deep auto-encoder combined with multi-feature fusion

Abstract

Many existing indoor localization systems use RSS as fingerprints, but RSS is a coarse-grained data, which not only fluctuates over time but also is not unique for a specific location due to rich multi-path effects and shadow fading in indoor environments. In order to further improve the localization accuracy, a robust and accurate indoor localization algorithm based on deep auto-encoder network combine with multi-feature fusion. To extract deep features hidden in CSI data and reduce the computational complexity of localization, an improved method is designed to achieve dimension reduction and feature extraction, which can avoid explicit information extraction of available CSI characteristics on the basis of effective representation of CSI fingerprint difference between different locations. Then, the fingerprint database is constructed by combination of RSS and CSI coding. Moreover, the Naive Bayer Classifier is adopted to improve the localization accuracy and stability. In the simulation experiment, the positioning effect of the proposed algorithm under different fingerprint libraries and the positioning effect under different positioning methods are mainly carried out. Experimental results show that the proposed method can effectively perform indoor positioning and has good practicability in the actual production environment.