Indoor Localization Using Bidirectional LSTM Networks

Abstract

Indoor localization witnessed the flourishing development in location based service for indoor environments. Regarding the availability of access points (AP) and its low cost for industry popularization, one of promising tool for localization is based on WiFi fingerprints. However, because of the interference of multi-path effects, the received signal strength data (RSS) are quite possibly to have fluctuated, thus they may result in propagation errors into localization results. In order to tackle this issue, We propose refined fingerprints based bidirectional long-short-term memory (bi-LSTM) neural network to learn the key features from the tested coarse RSS data, obtaining extracted trained weights as refined fingerprints(RFs). The extracted features of refined fingerprints are capable to demonstrate strong robustness with fluctuated signals and represent the environmental properties. The effectiveness of our bi-LSTM network is substantiated in the complex indoor environment, and accuracy is remarkably improved compared with our previous algorithm and other RSS-based approaches.