Open Space Indoor Navigation with RSS Sequence Transduction Neural Networks

Abstract

In conjunction to the extensive research conducted in outdoor positioning and navigation systems, indoor localization and navigation techniques have been gaining interest and attention with various technologies proposed in an attempt to achieve high positioning accuracy. Out of many proposed techniques, the received signal strength (RSS) fingerprinting-based approach remains widely adopted for indoor localization. One of the key challenges facing this approach is navigating an open space indoor location where the signal strength may vary indifferently from one position to the subsequent one. In this work, we propose a transduction neural network approach that takes as an input a sequence of past RSS fingerprints and accordingly predicts the current location, navigates the movement, and even predicts the future positions. The proposed models are based on Gated-Recurrent Unit (GRU) and Long-Short Term Memory (LSTM) recurrent neural networks (RNN). Moreover, we compare the accuracy of the proposed transduction neural networks to a non-recurrent neural network, then we evaluate the performance of all suggested models in a realistic environment with only WLAN RSS fingerprints, Cellular RSS Fingerprints, and both WLAN/Cellular RSS fingerprints. Our experimental results show that the LSTM-based architecture achieves accuracy of about one meter in an open area of 130m2 while only using Wi-Fi fingerprints.