Indoor Localization Using Smartphone Magnetic Sensor Data: A Bi-LSTM Neural Network Approach

Nowadays, smartphone-based indoor localization has gained significant attention across various research fields, industrial sectors, and practical applications. However, accurate indoor localization using smartphones is a challenging issue due to the presence of noise and crowded localization environments. To overcome these problems, we introduce a new system for localization in indoor locations using real magnetic data collected from a smartphone. In this system, we first normalize the dataset. After that, the normalized magnetic data are filtered by using the Empirical Mode Decomposition (EMD) method. Finally, we present a Bidirectional Long Short-Term Memory (Bi-LSTM) neural network for indoor localization using fingerprinting technique. We use three datasets in three different locations in the Faculty of Engineering. Results show that Bi-LSTM neural network has achieved the best localization performance among Feed Forward Neural (FFN) Network, Gated Recurrent Unit (GRU) neural network and LSTM neural network. Compared to the performance of the mentioned neural networks, the proposed system had the least mean and standard deviation (SD) root mean square error (RMSE) for indoor localization in the datasets. According to the results obtained with the proposed system, mean and SD RMSE for the network and embedded systems research laboratory location on the second floor of the Engineering Faculty is 0.1244 m and 0.1070 m, for the corridor location on the same floor is 0.0251 m and 0.0166 m, and for first floor location of this faculty is 0.0192 m and 0.0165 m, respectively. Therefore, the proposed method can be used practically in the real world.