A Low-Cost and Efficient Spatial–Temporal Model for Indoor Localization “H-LSTMF”

Abstract

The industrial Internet of Things (IIoT) requires location-aware services based on indoor localization. Wi-Fi-based indoor localization is a low-cost and straightforward method. Still, it does not perform satisfactorily because of the fluctuations of the signals due to the multipath effect, the movement of humans, objects, etc. Therefore, an optimized model is required to improve the performance by eliminating received signal strength indicator (RSSI) uncertainties and reducing noise and training time. The hierarchical-density-based spatial clustering of applications with noise (HDBSCAN) algorithm is available for clustering the fingerprint database to eliminate outliers. Furthermore, locally weighted regression (LWR) is available to smoothen the RSSI uncertainty at the input. Finally, the encoder-long short-term memory (LSTM)-encoder model has been proposed that can learn the temporal relationships among the data. A new model is proposed in this article that combines HDBSCAN, LWR, and encoder-LSTM-encoder methods and is named “H-LSTMF. ” The proposed model was evaluated in two distinct datasets. The first dataset was collected in a room of size <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$10\times $ </tex-math></inline-formula> 5 m (dataset I) with 50% fewer reference data points, and the other dataset was taken from the existing work for a room of size <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$35\times16$ </tex-math></inline-formula> m (dataset II). It has been found that the proposed method improved the localization accuracy by 20% and 60% in dataset I and dataset II, respectively, when compared with the state-of-the-art algorithms.