Real-time indoor localization using smartphone magnetic with LSTM networks

Due to the pervasiveness of geomagnetic fields and independence from external infrastructure, localization with smartphone-based magnetic has attracted considerable attention. However, existing approaches are still facing the problem of localization accuracy because of their low discernibility in geomagnetic features. Due to the need to traverse the magnetic database, these approaches cannot obtain localization results in real-time, especially in large indoor localization areas. To address these problems, a novel magnetic indoor localization approach based on long short-term memory networks (LSTMs) is proposed. The magnetic indoor localization is first formulated as a recursive function approximation problem. Based on the analysis of magnetic characteristics, a double sliding window-based dimension expansion approach is designed to generate a time-series magnetic feature dataset. The LSTMs are invoked for magnetic localization on this dataset by taking advantage of its benefits in time-series prediction and characterization. To evaluate the performance of the proposed algorithm, we implement the proposed LSTMs-based magnetic localization approach on a real Android-based smartphone and compare it with fingerprint-based localization algorithms. Extensive experiment results demonstrate the accuracy, response time, and robustness of the proposed algorithm in indoor localization.