Localizing pedestrians in indoor environments using magnetic field data with term frequency paradigm and deep neural networks

Imran Ashraf, Yousaf Bin Zikria, Soojung Hur, Ali Kashif Bashir, Thamer Alhussain, Yongwan Park

Research output: Contribution to journalJournal articlepeer-review

1 Scopus citations


Indoor environments are challenging for global navigation satellite systems and cripple its performance. Magnetic field data-based positioning and localization has emerged as a potential solution for ubiquitous indoor positioning and localization. The availability of embedded magnetic sensors in the smartphone simplifies the positioning without the additional cost of infrastructure. However, the data divergence due to smartphone heterogeneity circumscribes the wide applicability of magnetic field-based positioning approaches. This research proposes the use of term frequency (TF) extracted from the magnetic field data to alleviate the impact of smartphone heterogeneity. For this purpose, the magnetic field data are transformed into terms (words) and documents. Extracted TF vectors are used to train long short term memory and gated recurrent unit networks. A voting scheme is contrived to incorporate the predictions from these networks. Experiment results with three different smartphones like LG G6, Galaxy S8, and LG Q6 demonstrate that the use of TF mitigates the impact of the smartphones’ variability. Performance comparison with state-of-the-art approaches reveals that the proposed approach performs better than those of other approaches in alleviating the influence of using various smartphones for magnetic field-based indoor localization. Furthermore, the localization performance of the proposed is better than those of other approaches, even using a smaller amount of magnetic field data.


  • Deep neural networks
  • Indoor positioning and localization
  • Magnetic field data
  • Smartphone sensors
  • Term frequency


Dive into the research topics of 'Localizing pedestrians in indoor environments using magnetic field data with term frequency paradigm and deep neural networks'. Together they form a unique fingerprint.

Cite this