Indoor Floor Detection Using Magnetic Field Map and Deep Learning Model

This paper proposes a magnetic field map and deep learning model-based floor detection method to address the issues encountered with traditional floor detection methods in smartphone-based indoor multi-floor localization systems. The proposed method addresses the problems associated with using barometer sensors for detecting the user's initial floor by constructing magnetic field maps using the internal geomagnetic sensors of smartphones on each floor of a multi-floor building. The user's initial floor is then detected using a deep learning model. Additionally, to address the issue of data scarcity in deep learning training, a data augmentation method using magnetic field vector rotation transformations is proposed, which enhances model training and floor detection performance even with limited data. To validate the performance of the proposed floor detection method, experiments were conducted on two floors using paths opposite to those used for constructing the magnetic field maps. The results confirmed that the proposed method can detect the floor on which the user is located with high accuracy.