TSIG: A geomagnetic indoor positioning method leveraging temporal and spatial domain features

Numerous methods for geomagnetic positioning, using traditional or deep learning techniques, have made significant strides in indoor positioning. Nevertheless, geomagnetic mismatching still leads to considerable errors in positioning accuracy. To address this problem, we explore geomagnetic positioning in both the time and spatial domains concurrently. We propose a novel method named TSIG. In the temporal domain, we propose a multi-scale attention module with geomagnetic hierarchical embedding, named TIG. We first extract the temporal dependencies of geomagnetic subsequences at certain time scales from both local and global perspectives. Then, we use the extracted information to capture anomalies at different time scales and determine each scale’s contribution to the importance of features. In the spatial domain, we incorporate geomagnetic signal characteristics and propose a geomagnetic anomaly focusing and direction perception-driven feature extraction module named SIG. We reconstruct the geomagnetic sequences into images and further extract the spatial features using anomaly focusing and direction perception, thereby enhancing the representational capacity of the spatial features. Finally, the temporal and spatial domain features are input into the spatiotemporal feature fusion layer to achieve cross-domain feature fusion, generating the positioning result. The experimental results show that the proposed TSIG method can achieve accurate positioning.