Label Expansion through Walking Trajectories for Wi-Fi CSI-Based Indoor Localization

Wi-Fi fingerprint-based indoor localization methods rely on establishing a complete fingerprint database of the target area. However, fingerprint collection usually requires a significant amount of labor and time. We propose that a target area with only partially labeled fingerprints can be expanded to become completely labeled through human walking trajectories. Data collectors (trainers) are requested to move within the field to obtain unlabeled fingerprint trajectories. To enable label expansion, a series of corrections is made to ensure that the walking trajectories of trainers are correctly predicted and, consequently, labeled. This work greatly reduces the manual cost of fingerprint labeling.