ZSFA: Zero-Shot Fingerprint Augmentation for WiFi Fingerprint Based Indoor Localization
Tian Lan, Yuanqing Ye, Sihai Zhang
Abstract
The radio map construction is usually time-consuming and labor-sensitive in indoor fingerprint localization. Recently, some fingerprint augmentation methods based on super-resolution have been proposed to improve the accuracy of sparse radio maps. They use the Received Signal Strength(RSS) data at part of reference points(RPs) to construct RSS at other RPs, thus reducing the number of reference points needed to be collected in the offline phase. However, these methods require a large amount of historical dense radio map data for network training, which requires a lot of collection costs. To save these costs, we propose zero-shot fingerprint augmentation(ZSFA), a fingerprint augmentation scheme based on an unsupervised super-resolution method without historical data. Simulated and real data experiments show that compared with the sparse radio map, the proposed ZSFA explicitly reduces the localization error. The results further demonstrate that our method can reduce the number of reference points to be collected in the offline phase and save collection costs.