Fingerprinting-based Indoor Localization with Relation Learning Network

Recently, fingerprinting-based indoor localization in deep learning framework has attracted intensive interests with the satisfactory accuracy. However, the location performance counts on the sufficient and massive radio signal acquisition, which is impractical to realize real-time localization for a wide range of location based services. To address this issue, we develop ReFi location system that can localize the target by relation learning network with a small dataset. ReFi firstly learns to represent the appropriate features and then compare them with the similar relation from the training and testing location data. According to relation learning, the target location is associated with the training locations and estimated in a probabilistic method. The extensive experimental results demonstrate that the proposed location system can achieve decimeter level accuracy with a single transceiver pair.