AdaRF: Adaptive RFID-based Indoor Localization Using Deep Learning Enhanced Holography

Nowadays, RFID-based localization systems have been widely deployed in many factories and warehouses for sorting or locating products. These systems are mainly generalized schemes which might suffer severe accuracy degradation in multipath-rich scenarios. In order to suppress environmental interferences, we present a fine-grained RFID-based indoor localization system AdaRF, which leverages deep learning enhanced holography to create adaptive localization models for individual environments. The key idea is to optimize the localization model using signals from a small number of known location tags to achieve high positioning accuracy in its deployed environment. Based on this point, we propose Adjacent Differential Hologram (ADH) which yields a robust location-independent probability map for each tag. AdaRF subsequently leverages the neural network to create an effective hologram-based position estimation method, which estimates the target tag position by analyzing the whole hologram. And we introduce transfer learning technique to significantly lower training cost for the position estimation model while ensuring high accuracy simultaneously. Comparative experiments demonstrate AdaRF achieves cm-level positioning accuracy both in the lateral and radial direction with only one moving antenna even in complex scenarios.