An Adaptive and Robust Model for WiFi-based Localization

Abstract

The fluctuation of received signal strength (RSS) caused by environmental changing and heterogeneous devices severely degenerates the performance of WiFi fingerprint-based positioning methods. Deep Domain Adaptation (DDA) in transfer learning has proven to be an effective strategy to deal with this situation. However, the existing DDA methods show limited improvement in positioning accuracy in presence of the two factors simultaneously. In this study, we propose a new deep adaptation networks by adopting the joint constraints of mean and covariance to reduce domain discrepancy, which shows an excellent adaptability to environmental changing and heterogeneous devices. To further improve the robustness of our network, we design an exponential moving average method to update the parameters of the network, which can be further updated by unlabeled data from target domain, which is highly consistent with the actual application scenario and has practical significance. Experiment results show that the proposed model can reduce domain discrepancy effectively, and achieve lower positioning error than some other existing methods in real complex indoor environments.