DL-RNN: An Accurate Indoor Localization Method via Double RNNs

Abstract

Wireless fingerprinting localization method learns a mapping function from a fingerprint measurement to the estimated location, which is more suitable for complex indoor environments than the propagation model-based methods. However, most traditional methods only consider the location matching at single time or space points, but ignore the fact that correlations in the measurement sequences may improve the accuracy and robustness of mobile positioning. Recently, recurrent neural network (RNN) has been readily applied in many fields such as speech, language, and video processing, and it has flexible and powerful processing capabilities for nonlinear and high-dimensional sequence input. In this paper, we propose DL-RNN, a real-time wireless localization model, which consists of double RNNs: the first RNN estimates the location based on the historical observation signal, and the second RNN filters the location based on the historical estimated location, which further improves the localization performance. Both the simulation and measured experiments show the high accuracy and robustness of the proposed algorithm.