Fingerprinting-based Indoor Localization with Deep Neural Networks

Abstract

In recent years, more capabilities and applications have been added to existing wireless communication systems due to the rapid development of the Internet of Things(IoT). WiFi and RFID exhibits tremendous potential in this industry due to their prevalence and lowcost. Among the applications, indoor localization has been a popular field of research over the years, since it plays a vital role in resolving position-related challenges such as gesture recognition and human pose estimation. In the meantime, with the advancement of deep learning, researchers are attempting to integrate deep networks into indoor localization systems to take advantage of their superior ability to solve classification and regression problems. On the other hand, the fingerprint method emerges with its convenience and effectiveness, which transfers the localization problem into a feature matching to estimate the location of the signal. Thus, deep learning technique is a great complement to fingerprinting-based indoor localization systems. However, numerous intrinsic difficulties of fingerprinting-based localization systems remain unresolved even though the performance of indoor localization systems keeps improving with the iteration of deep networks. First, the distance between the stored fingerprints determines the minimum error of the fingerprinting-based localization system. To guarantee the lower-bound of the localization accuracy, as many fingerprints as possible have to be collected, which is laborious and time-consuming. Second, fingerprints are discrete signal space samples. As a result of the elimination of ambiguity between fingerprints, deep neural networks may produce counterintuitive location estimations, contrary to our expectations. To address such issues, the Deep Gaussian Process(DGP) is leveraged in this dissertation to generate a detailed radio signal map using a limited number of fingerprints. Then, the uncertainty information from DGP is adopted to train an LSTM model for enhancing the localization estimation by using the signal sequence. Furthermore, a novel network input, the hologram tensor, is employed for reserving the ambiguity between the fingerprints. In the last section, the threat of the adversarial attack to the fingerprints is investigated to promote the robustness of the localization system.