Indoor localization has been an active research field for decades, because of its wide range of applications. WiFi fingerprinting, which estimates the user’s locations using pre-collecting WiFi signals as references, is of particular interest as these days, every user can easily access to WiFi networks. Among numerous methods, Deep Neural Network (DNN) based methods have shown an attractive performance but their major drawback is the sensitivity to the fluctuation of received signals caused by multipaths. In order to ensure satisfactory performance, thus, a sufficiently large number of possible cases should be trained, which costs a lot. In this thesis, we address the above problem by presenting a Convolutional Neural Network (CNN) based localization method. As success in image classifications, the proposed method can be robust to the small changes of received signals as it exploits the topology of a radio map as well as signal strengths. Via experimental results, we demonstrate that the proposed CNN method can outperform the other DNN based methods using publicly available datasets provided in IPIN 2015.
