CellinDeep: Robust and Accurate Cellular-Based Indoor Localization via Deep Learning

Abstract

The demand for a ubiquitous and accurate indoor localization service is continuously growing. Current solutions for indoor localization usually depend on using the embedded sensors on high-end phones or provide coarse-grained accuracy. We present CellinDeep: a deep learning-based localization system that achieves fine-grained accuracy using the ubiquitous cellular technology. Specifically, CellinDeep captures the non-linear relation between the cellular signal heard by a mobile phone and its location. To do that, it leverages a deep network to model the inherent dependency between the signals of the different cell towers in the area of interest, allowing it achieve high localization accuracy. As part of the design of CellinDeep, we introduce modules to address a number of practical challenges such as handling the noise in the input wireless signal, reducing the amount of data required for the deep learning model, as avoiding over-training. Implementation of CellinDeep on different Android phones shows that it can achieve a median localization accuracy of 0.78m. This accuracy is better than the state-of-the-art indoor cellular-based systems by at least 350%. In addition, CellinDeep provides at least 93.45% savings in power compared to the WiFi-based techniques.