A WiFi Indoor Localization Method Based on Dilated CNN and Support Vector Regression

A novel method is proposed to improve positioning real-time property while ensuring the accuracy. Firstly, a dilated convolutional neural network (D-CNN) model is trained with images formed by the received signal strength (RSS). Secondly, the errors of the predicted results of D-CNN are used to train a support vector regression (SVR) model. Experiments are conducted using the public database collected from a library of Universitat Jaume I in Spain. The results demonstrated the superior performance of D-CNN. Moreover, the results proved that the average runtime of the proposed D-CNN + SVR algorithm was only 0.612s, which was reduced by 86.27% compared with P-CNN + Gaussian process regression (GPR), when ensuring the localization accuracy in the indoor environment.

I. Introduction

Location awareness is the foundation of location-based service, so how to quickly and accurately obtain the location of users has become an urgent problem for these applications. Outdoor, global navigation satellite systems, such as GPS, BDS, GLONASS and Galileo, have been extremely successful and commercialized over the past few decades. However, satellite signals are sensitive to obstacles, which make it impossible to provide positioning services that meet the requirements for indoor users.