Deep Regression Model for Received Signal Strength based WiFi Localization

This paper propose a deep regression model for WiFi localization using received signal strength (RSS). In the offline phase, we first construct RSS fingerprints at all grid points in a residential area by searching some detectable access points (APs). Based on the RSS fingerprints, we propose a deep regression model, namely DNN-CNN-DS, which consists of Deep Neural Networks (DNN), Convolutional Neural Network (CNN), and Dempster-Shafer, in which the initial weights of DNN is determined by AutoEncoder. The optimal weights of DNN-CNN-DS are calculated by minimizing the means square error between the output of the model and real location. In the online phase, our proposed DNN-CNN-DS regression model can accurately predict the location of user when inputting an RSS testing sample instantaneously. Compared with the existing models, DNN-CNN-DS can effectively improve the positioning accuracy by fully leveraging the complementarity between the three techniques. Experimental results demonstrate that our proposed model outperforms other methods in accuracy and robustness.