CNN based approach for Indoor Positioning Services using RSSI Fingerprinting Technique

Nowadays Indoor Positioning Systems (IPS) play a crucial role in providing information about the location of a smartphone inside covered areas (e.g, buildings), especially with the widespread deployment of WiFi technology. In this paper, we propose an IPS based on WiFi and Convolution Neural Network (CNN). We focus on the WiFi RSSI fingerprinting technique as it is the most widely used method, which principle is to return the smartphone's location using a learning algorithm that best matches the real-time recorded RSSI vector with the previous stored data. In our model, CNN was used as a learning algorithm owning to its ability to capture the structure of the unstable WiFi RSSI data. CNN was employed to match RSSI image corresponding to one Reference Point (RP) to its location. However, the correlation between RPs leads to RP ambiguity for RSSI images. To overcome this issue, we opt for Distance-Based Label Distribution (DBLD) method for RP encoding that takes into consideration the distribution of RP's neighbors while encoding one RP. To further improve the performance, we apply Random Forest Regression as an Error Correction Model (ECM) that adjusts the predicted position of the target point. Our implemented model was evaluated using a public dataset collected in the library building of Jaume I University in Spain. The experimental results show that our approach achieves a high accuracy with a mean positioning error equal to 0.72 m.