Zone-Based Indoor Positioning System in Faculty Building with Neural Networks

Abstract

Pervasive Wi-Fi deployment has made Wi-Fi an economically convenient wireless platform for developing an Indoor Positioning System (IPS). This paper presents a zone-based IPS developed on Wi-Fi using fingerprinting technique with Probabilistic Neural Network (PNN) and Radial Basis Function Neural Network (RBFNN) to predict target positions. The zone-based IPS is deployed in an indoor environment (a faculty building) with four Wi-Fi modules separately placed. The indoor environment consists of office rooms and laboratories separated by concrete walls. A two-dimensional coordinate system and zone label are deployed to define each location. After that, data collection is performed on each location. The Wi-Fi Received Signal Strength (RSS) for every Wi-Fi module at each location is discovered, labelled with the location coordinate and zone value to form a fingerprint and finally stored in a database. Fingerprints in the database are then separated into training and testing sets for training and testing of PNN and RBFNN. The testing result shows that the mean positioning error for coordinate prediction of PNN and RBFNN is 3.84m and 6.91m, respectively. Although RBFNN has a large mean positioning error, RBFNN presented a zone positioning accuracy of 78.7%, which is close to the 82.2% accuracy presented by PNN.