Smart indoor Positioning Model for Deterministic Environment

Abstract

Location based services are very popular nowadays. However, indoor location identification is a big challenge. Many different solutions are proposed for indoor mapping however, most of these require additional hardware are also not very accurate. One of these solutions is use of received signal strength indication (RSSI) values for location mapping. Theoretically, it is possible to triangulate the location by tagging specific location point with different signal strengths received from multiple access points however this technique has many issues. It is affected by many factors like layout of the building, furniture, number of people in the area. Moreover, reflection of signals from the walls also add huge noise to the signals received. Since, many machines learning algorithms are successfully being applied in many real-world scenarios for predicting and inferring information, in this paper, we propose use of machine learning techniques on the Wi-Fi received signal strength indication (RSSI) for indoor localization. This study belongs to range-free and fingerprinting of localization. The research is divided into off-line training and on-line predicting stages. In the off-line training stage, a signal map with two-dimensional array structure is established by capturing the Wi-Fi RSSI of different reference locations, and different access points (Aps), Each output node of the neural network (NN) represents the probability that a signal vector occurs at the corresponding reference location. In the on-line predicting stage, instantaneous RSSI is recorded at an unknown location. The trained NN can accurately predict indoor positions even far from network devices.