Indoor location estimation and location tracking with dynamic artificial neural network

The positioning systems are grouped into two categories: indoor and outdoor positioning. The global positioning system (GPS), location detection and location tracking, is successfully implemented at outdoors thanks to satellite-based systems and cellular systems. However, these systems are limited in terms of location detection or location tracking at the indoors due to the weakening of signals due to environmental effects indoors. Therefore, the methods and systems that works effectively indoors are required and have been investigated by researchers. Among these systems, mostly radio frequency (RF) signal-based positioning systems are emphasized in the literature. In this thesis, the triangulation method, Artificial Neural Network (ANN) and dynamic ANN methods are used to locate an indoor object. The fingerprint-based positioning method, which is one of the mostly used methods of indoor positioning, is implemented by using the received signal strength indicator (RSSI) data. In this method, the radio map of the environment is obtained by the fingerprint of the location (room, corridor etc.). In order to reduce the effect of the multipath fading, the size of the radio map is reduced (dividing into cells). Support vector machine (SVM) is used to divide into cells based on fingerprint method. For the location determination and location tracking of the mobile device, RSSI data is obtained separately from each cell and based on these RSSI data, an ANN is trained offline for each cell. After the offline training, the location of the mobile device is determined and tracked online (real time) by using these trained ANNs based on RSSI data which is received by fixed devices (receivers). The triangulation, the ANN and dynamic ANN methods are examined separately, and the results are compared.