Generalized Regression Neural Networks and Kalman Filtering Based Indoor Target Tracking Using Wireless Sensor Network
Satish R. Jondhale, Rajkumar S. Deshpande
Abstract—Traditional received signal strength indicator (RSSI) based mobile target tracking using wireless sensor network’s (WSN’s) generally rely on simple geometry based techniques such as lateration or angulation. However for the indoor environment, significant localization errors are involved with these techniques due to highly nonlinear relationship between RSSI and distance or angles because of uncertainty in the measurement noise due issues such as NLOS, multipath propagation and abrupt variations in the target velocity. Being a one-pass learning algorithm, the Generalized Regression Neural Network (GRNN) can be a very attractive alternative for characterizing the given indoor environment. This paper introduces the application of the GRNN as an alternative to the traditional trilateration and angulation techniques. The developed GRNN architecture is employed to give first 2-D location estimate of the single mobile target moving in WSN, which is then further refined using kalman filtering (KF) framework. Two algorithms namely, GRNN + kalman filter (KF) and GRNN + unscented kalman filter (UKF) are proposed in this research work. The simulation results demonstrate the efficacy of the proposed algorithms over the conventional strategies.
Keywords: Generalized Regression Neural Network (GRNN), Received Signal Strength Indicators (RSSIs), Target Tracking, Unscented Kalman filter (UKF), Wireless Sensor Networks (WSNs)