Target tracking system using wireless sensor networks

Wireless sensor networks have gained considerable attention from both academia and industry in last few years due to their crucial and advantageous applications including localising and tracking variety of objects and phenomena in different environments. Using a fleet of inexpensive sensor nodes is substituted the old fashion of the expensive nodes in the technology of autonomous sensing and tracking. The existing works are heavily based on state-space systems of a linear motion model along a nonlinear measurement. Due to nonlinearity in measurements, non-linear state estimation methods, like particle filtering, has become more popular in this field. There are three issues highlighted on using such tools in wireless sensor networks consist of heavy computation, lack of good knowledge of motion model, and initialization. This research has proposed solutions to all three problems. The tracking problem is divided into two categories of static and dynamic tracking. In the former, the main proposed method is a maximum likelihood estimator which is initialised via a least squares estimation. In this case, there is no assumption about target motion and it is independent from sampling rate. For the latter group, a nearly-constant-acceleration linear model is assumed for target movement along a nonlinear measurement model. First, this work derives a linear form of the nonlinear measurement. Then, a Kalman filter estimator initialised by maximum likelihood is designed for tracking purposes. For analogy to Kalman model, a particle filter estimator is also designed on the original nonlinear measurement. During this research, it is also found that most of existing works are assuming independent measurement noise, probably for problem simplification. Therefore, a similar tracking system is proposed considering dependent noise measurement. The analysis and numerical proof is presented in this research and both cases are compared together. The results show that the proposed tracking system with dependent measurement noise highly overtake the other method with independent measurement noise. At last, due to importance of distributed techniques in wireless sensor networks, a quasi distributed tracking method is developed through applying a naive extended Kalman filtering on each sensor’s measurements. However, the final coordinate estimation is done in the fusion center using standard least squares estimation. The simulation results show a great improvement on the accuracy of estimation over the standard least squares.