Particle Filter based Target Tracking in Wireless Sensor Networks using Support Vector Machine

Target tracking is estimating the state of moving targets using noisy measurements obtained at a single observation point or node. Particle filters or sequential Monte Carlo methods use a set of weighted state samples, called particles, to approximate the posterior probability distribution in a Bayesian setup. During the past few years, Particle Filters have become very popular because of their ability to process observations represented by nonlinear state-space models where the noise of the model can be non-Gaussian. There are many Particle Filter methods, and almost all of them are based on three operations: particle propagation, weight computation, and resampling. One of the main limitations of the previously proposed schemes is that their implementation in a wireless sensor network demands prohibitive communication capability since they assume that all the sensor observations are available to every processing node in the weight update step. In this paper, we use a machine learning technique called support vector machine to overcome this drawback and improve the energy consumption of sensors. Support Vector Machine (SVM) is a classifier which attempts to find a hyperplane that divides two classes with the largest margin. Given labeled training data, SVM outputs an optimal hyperplane which categorizes new examples. The training examples that are closest to the hyperplane are called support vectors. Using our approach, we could compress sensor observations and only support vectors will be communicated between neighbor sensors which lead to cost reduction in communication. We use LIBSVM library in our work and use MATLAB software to plot the results and compare the proposed protocol with CPF and DPF algorithms. Simulation results show significant reduction in the amount of data transmission over the network.