Tracking Moving Object via Unscented Particle Filter in Sensor Network

Moving object tracking is one of the typical applications in wireless sensor network (WSN). As a result, a lot of important solutions have been proposed in the last decade, toward addressing different aspects of object tracking in WSN settings. This work describes an Unscented particle filter (UPF) based Moving Object Tracking algorithm, UMOT, in WSN settings, where the sensor nodes are clustered dynamically to provide sensing and data fusion tasks. The key idea of UPF is to capture accurately the posterior mean and covariance of non-linear Gaussian variable up to the second order, through propagating a set of sample points in the state system. It has been demonstrated that UPF addresses the deficiency of using transition prior as the proposal distribution, which results in biased posterior estimation due to excluding the recent observations. We present the design and implementation of UMOT, together with comprehensive simulations, conducted to evaluate the proposed methods. The simulation results show that UMOT achieves significant improvement over existing schemes in various network settings.