Target Tracking Based On RTS Smoother For Wireless Sensor Networks

In this paper, we focus on the forms of the Kalman filter algorithm and Kalmansmooth algorithm applied in non-linear systems. The following filtering algorithm isdisscussed: Extended Kalman filter (EKF), Unscented Kalman filter (UKF), CubatureKalman filter (CKF). The following smoothing algorithm is disscussed: Two-fltersmoother, Forward–backwar smoother and Cubature Rauch-Tung-Striebel smoother.This paper discuss the method which using the improved smoothing method forimprove the accuracy of Cubature Kalman filter. At the same time, we discuss themethod which using the improved Unscented Kalman filter optimize the initial valueof Unscented Rauch-Tung-Striebel smoother, the algorithm improve the accuary ofestimates. The main contribution in this paper is described as follows:（1）As the accuracy of Cubature Kalman filter is not enough high, andtraditional Cubature Rauch-Tung-Striebel smoother is the form of fixed-interval,which can not meet the requirement of the promptness in target tracking. A newRauch-Tung-Striebel form of the fixed-lag cubature Kalman smoother is developedfor nonlinear state-space models by adopting cubature transformation for optimalsmoothing. In this research, we introduce a fixed-lag cubature Kalman smoother thatuses all filtering estimates from step k-n to k to correct the estimate in step k-n bybackward iteration, given that n is the lag, and k is the current step. Derivedsmoothing equations here are approximated to the formal Bayesian optimal smoothingequations. The effect of different n values on smoothing estimate is investigated andan example is used to demonstrate the new algorithm’s performance.（2）Considering the estimates of C-UKF are dispersed for a relatively longperiod. In this study, a new Unscented Kalman consensus filter based onRauch-Tung-Striebel Smoother (C-URTS) is proposed. This new algorithm andUnscented Kalman consensus filter (C-UKF) differ in their reliance on a traditionalUnscented Kalman filter. Unscented Kalman consensus filters are used to apply stateestimates for the initial smooth value of URTS. The C-URTS uses a neighbor’sinformation to reach a consensus, and then runs a backward smoother with theobtained Kalman consensus filters’ value for better accuracy. Simulation resultsdemonstrate the superiority of C-URTS in sensor networks.(3)At last, we present a smoothing algorithm which is applied to target tracking in wireless sensor network, then, we discuss the application area of the threesmoothing algorithm forms. The algorithm uses a neighbor’s information with theUnscented Kalman consensus filters to reach a consensus, and then the obtainedKalman consensus filters’ value will be the initial value of the Rauch-Tung-Striebelsmoother. Because the three optimal smoothing forms is suitable for differentapplications, we exposition three optimal smoothing forms of the UnscentedRauch-Tung-Striebel Smoother based on Kalman consensus filter. Simulation resultsare provided to demonstrate the algorithm which improve the accuracy of the wholetracking.