Target State Estimation And Fusion Based On Wireless Sensor Networks

With the signifcant advances in networking, wireless communications,microfabrication, and microprocessors, the topic of wireless sensor networks (WSNs)has become a fast-growing research area. As a typical application of WSNs, targetlocation and tracking has its own advantages such as low cost, high reliability andself-organization.Starting from the classical Kalman Filter, this paper studies how to usesmoothing technology to improve estimation precision of the Sigma Point KalmanFilter. Later, depending on the structural advantage of Information Filter in datafusion, a method seeking to reduce overall energy consumption under the condition oftracking accuracy is also proposed in this dissertation. The main contributions can beconcluded as follows:(1) Smoothing is a method that uses features of the first two moments tocontinuously corrects a markov or gauss-markov sequences in order to get the optimalsmoothing value. This paper proposes a new method which combines fixed lagsmoothing algorithm and sigma transformation strategy. As a result of embedding abackward calculating smoother to a forward calculating filter, it can significantlyreduce the time delay and achieve a nearly same estimation precision with traditionalsmoother.(2) Considering the limitation of sensor node itself, a cluster strategy is alsopresented, which concerns not only the overall energy consuming but also theremainder energy. This method firstly predicts the target’s next-step position, thendecides which nodes should be choosed as members of next cluster by using thenearest neighbor method. With energy equilibrium strategies, the cluster head is alsoelected. Because only the nearest nodes participating in target tracking at every step,the cluster strategy can effectively reduce energy consumption and guarantee thetracking accuracy at the same time.(3) The Information Filter could be easily distributed because of its particularinverse covariance form. By embedding the newly proposed cluster strategy to theDistributed Sigma Point Information Filter, a novel distributed tracking algorithm hasbeen presented. In this method, each member of the cluster sents its local informationstate contribution to the fusion center. Because each node has the ability of datacalculation, the cluster head doesn’t need to process all members’ original measurements. Thus, it can obtain a lower computational complexity than centralizeddata fusion strategy.(4) In order to further improve tracking precision, a Smoothing-BasedDistributed Sigma Point Information Filter is also proposed in this paper. This methodusing the fused filtering results as inputs of a backward smoother to iterativelycalculate the smoothed state estimation value.