Consensus-Based Distributed Mixture Kalman Filter for Maneuvering Target Tracking in Wireless Sensor Networks
Yihua Yu · IEEE Transactions on Vehicular Technology · 2015
We consider the distributed state estimation for conditional dynamic linear systems (CDLSs) in a wireless sensor network without a data fusion center. Each sensor only exchanges its local information with its neighbors. The mixture Kalman filter (MKF) is an effective technique for the state estimation in the CDLSs. We extend the MKF and develop a distributed MKF for the CDLSs. Each sensor in the network runs a local MKF and interacts with other sensors to generate a mixture Gaussian representation of the global posterior distribution. Since the global likelihood in the CDLSs is not suitable for the distributed computation, we apply the cubature rule for the calculation of likelihood, to enable the distributed implementation of the likelihood computation. Consensus algorithms are executed to fuse the observations from all sensors. For the distributed MKF, the number of particles can be significantly reduced, compared with the distributed particle filtering. Finally, we apply the distributed MKF to the maneuvering target tracking. We consider that the maneuvering variable is discrete and continuous. We also consider that the observation equation is linear and nonlinear. Simulation results illustrate that the performance of the distributed MKF is close to that of the centralized MKF in various scenarios of maneuvering target tracking.