Bayesian method with optimal sensors selection

Due to on-line processing and transmitting limitations, it may be infeasible to handle all independent measurements at once. This paper proposed a technique to select the likelihood function of a sequential Bayesian estimator from a set of available sensors. The selection criterion is based on maximizing the conditional entropy of the state vector assuming one measurement can be handled at each time. We evaluated the conditional entropy for all candidate sensors by using a weighted approximation of the Probability Density Function (PDF) of the state vector. In turn, the weights of the unknown PDF are updated using a particle filter. The optimal likelihood is expected to have widest dispersion, which should mitigate sample impoverishment in the particle filter. Result from Numerical simulation experiment shows the performance of the proposed algorithm, which has potential applications on sensor fusion.