A novel distributed fusion framework via an adaptive Bernoulli filter

Title: A novel distributed fusion framework via an adaptive Bernoulli filter

Abstract:

Target detection probability and noise statistics are critical in distributed target tracking problems. This information, however, is often unknown or time-varying in practice. To accommodate this situation, this work first proposed an adaptive Bernoulli filter to jointly estimate the unknown parameters regarding the target state, detection probability and noise statistics. A mixture of exponential-class distributions, i.e., Gaussian inverse Gamma inverse Wishart mixture (GIGIWM), is adapted to provide a closed-form implementation. Wherein, the unknown parameters related to the target state and noise statistics are computed by the variational Bayesian (VB) approach with constrained noise covariances. On the basis of the proposed filter and generalized covariance intersection (GCI), a novel distributed fusion framework is designed to obtain an improved and robust estimation result.