Conditional Posterior Cramer-Rao Lower Bound and its Applications in Adaptive Sensor Management

For a general nonlinear non-Gaussian tracking problem, the new concept of conditional posterior Cramer-Rao lower bound (PCRLB) is introduced as a performance metric for adaptive sensor management. Both the exact conditional PCRLB and its recursive evaluation approach are presented. The recursive conditional PCRLB can be computed efficiently as a by-product of the particle filter which is often used to solve nonlinear tracking problems. Numerical examples are provided to illustrate that the conditional-PCRLB-based sensor management approach leads to similar estimation performance as that provided by the state-of-the-art information theoretic measure-based approaches. Analytical results show that the complexity of the conditional PCRLB is linear in the number of sensors to be managed, as opposed to the exponentially increasing complexity of the mutual information. Future work is proposed to develop conditional-PCRLB-based sensor management approaches in camera networks.