Bayesian target detection and localization using a dual-mode sensor

We investigate Bayesian methods and heuristics for configurable sensor management in a simple target detection and localization problem. In this problem, a target (if present) is located in one of M cells. A dual-mode sensor repeatedly interrogates either the entire search area (Mode A) or a single cell (Mode B); its performance in both modes is characterized by probabilities of correct detection and false alarm. We investigate several sensor control strategies, including the myopic optimal strategy which minimizes the probability of error for a single observation. All strategies are closed loop; the current sensor configuration depends on previous observations. Monte Carlo simulations show that the myopic optimal strategy gives the lowest probability of error for a fixed number of observations, while interrogating the cell with the highest probability of target present gives the lowest average number of observations needed to guarantee a fixed error probability.