Risk-based sensor resource management for field of view constrained sensors

This paper introduces a statistical risk based metric for field of view constrained sensor management in a target tracking scenario. This metric is based on a Bayesian estimate of both the target position and the target classification. Due to field of view restrictions it is assumed that more targets exist than the given sensor is capable of tracking simultaneously. It is also assumed that initially all target classifications are unknown and that a cost exists for incorrectly classifying a target track. This cost is higher for certain classes of targets than it is for others. To account for uncertainty in both the kinematic and classification state estimates, the proposed metric treats the cost as a random variable and uses a hierarchical statistical model to calculate the expected value of this cost when conditioned on the event of losing a target track. This metric is then applied to a simulated radar sensor manager to maintain an acceptable level of kinematic accuracy on targets of high cost. It is shown through empirical statistical tests that this sensor manager maintains track on high priority targets significantly better than other common methods.