CASI - TR - 98 - 10 Controlling Target Estimate Covariance in Centralized Multisensor Systems

Current multisensor fusion tracking systems can be easily overwhelmed by incoming data, especially as the number of targets and sensors increases. Sensor management schemes have been proposed to reduce the computational demand of these systems while minimizing the loss of tracking performance. This report presents a system that will maintain a desired covariance level for each target while reducing the resource demands on the tracking system. Other functions performed by a sensor manager like prioritizing and scheduling are assumed to be done elsewhere, but result in delays in the execution of sensing requests made by the system. Three sensor selection algorithms are presented based on different resource and performance metrics and show a dramatic improvement over " dumb " sensing systems in simulation. Since the proposed system is sensitive to delays in the execution of sensor assignments, we further analyze the effect of that delay and examine methods of eliminating that effect. Because of the lack of a closed form solution for the covariance propagation of the discrete-time Kalman filter, the analysis centers on the performance of the continuous-time scalar Kalman-Bucy filter and then extends those results to the discrete-time case. The analysis shows that for all stable systems and unstable systems under certain conditions, the sensitivity of the covariance estimate to delays of sensing actions decreases steadily with time. Furthermore, when attempting to estimate unknown delays, overestimating the delay will produce smaller covariance prediction errors than underestimating the delay by a similar amount.