Adaptive sensor management for feature-based classification

Modern surveillance systems often employ multiple steerable sensors that are capable of collecting information on selected objects in their environment. In this paper, we study the problem of managing these sensors adaptively to classify a collection of objects using information on their observed features. We develop a new theory for sensor management that models sensors as providing observations of object features subject to degradation by noise, obscuration, missed detections and background clutter. We establish a statistical framework, based on random sets, to characterize the relationship between observed features and object types. This model uses Bhattacharyya distances to generate off-line apriori estimates of the discrimination value of measurements. These estimates are combined with real time information to generate predictions of the usefulness of measurements. Using these predictions, we develop assignment algorithms to compute sensor management strategies. The resulting sensor management algorithms are capable of solving problems involving a large numbers of objects in real-time. We show simulations of the resulting algorithms for classifying 3-dimensional objects from 2-dimensional noisy projections, and show how the algorithms select complementary views to overcome obscuration and provide accurate classification. Our real-time algorithms achieve comparable classification accuracy to on-line approaches that also evaluate the value of information, while requiring nearly five orders of magnitude less computation.