Application of sequence comparison methods to multisensor data fusion and target recognition

This research addresses methods for exploiting the spatio-temporal joint likelihood of observed kinematic and nonkinematic (sensor signature) physical events to improve dynamic object and target recognition. A principal direction is the application of dynamic programming sequence comparison techniques to condition matching of object signatures to known models according to observed kinematics--that is, to use information from observed kinematics in determining allowable aspect angles with which observed signatures may be matched on models for candidate objects. A second direction is the application of kinematic/aspect-angle Kalman filter trackers to condition kinematic tracking according to observed signatures. These conditioning processes dramatically reduce ambiguity in object recognition, and can be used together or separately to allow computation of a posteriori probabilities of object class membership using Bayesian methods. Proposals are supported by results of simulated target tracking and high range resolution radar signature analysis. The original contributions of this effort include: (1) new approaches for and theoretical understanding of syntactic methods in multisensor fusion and dynamic object recognition; (2) extension of estimation and tracking techniques to allow object recognition; and (3) introduction of a new performance evaluation technique and approach for establishing performance bounds in dynamic object and target recognition.