STATISTICAL MODELS FOR ATTRIBUTED-BASED TRACK CORRELATION/ CLASSIFICATION

WE ADDRESS THE NEED TO INTEGRATE ATTRIBUTES INTO ALGORITHMS USED TO TRACK AND CORRELATE TARGET CONTACTS. EXISTING SYSTEMS ARE GENERALLY BASED ON OBSERVATIONS OF TARGET KINETICS. WE PROPOSE TO DEVELOP DURING PHASE I AN ALGORITHM FOR CORRELATION AND CLASSIFICATION OF SURFACE CONTACTS IN WHICH THE TREATMENT OF SELECTED TARGET ATTRIBUTES IS FULLY INTEGRATED WITH THE TREATMENT OF TARGET POSITION AND VELOCITY. THIS ALGORITHM WILL BE BASED ON MATCH, A SYSTEM PREVIOUSLY DEVELOPED BY THIS FIRM FOR TRACKING MULTIPLE TARGETS USING POSITION AND ELINT OBSERVATIONS. MOST OF OUR EFFORT DURING PHASE I WILL BE DEVELOPING PRACTICAL STATISTICAL MODELS FOR RELEVANT ATTRIBUTES: ELINT,CALL SIGNS, SONAR-RELATED ATTRIBUTES, AND IMPERFECTLY REPORTED TARGET NAMES AND TYPES. OUR PROPOSAL DESCRIBES A UNIFIED FRAMEWORK FOR ATTRIBUTE MODELING BASED ON THE STATISTICAL THEORY OF CONJUGATE PRIOR DISTRIBUTIONS, WHICH WE HAVE USED SUCCESSFULLY FOR ATTRIBUTE-BASED CORRELATION IN OTHER CONTEXTS. IN PHASE II WE HOPE TO USE THE NEWLY DEVELOPED ATTRIBUTE MODELS AS A BASIS FOR IMPROVEMENTS IN AN OPERATIONAL NAVY SYSTEM AND TO INVESTIGATE, WITHIN THE CONTEXT, THE POSSIBLE USE OF AN EXPERT SYSTEM TO INTEGRATE CORRELATION WITH TARGET CLASSIFICATION.