CLUTTER REJECTION SIGNAL PROCESSING

INFARED SENSORS USED FOR SURVEILLANCE MUST BE ABLE TO DISCRIMINATE UNRESOLVED TARGETS AGAINST A CLUTTERED BACKGROUND WHICH MAY CONTAIN CLOUDS, SEA OR TERRAIN. ONE OF THE IMPORTANT PROBLEMS ASSOCIATED WITH THIS TASK IS TEMPORAL PROCESSING IN WHICH THE TARGET MOVEMENT RELATIVE TO BOTH BACKGROUND AND FOREGROUND CLUTTER IS USED TO ENHANCE DETECTION PROBABILITY. IN THIS PROPOSAL, WE PRESENT AN INTEGRATED AND POWERFUL APPROACH TO THIS PROBLEM WHICH HAS THE POTENTIAL TO YIELD SUPERIOR PERFORMANCE TO CURRENT TECHNOLOGY SUCH AS MTI. CURRENT ALGORITHMS TEND TO BE AD HOC, CONSISTING OF A CASCADE OF PROCESSORS WHICH MAY WORK AT CROSS PURPOSES. A MORE SUITABLE APPROACH WOULD BE TO COMBINE MOTION DETECTION AND CLUTTER REDUCTION IN A SINGLE STEP. WE PROPOSE A PROBABILISTIC FRAMEWORK IN WHICH THE IMAGE IS MODELED AS A RANDOM FIELD TO BE ESTIMATED IN REAL TIME FROM NOISY AMBIGUOUS MEASUREMENTS FROM MULTIPLE SENSORS. A BAYESIAN VIEWPOINT S ADOPTED, IN WHICH THE PRIOR KNOWLEDGE IS EXPRESSED AS A PROBABILITY DISTRIBUTI ON. USING A PROBABLISTICS DESCRIPTION OF THE OBSERVATION NOISE, THE POSTERIOR DISTRIBUTION OF THE RANDOM FIELD CAN BE COMPUTED. THESE MODELS ARE BASED ON MARKOV RANDOM FIELDS AND THE GIBBS DISTRIBUTION. SIGNIFICANTLY, THESE ASSUMPTIONS LEAD TO COOPERATIVE DISTRIBUTION ALGORITHMS WHICH MAY BE IMPLEMENTED ON PARALLEL PROCESSORS. IT IS POSSIBLE TO MODEL BOTH PIECEWISE CONTINUOUS SURFACES AND THE BOUNDARIES BETWEEN SMOOTH PATCHES (TARGETS, CLOUDS, OBJECTS, E.G.). THE PARAMETERS THAT APPEAR IN THE RECONSTRU CTION ALGORITHMS HAVE A PRECISE STATISTICAL INTERPRETATION WHICH MAY BE VALIDATED ON PHYSICAL GROUNDS.