FOVEAL MACHINE VISION SYSTEMS

MILITARY APPLICATIONS OF COMPUTER VISION SYSTEMS REQUIRE ORDERS OF MAGNITUDE IMPROVEMENT IN CAPACITY AND SPEED OVER CURRENT SYSTEM PERFORMANCE. THE PRINCIPLE OBSTACLES TO HIGHER PERFORMANCE ARE: THE LARGE AMOUNTS OF IMAGE DATA TO BE PROCESSED; IMAGERS UNOPTIMIZED FOR THE VISION TASKS (TARGET DETECTION, CLASSIFICATION, RECOGNITION, IDENTIFICATION AND TRACKING); AND OPEN-LOOP SYSTEM ARCHITECTURE. IN THIS STUDY, A CLASS OF MUTUALLY OPTIMIZED IMAGERS AND IMAGE PROCESSING ALGORITHMS COLLECTIVELY CALLED FOVEAL VISION SYSTEMS ARE INVESTIGATED. THESE SYSTEMS, MODELED AFTER ADVANCED BIOLOGICAL VISION, FEATURE SPACE-VARIANT (VARIABLE RESOLUTION) IMAGER TOPOLOGIES AND A CLOSED-LOOP SYSTEM ARCHITECTURE. THE IMAGER TOPOLOGY IS CHARACTERIZED BY HIGH RESOLUTION SAMPLING AT THE CENTER OF THE SAMPLING PATTERN AXIS IS CONTROLLED BY FEEDBACK FROM THEIMAGE PROCESSING ALGORITHMS, ALLOWING THE ALLOCATION OF SAMPLING RESOURCES TO THE REGION(S) OF INTEREST AND RESULTING IN GREATER RELEVANT INFORMATION FROM THE IMAGER AT ORDERS OF MAGNITUDE REDUCTION IN DATA. CLOSED-LOOP OPERATION ALSO PERMITS THE FOVEAL VISION SYSTEM TO ADAPTIVELY OPTIMIZE ITS PERFORMANCE TO THE GIVEN VISION TASKS. IMAGER TOPOLOGIES WITH DIFFERENT SAMPLING ELEMENT GEOMETRIES, CORE CONFIGURATIONS, AND RESOLUTIONS ROLL-OFF FUNCTIONS ARE INVESTIGATED WITH DIFFERENT FEEDBACK CONTROL FUNCTION (BAYESIAN LEARNING, MAXIMUM LIKELIHOOD, NON-PARAMETRIC). THEIR RELATIONSHIP WITH OVERALL VISION SYSTEM PERFORMANCE IS BEING ASSESSED. WHEN FULLY DEVELOPED,THE CLASS OF COMPUTER VISION SYSTEMS CAN PERFORM HIGHER LEVEL VISION TASKS THAN CONVENTIONAL VISION SYSTEMS, WITH LESS PROCESSING TIME, HARDWARE, AND DATA. SYSTEM PERFORMANCE COULD BE SUBSTANTIALLY IMPROVED THROUGH ADAPTIVE SENSOR AND PROCESSING RESOURCE ALLOCATION. ... MOREOVER, THE SAVINGS FACTOR INCREASES WITH FIELD-OF-VIEW AND RESOLUTION WOULD MAKE SUCH SYSTEM PARTICULARLY ATTRACTIVE IN LARGE-SCALE FOCAL PLANE ARRAY APPLICATIONS.