NEURAL NETWORK RETINAL MODEL REAL TIME IMPLEMENTATION

HNC PROPOSES TO IMPLEMENT AND DEMONSTRATE A REALTIME NEURAL NETWORK MODEL OF THE RETINA FOR TARGET DETECTION AND TRACKING. THE ALGORITHM FOR THE NEURAL NETWORK PROCESSING WAS INVENTED AND INITIALLY DEMONSTRATED ON A SUN SPARCSTATION TM AT LAWRENCE LIVERMORE NATIONAL LABORATORY WITH GOOD RESULTS. HNC PROPOSES TO USE THE LATEST VLSI NEURAL NETWORK HARDWARE TO DEMONSTRATE AN IMPROVED AND OPTIMIZED VERSION OF THE ALGORITHM FOR REAL TIME TARGET DETECTION AND TRACKING. THE REAL TIME ALGORITHM CAPABILITIES WILL BE ACHIEVED BY A NEW, HNC DESIGNED, IMAGE PROCESSING NEURAL NETWORK CHIP SET. THIS CHIP SET, CALLED THE VISION PROCESSOR (VIP)TM, WAS INITIALLY FUNDED ON A DARPA CONTRACT, AND IS BEING FABRICATED BY HNC AND A COMMERCIAL PARTNER. THECHIP SET IS IMPLEMENTED AS A VME IMAGE COPROCESSOR FOR THE SUN WORKSTATION. THE OBJECTIVE OF PHASE I IS TO IMPLEMENT ON THE VIP COPROCESSOR HARDWARE THE NEURAL NETWORK ALGORITHM THAT PRESENTLY RUNS IN SOFTWARE ON THE SUN SPARCSTATION. SIMULATIONS OF THE VIP CHIP SET HAVE SHOWNTHAT IT CAN EXECUTE THE TRACKING ALGORITHM'S OPERATIONS OVER 1000 TIMES AS FAST AS THE SOFTWARE CAN EXECUTE THEM. ANTICIPATED BENEFITS/POTENTIAL COMMERCIAL APPLICATIONS - IN PHASE II, HNC PLANS TO DEVELOP AND DELIVER A PROTOTYPE TRACKING SYSTEM FOR A SPECIFIC TRACKING DATA SENSOR FOUND IN A RADAR, OPTICAL, OR INFRARED DEFENSE SYSTEM. THE PROTOTYPE WILL BE ABLE TO TRACK MANY OBJECTS IN A CLUTTERED ENVIRONMENT IN REAL TIME. SINCE THE PROTOTYPE SYSTEM IS BASED ON A SMALL, LIGHTWEIGHT VLSI CHIP SET, IT CAN ALSO BE INTEGRATED INTO SMART WEAPONS FOR TARGETTRACKING AND RECOGNITION. THE RETINAL MODEL CAN BE EXPANDED IN THE FUTURE TO INCLUDE VISION FUNCTIONS OTHER THAN TRACKING, SUCH AS SEGMENTATION AND ATTENTION FOCUSING. THIS SYSTEM CAN SERVE AS THE BUILDING BLOCK OR MODULE IN A COMPLETE VISION SYSTEM FOR AUTONOMOUS VEHICLES OR ROBOTIC SYSTEMS.