THOUSAND-OBJECT TRACKING WITH NEURAL NETS

NEW COMPUTATIONAL MEANS NEED TO BE DEVELOPED FOR THE PRACTICAL SOLUTION OF REAL-TIME TRACKING OF A THOUSAND OR MORE SPACE-BORNE OBJECTS AT ONE TIME. IT IS ASSUMED THAT, DURING AN ENGAGEMENT, DATA WOULD ARRIVE FROM A NUMBER OF SENSORS WHICH MAY BE LOCATED IN A NUMBEROF PLACES, OBSERVING OBJECTS IN SPACE WITH AND WITHOUT OVERLAPPING COVERAGE. THE SENSOR DATA WOULD BE FED TO COMPUTER EQUIPMENT DEDICATED TO THE TRACKING TASK, AND THE OUTPUT WOULD BE THE TIME HISTORIES OF DETECTED OBJECTS GIVING THEIR TRACK NUMMBERS AND POSITIONS VERSUS TIME. CONVENTIONAL COMPUTERS CANNOT PERFORM THE TRACKINGFUNCTION AT REAL-TIME RATES WITH A LOAD OF A THOUSAND OR MORE OBJECTS TO TRACK. MANY OF THE FUNCTIONS CAN BE DONE WITH NEURAL NETWORKS THATCAN BE IMPLEMENTED IN A HIGHLY PARALLEL STRUCTURE. SYSTEM ARCHITECTURE IS BEING DEVELOPED TO IMPLEMENT THIS TRACKING APPROACH. EFFECTS OF SENSOR NOISE AND SENSOR SIGNAL QUANTIZATION ON PROBABILITIES OF FALSE TRACK INITIATION AND LOSING A TRACK ARE BEING DETERMINED. A PRELIMINARY HARDWARE DESIGN IS BEING UNDERTAKEN FOR THE TRACKING APPROACH. WHEN SUCCESSFUL, THIS RESEARCH WOULD LEAD TO THE DESIGN OF A PRACTICAL, RELIABLE, ECONOMICAL SYSTEM FOR REAL-TIME SIMULTANEOUS TRACKING OF A THOUSAND PLUS SPACE-BORNE OBJECTS. SOLUTION OF THIS PROBLEM ALSO WOULD ADDRESS THE NATION'S TRACKING PROBLEM IN THE AIR TRAFFIC CONTROL AREA.