Department of Transportation · Federal Railroad Administration
IGF::OT::IGF RAILROAD ARTIFICIAL INTELLIGENCE INTRUDER LEARNING SYSTEM (RAIILS) PROPOSAL FOR THE PROCESSING OF DATA FROM HETEROGENEOUS SENSING PLATFORMS FOR THE PURPOSE OF IMPROVED DETECTION OF RAILROAD TRESPASSERS IN REAL-TIME USING CONVOLUTIONAL NEURAL NETWORKS (CNNS). FOR A SYSTEM EQUIPPED WITH CO-LOCATED NATURAL COLOR RED/GREEN/BLUE (RGB) AND THERMAL INFRARED CAMERAS, WE PROPOSE A DEEP LEARNING AI ARCHITECTURE TO DO ON-PLATFORM SENSOR FUSION. SUCH A SYSTEM WILL ALLOW FOR THE DETECTION OF RAILROAD TRESPASSERS BOTH IN REAL-TIME AND DURING THE ENTIRE 24HR CYCLE. THE PROPOSED ARCHITECTURE WOULD LEVERAGE BOTH THE RGB AND INFRARED DATA WHEN AVAILABLE AND WOULD CHANGE TO INFRARED ONLY WHEN ENVIRONMENTAL/TIME-OF-DAY CONDITIONS DO NOT PERMIT THE USE OF THE VISIBLE-SPECTRUM DATA.