Vibrometry classification of moving vehicles using throttle signature analysis

Any operating vehicle will emanate a degree of vibration due to the running engine and other machinery. These vibrations differ based on engine type, throttle-level, and vehicle structure and as such they constitute a unique signature for a given vehicle. The availability of remote vibrometry sensors, such as pulse laser radar, makes it possible to utilize this signature for remote automatic target identification. We propose a novel classification technique that is specifically tuned to identify vehicles whose engines have a variable amount of applied throttle based on a limited training set with known idle and full throttle values. The presented algorithm is real-time and operates reliably with short time duration samples of input data. Other benefits include a small training set and simple implementation based on linear algebraic techniques. It is also possible to adjust the technique to indicate the degree of confidence in the result. We have achieved classification accuracy for a two class problem of 90% to 98%.