RCS based target recognition with real FMCW radar implementation

Abstract

In this article, we investigate the methods that can realize automatic target recognition and tracking by exploiting signal distribution of radar cross section (RCS) with frequency modulated continuous wave (FMCW) radar. In doing this we use the real RCS data measured from the short-range FMCW vehicle radar. We estimate the continuous valued degree of freedom and the mean of RCS distribution using maximum likelihood estimation (MLE) assuming that RCS follows gamma distribution. The experiments with real radar verify that parameterized gamma distributions for three targets of man, vehicle, and drone closely follow the empirical distributions. Then, we apply maximum a posteriori criterion (MAP) for target recognition. The average recognition probabilities for man, vehicle, and drone using MAP are 85%, 100%, and 92%, respectively. Since the vehicle has distinct RCS and thus perfectly recognizable, we apply a support vector machine (SVM) hoping to better classify the man and the drone. The man is recognized with similar accuracy, but the drone is not due to the lack of training samples, of which constraint is imposed by real implementation and experiment. © 2016 Wiley Periodicals, Inc. Microwave Opt Technol Lett 58:1745–1750, 2016