A Novel Method for CFAR Detection Threshold by Right-tail Local Cumulative Density Function Estimation

In Constant False Alarm Ratio (CFAR) detection, the traditional method for detection threshold, which is based on Global Probability Density Function (GPDF), may suffer remarkable performance loss due to model mismatch and parameter estimation error. In order to solve this problem, a novel method independent of clutter model is proposed. Without clutter GPDF assumption and parameter estimation, we obtain detection threshold by fitting the right-tail Local Cumulative Density Function (LCDF) with polynomials. Experiments with both simulation and real data confirm good CFAR performance and validity of this method.