Robust L1/geometric covariance matrix estimator: Comparison with huber-type M-estimator

Target detection in dense and inhomogeneous clutter requires a specific approach. An adapted modelization of the clutter is necessary and some have been proposed in the radar literature. Models family like SIRV (Spherically Invariant Random Vector) have been extensively used for their flexibility and their ability to accurately approximate real clutter. The key point for the detector is its robustness to be adaptative enough but without losing precision. The detectors that this paper deals with are compared regarding this particular trade-off.