The purpose of this work is to propose a reliable method to detect outliers, continuing the work done in a previous article. The previous work was affected by a serious defect related to a rather unstable local estimate of a dispersion index. This paper relies on a robust estimation of a relevant parameter that, before being used for testing the data, is processed with a low-pass filter. The many advantages, obtained in this way, are highlighted in an experiment on a real D.T.M. dataset.