A new advanced and effective indicator is proposed and validated to assess the safety of road infrastructures. The indicator is based on the following assumption: a subject driving on a self-explaining road assumes a correct and safe trajectory and the local transversal accelerations depend only by the curvature of road geometry. If driver corrects more often the vehicle's trajectory rather than what road curvature imposes, the road is not self-explaining and, consequently, it can be unsafe. If the local transversal accelerations do not depend only by curvature, they are biased by driver's corrections of trajectory. The proposed indicator takes into account the frequency and the amplitude of anomalous corrections of trajectory. The theoretical hypothesis of high correlation between the proposed indicator and the observed accident rate has been verified using an advanced driving simulator. Two Italian case studies are presented. The numerical results confirmed such a theoretical hypothesis. The values of correlation parameters are much more high than any expectation. These outcomes are exceptionally promising but validations to other case studies are suggested before generalization.