In this paper, a visual-based surveillance system for monitoring motorway infrastructures is presented. The system receives in input a sequence of color images and makes out an interpretation of the observed scene according to a set of possible events previously defined.The considered events are divided into four classes and range from normal situations (any object is present on the surveilled overpass) to very dangerous situations (persons performing harmful behaviors). Experimental results are presented on a long real image sequence containing two persons with suspicious behavior moving into a limited area of a motorway overpass.