This paper presents a method to detect road markings on top-view road images, which are pseudo-top-view images created by transforming each frames in foreground drive videos under influence of difference of brightness caused by varying photography environment. Road markings in pseudo-top-view images appear differently due to the variation of shadow and markings' line faded. Therefore, the recognition process of road markings is difficult in real environment. Our method used shadow detection and run-length compensation for robust recognition. Detected markings are registered to database for car navigation system, and then it will enable to sound a warning when approaching to a point of caution.