In this paper a vision-based landmark extractor for goal-oriented navigation of a mobile robot is proposed. An indoor space scene analysis allows to detect natural structures relevant for navigation like doors or the floor in monocular images of the environment. The detection is based on a priori knowledge of the shape and functionality of searched structures. This algorithm works in real-time and is stable against variation of illumination. The approach is applicable for indoor environments which have clear line structures and large homogenous colour surfaces. The core of this method combines region based as well as edge based elements. The segmentation begins with an orientation selective Hough-Transform (OHT) including line segment detection and generates a reticule of convex polygons. Homogeneous colour similar polygons are segmented and merged by an region growing process. Finally a feature extraction and identification is performed to assign regions to known objects.