In this paper we present and evaluate several methods for real-time environment representation by extracting object delimiters from the traffic scenes using a Dense Stereovision System. The delimiters detection is based on processing the information provided by a 3D classified occupancy grid obtained from the raw dense stereo information. One of the problems in representing the environment through the occupancy grid is a large volume of data. Therefore we propose a more compact 2.5D model by representing the environment as a set of polylines with associated features. Two approaches to extract object delimiters are presented: an improved contour tracing called 3A Tracing and a polyline extraction method through the radial scanning of the occupancy grid. We discuss the advantages and drawbacks for each of these methods.