Road Network Extraction in Suburban Areas

Abstract

Abstract In this paper, an algorithm for the extraction of road networks in suburban areas is presented. The algorithm is region‐based and uses high‐resolution colour infrared images as well as, optionally, a digital surface model (DSM). The road extraction starts with a segmentation using the normalised cuts algorithm; afterwards the segments are grouped. Road sections are extracted from the grouped segments. Road sections that are likely to belong to the same road are connected to subgraphs in the next step. To eliminate false connections in the subgraphs, context objects such as vehicles, buildings and trees are employed. The remaining road strings, represented by their centre lines, are connected to a road network. The process employs combinations of radiometric and geometric features, derived from knowledge about the appearance of roads in suburban areas. Results are presented for two test data‐sets, acquired by different sensors. A quantitative analysis is performed for the quality of the road extraction as well as the topological quality of the extracted network.