Semi-Automatic Extraction of Straight Roads from Very High Resolution Remotely Sensed Imagery by a Fusion Method

Abstract

Semi-automatic road tracking methods are regarded as more practical than automatic methods to obtain and update the vector road data. However, road tracking methods lose efficiency when they are applied to extract the straight roads compared to manual digitalizing. An efficient semi-automatic method is proposed for extraction of the straight roads in this paper. Particularly, spoke wheel algorithm and region growing algorithm are combined and fused together to extract a rectangle corresponding to a short road segment. The above process is performed at each pixel along the line segment between two seed points, which derives some rectangles. Moreover, statistical analysis of histograms of directions and widths of the obtained rectangles is made to determine the optimal rectangle. Then a precise road model is determined by the above optimal rectangle. Experiments show that our fused method is capable of efficiently extracting straight roads with a high spatial accuracy, even various types of noises existing.