Research on road information extraction from high resolution imagery based on global precedence

Abstract

Semi-automatic/automatic road extraction from remote sensing imagery is one of the hot topics in the field of remote sensing, surveying and mapping and computer vision, etc. Traditional methods based on Marr's Computation Theory of Vision follow the pattern of local-to-global features extraction. However, in high resolution image, the local features such as road boundary and road width are easily affected by noise and make the traditional extraction process based on edge extraction and template matching more difficult. At the same time, it is inconsistent with the human cognitive process during the visual interpretation. For all of the above reasons, the paper develop a new road extraction method according with human cognitive process. It is a kind of top-down road extraction strategy based on global precedence in the background of GIS data updating. The proposed methodology consists of four parts: 1) extract road priori topological information from the now available GIS data; 2) extract road morphological skeleton; 3) built the global features of road in the image space using automatic approximation conflation between road vector and road skeleton; 4) extract the local features in the high resolution images under the constraints of global features. The global topological approximation method based on Network Snakes algorithm is the focus in the paper. With the experiment on IKONOS image of Weihai city, the method was confirmed to be able to produce acceptable road global characteristics and local features (such as centerlines).