Research On Road Information Extraction From High Resolution Imagery Based On Global Precedence

Automatic road extraction from remote sensing imagery is one of the hot topics in the area of remote sensing, surveying and mapping and computer vision, etc. Traditional semi-automatic/automatic road extraction methods extract roads based on local features by edge tracking or template matching, etc, while human always extract roads by global features. For improving the efficiency of road collection and updating from high resolution remote sensing images, it is a great need for developing an automatic or semi-automatic road extraction method according to human cognitive principles. This research studies the road extraction method from high resolution images in the background of GIS data updating. The main works and research fruits can be summed up as follows:(1) A global precedent approach for road extraction from high resolution imagery is proposed. The research status of road extraction is firstly analyzed and summarized. The basic process of human interpretation of images is analyzed. It is pointed out that the human can recognized the road network directly by eyes and maintain the connectivity of the road network in road tracking manually without the influence of the noise. It was revealed by cognitive psychology that human perceptual processing is from global to local in visual cognition. This global precedence concept is introduced by the authors to the road extraction from high resolution imagery by applying global topological perception prior to the perception of other locally featured properties. Topological information of roads is derived from vector data and used for global topological control of the connection of local road primitives. A framework for road extraction based on the global precedence is proposed in the thesis.(2) A method based on shape characteristics for road skeleton extraction from high resolution imagery is developed in the paper. An image segmentation method by region growing according to the similarity of the pixel gray level is adopted. For selection the proper road regions from the segmentation results, a road region extraction method based on the shape features of regions is designed using region area, the major axes of the ellipse, and maximum inscribed circle. An expression of road global structure features by morphological skeleton is achieved.(3) A global topological conflation method based on network snakes is proposed to build the global topology relation of roads in imagery space. Topological information of roads is derived from vector data and set as the beginning of the vision. A network snakes method has been developed for the automated conflation and building topological relations among road skeletons. After that, the topological information derived from the vector data is reset up in the imagery.(4) A method for extracting local features of roads under the global control is designed. The barriers separating the traffic flow is a sign the road existence and also expresses the central line information about road. The road centerlines are expressed based on the barrier information. An approach for extracting road barriers based on the Burns algorithm is achieved.(5) Several experiments are made to examine the method proposed in this dissertation. Experimental results show the reasonability and practicability of the global precedent approach for road extraction. The global topological conflation method proposed in this paper has been applied in the production of image map for the conflation of vector data and image data.