Research On High Resolution Remote Sensing Road Extraction Methods

Road extraction from high resolution remote sensing images has been widely used in many applications including retrieve and update of GIS data, target detection, image matching, digital mapping and so on. It is a research focus in the fields of artificial intelligence, computer vision, remote sensing and photogrammetry, which is scientifically and practically important. This thesis studied many existing methods on road extraction from high resolution remote sensing images and modified some of them considering the characteristics of the high spatial resolution images.First, the thesis presented a method for automatic road extraction from high-resolution images based on GIS data and Ribbon Snake model. The method created B-spline curves using GIS data as the initial road centerline, and constructed the regional energy function according to the texture feature of road region. The method could adjust the weight for each part of the energy function and the searching space of control points automatically during contour evolution. The experiments show that it is effective even though the initial curve is far away from target location.Then, the thesis modified the adjacency graph in path opening algorithm to detect roads with larger curvature, and combined the path opening algorithm with Ribbon Snake model for automatic road extraction from high-resolution remote sensing images. The initial road centerline achieved by path opening is successively processed by Ribbon Snake to evolve to the accuracy position. Eventually we obtained the smooth road centerline and road boundaries. The experimental results validated the feasibility and effectiveness of the method.Finally, the thesis proposed a novel method for extracting damaged roads based on modified path opening algorithm. The method combined the length of the longest path in path opening algorithm with the geometrical feature of the extracted area to set up road decision rules, which could automatically set the parameter values. Experimental results demonstrated the accuracy and efficiency of our proposed method.