A Model-Driven-to-Sample-Driven Method for Rural Road Extraction

Road extraction in rural areas is one of the most fundamental tasks practical application remote sensing. In recent years, sample-driven methods have achieved state-of-the-art performance road tasks. However, are prohibitively expensive and laborious, especially when dealing with roads irregular curvature changes, narrow widths, diverse materials. The template matching method can overcome these difficulties to some extent achieve impressive results. This also has advantage vectorization results, but automation limited. Straight line sequences be substituted for curves, use color space increase recognition nonroads. A model-driven-to-sample-driven a much higher degree than existing proposed this study. Without prior samples, on basis geometric characteristics long using advantages straight lines instead curved lines, center point model established through length constraints gray mean contrast sequences, completed topological connection analysis. addition, we take extracted manual input data as local improved segment histogram determine direction, panchromatic hue, saturation, value (HSV) interactive measure complete tracking extraction. Experimental results show that, different types scenarios premise, accuracy recall rate evaluation indicators reach more 98%, and, compared other methods, algorithm increased by 40%.