This paper presents a robust approach for generating digital maps from map's blue prints. According to the characteristics of human computer interaction, the paper developed reliable algorithms to extraction man-made object such as road and building. The extraction process of road consists of three modules of rolling ball, road intersection detecting, and road network searching. The extraction of building is dividend into two classes: closed graphs and open graphs. The paper concentrated on the building extraction of open image graphs. Given a starting point, we are able to generate hypotheses of the building seeds and eliminate error seeds by considering the geometric and topologic relations among seeds, and then obtain closed building boundary graph. The building map data are approximated by using the polygonalization method based on dominant points. Graph structure was used to detect and reconstruct map data of building and road, which made our algorithms be extended robustly and efficiently to different original image.
Abstract