Study On Automatic Building Information Extraction From Aerial Images In Urban Areas

Geographic information acquisition of city by using high-resolution aerial images is one of the primary technologies for city programming. How to extract building information from aerial images is a key technical issue desiderated to be solved. In this paper, some research about extraction of 3-D information of building from high-resolution urban aerial images is implemented.Some analyses of data feature in urban aerial image are done first, and a split-block-clustering based on FCM method is discussed for the coarse segmentation of building groups, which can reduce the search region for buildings extraction and decrease calculation much more. In a general way, aerial image is not orthograph, most buildings exist slope angle as well as vertical lines in aerial images. In this paper, Combined with vanishing point geometrical restriction given by the aerial photogrammetric modeling, a method for extracting vertical lines based on the adaptive fuzzy Hough transform is presented. Then the extracted vertical lines are validated by analyses of window texture. Based on the extracted vertical edges, a method for roof boundary extraction in monocular aerial images using line Snakes model is proposed. In addition, to detect buildings which have smooth gray-value in roof., a method for extracting building roof region is presented, which is based on fitting ground points of vertical lines from multiple aerial images. Both the two methods mentioned above mainly apply to the analysis of high-rise buildings with slope angle of view, and are basically unaffected by the complexity of building shapes. The results of experiment on actual image are good, which show that the methods are effective.