BUILDING STRUCTURAL CORNER DETECTION USING HIGH RESOLUTION OBLIQUE AIRBORNE IMAGES

摘要

Detecting high accuracy building structural corners promotes building three dimensional (3D) reconstruction and measurements considerably. In this paper, a method which can detect building structural corners from high resolution oblique airborne images was proposed. This corner detector can extract building corners as completely as possible so that those sparse corners can define a intact building without uncertainty caused by key points missing. The premise of this method is that straight lines are rich on the surface of building. Then, two of them intersect each other and generate a corner. In this method, straight lines were extracted firstly. Then a rule named "be mutually closest endpoints" was proposed and employed to find straight line pairs for generating corners. Corners detected by this method are not only points with gray meaning but also points which can describe the structure of building surface in certain level with topological relation. In addition, three indexes of accuracy evaluation were proposed in this paper. And results obtained from our method and another contour based method were compared according to these three indexes. Experiments conducted on different areas of images, such as image areas which contain building facade, in this paper show that this method is effective to detect building structural corners with high accuracy of subpixel. And at the same time, this method can make effective suppression on the generation of points which are not building structural corners.