INTERPRETING BUILDING FACADES FROM VERTICAL AERIAL IMAGES USING THE THIRD DIMENSION

Information is being extracted from vertical aerial photography and various data products for an efficient interpretation of terrain objects. Our focus lies on characterizing individual properties using aerial imagery as accurately as possible. We want to determine the size of buildings, their number of floors, the number and size of windows, the existence of impervious surfaces, status of vegetation, roof shapes with chimneys and sky lights etc. To achieve robust results it is very important to incorporate all data that a set of images can offer. A key aspect therefore is the inclusion of the 3rd dimension when interpreting facade images and to deal with the different imaging angles when interpreting the details of building facades from vertical aerial photography. This paper first addresses the question which incidence angles are sufficiently large to get useful results. Secondly we show that novel oblique imagery suffers from excessive occlusions that prevent the floor and window detection to produce meaningful results. We finally explain the use of 3D point clouds to deal with complex facades with balconies, awnings and arches. Furthermore, we obtain from the 3d representation of the facades also their exact footprint. We expect to achieve an enhanced quality of the floor counts and window detection results. The topics are closely related since one first needs to understand which facade images have been taken under too small angles so that the facades are grossly distorted. Second, the plane sweep algorithm needs images with as small a distortion as possible, and with a pruned-down set of images.