Initial work focused on planar facades and resulted in a correct floor count at a rate of 93% and a correct window count of 86%. A study of the dependence of these counts on the angle off-nadir in the images reveals that best results are achieved in the range of 20° to 27° off nadir. At greater angles, such as those used in oblique aerial photography, occlusions start to become distinctive so that floor and window counts fail. Such counts will also fail if a facade is complex, yet the analysis method is unaware of that complexity. One needs then to consider the 3D shape of building facades, for example with extruding stair cases, balconies, awnings. We explain in this paper an approach based on 3D clouds of facade surface points, all extracted from overlapping vertical images. We have explained the approach by example and will have to embark on a systematic study of its performance with many buildings and imaging configurations. It seems not only important to test the work flow on many buildings, but also that the buildings represent a variety of sizes and styles and come from different regions (cities and rural areas) and parts of the world.