Given a set of airborne imagery acquired with a drone, we would like to characterize the visual architectural elements that are most distinctive for a certain neighbourhood. Since distinguishing architectural features of different neighbourhoods could be quite complex, it is a challenging task. We build a 3D point cloud of the neighbourhood from the collected images. We propose to characterize the architectural structure of the neighbourhood by extracting 3D visual features from the point cloud. Since our objective is to find representative architectural elements that is likely to define a slum, or a suburb, we test the performance of several 3D feature desciption methods for this purpose. A dataset is collected from two neighbouring boroughs of Istanbul, namely Ayazaga and Resitpasa, by flying a drone over these two pilot areas. Although Ayazaga and Resitpasa are very close to each other and separated by only a valley, their architecture are totally different. While the buildings in Ayazaga region have a very regular architectural structure, Resitpasa area just across the valley can be considered as a slum neighbourhood being predominantly composed of squatter houses. We first detect the roofs in the pilot regions and than extract the 3D features of the vertical surfaces beneath the roofs.