In this paper, we describe a set of methods for generating façade meshes of cities from a series of vertical 2D surface scans. These series of scans are obtained while driving continuously on public roads under normal traffic conditions. The driven path is cut into easy-to-handle segments with approximately straight driving direction, such that each segment is quasi-linear, its scan columns have a sequential topological order and it can be processed by transforming it into a depth image. Dominant building structures are detected in the depth images and points are classified and separated into different scenery layers, e.g. foreground and main structure layer. Large holes in the main layers, caused by occlusion from front layer objects, are filled by planar or horizontal interpolation. The depth image is cleaned up by removing isolated points and filling remaining smaller holes, and almost watertight textured surface meshes are generated. The results on a large set of data, with several million 3D points, are shown and discussed.
