Generation of Building Facade from Sequential Urban Images for Polygonal Buildings on 3D Map

In this paper, a method to generate building wall textures from an in-vehicle camera is proposed as an aid to construct 3D maps. The building wall textures are required to attach to 3D polygons which are obtained through 3D measurements in urban space. We assume that the in-vehicle camera is under linear uniform motion and building walls are planar regions perpendicular to the optical axis. Under the assumption, the same building wall region has the same depth, or disparities, over the region among successive images. Since disparities derived from foreground objects are different from the disparities derived from the building wall, we can use the disparity differences as a clue to effectively distinguish the building walls from the foreground objects. We formulate the extraction of building wall textures incorporating these considerations as an optimization problem which can be solved by graph-cuts algorithm. To show the effectiveness of the proposed method, it is applied to a sequential scene in a miniature model street and an actual street.