RECONSTRUCTING CCTV, BEIJING

Abstract. This paper deals with the reconstruction of a building, starting from a point cloud. The shape of this building is a non-stellar concave and multi-connected structure, composed of sowns and chains. A sown is the representation of a horizontal plane formed by dense points. A chain is a planar loop modeled by rare points. CCTV structure is defined only by the three orthogonal Cartesian coordinates. The reconstruction uses a sequence of procedures and the desired output is a consistent 3D model. The first procedure is devoted to attributing points to their voxel and to estimating the three values needed afterwards. The second procedure is devoted to analyzing clusters vertically and horizontally, to preliminarily distinguishing chains from sowns and to generating relational matching. The third procedure is devoted to building closed loops between all chains and all their projections on sowns. The fourth procedure is devoted to connecting points with triangles. The fifth procedure, still being implemented, is devoted to interpolating triangles with triangular splines.