News from CyberCity-Modeler

Semi-automated object extraction has become a viable concept for the generation of 3-D city models. CyberCity-Modeler (CCM) has been developed with the aim of creating not only buildings, but also other objects pertaining to a city model efficiently and with a high degree of flexibility concerning the level of detail. In its commercial implementation, CCM has been confronted with a number of user requirements which needed to be observed. This led to some extensions in functionality, which are addressed in this paper: Geometrical regularization of buildings, editing functions for topology adjustment, integration of façades and other vertical walls and modeling of roof overhangs. These extensions of the original concept make CyberCity-Modeler an even more powerful tool for 3-D city modeling. 2 CCM – THE ORIGINAL CONCEPT CyberCity-Modeler, as the name suggests, was designed as a tool for data acquisition and structuring for 3-D city model generation. From the very beginning, CCM has been devised as a semi-automated procedure. This was done in view of the need to observe the following constraints: − Extract not only buildings, but other objects as well, like traffic network, water, terrain, vegetation and the like − Generate truly 3-D geometry and topology − Integrate natural (real) image textures − Allow for object attributation − Keep level of detail flexible. Accept virtually any image scale − Allow for a variety of accuracy levels (5 cm to 2 m) − Produce structured data, compatible with major CAD and visualization software In site recording and modeling, the tasks to be performed may be classified according to: − Measurement − Structuring of data − Visualization, simulation, animation − Analysis In CCM, the image interpretation and even the measurement task is done by the operator. The software does the structuring. For visualization, simulation, animation and analysis we largely resort to other parties’, mostly commercial, software. Figure 1 describes the workand dataflow of CCM. The operator measures on an Analytical Plotter or on a Digital Station in the stereomodel individual points that fully describe the visible part of an object, i.e. the roof of a building. The measurement sequence of these points is only weakly structured. Figure 1. Workand dataflow of CCM. CCM presents a new method for fitting planar faces to the resulting 3-D point cloud. This face fitting is defined as a consistent labeling problem, which is solved by a special version of probabilistic relaxation. In theory there are various labeling methods available, but only one solution is desired, which meets the inherent topological constraints of the object. From a geometrical point of view, the inherent topological constraints can be summarized as: (1) a 3-D object is a closed multiple-plane object; (2) planes are not supposed to pierce each other; (3) Invited Paper, Third International Workshop on Automatic Extraction of Man-Made Objects from Aerial and Space Images, Monte Verita, Ascona, Switzerland, 10-15 June 2001 every two adjacent boundary points are always part of a face. As an automatic topology generator, CCM is generic in the sense that any object, which is bounded by a polyhedral surface, can be structured. With this technique, hundreds of objects may be measured in a day. The computation of the structure is much faster than the measurements of the operator, such that the procedure can be implemented in on-line mode. If overlay capabilities are available on the stereo device the quality control and the editing by the operator becomes very intuitive and efficient. The DTM, if not given a priori, can also be measured and integrated. Texture from aerial images is mapped automatically on the terrain and on the roofs, since the geometrical relationship between object faces and image planes has been established. Façade texture is produced semi-automatically via projective transformation from terrestrial images, usually taken by camcorders or still video cameras. The system produces its own internal 3-D data structure, including texture. Interfaces to major public data formats are available. For a detailed description of CCM see Gruen & Wang (1998). The system and software are fully operational. In the order of 100,000 buildings at high resolution have been generated already with this approach. Figure 2 shows one of the models, the Congress Center RAI, Amsterdam, location of the XIXth ISPRS Congress 2000. Figure 2. 3-D model of the Congress Center RAI, Amsterdam, produced with CCM (vector data, overlaid with natural texture). 3 SOME RECENT EXTENSIONS What looks like a complete approach and system from a scientific point of view may not necessarily fulfill some specific practical requirements efficiently. This is the case with the original approach of CCM. Whereas it does the job in most projects very well there are always some specs which require modifications and additions. One of those is geometric regularization. While CCM was built to model the objects as close to their existing size and shape as possible, there arises sometimes the need to regularize the geometry. Under these constraints do fall the requests to make straight lines parallel and perpendicular where they are actually not, or to have all points of a group (e.g. eaves or ridge points) at a unique height. Another problem grew from the fact that CCM was designed to handle individual buildings sequentially and independent of each other. Building neighbourhood conditions were not considered. The geometrical inconsistencies originating from that fact, like small gaps or overlaps between adjacent buildings (in the cm/dm range), are not dramatic and tolerable in many applications, especially those which are purely related to visualization. However, the topological errors constitute a serious problem in projects where the 3-D model is subject to legal considerations or some other kind of analysis which requires topologically correct data. Another significant extension refers to the precise modeling of building façades. Façades are usually not visible in aerial images, but available in cadastral maps. We combine this façade information with the roof landscape modeled with CCM in order to be able to represent the roof overhangs. We also show that we can model other vertical walls explicitly. In the following, we will describe all these extensions in more detail. Figure 3 shows the flowchart of the processes mentioned above, which are executed after the face definition by probabilistic relaxation is done. Figure 3. Flowchart of CCM extensions. The dashed connection between „Vertical wall integration“ and „V3D“ is under development. 3.1 Geometrical regularization and neighbourhood topology Geometrical regularization refers to the task of modifying the geometry in such a way that regular structures are obtained. Measurements from images are always erroneous, although the errors may be very small. In addition, in particular with older buildings, the geometry deviates from regular patterns sometimes significantly. Edges are not parallel, intersections not perpendicular, roof faces not planar. We therefore have developed two strategies for regularization: A fully automatic adjustment based on least squares and a semi-automated approach of CAD editing. Both approaches are integrated in the software package CC-Edit. The requirements for geometrical regularization are as follows: − Same height for groups of eaves points, ridge points and other structure points − Roof patches containing more than 3 points should form planar faces − Parallelity of straight edges − Right angles of intersecting roof edges − Collinearity of edge points 3.1.1 Least squares adjustment We solve these requirements by formulating the constraints as stochastic constraints, i.e. as weighted observation equations in a least squares context. This results in the following system for each roof unit: V3D initial Interface Geometric regularization Automatic by LSA adjustment Semi-automatic by CAD editing Neighbourhood topology correction Vertical wall integration Interface V3D corrected Invited Paper, Third International Workshop on Automatic Extraction of Man-Made Objects from Aerial and Space Images, Monte Verita, Ascona, Switzerland, 10-15 June 2001 (1) In a first step all points which form a group whose height should be a unique value are identified and an average height is computed as ; n z z z ij j ∑ = group point of D I group per points of no. = = = j n ; n ,..., 1 i z z (1) This height is introduced as approximate value in the following computations. There will be only one correction value dz for all points of this group. The group definition is subject to a tolerance value set by the user (e.g. ± 0.2 m). (2) Parameters of planar faces The parameters of the planar faces are derived from the observation equations faces of ID face planar per points of no. = = = = + + + k n ; n ,..., 1 i w ; v D z C y B x A p p ik ik k ik k ik k ik k (2) v = residuals w = weights Since the system is linear in A, B, C, D, we get these parameters in just one least squares adjustment (LSA) step. (3) Parallel edges constraints For each group of parallel straight edges, the parameters of a 2D straight line are derived from the observation equations line th group line th line th in points of no. m m k k m n ; n ,..., 1 i w ; v c y b x a p p im im km im k im k