Zuxun Zhang
THREE DIMENSIONAL RECONSTRUCTION AND VISUALIZATION
OF REGULAR HOUSES AND THEIR TEXTURE FROM IMAGE PAIR
Zuxun Zhang, Jianqing Zhang, Yinghao Zhu, Yu Zhang
Wuhan Technical University of Surveying and Mapping, P.R.China
zxzhang@supresoft.com, jianging@supresoft.com
KEY WORDS: 3D, Reconstruction, Visualization, House, Texture, Image Pair
ABSTRACT
Three-dimensional reconstruction and visualization of houses with their texture rendering are very useful in urban
planning, communication and tourism etc. An automation of house reconstruction is tested. The digital surface model
(DSM) is first created by image matching, based on hierarchical relaxation, and the approximate positions of houses are
extracted by detecting the lumps or hills. Within these areas single image analysis and perceptual organization are made
to detect the boundaries of house. A mathematical model based on least squares matching with geometrical constrains
in object space has been used to recover the house. Finally, texture rendering of visual surface on the buildings is
achieved based on the complete 2.5D triangulated irregular network (TIN) of house surface. The real landscape of urban
is constructed.
The theory and method used in the reconstruction procedure mentioned above are introduced in this paper, and the
relative experiment results are shown and analyzed in each parts.
1. INTRODUCTION
Current methods of photogrammetry have solved most parts of the problem in automatic aerial triangulation and
automatic digital elevation model (DEM) acquisition. Nevertheless, because of the limitation of visual computational
theory and the hardware equipment as well as the complexity of the image data, the research work of the automatic
measurement for space objects is in the primary stage. The existing research results demonstrate that the problem of
automatic measurement for space objects is difficult to be fully solved at current stage. It is a bottleneck in full
automation of digital photogrammetric systems. In view of this situation, researches on the methods of semi-automatic
measurement for space objects have invited much attention. And these researches have achieved some progress and
paved the way for the final solution to the problem.
The paper introduces some elementary experiments of automatic extraction, 3D-reconstruction ad visualization of
houses with the basic procedures as follows: Using VituoZo digital photogrammetry system to begin stereo matching of
stereo image pairs and acquiring digital surface model (DSM), sequentially Extracting the approximate range of the
houses. Secondly, analyzing the single image and extracting the boundaries of the houses. Then, reconstructing 3D
geometrical models of houses by least squares matching with geometrical constraint conditions in object space (Zhang,
2000). And finally realizing the reconstruction of visible surface texture based by considering the full 2.5D triangular
irregular network (TIN) in the building surface and based on the original image, and creating real 3D landscape of a
city.
2. GENERATION OF DSM AND DETERMINATION OF APPROXIMATE LOCATION OF HOUSE
2.1 Image Matching and DSM Generation
The use of DSM and DEM in house detection is getting more and more important (Baltsavias, 1995; Haala, 1995;
Weidner, 1997) . Image matching, which is a process to search homologous images in two images of stereo pairs, is a
key technique to DSM generation. Generally, it may adopt feature-based matching or area-based matching. However,
the isolated single point matching, no matter whether feature-based or area-based matching, results in a rather low
reliability. Hence, more researches are focused on global matching (Heipke 1992. The generation of DSM presented in
the paper is realized in VirtuoZo, a commercial digital photogrammetry system (DPW).
VirtuoZo DPW is an image matching system based on image space, adopting the technique of pyramid data structure
and multi-level global image matching. It is a global matching algorithm with probability relaxation based on the
bridge-mode.
1016 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.