Knyaz, Vladimir 
  
APPROACH TO ACCURATE PHOTOREALISTIC MODEL GENERATION FOR COMPLEX 3D 
OBJECTS 
Vladimir A. Knyaz, Sergey Yu. Zheltov 
State Research Institute of Aviation System (GosNIIAS), 
Victorenko str., 7, Moscow, 125319, Russia 
knyaz@ gosniias.msk.ru 
  
Working Group V/1 
KEY WORDS: photogrammetry, spatial data, surface reconstruction, textures mapping, orthophoto, VR 
ABSTRACT 
Due to the progress in computers and sensors development the number of virtual reality application (VR) grows 
significantly. Different sensors and techniques such as laser range finders, still and video cameras, mechanical digitizers 
are used to obtain spatial object coordinates for producing 3D models. The two main requirements to the produced 
model are the accurate geometry reconstruction and the realistic texture generation. The means of digital close-range 
photogrammetry are quite enough to satisfy those requirements. The same images can be used as for accurate 3D- 
geometry reconstruction as for adequate photometry modeling - generation of realistic object texture. 
The close-range photogrammetry technique for accurate geometric and texturing modeling is well-developed for 2.5D 
(depth) object models. But for complex 3D objects that have no single valued plane projection there are some problems 
with surface reconstruction and texture generation. In this case uniform surface can not be reconstructed by traditionally 
used Delaunay triangulation procedure for all points of objects. So the orthophoto can not be generated because of 
surface singularity. The paper considers the approaches for these problems solution. It describes the method for 
reconstruction of the uniform surface for some classes of complex 3D objects represented in form of object space 
coordinates cloud. Also the method for accurate texture generation from a set of object images obtained from various 
viewpoints is presented. The results of object reconstruction and texturing are given along with the description of 
developed close-range photogrammetric system. 
1 INTRODUCTION 
Last years the growing number of applications has a demand for accurate and realistic 3D models. Along with the 
requirement of metric object geometry description many tasks have a need for photorealistic model presentation 
(another words accurate texture mapping). The latter requirement is important for some kind of expertise task such as 
recognition, identification based on 3D models when texture information significantly contributes for valid solution. 
The first step of 3D-model reconstruction is to obtain cloud of object spatial coordinates. It can be performed using 
various means beginning from mechanical measurement machine to laser range finder (Chikatsu H., 1997, Wehr A., 
1997.). If spatial coordinates cloud represent depth (so called 2.5D) model such as digital terrain model (DTM) object 
surface can be presented as single valued function z from x and y arguments z= f(x,y). So the surface reconstruction 
from point cloud can be produced automatically by common Delaunay triangulation procedure. In case of generating 
3D coordinate cloud by digital photogrammetry technique model texturing can be performed by well-known image 
orthotransformation procedure using the same images as for DTM generation. 
For complex 3D object which surface can not be presented as single valued function of two arguments in Cartesian 
system of coordinate the problem of surface reconstruction becomes more complicated. If object model in coordinates 
cloud form is obtained it can be divided into some number 7 of fragments so that for each fragment its surface can be 
presented as single valued function z; — f(x, y );i — 1,...,n . Then the fragments can be integrated in united surface. 
This approach requires a lot of manual operation such as points model division into fragments and fragments 
integration. Besides precision of reconstruction can decrease while dividing into fragments and assembling. 
The paper presents the method for uniform surface reconstruction for some classes of complex 3D objects. The 
approach is based on choice of special type projection which is single valued for all points of 3D coordinates cloud. It 
  
428 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000.