Istanbul 2004 
. Pitman Publ. 
assadenentzer- 
dtviertel. Zeit- 
(8):431-436. 
ent of straight- 
try. Int. Arch. 
point detection 
p. 382-391. 
interactive 3D 
single images. 
a single archi- 
"roc. CIPA 18^ 
tion for archi- 
Sens., 32(B5), 
. 3D modeling 
ised approach. 
amera calibra- 
CAMERA PLACEMENT FOR NETWORK DESIGN IN VISION METROLOGY 
BASED ON FUZZY INFERENCE SYSTEM 
M. Saadat-Seresht*", F. Samdzadegan®, A. Azizi", M. Hahn” 
"Dept. of Surveying and Geomatics, Faculty of Engineering, University of Tehran, Tehran, Iran — (samadz-aazizi@ut.ac.ir, 
mostofi_n@yahoo.com 
"Dept. of Geomatics, Computer Science and Mathematics, Stuttgart University of Applied Sciences, Stuttgart, Germany - 
michael.hahn@hft-stuttgart. de 
Commission V, WG V/1 
KEY WORDS: Close Range, Design, Fuzzy Logic, Automation, Industry, Vision, Artificial Intelligence, Robotics 
ABSTRACT 
For measuring complex industrial objects using vision metrology systems, automatic optimum network design is a real challenge. In 
the absence of given or simulated 3D CAD models of the objects and the workspace, the complexity of objects introduces several 
uncertainty factors into the camera placement decision making process. These uncertainty factors include the vision constraints such 
as visibility, accessibility and camera-object distance. 
For more complex objects, visibility is vastly influenced by hidden areas, the incidence angle of a target and the camera orientation. 
Mutual dependency of these factors increases the difficulty of camera placement. Further these factors directly influence the 
mensuration quality, in particular, precision and reliability. If an a priori 3D CAD model of the object is available, the 
aforementioned ambiguities can be tackled. However, a 3D model is often not available which makes the camera placement problem 
a nondeterministic process. An answer to this problem is to develop a fuzzy logic inference approach for camera placement and 
network design. The idea is to deal with the vision constraints in a fuzzy manner. 
In this paper a novel method based on fuzzy logic reasoning strategy is proposed for the camera placement. The system is designed 
to make use of human type reasoning strategy by incorporating appropriate rules.The paper reports on the results achieved by testing 
the fuzzy based camera placement approach on simulated and real objects. The results indicate that this new conceptual approach has 
a remarkable strength for automatic sensor placement in vision metrology. 
1. INTRODUCTION 
Close range photogrammetry or vision metrology has 
demonstrated its capability as a precise measurement technique 
for 3D object acquisition with lots of applications (Atkinson, 
1998; Fraser, 2001; Ganci and Handley, 1989). Users appreciate 
vision metrology as a technique with high flexibility, 
considerable accuracy, relatively low cost, and a high level of 
automation compared to other optical and mechanical methods. 
Automation in vision metrology systems has shown fast 
progress, c.g. regarding efficient self-calibration models, 
exterior orientation devices, coded target measurements, image 
matching algorithms and others (Hattori et al., 2002). So far less 
attention is paid to automatic network design which should be a 
first step in various vision metrology projects. 
This paper focuses on intelligent network design based on fuzzy 
inference systems (FIS) and proposes a concept for automated 
camera placement. Many researchers in both fields of 
photogrammetry (e.g. Mason, 1995; Fritsch and Crosilla, 1990; 
Olague and Mohr, 1998) and machine vision (e.g. Sakane et al., 
1992; Cowan and Kovesi, 1988) have carried out investigations 
on network design or sensor (camera) placement. Our aim is 
deriving accurate coordinates of some object points by a 
network of multi-image convergent camera stations. In contrast 
to Olague and Mohr (1998) and Mason (1995) we aim at a 
strategy of local improvement rather than looking for a global 
network design. We not expect that a 3D simulated model is 
available. Most unique for this research is that for the first time 
fuzzy inference in introduced for camera placement. 
Figure 1 illustrates the concept of the proposed network design 
process. A primary or draft network design can be performed 
by users maybe even in the field. The output of this first step is 
a network with, for example, ten or more images. These images 
may allow to measures a high quota, e.g. of 95% of object 
points with good quality but for the remaining this is not 
guaranteed. The loop in Figure | indicates the iterations which 
  
Perform primary network 
based on simple heuristic 
network rules 
Y 
  
  
  
  
  
Photography from y 
: : Automatic image 
desiened camera stations 
measurement 
  
  
  
Automatic camera 
placement to improve 
weak points 
5 
Updating visibility model 
  
Bundle adjustment 
(3D object coordinates and 
their accuracy) 
  
  
  
  
  
  
   
  
     
  
  
  
  
  
   
Is there any 
weak points? 
End d. 
Figure 1. Proposed network design process