1ggested, 
CAD, 
attempt 
ild up a 
using 
his sub- 
E(Space 
|. 1 here 
in. 4/260 
li video 
] on the 
id while 
362), a 
camera 
the two 
s can be 
led. The 
array of 
intensity 
1puter is 
d 
  
  
  
ation of 
language 
want to 
a called 
system, 
’omputer 
niversity. 
ted with 
apability, 
ects. An 
17; three 
nany an 
results of 
pond to 
> overlap 
Now we 
SSeS. 
The original image is firstly put into computer ® Detecting line, which is to detect the line is 
after CCD grabbing, then filtering, enhancing straight or curve. 
and smoothing are implemented. Finally, the e Face coding, after being operated above, face 
image is segmented into faces. Several main coding is completed in the end. 
processing are dealt with as follows: 3) Constructing attribute graph: We first put in 
order the faces with order number 1,2,3,---, then 
[ I I define the attribute name and attribute value 
  
according to regulations above. The connected 
  
  
  
  
LP Soft | | CAD Aspect |: | Input Image 
relation is determined by neighbor faces which 
have the same coordinates. 
4) Constructing volume primitive graph and 
hypergraph: The face interpretation by seeking 
for consistent face code is done, and then 
Segment constructing the volume primitive graph and 
attribute hypergraph according to face coding 
labeling, face code mergence and face code 
splitting principle. 
5) Model attribute graph: data structure from 
CAD system is translated into attribute relational 
graph including attribute hypergraph, primitive 
[Attribute Graph | Face Matching | graph and attribute relational graph. 
i 6) Model matching: an unique primitive is 
Model Matching | Straight Matchinb selected as start research point. Matching is 
adopted tree-back research. 
7) Recognition of primitives: various 
o Automatic Measurement | primitives is recognized by operations above. 
8) Reconstruction of 3D object: Recognized 
object is measured by LP, and then is 
represented with contour. 
Fig. 14 Flow chart for soft component 
1) Image segment: Man-machine interacting 8. CONCLUSION 
is designed for non-perfect segment since weak The presented system is an initial attempt, 
reflect edge, puzzle edge as well as which integrates CAD, Photogrammetry and CV. 
discontinuous edge in image segment. However, it is a good start for photogrammetry 
2) Face coding: In order to code for face, it is in industrial application. It builds up a bridge 
necessary to implement below steps: between CAD and CV using photogrammetry. 
e Face filling, which is adopted 4-neighbor we should further perfect its functions. 
seed point algorithm, it can obtain the area. 
e Detecting boundary, after being filled with Acknowledgments 
above, the face is binary gray image. Detecting, Most part of the work described in this paper was done 
thinning and deleting burr are implemented for When author is in pursuing Ph.D. in Dept. of Photogrammetry 
and Remote Sensing at Wuhan Technical University of 
this boundary. Consequently, the boundary is Surveying and Mapping (WTUSM). So the author is certainly 
determined with a pixel width. very greatly to my advisor, Prof. Li Deren, for the novelty idea 
e Boundary vector, which is adopted 8- and broad knowledge inspiring me, as well as enthusiasm and 
continued support. The author also wants to thank my 
neighbor tracking operator. It can obtain vector colleagues, from particularly CAD and vision research group 
data of boundary and its perimeter. in Dept. of computer science and technology at Tsinghua 
e Detecting corner, which is adopted curvature University for many stimulating and helpful discussions in 
module of model transfer, which is how to transfer data 
difference maximum between forward K steps structure in CAD system into vision model( attribute relational 
and backward K steps (K=3). It is usually graph). They fully appreciate the work, and allow to mount our 
necessary to suppress local non maximum. system into GEMS system and help me test whole system. 
e Put in order again for vector data, in order to 
detect whether the line is straight or curve or not REFERENGFS 
. 1. I. Biederman, Human Image Understanding: recent research 
between neighbor two corners, vector data have nd theory, CVGIP, Vol. 37, No. 1, PP.29-73, 1985. 
to be put in order again. 
  
  
  
  
  
  
  
  
ckage System Code Bas 
  
  
Knowlädge Base 
  
  
Model 
Transfer 
  
  
  
  
  
  
  
  
  
  
  
  
  
1025 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996