e circle 
ted. Of 
"ü P). 
> of the 
and to 
; called 
ities of 
jacities 
a pre- 
a high 
) lower 
ittle bit 
bout 5 
octor’s 
nalysis 
iologic 
| Chest 
onnier- 
density 
Chest 
ysis of 
  
Kunii, Yoichi 
AUTOMATIC STEREO MATCHING USING OPTICAL FLOW FOR 3D OBJECT MODELING 
Yoichi KUNIT , Hirofumi CHIKATSU 
Tokyo Denki University, Japan 
Department of Civil Engineering 
99smg08@g.dendai.ac.j 
** 
chikatsu(g)g.dendai.ac.]p 
Working Group IC V/III 
KEY WORDS: Optical Flow, 3D Object Modeling, Stereo Matching, Coplanarity Condition, Line Tracking, 
Sequential Images. 
ABSTRACT 
In order to acquire 3D spatial data for objects using stereo image, stereo matching such as SSDA, area based matching 
and so on have been proposed. However, there are some issues for efficient 3D object modeling. In particular, efficient 
line matching for reconstruction of the objects such as buildings is needed to be resolved. With this objective, this paper 
investigates automatic stereo matching method using optical flow estimation. Furthermore, this paper investigates a 3D 
spatial data acquisition method using coplanarity condition, and shows 3D modeling of a building which was performed 
automatically. 
1 INTRODUCTION 
Recently, efficient spatial data acquisition and visualization have been received more attention from the view point of 
3D-GIS, city modeling, and so on. Generally, 3D spatial data is acquired by stereo matching, and there are many stereo 
matching methods, e.g. SSDA, area based matching. However, automated segmentation of feature areas or extraction of 
feature lines of objects are still issues which are needed to be resolved for efficient modeling. With this objective, this 
paper investigates an automatic stereo matching method using optical flow. The optical flow is estimated using 
sequential images for objects, and the optical flow has ability for line tracking. The line tracking is performed 
automatically in the sequential images from the first frame to the last frame. Then, this paper shows that the stereo 
matching can be performed automatically, and the feature lines for objects can be acquired simultaneously. 
Furthermore, 3D spatial data acquisition method using coplanarity condition is investigated as follows: Firstly relative 
ground coordinates are calculated by the relative orientation using coplanarity condition. Secondly, the relative ground 
coordinates are transformed to absolute ground coordinates. Finally, the absolute ground coordinates were set as 
approximate values for the bundle adjustment, and 3D coordinates of the each point on the object are calculated. 
Therefore, 3D spatial data can be acquired efficiently, and shows that 3D modeling of a building is performed 
automatically in this paper. 
2 AUTOMATIC 3D OBJECT MODELING METHOD 
2. Automatic Stereo Matching 
Stereo matching was performed automatically using optical flow in this paper. The detail procedures of the automatic 
stereo matching method are as follows. 
2.1.1 Line Extraction: The line extraction was performed by Canny operator (CANNY, J., 1986.) with 2 threshold 
values relative to 2 components which called the height and reliability of edge in this paper (YOKOYAMA, H. etal, 
1998.). The height of edge is a variation of the gray level periphery at the point, and the reliability is an index for 
representing influence of noise. Threshold value of height was set at 10, and reliability was set at 0.1 in this paper. In 
order to perform line tracking, each both ends for the line was connected by straight line. Figure 1 shows original image 
of the house model and figure 2(a) shows filtering image land (b) shows straight line image. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 459