Kunii, Yoichi 
2.2.2 Experimental Results: R.M.S.E. were 
investigated by the coplanarity condition and combined 
  
the 3D adjustment with the coplanarity condition and the Table 1. RMSE of the each methods 
bundle adjustment. 
Table 1 shows the R.M.S.E. for check points by each 
method. It may be seen from the results of this Method > (mm) 2 (mm) 
  
experiment that the 2D accuracy for the each methods 
ing the are high value, and the accuracy of Z-coordinate for the Coplanarity Condition 9.030 110752 
combined adjustment is almost equal to the 2D accuracy. 
It means that the combined adjustment is sufficient for 
  
  
  
  
  
  
and the i | 
3D object modeling. Therefore, the combined Combined Adustment 31206 31.196 
ates of adjustment is useful method for 3D spatial data 
acquisition. 
2.3 3D Object Modeling 
using a : 
3 Te 3D spatial data of the house model could be acquired using the combined adjustment in this paper. Then, wire frame 
k circle model of the house model could be reconstructed automatically. Figure 8 shows the wire frame model of the house 
model. Furthermore, texture mapping was performed by following procedures: firstly one square surface was selected 
on the wire frame model on manually. Secondly, color information which was corresponded to the square surface was 
obtained from an image. Finally, the color information was put on the square surface, and repeating this procedures. 
Figure 9 shows the texture mapping model. 
h ratios 
  
   
Figure 8. Wire frame model Figure 9. Texture mapping model 
  
The detail procedures of the automatic 3D object modeling method 
are as follows: 
1. Line extraction is performed for the first image in the sequential 
images. 
2. The image coordinates of the both ends for each extracted lines 
are calculated, and line tracking start. 
3. Optical flow is estimated using the first image and the next 
image. 
4. Similarly, line extraction for the next image is performed, and 
image coordinates of the both ends for each extracted line are 
calculated. 
5. Each line positions in the first image are moved to amount of 
the optical flow. 
6. Moved lines and the lines in the next image are corresponded 
by similarity function, and just corresponded lines are remained. 
7. Above procedure is successively repeated to the last image, and 
the matching points between the first image and the last image 
are automatically acquired. 
Line Extraction for the First Image | 
    
       
      
    
   
  
  
Estimation of Optical Flow 
  
  
Line Extraction for the Next Image 
  
  
Line Position Movement 
  
Each Line Correspondence 
Stereo Matching 
  
     
  
  
  
3D Spatial Data Acquisition 
8. 3D spatial data is acquired using the combined adjustment. 
9. 3D modeling of an object is performed. ES 
These procedures are shown in Figure 10. Figure 10. Automatic 3D modeling method 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 463