The testfield in the background contains many similar targets 
(repeated pattern) and they are used just as features for the 
   
  
  
   
processing. No 3-D information is available. As the sequence is In thi 
self-acquired, the parameters of the camera are known, but they from 
are recovered as described in section 3.2. The process presented proce 
in section 3.1 is applied to find the required tie points for the impor 
orientation of the sequence; four control points measured on the or an 
body are used for the bundle. The final configuration of the identi 
system after the adjustment is shown in Figure 10. 1997] 
matru 
: ogre of the 
a we musa degre 
» : Eo and tl 
P. will t 
Figure 7: 3-D point cloud (8523) of the human body . 
. I wo 
Therefore, for realistic visualization of the results, each point of Tech 
the point cloud is re-projected onto the central image of the toolk 
sequence to get the related pixel colour. The results are 
presented in Figure 8. 
Figure 10: Recovered camera positions and object coordinates Beye 
Then the matching procedure is performed with all consecutive er 
triplets of images and the obtained and filtered point cloud 
(5604 points) is shown in Figure 11 and 12. Bhat 
Or 
23 
Brov 
Vi 
Capt 
ca 
N 
Figure 8: Visualization of the point cloud with pixel intensity Coll 
0 
S. ANOTHER EXAMPLE m 
The reconstruction of the human body is also performed using a 
sequence acquired with a video camera handycam Sony DCR- D'A 
VX700E. The artefacts created by the interlacing during the m 
digitization process have to be removed: therefore one field of ol 
the video is deleted and then the remaining lines are 
interpolated. A less smooth sequence is obtained as the D'A 
resolution in vertical direction is reduced by 50 per cent. pl 
Instead of removing all the odd (even) lines, another possible A 
approach could be to remove lines just in portions of the video 
where interlace artefacts are present. Fau 
For the process 12 frames are selected out of a one minute T 
sequence around a standing person (Figure 9). 5 
Fisc 
p 
a 
N 
Fitz 
a 
| S 
p ( 
à Figure 12: 3-D point cloud (left) and visualization 
Figure 9: Three frames (out of 12) of the acquired video with grey scale pixel intensity (right) Gri 
sequence. Images have a resolution of 576x720 pixel 
—594—