3. Istanbul 2004 
. View 1 of the 
ase. The rota- 
pairwise regis- 
vere calculated, 
» present in the 
rding to Eqn. 3 
  
senting the dif- 
is derived from 
ount of rotation 
tation matrices 
ation. The dif- 
of each view n 
nd global regis- 
S. 
©) 
lation vectors is 
ition in order to 
ments the mesh 
sh resolution of 
vs the t,, for all 
10ws the 0, and 
4, for view | of 
view 1 is taken 
bunny (Fig. 4), 
ation (1.2°) and 
because view 4 
ence chain (see 
by the pairwise 
is shown by the 
view and an arc 
resent overlaps 
tion (since pair- 
raph). Note that 
ps for the global 
‘ig. 3. The over- 
anslation result- 
)bal registration 
the case of the 
d the average tn 
ase of the robot, 
rage t,, is equal 
since there was 
1, the global reg- 
small amount of 
lerived from the 
; that the corre- 
:orrespondences 
  
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
n us. re 
LE I-A YEN cS 2. 7 \ 
(s) X X 8) p A A 
7 / N N / vx (10) 7 
| (5): À f o MO E 
f ae À, (i y (11) © 
(7) (Tyr, A 
“og” NE V j 
\ / / 
IN ud V gu 
(6) (4) (6) (4) 
ae =, ps: ThE = T 
e 8 ern Ll en 
p > (b) (8) 
Figure 3: Graphs showing the limited overlap information 
of the views of the bunny (a) and the robot (b), used by 
pairwise correspondence and registration. 
  
  
Figure 4: The difference in rotation angles (a) and normal- 
ized translation (b) of the views of the bunny derived with 
pair wise and global registration. 
been inaccurate the errors in the resulting pairwise registra- 
tion would have been large. A very large error would have 
accumulated between views that are far apart in the graph 
of Fig. 3 and hence global registration would have had to 
distribute these large errors resulting in much greater dif- 
ferences between the pairwise and global registrations. 
6 CONCLUSION 
We have presented an automatic 3D modeling technique 
using our automatic correspondence algorithm combined 
with global registration. Our technique is fully automatic 
and only assumes the prior information of the ordering of 
the views which is generally available from the sequence of 
acquisition. We have also presented qualitative and quan- 
titative analysis of our technique. Qualitative analysis was 
performed by visual inspection of the registered 3D mod- 
els. The quantitative analysis was performed by comparing 
the results of pair wise registration with the global regis- 
tration results. In future work, we plan to extend our tech- 
nique to be able to construct a 3D model from an unordered 
set of views. 
ACKNOWLEDGMENTS 
We would like to thank The Robotics Institute, Carnegie 
Mellon University, USA for providing the range data used 
in our experiments. This research is sponsored by ARC 
grant number DP0344338. 
REFERENCES 
Ashbrook, A., Fisher, R., Robertson, C. and Werghi, N., 
1998. Finding Surface Correspondence for Object Recog- 
121 
I 
il 
& | 
= 
* 
* | 
zZ 08} 
  
1 = 07} 
8 965} 
s: $4 | 
“as 1 4 i | 
s & 94. 1 
Ë 0.4 i 2 03} i 
Ë j a2} j 
02 1 sut | 
oir 1 
I 1 
ol aula beh J 
2.314485 5.5 7,18. 095 40) 1 x POST 545..5 7:98 92-10 M 
(a) View Number (b) Veram 
Figure 5: The difference in rotation angles (a) and normal- 
ized translation (b) of the views of the robot derived with 
pair wise and global registration. 
nition and Registration Using Pairwise Geometric His- 
tograms. IJPRAI 2, pp. 674—686. 
Benjemma, R. and Schmitt, F., 1997. Fast Global Registra- 
tion of 3D Sampled Surfaces Using a Multi-Z-Buffer Tech- 
nique. In: Internation Conference on Recent Advances in 
3D Digital Imaging, pp. 113-120. 
Besl, P. and McKay, N., 1992. Reconstruction of Real- 
world Objects via Simultaneous Registration and Robust 
Combination of Multiple Range Images. IEEE TPAMI 
14(2), pp. 239—256. 
Chen, C., Hung, Y. and Cheng, J., 1991. RANSAC-Based 
DARCES: A New Approach to Fast Automatic Registra- 
tion of Partially Overlapping Range Images. IEEE TPAMI 
21(11), pp. 1229-1234. 
Chen, Y. and Medioni, G., 1991. Object Modeling by 
Registration of Multiple Range Images. In: IEEE ICRA, 
pp. 2724-2729. 
Chua, C. and Jarvis, R., 1996. 3D Free-Form Surface Reg- 
istration and Object Recognition. [JCV 17, pp. 77-99. 
Higuchi, K., Hebert, M. and Ikeuchi, K., 1994. Building 
3-D Models from Unregistered Range Images. In: IEEE 
ICRA, Vol 3, pp. 2248-2253. 
Johnson, A. E. and Hebert, M., 1997. Surface Registra- 
tion by Matching Oriented Points. In: IEEE Int. Conf. on 
Recent Advances in 3-D Imaging and Modeling, pp. 121— 
128. 
Mian, A. S., Bennamoun, M. and Owens, R. A., 2004. 
Matching Tensors for Automatic Correspondence and Reg- 
istration. In: ECCV. To appear. 
Rangarajan, A., Chui, H. and Duncan, J., 1999. Rigid Point 
Feature Registration Using Mutual Information. Medical 
Image Analysis 3(4), pp. 425-440. 
Roth, G., 1999. Registering Two Overlapping Range Im- 
ages. In: IEEE Int. Conf. on 3-D Digital Imaging and Mod- 
eling, pp. 191—200. 
Williams, J. and Bennamoun, M., 2001. Simultaneous 
Registration of Multiple Corresponding Point Sets. CVIU 
81(1), pp. 117-142. 
Wyngaerd, J., Gool, L., Koth, R. and Proesmans, M., 1999. 
Invariant-based Registration of Surface Patches. In: IEEE 
ICCV, Vol. 1, pp. 301—306.