ELIO E. 
i < J nn -— (0D 
ow ud 
"UO m.s. 
175 
5. EXPERIMENTS 
Significant experimental verifications of registration algorithms are usually difficult to achieve, since the result of 
applying a given method to a pair of images is a function of the geometry of the object in the image, the amount of 
overlap between views, the surface sampling densities and the initial estimate on the transformation. 
A first experiment aims at demonstrating the gain in convergence speed due to the constraint introduced by curvature 
compatibility. For this purpose, the range image in Figure 1(a) is translated and rotated by a known amount: the 
registration algorithm is applied, with and without the use of the curvature compatiblility. For a rotation of the image 
of 30° around the x axis and a translation equal to the width of the object, 28 iterations are required to recover the 
transformation with the curvature information, while it takes 40 iterations if the curvature labels are not used. In a 
second experiment, the two range images shown in Figure 1 (a) and (b) are registered. The object in the images was 
rotated by 20°. The images differ due to self-occlusion of the surface from the rotation as well as clipping from the 
limited field of view. 32 iterations were required here. 
6. CONCLUSION 
This paper described a method for the registration of range images which uses an invariant method for the computation 
of curvatures. Curvature labels derived from the signs of the mean and Gaussian curvatures are used to constrain 
the matching problem. An algorithm, called the Iterative Closest Compatible Point, was introduced: it operates by 
iteratively minimizing the distance between closest pairs of points that are compatible, based on the curvature labels. 
We are working to integrate this approach in a framework that also uses invariant reflectance properties computed 
from the intensity measurements of the laser range sensor [7]. 
ACKNOWLEDGEMENTS 
The authors would like to thank Luc Cournoyer for the range image acquisitions. Thanks also to Gerhard Roth and 
Takeshi Masuda for helpful discussions. 
REFERENCES 
[1] P.J. Besl, R.C. Jain, "Invariant Surface Characteristics for 3D Object Recognition in Range Images”, Computer 
Vision, Graphics, and Image Processing, 33, pp. 33-80, 1986. 
[2] P. J. Besl, N. D. MacKay, "A method for the registration of 3-D shapes", IEEE Trans. on Pattern Analysis and 
Machine Intelligence, vol. 14, no. 2, pp. 239-256, February 1992. 
[3] E. Bittar, S. Lavallée, R. Szeliski, "A method for registering overlapping range images of arbitrarily shaped 
surfaces for 3-D object reconstruction" , Proc. of SPIE Sensor Fusion VI, vol. 2059, pp. 384-395, Nov. 1993. 
[4] P. Boulanger, P. Cohen, "Adaptive smoothing of range images based on intrinsic surface properties, Proc. of 
SPIE Hybrid Image and Signal Processing Il, vol. 1297, Orlando, FL, April 16-20, pp. 254-263, 1990. 
[5] P. Boulanger, P. Cohen, "Viewpoint invariant computation of surface curvatures in range images”, Proc. Vision 
Interface 94, Banff, Canada. May 16-20, 1994. 
[6] O. Faugeras, M. Hébert, “The representation, recognition, and locating of 3-D objects”, International Journal 
of Robotics Research, vol. 5, no. 3, pp. 27-52, 1986. 
[7] G. Godin, M. Rioux, R. Baribeau, "Three-dimensional registration using range and intensity information”, Proc. 
SPIE Videometrics Ill, vol. 2350, Boston, 2-4 November, pp. 279-290, 1994. 
[8] R. Haralick, H. Joo, C. Lee, X. Zhuang, V. Vaidya, M. B. Kim, “Pose estimation from corresponding point 
data", in Machine Vision for Inspection and Measurement (H. Freeman, ed.), pp. 1-83, 1989. 
[9] B.K.P. Horn, "Closed-form solution of absolute orientation using unit quaternions , Journal of the Optical Society 
of America A, vol. 4 no. 4, pp. 629-642, April 1987. 
[10] J. Hoschek; " Intrinsic parametrization for approximation". Computer Aided Geometric Design, 5, pp. 27-31, 
1988. 
[11] N. Kehtarnavaz, S. Mohan, "A framework for estimation of motion parameters from range images”, Computer 
Vision, Graphics and Image Processing, vol. 45, pp. 88-105, 1989. 
[12] T. Masuda, N. Yokoya, "A robust method for registration and segmentation of multiple range images”, Pro- 
ceedings 2nd CAD-Based Vision Workshop, Champion, PA, pp. 106-113, 1994. 
[13] B. Sabata, J. K. Aggarwal, "Estimation of motion from a pair of range images: a review", CVGIP: Image 
Understanding, vol. 54, no. 3, pp. 309-324, 1991. 
[14] D. A. Simon, M. Hebert, T. Kanade," Real-time 3-D pose estimation using a high-speed range sensor”, IEEE 
Int. Conf. on Robotics and Automation, San Diego, CA, May 8-13, pp. 2235-2241, 1994. 
[15] G. Soucy, F. P. Ferrie, “Motion and Surface Recovery Using Curvature and Motion Consistency”, 2nd European 
Conf. on Computer Vision 1992, Santa Margherita Ligure, May 19-22, 1994. pp. 222-226. 
[16] Z. Zhang, "Iterative point matching for registration of free-form curves and surfaces", Int. Journal of Computer 
Vision, vol. 13, no. 2, pp. 119-152, 1994. 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995