Surface registration by markers guided nonrigid iterative closest points algorithm - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 416

FIGURE 4 Distance map histogram [mm] (a) and correspondence map histogram [number of
units], (b) for different modifications of the ICP computing correspondence algorithm: static
marker vector (SM) and static marker vector with 5, 10, and 15% radius constraints for the
ToF camera.
The convergence of ICP algorithm is topic of research [ 10 ] . In presented approach no reli-
ability weighting is used (weighting is always equal 1), the residual in an optimal step ICP is
always decreased, because neither finding a new deformation, nor finding new closest points
can increase residual. A formal proof for rigid case, which can be applied to presented ap-
proach can be found in Ref. [ 2 ] . Danilchenko and Fizpatrick [ 11 ] proposed principal access
approach to taking into account local character of anisotropic noise in rigid point registration
and Maier-Hein et al. [ 10 ] used this approach to rigid anisotropic weighting ICP algorithm
and ToF data. In future work this approach can be generalized to nonrigid ICP.
Presented approach may be used in different medical, entertainment, and industrial ap-
plications, where nonrigid point clouds should be registered, when initial relative position
of clouds is that finding correspondences by Euclidean distance or normal shouting is not
enough. The proposed changes do not introduce complex calculations. Initial calculation of ri-
gid registration allows to solve the problem of unknown transformation matrix initialization.
Comparing to classical nonrigid ICP the disadvantage of proposed approach is that initial cor-
responding positions of markers in source and target point clouds are needed.
References
[1] Wyawahare M, Patil P, Abhyankar H. Image registration techniques: an overview. Int
J Signal Proc Image Process Patern Recogn. 2009;2(3):11-28.
[2] Salvia J, Mataboscha C, Fofib D, Forest J. A review of recent range image registration
methods with accuracy evaluation. Image Vision Comput. 2007;25:578-596.
[3] Besl P, McKay N. A method for registration of 3D shapes. Patern Anal Mach Intell.
1992;14(2):239-256.
[4] Chen Y, Medioni G. Object modeling by registration of multiple range images. Image
Vision Comput. 1992;10(3):145-155.
[5] Rusinkiewicz S, Levoy M. Efficient variants of the ICP algorithm. In: Proc. 3rd in-
ternational conference on 3D digital imaging and modeling; Stanford, CA: Stanford
University; 2001:145-152.
[6] Amberg B, Romdhani S, Veter T. Optimal step nonrigid ICP algorithms for surface re-
gistration. In: Proc. IEEE conference of computer vision and patern recognition CVPR;
2007:1-8.
[7] Horn B, Hilden H, Negahdaripour S. Closed form solution of absolute orientation us-
ing orthonormal matrices. J Opt Soc Am A. 1988;5:1127-1135.
[8] htp://www.mesa-imaging.ch/swissranger4000.php - Mesa-Imaging - manufactor
website: SR4000 Data Sheet.