A new efficient EM-ICP algorithm for non-linear registration of 3D point sets

Benoît Combès 1 Sylvain Prima 1
1 VisAGeS - Vision, Action et Gestion d'informations en Santé
INSERM - Institut National de la Santé et de la Recherche Médicale : U746, Inria Rennes – Bretagne Atlantique , IRISA-D5 - SIGNAUX ET IMAGES NUMÉRIQUES, ROBOTIQUE
Abstract : In this paper, we present a new method for non-linear pairwise registration of point sets. In this method, we consider the points of the first set as the draws of a Gaussian mixture model whose centres are the points of the second set displaced by a deformation. Next we perform {\it maximum a posteriori} estimation of the parameters (which include the unknown transformation) of this model using the expectation-maximisation algorithm. Compared to other methods using the same ''EM-ICP'' paradigm/framework, we propose three key modifications leading to an efficient algorithm allowing for fast registration of large point sets: 1) symmetrisation of the point-to-point correspondences; 2) specification of priors on these correspondences using differential geometry; 3) efficient encoding of deformations using the RKHS theory and the Fourier analysis. The resulting algorithm is efficient and is able to register large data sets. We evaluate the added value of the modifications and compare our method to the state-of-the-art CPD algorithm on synthetic data.
Document type :
Reports
Complete list of metadatas

Cited literature [56 references]  Display  Hide  Download

https://hal.inria.fr/hal-00656388
Contributor : Benoît Combes <>
Submitted on : Wednesday, January 4, 2012 - 10:42:09 AM
Last modification on : Monday, April 15, 2019 - 6:09:18 PM
Long-term archiving on : Tuesday, December 13, 2016 - 7:33:18 PM

File

RR-7853.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-00656388, version 1

Citation

Benoît Combès, Sylvain Prima. A new efficient EM-ICP algorithm for non-linear registration of 3D point sets. [Research Report] RR-7853, INRIA. 2012. ⟨hal-00656388⟩

Share

Metrics

Record views

849

Files downloads

1390