A Method for Registration of 3D Surfaces ICP Algorithm

Report
A Method for Registration of 3D
Surfaces
ICP Algorithm
Erhan Avinal
Introduction

Building 3D models of real world objects



Data capture using range camera
Registration
Data merge
Registration - Goal


To transform sets of
surface measurements
into a common
coordinate system
A model shape and a
data shape
Registration

If we know correct correspondences, we can
find correct translation and rotation
Registration


Issue : Finding corresponding points
ICP : Assume closest points correspond to
each other, compute the best transform
Rotation and translation
ICP Algorithm
Can be used with
Distance

Euclidian distances

Point to point set

Point to line segment

Point to triangle

Point to parametric entity

Point to implicit entity
Algorithm


Point Set P with Np points, model shape X
Iterate until convergence

Compute closest points



Squared Euclidian distances
Compute registration (rotation and translation)
Apply the registration

New point set
Other Issues

Color matching




Sharp, 2002
Godin, 1995
Johnson, 1997
Orientation


Godin 2001
Schutz 1998
Other Issues

Weighting


Initial point selection




Important nodes
Uniform sampling
Random
Select in regions of high curvature
k-d trees to find closest points
Results
Results
Results
Results
Results
Conclusions





ICP can register a data shape to a model
shape
Independent of shape representation
Does not require preprocessing of 3D data
A good initial estimate of transformation is
required
High computation cost
Future Work



Computational speedup
Parallel testing
Allow deformations
References

Original Paper:




P.J. Besl, N.D. McKay,A Method of Registration of 3D
Shapes, 1992
T. Jost, Fast Geometric Matching for Shape
Registration, 2002
S.M. Rusinkiewicz, Real time Acquisition and
Rendering of Large 3D Models, 2001
http://www.ee.surrey.ac.uk/Research/VSSP/3DVisio
n/model_building/model.html
Thanks

similar documents