179
A METHOD FOR ESTIMATING 3D SHAPES MOTION BY A FREQUENCY DOMAIN TECHNIQUE
G.M. Cortelazzo A. Guamieri
Dipartimento di Elettronica ed Informatica Dipartimento del Territorio - TESAF
Università degli Studi di Padova - Italy Università degli Studi di Padova - Italy
e-mail: corte@dei.unipd.it
A. Vettore
Dipartimento del Territorio - TESAF
Università degli Studi di Padova - Italy -
e-mail: vettoan@uxl.unipd.it
Commission VI, Working group 3
ABSTRACT
Free-form 3-D surfaces registration is a fundamental problem in 3-D imaging, tipically approached by extensions or variations of the
ICP algorithm [1, 2]. This work suggestes an alternative procedure for 3-D motion estimation based on the Fourier transform of the
3-D intensity function, implicitly described by the registered time-sequences of range data. Similar to the frequency domain techni
ques for estimating motion parameters, this is a non feature-based method suitable for unsupervised registration of 3-D views. This
method has several advantages related to the fact that it uses the total avalaible information and not sets of features. With respect to
memory occupancy, the use of a time-sequence of a 3-D intensity function represents a considerable data reduction with respect to a
pair of time-sequences of 2-D functions. The proposed technique, wich extends to the 3-D case previous frequency domain estima
tion algorithms developed for planar case, retain their robustness.
1. Problem statement
Let /,(x), xeR J , be a 3-D object and let / 2 (x) be a rigidly rotated
version of /,(x) (these data can be obtained from registered
range and intensity data captured at different times). It can be
shown that /,(x) and / 2 (x) relate as:
l 2 (x) = /, (R~X -1) where 1 ReS0(3) , teR J (1)
According to (1) / 2 (x) is first translated by the vector t and then
rotated by the matrix R.
Denote as
and (5) can be used in order to determine R. Therefore in the
frequency domain the estimation of R and t can be decoupled
andone can estimate first R from (5) and t from (3), according to
the follow'ing two passes procedure.
2. Estimation of the 3-D Rotational Matrix R
Write the 3-D rotational matrix R as follow [3]
R = R(co,e)=e" 9 (6)
i,(k) = F[l, (*)| k]= JJJ ~l,(x)e- J ** T ‘dx (2)
with k=[k x , k y , k 2 ] T
the 3-D Fourier transform of /¡(x), i= 1,2; it is straightforward to
prove that the two transforms are related as
L 2 (k) = L { (R- [ k)e- J2l * TRl (3)
From (3) one sees that the translation t affetcs only phases and
not magnitudes. However, if the 3-D surfaces were modeled by
impulsive functions supported on them, their Fourier transforms
would have a lot of spurious high frequency content not suited
to the frequency domain techniques to be presented. For this
reason we preliminarily build 3-D solids from the range data
defined by the 3-D surfaces and we apply the proposed techni
que to them.
Let S,(x) and S 2 (x), xeR 3 , be the range data defining the sur
faces of a rigid body moving in R J , at times t, and t 2 respecti
vely. From Si(x) and S 2 (x) define /|(x) and / 2 (x) as:
/ i( x)=jl if x&Int(S i (*)) i=l 2 (4)
[0 elsewhere
where Int(5j(x)) denotes the interior of the surface S^x).
Magnitudes are related as
|i,(k)| = |l,(«-'k)| (5)
1
S0(3) is the group of the 3x3 special orthogonal matrices
where
-CO
CO
0
E so(3) 2
(7)
co = [co x , co y , co z ] T E R 3 , is a unit vector determining the rota
tional axis, CO is a skew-symmetric matrix obtained from the
vector to and 0 E R is the rotational angle in radians.
Define the difference function A(k) between tyransforms as
A(k) = A(k r ,k v ,k.) =
lAOOl
Kool
|Mk)|
1,(0)
¿2«»
¿,(0) ¿,(0)
It can be proved [3] that R, as rotational matrix, has eigenvalues
X, = 1, X 2 = e jQ and X 3 = e 70 . Call co the eigenvector corre
sponding to X] = 1 . The vector co is a solution of the equation
A(k) = 0, indeed A(k) = 0 if R' ] k = k or equivalent /?k = k . In
other words the locus A(k) = 0 includes a line through co. For
objects without special symmetries (as natural objects typically
are) this property of the function A(k) can be exploited in order
to determine the versor co and then the angle 0 by means of the
following procedure:
2 so(3)={S E R jx3 : S T = - S} is the space over reals of the 3x3
skew-symmetric matrices.The map so(3) —> S0(3) is suriective.