Let X be a centered stationary Gaussian stochastic process with a d-di
mensional parameter (d greater-than-or-equal-to 2), F its spectral mea
sure, integral R(d) parallel-to x parallel-to 2 F(dx) = +infinity (par
allel-to x parallel-to denotes the Euclidean norm of x). We consider r
egularizations of the trajectories of X by means of convolutions of th
e form X(epsilon)(t)=(PSI(epsilon)X)(t) where PSI(epsilon) stands for
an approximation of unity (as epsilon tends to zero) satisfying certa
in regularity conditions. The aim of this paper is to recover the loca
l time of X at a given level u, as a limit of appropriate normalizatio
ns of the geometric measure of the u-level set of the regular approxim
ating processes X(epsilon). A part of the difficulties comes from the
fact that the geometric behavior of the covariance of the Gaussian pro
cess X(epsilon) can be a complex one as epsilon approaches 0. The resu
lts are on L2-convergence and include bounds for the speed of converge
nce. L(p) results may be obtained in similar ways, but almost sure con
vergence or simultaneous convergence for the various values of u do no
t seem to follow from our methods. In Sect. 3 we have included example
s showing a diversity of geometric behaviors, especially in what conce
rns the dependence on the thickness of the set in which the covariance
of the original process X is irregular.Some technical results of anal
ytic nature are included as appendices in Sect. 4.