Gershgorin discs of gramian matrices for target tracking

We study the stochastic behavior of the eigenspectrum of time-varying Euclidean-Gramian matrices typically associated with the MDS algorithm using the Gershgorin theorem. Relying on an empirical model, we quantify the probability that premultiplying the kernel at instant k by the eigenvectors of the kernel at instant k-1 preserves the principal component structure of the kernel. The study, which shows that the above probability remains large even for fast-varying kernels, indicates that subspace-tracking methods are adequate to fast-moving multitarget tracking.