Motion estimation from image sequences: classical vs. neural-network-based approaches

Abstract

Motion information is used for many image sequences filtering algorithms and coding. In the paper the bases of apparent motion description are presented. Next, on the basis of available literature, the review of existing classical, deterministic motion estimation algorithms and algorithms based on the artificial neural networks is presented. The algorithms based on block-matching, gradient algorithms, spatio-temporal energy methods and model-based methods are described in details. The different classes of artificial neural networks which can be used for 2D motion estimation problems' solution are also presented.