Parallel visual motion analysis using multiscale Markov random fields

Abstract

The use of Markov Random Field (MRF) models within the framework of global bayesian decision has brought new powerful solutions to visual motion analysis. The efficiency of MRF models for image sequence analysis has been proved on various classes of real-world sequences: outdoor and indoor scenes including several moving objects and camera motion. The authors extend this work by investigating new multiscale motion analysis algorithms based on MRF models. These algorithms are related to a new class of consistent multiscale MRF statistical models. The multiscale paradigm exhibits fast convergence properties towards quasi optimal estimates. Its performances are compared to standard relaxation in the case of optical flow measurement.<>