“A set of motion features for tracking and representation of dynamic video contents”

TOMÁS CRIVELLI; BRUNO CERNUSCHI FRÍAS; PATRICK BOUTHEMY

Mar del Plata, Argentina

Congreso; Argentine Symposium on Technology (AST 2009), 38 JAIIO, Mar del Plata, Argentina. 24-28 Agosto 2009.; 2009

SADIO

In this work we are interested in dynamic or motion textures,which refer to any dynamic visual information that can be classified asa texture with motion (fire, smoke, crowd of people). Experimentallywe observe that such motion maps exhibit values of a mixed type: adiscrete component at zero (absence of motion) and continuous motionvalues. We thus introduce a mixed-state Markov chain model for thecharacterization of motion textures from which a set of 13 parametersis extracted as the descriptive feature of the dynamic content. Then amotion texture tracking strategy is proposed based on the computationof the Kullback-Leibler divergence between mixed-state probability den-sities. This permits to estimate the displacement of a motion texture anddeal with the problem of tracking.