INSTITUTO ARGENTINO DE MATEMATICA ALBERTO CALDERON
Unidad Ejecutora - UE
congresos y reuniones científicas
Learning mixed-state Markov models for statistical motion texture tracking.
TOMÁS CRIVELLI; PATRICK BOUTHEMY; BRUNO CERNUSCHI FRÍAS; JIAN-FENG YAO
Congreso; 2nd International Workshop on Machine Learning for Vision-based Motion Analysis (MLVMA09); 2009
A motion texture is the instantaneous scalar map of ap-parent motion values extracted from a dynamic or temporaltexture. It is mostly displayed by natural scene elements(fire, smoke, water) but also involves more general texturedmotion patterns (eg. a crowd of people, a flock). In thiswork we are interested in the modeling and tracking of mo-tion textures. Experimentally we observe that such motionmaps exhibit values of a mixed type: a discrete compo-nent at zero and a continuous component of non-null mo-tion values. Thus, we propose a statistical characterizationof motion textures based on a mixed-state causal modeling.Next, the problem of tracking is considered. A set of mixed-state model parameters is learned as a descriptive featureof the motion texture to track and displacement estimationis solved using the conditional Kullback-Leibler divergencefor statistical window matching. Results and comparisonsare presented on real sequences.