Learning mixed-state Markov models for statistical motion texture tracking

Abstract

A motion texture is the instantaneous scalar map of apparent motion values extracted from a dynamic or temporal texture. It is mostly displayed by natural scene elements (fire, smoke, water) but also involves more general textured motion patterns (eg. a crowd of people, a flock). In this work we are interested in the modeling and tracking of motion textures. Experimentally we observe that such motion maps exhibit values of a mixed type: a discrete component at zero and a continuous component of non-null motion values. Thus, we propose a statistical characterization of motion textures based on a mixed-state causal modeling. Next, the problem of tracking is considered. A set of mixedstate model parameters is learned as a descriptive feature of the motion texture to track and displacement estimation is solved using the conditional Kullback-Leibler divergence for statistical window matching. Results and comparisons are presented on real sequences.

Perceptual Principles for Video Classification With Slow Feature Analysis

“…This type of stochastic models can be successfully applied in various contexts, from dynamic texture classification to motion segmentation [9] or tracking [8]. However, not unlike optical flow, LDS must respect constraints which are not easily satisfied in complex natural scenes.…”

Simultaneous Motion Detection and Background Reconstruction with a Conditional Mixed-State Markov Random Field

“…It is demonstrated that the normal flow scalar motion observations extracted from these video sequences, show a discrete value at zero (null-motion) and a Gaussian continuous distribution for the rest of the values. This model was extended in Crivelli et al (2006Crivelli et al ( , 2009 and applied to the problems of motion texture segmentation, recognition and tracking. For these applications, the issue is different than for simultaneous decisionestimation problems.…”

Dynamic Scene Classification: Learning Motion Descriptors with Slow Features Analysis

“…In order to explicitly model texture dynamics, linear dynamical systems (LDS) have been proposed in [32]. Such stochastic models have been successfully applied in various contexts, from dynamic texture classification to motion segmentation [6] or tacking [5]. However, LDS is intrinsically limited by the first-order markov property and linearity assumption.…”

“…To do this, we consider N training videos of duration T on a p × p grid as illustrated in red in figure 2. We define v n xy (t) 5 as the V1 feature for video n at spatial position (x, y) and time t. We compute all possible features v n xy (t) and compute the temporal derivativesv n xy (t). The temporal covariance matrix of equation 2 is then computed by…”