Non-Stationary "Shape Activities"

摘要: The changing configuration of a group moving landmarks can be modeled as and deforming shape. defining the shape could objects(people/vehicles/robots) or rigid components an articulated like human body. In past work, term "shape activity" has been used to denote particular stochastic model for deformation. Dynamical models have proposed characterizing stationary activities (assume constant mean shape). this work we define dynamic non-stationary show that activity follows special case this. Most performed by (here, objects) are not hence more general is needed. We also piecewise with transitions which segment out track sequence activities. Noisy observations coming from these tracked using particle filter. discuss applications our framework abnormal detection, tracking segmentation.