CONDENSATION - CONDITIONAL DENSITY PROPAGATION FOR VISUAL TRACKING

The problem of tracking curves in dense visual clutter is challenging. Kalman filtering is inadequate because it is based on Gaussian densit ies which, being unimodal, cannot represent simultaneous alternative h ypotheses. The CONDENSATION algorithm uses ''factored sampling'', prev iously applied to the interpretation of static images, in which the pr obability distribution of possible interpretations is represented by a randomly generated set. CONDENSATION uses learned dynamical models, t ogether with visual observations, to propagate the random set over tim e. The result is highly robust tracking of agile motion. Notwithstandi ng the use of stochastic methods, the algorithm runs in near real-time .