Contour Tracking by Stochastic Propagation of Conditional Density

摘要: The problem of tracking curves in dense visual clutter is a challenging one. Trackers based on Kalman filters are limited use; because they Gaussian densities which unimodal, cannot represent simultaneous alternative hypotheses. Extensions to the filter handle multiple data associations work satisfactorily simple case point targets, but do not extend naturally continuous curves. A new, stochastic algorithm proposed here, Condensation — Conditional Density Propagation over time. It uses ‘factored sampling’, method previously applied interpretation static images, distribution possible interpretations represented by randomly generated set representatives. combines factored sampling with learned dynamical models propagate an entire probability for object position and shape, result highly robust agile motion clutter, markedly superior what has been attainable from filtering. Notwithstanding use methods, runs near real-time.