The problem of tracking curves in dense visual clutter is challenging. Kalman filtering is inadequate because it is based on Gaussian densities which, being unimodal, cannot represent simultaneous alternative hypotheses. The PF(Particle Filtering) algorithm uses “sequential importance sampling”, previously applied to the posterior of static signals, in which the probability distribution of possible interpretations is represented by a randomly generated set. PF uses learned “sequential Monte Carlo” models, together with practical observations, to propagate and update the random set over time. The result is highly robust tracking of agile motion. Not withstanding the use of stochastic methods, the algorithm runs in near Real-Time.
An Auxiliary Particle Filtering Algorithm With Inequality Constraints
