On sequential Monte Carlo sampling methods for Bayesian filtering