On the use of Bayesian networks for real-time urban traffic measurements: a case study with low-cost devices