Theoretical and practical capabilities of probabilistic data fusion with Bayesian networks

Abstract

The increasing demand for high quality data in context of intelligent transportation systems more and more facilitates the use of data fusion methods in order to derive as much information as possible from existing sensors and sensor technologies. This paper discusses the application of simple Bayesian networks for this task with regard to fusing traffic state measurements, in particular. Theoretical aspects like model calibration and statistical quality of the results are discussed in a mathematically exact way. Moreover, a real-world example based on floating car data from two independent vehicle fleets is described in order to evaluate the Bayesian approach also from a practical perspective. Chances and restrictions of the presented model – also with regard to possible modifications – are critically discussed.