Microscopic driving behavior modelling at highway entrances using Bayesian network

Entrances into highway represent critical sections, as entering vehicles are bound to merge with the existing traffic at a fixed junction point almost independently from the traffic conditions-in some sense a “constrained” cut-in manoeuver. Advanced driver-assistance systems (ADAS) need to perform correctly under these conditions as well, but could also be used to facilitate merging and reducing risks. In particular, vehicles on the main road could adapt their speeds-or even change lane-provided an estimate of the entering vehicle's time to merge is available, exactly as a human driver would do. This paper is concerned with providing such an estimate. To this end, we observe that the speed profile on the entering ramps is rather well predictable if the acceleration of the entering vehicles is described as a function of their actual distance from the junction point. Still, uncertainties remain and to cope with them, we use a stochastic prediction model based on Dynamic Bayesian Network. The result are probability distribution functions of the time to merge based on the observation of speed and distance of the entering vehicle once they become visible to the traffic on the main road. Experimental data are used to illustrate the model structure and the parameter determination. The model quality is then assessed by comparing statistics from simulations with the one recorded on road. The possible use of the model as a tool for traffic prediction algorithm embedded in ADAS and an extension to existing highway stochastic traffic models are shortly discussed at the end of the paper.