A Driver Behavior Learning Framework for Enhancing Traffic Simulation

Abstract — Traffic simulation provides an essential support for developing intelligent transportation systems. It allows affordable validation of such systems using a large variety of scenarios that involves massive data input. However, realistic traffic models are hard to be implemented especially for microscopic traffic simulation. One of the hardest problems in this context is to model the behavior of drivers, due the complexity of human nature. The work presented in this paper proposes a framework for learning driver behavior based on a Hidden Markov Model technique. Moreover, we propose also a practical method to inject this behavior in a traffic model used by the SUMO traffic simulator. To demonstrate the effectiveness of this method we present a case study involving real traffic collected from Timisoara city area