Towards a unified traffic situation estimation model — Street-dependent behaviour and motion models

For Advanced Driver Assistance Systems and Autonomous Driving, estimating and predicting traffic situations becomes more and more essential. Many approaches focus on one specific application like vehicle state estimation from sensor data or road model estimation from environment perception. To integrate single approaches to one coherent system, one unified model is needed where the existing applied algorithms can be grounded to. In this paper we propose a Unified Traffic Situation Estimation Model that describes the probabilistic dependencies between road elements. While its independence from time makes it usable for offline mapping tasks, we show that online prediction capabilities can be achieved by applying the model to a longitudinal vehicle state estimation problem: Using the Markov assumption and appropriate state spaces the general unified model can be specialized to an Interacting Multiple Model Filter. Finally, experiments show an improvement in state estimation and prediction over standard models, which only consider vehicle dynamics. Additionally the unified model allows the prediction of street related routes of vehicles.