Action recognition and prediction for driver assistance systems using dynamic belief networks

Abstract

The design of advanced driver assistance systems always aims at enabling the driver to master today’s traffic in a more safe and comfortable way. In order to judge the risks in a situation and initiate precautionary actions, future systems have to possess the capability to predict the behavior of surrounding traffic participants. This paper outlines an approach to predictive situation analysis for driver assistance systems and discusses one key issue in more detail-namely the predictive action recognition. In this context, a situation representation formalism will be introduced that exploits time as a compact physical measure. Furthermore, it will be shown how probabilistic networks can be used for reasoning about driver (action) intentions and how such networks can help to cope with uncertainty resulting from inaccuracy in models and sensor data. First results are shown in simulation for highway overtake scenarios. In the situations presented the prediction for an upcoming lane change can be made by the assessment of the time gaps to the nearest neighbors of that specific vehicle.