A driving path based target object prediction

The determination of the target object has been a challenging task since the development of Adaptive Cruise Control and is subject to numerous scientific contributions, providing improvements towards predictive detection. Autonomous driving functions have to assure safe and comfortable traveling under all conditions. The recognition of a future target object has to be reliable, independently of the surrounding situation. State-of-the-art systems however, still show difficulties detecting close cut-in maneuvers of other traffic participants in time. In this paper we propose a probabilistic approach to predict the relevance of surrounding vehicles as target object. We model the driving path of the ego vehicle, determined from the vehicle dynamics. By the use of an object oriented Bayesian network we are able to model relations between the derived path and the surrounding vehicles to determine their relevance as target object. The experimental results demonstrate the outcome of our implementation under real world conditions. An average decrease of 40% in detection time, compared to a visible lane marking crossing during a lane change of the vehicles is achieved.