Over-Approximation of the Driver Behavior as Occupancy Prediction

The prediction of the future behavior of drivers is a challenging research topic. Therefore, this paper presents a new approach for occupancy prediction of the surrounding vehicles based on a static overapproximation of the driver behavior in longitudinal direction and a situation specific overapproximation of the driver behavior in lateral direction. Compared to existing probabilistic motion prediction approaches no prior knowledge of the situation is necessary. Therefore, the presented approach is not limited to specific situations and can be used to predict the occupancy in unstructured environments. The evaluation of the approach with real world data from the common road benchmark dataset shows the reduction of the occupancy area size up to 70% compared to a baseline method. Nevertheless, the prediction is accurate up to a prediction time of 2 seconds whereby the safety of the autonomous vehicle is ensured. The presented approach successfully handles the trade-off between occupancy area size and prediction safety while being applicable to all situations.