Motion Pattern Recognition for Maneuver Detection and Trajectory Prediction on Highways

Abstract

Intelligent automated driving functions require a deep understanding about the current traffic situation and its likely evolution. For highly automated driving on highways, predicting trajectories of traffic participants is a crucial task for collision-free trajectory planning and risk-aware maneuver choice. For a prediction horizon of a few seconds the execution of those trajectories is fuzzy and highly dependent on the maneuver choice of the driver. This paper presents a new online-capable statistical approach for maneuver detection and uncertainty-aware trajectory prediction in highway scenarios based on detecting and clustering typical motion patterns in real highway footage and deriving prototypical trajectories for each cluster. The cluster prototypes are utilized for maneuver detection by evaluating their proximities to incomplete tra- jectory records while identifying for each prototype its most similar section. The remaining segment of the best fit is used as an estimate for the future motion of the traffic participant. Quantitative evaluation results demonstrate the potential of the proposed concept for maneuver detection and maneuver-based trajectory prediction.