Map‐Matching With Recursive Bayes Filter and Context‐Dependent Transition Dynamics for Scene Modeling in Automated Driving

ABSTRACT We present an efficient multi‐object map‐matching approach for scene modeling in the context of automated driving. The mapping of traffic participants to the road topology is a key component of the scene modeling step in the processing pipeline for situation‐aware automated driving. Symbolic representations combined with knowledge graphs are essential for safe and interpretable planning and decision‐making. They enable the structured encoding of relationships between dynamic objects and the road topology, forming a coherent knowledge‐graph representation of the situation. To take advantage of the partially structured nature of the driving environment, particularly the road topology, as prior knowledge for improving lane‐level object estimation, a discrete recursive Bayesian filter with context‐dependent, time‐varying transition dynamics is proposed. The dynamic transition probability matrix is derived from the road topology and the observation horizon and encodes feasible lane transitions such as lane changes, merges, and splits through the connectivity of the lane graph. This approach explicitly captures temporal consistency and measurement uncertainty while integrating structural knowledge of the road network, and is validated in simulation across complex traffic scenarios.