Maneuver recognition using Probabilistic Finite-State Machines and Fuzzy Logic

This paper presents a general approach for recognition of driving maneuvers in advanced driver assistance systems (ADAS). Such systems often rely on the identification of driving maneuvers (overtaking, left turn at intersections, etc.) to improve the prediction of potential collisions or to trigger appropriate support for the driver. The proposed maneuver recognition approach combines a fuzzy rule base to model basic maneuver elements and probabilistic finite-state machines to capture all possible sequences of basic elements that constitute a driving maneuver. The proposed method is specifically tailored to ADAS requirements because of its low computational complexity, its flexibility and its straight-forward design based on easily comprehensible logical rules. In addition, we propose a suitable training method to optimize the fuzzy rule base. Our approach is evaluated on the recognition of turn maneuvers. Experiments on real data from a test vehicle demonstrate the feasibility of the proposed method.