Driver intention inference with vehicle onboard sensors

This paper investigates methods to infer the driver´s intention to leave the lane- or car-following state and start a specific manoeuvre, such as a lane change or a turn, from easily accessible vehicle onboard sensors, such as the vehicle data bus, GPS localisation, digital maps, and optionally camera-based lane detection. Pattern recognition and classification is done with Hidden Markov Models. Starting from suitable model grammars, training and evaluation processes are described with the focus on the task of early recognition during manoeuvre start. Information about imminent manoeuvres can then be used in driver assistance systems to predict trajectories or assess risk.