Neural network for lane change prediction assessing driving situation, driver behavior and vehicle movement

Abstract

With the steady progress in advanced driver assistance and partial automation of the task of driving it is also increasingly important to put the vehicular systems in the position to autonomously identify and assess the driving situation as well as the needs and intentions of the road users. This is in particular relevant for driving maneuvers such as lane changes. To predict them features have to be taken into account covering a wide range of situations and drivers. Against this background, an algorithm is proposed predicting situations of upcoming lane changes based on assessments of the driving situation, the driver's behavior and the vehicle's movement. It relies on a 360° sensory perception of the vehicular surroundings and on the analysis of the driver's gaze behavior preparing lane changes. The information gained is fused and used for classification by means of an artificial neural network that was parameterized by applying machine learning. The resulting prediction algorithm is working in real-time as a vehicular application. The parameterization as well as the evaluation of the whole system were done using naturalistic driving data obtained by a driving study.