Neural network based lane change trajectory predictions for collision prevention

Abstract

Lane change is a vital maneuver which disturbs the traffic equilibrium and is a major cause of collision on the road. This process involves decision to change lanes followed by the actual lane change. The lane change trajectory is, thus, influenced by neighborhood traffic conditions and driver's behavior and perception. However, most of the existing lane change models do not consider the uncertainties and human behavior involved in lane change maneuver. In the present paper, neural network is used to predict the lane change trajectory to reflect these uncertainties and perceptions to represent lane changing behavior more realistically. The neural network is employed to perform short term and long range prediction of the Lane change trajectories of a vehicle. The network is first trained using past trajectory of the lane changing vehicles and other vehicles in its neighborhood. The trained network is then used for prediction. The comparison of simulated results with observed data indicates that neural network is able to learn the driver behavior more realistically than other standard modeling and is able to perform short term prediction with sufficient accuracy.