Prediction of Lane Change Trajectories through Neural Network

Abstract

Lane changing process is an essential maneuver, however, the process is responsible for large number of collisions and traffic instability. In this work, the effectiveness of neural network for prediction of future lane change trajectory based only on the past vehicle path is presented. Existing lane change models and lane change process do not consider the uncertainties and perceptions in the human behavior that are involved in lane changing. A neural network may learn and incorporate these uncertainties to predict the lane changing trajectory in the near future more accurately. A multilayer perceptron (MLP) has been employed to train itself from existing NGSIM field data and predict the future path of a lane changing vehicle. The impact and effectiveness of the proposed technique is demonstrated. Prediction results show that an MLP is able to give the future path accurately only for discrete patches of the trajectory and not over the complete trajectory. The results confirm to the observation that a vehicle trajectory has immediate influence from its neighborhood whose information is imperative for trajectory prediction.