Driver Modeling and Lane Change Maneuver Prediction

Accurate driver models can be used to study new infrastructure components, new vehicle interfaces or congestion reduction methods. The aim of this study was twofold: to model driver following behavior, and to create a classifier for predicting lane change maneuvers. A preprocessing method using cubic smoothing splines was proposed to extract smooth vehicle trajectories from video monitoring data of highway traffic. Reference measurements showed that the position information was obtained with high accuracy, which decreased with every differentiation. The driver models are based on the Intelligent Driver Model (IDM) and the Human Driver Model (HDM) for which a Genetic Algorithm was used to find the optimal parameters. HDM was not able to significantly more accurately fit observed trajectories than IDM, but it did show more realistic traffic congestion patterns. The most useful features for maneuver prediction were the lateral position and its derivatives. Information of neighboring vehicles only improved the performance when a larger training set is used. The lane change maneuver classifier was able to predict 90% of the exit maneuvers 1.44 seconds in advance. Although the classification accuracy varies for different maneuvers, in overall it was able to classify the vehicle maneuvers with high accuracy.