A maneuver-prediction method based on dynamic bayesian network in highway scenarios

Abstract

The accurate maneuver prediction for dynamic vehicles can enhance driving safety in complex environments. This paper presents a maneuver prediction method for dynamic vehicles in highway scenarios. The method effectively combines multi-frame vehicle states, road structures and interactions among vehicles. With a novel extraction algorithm of environment feature, the method infers the probability of each driving maneuver by using a Dynamic Bayesian Net­work. The experimental results demonstrate that our method can predict lane-change maneuvers at least 2 seconds before they occur in real environments with an accuracy of 84.9%.