Driver intention prediction at roundabouts

Abstract

In this paper, a strategy to estimate the intention of the driver seconds before entering a roundabout is presented. In real situations, detecting that a driver is going to turn in a particular direction at the time of entering the roundabout is not enough to implement risk reduction strategies. It is then critical to recognize in advance the possible actions of the drivers. The proposed method uses kinematic and position information of a vehicle a few meters before the beginning of the roundabout to feed a neural network based on Long Short-Term Memory or LSTM models, allowing to predict the probability that the vehicle will turn right, left, or go straight before entering the roundabout.