A comparison of selected simple supervised learning algorithms to predict driver intent based on gaze data

Gaze behaviour is known to indicate information gathering. It is therefore suggested that it could be used to derive information about the driver's next planned objective in order to identify intended manoeuvres without relying solely on car data. Ultimately this would be practically realised by an Advanced Driver Assistance System (ADAS) using gaze data to correctly infer the intentions of the driver from what is implied by the incoming gaze data available to it. Neural Networks' ability to approximate arbitrary functions from observed data therefore makes them a candidate for modelling driver intent. Previous work has shown that significantly distinct gaze patterns precede each of the driving manoeuvres analysed indicating that eye movement data might be used as input to ADAS supplementing sensors, such as CAN-Bus (Controller Area Network), laser, radar or LIDAR (Light Detection and Ranging) in order to recognise intended driving manoeuvres. In this study, drivers' gaze behaviour was measured prior to and during the execution of different driving manoeuvres performed in a dynamic driving simulator. Artificial Neural Networks (ANNs), Bayesian Networks (BNs), and Naive Bayes Classifiers (NBCs) were then trained using gaze data to act as classifiers that predict the occurrence of certain driving manoeuvres. This has previously been successfully demonstrated with real traffic data [1]. Issues considered here included the amount of data that is used for predictive purposes prior to the manoeuvre, the accuracy of the predictive models at different times prior to the manoeuvre taking place and the relative difficulty of predicting a lane change left manoeuvre against predicting a lane change right manoeuvre. (C) 2013 Elsevier B.V. All rights reserved.