Abstract
Keywords- Driving Manoeuvres, Eye Movements, Driver Intention, Driver Monitoring, Advanced Driver Assistance Systems
An advanced driver assistance system (ADAS) should ideally be capable of correctly inferring the intentions of the driver from what is implied by the incoming data available to it. Mismatches between the driver´s intention and the system´s reaction can occur. For example, if a driver intends to overtake another vehicle a collision warning could be incorrectly given by an assistance system as the overtaking vehicle closes in on the vehicle ahead. This false warning is a potential source of confusion to the driver. The false warning of the assistance system in the preceding example occurs because the function `collision warning´ is designed for the manoeuvre `follow car´. The fast approach to the vehicle ahead is interpreted as a critical situation that might lead to a collision. It is not clear to the system that the driver intends to change lanes before the vehicle ahead is hit. It is, therefore, important to establish which data and methods of interpreting that data allow drivers´ intentions to be accurately inferred by an ADAS.
Eye movements have been found to be an indicator of information gathering, and therefore could, in theory, be used to derive information about a driver´s next planned objective. The aim of this study was to establish whether characteristic eye movement patterns precede particular driving manoeuvres in order to provide a means for an ADAS to identify intended manoeuvres without relying solely on CAN-bus data.
Drivers´ gaze behaviour was measured prior to and during the execution of five different driving manoeuvres (follow road, follow car, change lane left/right, overtake) performed on different types of road in real traffic using the DLR `ViewCar´, a research vehicle equipped with sensors and cameras. The interior of the car was subdivided into non-overlapping viewing zones (windscreen, rear view mirror, left wing mirror, speedometer, etc.). The eye movement data were analysed with Markov matrices in order to obtain the fixations´ transition frequencies from one viewing zone to another. Using a Chi-square test, it was found that significantly distinct gaze patterns precede the driving manoeuvres analysed. This study indicates that eye movement data could be used as a useful additional input to an ADAS by supplementing CAN-bus data. Future studies will examine the robustness of an ADAS composed of CAN-bus data combined with data from drivers´ gaze behaviour.