A field test of eye tracking systems with one and three cameras

Driver state monitoring by eye tracking is an increasingly used method to assess driver distraction and sleepiness. The aim of this study is to compare a 1-camera eye tracking system with a 3-camera system in a field setting and evaluate how the choice of eye tracker impacts (a) the quality of the recordings and (b) behavioural performance indicators. Eighteen drivers drove an instrumented vehicle on a motorway for about 90 minutes while their gaze patterns were recorded with the two systems in parallel. The results show that neither quality nor performance indicators deteriorate over time. Tracking is available for 96 ± 6 % of the time for the 3-camera system and 77 ± 22 % for the 1- camera system, and both the frequency and the duration of lapses are considerably higher in the 1- camera system. The 3-camera system provides better vertical accuracy and increased horizontal availability. These properties cause performance indicators calculated from the two systems to deviate. For example glance durations are 0.76o ± 0.34o for the 3-camera system and 0.60o ± 0.48o for the 1-camera system. When comparing the output from the systems, it becomes clear that not all differences can be explained by the larger coverage of multiple cameras. In conclusion, the 3-camera system shows better performance with respect to data availability, working environment and accuracy. In many applications it is sufficient to use one camera, but special focus on the handling of lapses is essential.