Eye Movement-enhanced Visual Looming Model for Driver Emergency Break Reaction Time

The accurate prediction of driver brake reaction time (BRT) in emergency is essential for advanced driver assistant system design. However, the current BRT models lack sufficient consideration of driver's visual attention, leading to bias in the estimation. This research focuses on a more accurate BRT prediction model based on an enhanced visual looming model which incorporates eye movement parameters. Firstly, 32 participants’ brake reaction data under four types of distraction were collected through a driving simulation experiment. Twelve eye movement parameters were extracted, including Gaze Concentration, Glance Behavior Parameter, and Attentional Focus. Secondly, based on the visual looming theory, two basic threshold models were constructed by using the rate of change of angle and the ratio of the rate of change of angle to the angle itself as the trigger conditions for braking. Then, two eye-movement enhanced models have been constructed by incorporating eye movement variables into the model using addition and multiplication methods. Thirdly, the Akaike Information Criterion was used to compare the four models, obtaining the best eye movement parameter combination for prediction performance. The results confirm that the model using the ratio as the threshold provides a better goodness of fit. Moreover, incorporating eye-movement parameters into the visual looming model improves the goodness of fit. Specifically, gaze switching frequency, proportion of on-road gaze, gaze concentration in the last two seconds before lead vehicle deceleration, and gaze switching frequency in the last five seconds before lead vehicle deceleration are the most significant parameter for improving the model's predictions.