Characterisation of Drowsy Driver Behaviour and Drowsiness Baseline Data Set in a Dynamic Driving Simulator Cristina Periago*, James Jackson, Clara Cabutí, Fiona Azcarate, Elena Castro, Francesco Deiana, Adrià Roig, IDIADA Automotive Technology SA, Spain KEYWORDS – Drowsiness, driving simulator, KSS, eye tracking, physiological sensors. ABSTRACT The objective of this study is to investigate how extended periods of monotonous driving affect driver behaviour, with the goal of developing driver monitoring systems that can identify signs of fatigue and mitigate its adverse effects on the road. The study consists of two phases. The first, with 10 participants, aims to validate the test method. The second, involving 20 participants, will attempt to define sleep behaviour patterns in relation to the different levels defined by the KSS. The experiment is conducted in a dynamic driving simulator, in which the conditions were set up according to the objective of having a monotonous environment free of distractions. Participants drive for 90 minutes and every 5 minutes the experimenter asks them about their level of sleepiness, using the Karolinska Sleepiness Scale (KSS), a standardized instrument that measures the participant's subjective perception. In addition, participants are instrumented to collect physiological data (ECG, EEG, EDA and respiration), and an eye-tracking system monitors other sleepiness-related behaviours, such as blinking or yawning. The test ends when 90 minutes elapse or participants reach an advanced level of sleepiness on the KSS, 9. The results of a preliminary analysis of the first phase, with 10 participants, reveal patterns in the trends that are consistent with the expected results, so the method is considered validated. In the second phase, consisting of 20 participants, a detailed analysis of the data is performed, which corroborates the trends observed in the first analysis, in which blinking, microsleeps and PERCLOS have a positive trend. Following the analysis of the remaining signals, which is currently under development, we expect that negative trends in respiration and heart rate will be confirmed and that the number of times the participant suffers a deviation from the lane will increase given the momentary loss of control that can be suffered as a result of drowsiness. We dispose of a high variability of physiological profiles, which makes it difficult to extract common patterns for each of the levels of the KSS scale. Expanding the sample of participants would probably give us a profile that could be extrapolated to the population, or would indicate the existence of different profiles to be analyzed. This paper presents an adaptation of the method already tested with a real vehicle on tracks to the dynamic driving simulator. The fact of performing a simulation contributes to improve the conditions of replicability and safety of the experiment. In this particular case, it has made it possible to obtain additional data, such as physiological data at KSS levels higher than those allowed in real conditions. A first phase of testing validated that parameters such as the duration of the test, the ability to induce sleep in the participants and the achievement of sleepiness level 9 were feasible. From the first phase we conclude a valid method for drowsiness analysis. Furthermore, from the second phase we can clearly state that the results being obtained from the data analysis indicate that we understand the effects of drowsiness while driving, as we obtain predictions of KSS levels based on the raw data that are close enough to reality to confirm that the assumptions made are true.

Ms. Cristina Periago, Human Factors, Applus+ IDIADA