As a cause of accidents, drowsiness can cause economical and physical damage. A range of drowsiness estimation methods have been proposed in previous studies to aid accident prevention and address this problem. However, none of these methods are able to improve their estimation ability as the length of time or number of trials increases. Thus, in this study, we aim to find an effective drowsiness estimation method that is also able to improve its prediction ability as the subject’s activity increases. We used electroencephalogram (EEG) data to estimate drowsiness, and the Karolinska sleepiness scale (KSS) for drowsiness evaluation. Five parameters (α, β/α, (θ+α)/β, activity, and mobility) from the O1 electrode site were selected. By combining these parameters and KSS, we demonstrate that a typical support vector regression (SVR) algorithm can estimate drowsiness with a correlation coefficient (R²) of up to 0.64 and a root mean square error (RMSE) of up to 0.56. We propose a “recurrent SVR” (RSVR) method with improved estimation performance, as highlighted by an R² value of up to 0.83, and an RMSE of up to 0.15. These results suggest that in addition to being able to estimate drowsiness based on EEG data, RSVR is able to improve its drowsiness estimation performance.

Traffic accidents remain one of the most critical issues in many countries. One of the major causes of traffic accidents is drowsiness while driving. Since drowsiness is related to human physiological conditions, drowsiness is hard to prevent. Several studies have been conducted in assessing drowsiness, especially in a driving environment. One of the common methods used is the electroencephalogram (EEG). It is known that drowsiness occurs in the central nervous system; thus, estimating drowsiness using EEG is the promising way to assess drowsiness accurately. In this study, we tried to estimate drowsiness using frequency-domain and time-domain analysis of EEG. To validate the physiological conditions of the subjects, the Karolinska sleepiness scale (KSS), a subject-based assessment of drowsiness condition; and an examiner-based assessment known as facial expression evaluation (FEE) were applied. Three categories were considered; alert (KSS <; 6; FEE <; 1), weak drowsiness (KSS 6-7; FEE 1-2) and strong drowsiness (KSS > 7; FEE > 2). The six parameters (absolute and relative power of alpha, ratio of β/α and (θ+α)/β, and Hjorth activity and mobility parameters) had statistically significant differences between the three drowsiness conditions (P <; 0.001). By using both KSS and FEE, these parameters showed high accuracy in detecting drowsiness (up to 92.9%). Taken together, we suggest that EEG parameters can be used in detecting the three drowsiness conditions in a simulated driving environment.

It is widely known that many traffic accidents occur every year not only in Japan but also throughout the world. Sleepiness or drowsiness, which is the cause of dozing at the wheel, happens regardless of the physical condition of the driver at the time such as after having had meals or at midnight. This indicates that it is too difficult to expect the driver to avoid sleepiness or drowsiness by themselves. Therefore, various systems have been proposed to prevent traffic accidents caused by dozing at the wheel. In this study, we examined the relationship between subjective sleepiness during driving, which was evaluated by the Japanese version of the Karolinska sleepiness scale (KSS-J) and physiological parameters extracted from thoracic respiration signals. Then we tried to classify the existence of heavy, light, and no sleepiness using a support vector machine on those parameters. In this study, we determined a KSS-J score of 8 or 9, 6 to 8, and from 1 to 5 as signifying heavy, light, and no sleepiness states. The support vector machine was trained using three-quarters of the data for each subject and the remaining data was used as the testing data. This approach enabled us to obtain an accuracy of 89.4%. Therefore, it was suggested that thoracic respiration parameters were relevant to sleepiness.

Gazing Time Analysis for Drowsiness Assessment Using Eye Gaze Tracker

Traffic accident is one of the most important problems in many countries. One of traffic accident causes is drowsiness. Many studies have been done to solve this problem. One of them is to capture the driver's face expression and estimate the driver's drowsiness. Besides measuring the driver's physiological condition and use it to predict one's drowsiness is also considered. Since drowsiness is strongly related to human brain, by assessing drowsiness using biological signal especially brain signal become the most promising approach to evaluate drowsiness. We proposed a study to investigate the relationship between drowsiness and physiological condition by employing electroencephalogram (EEG) signal and Japanese version of Karolinska sleepiness scale (KSS-J) in driving simulator environment. The result showed alpha band power of EEG signal from occipital lobe in drowsy condition (KSS-J ≥ 7) increased significantly compared with that in alert condition (KSS-J <; 7) with P <; 0.001. Therefore, it is suggested that EEG is effective to find the drowsiness in driving simulator environment.

Many traffic accidents are due to drowsy driving. However, to date, only a few studies have been conducted on the gazing properties related to drowsiness. This study was conducted with the objective of estimating the relationship between gazing properties and drowsiness in three facial expression evaluation (FEE) categories: alert (FEE = 0), lightly drowsy (FEE = 1−2), heavily drowsy (FEE = 3−4). Drowsiness was investigated based on these eye-gazing properties by analyzing the gazing signal utilizing an eye gaze tracker and FEE in a driving simulator environment. The results obtained indicate that gazing properties have significant differences among the three drowsiness conditions, with p < 0.001 in a Kruskal–Wallis test. Furthermore, the overall classification accuracy of the three drowsiness conditions based on gazing properties using a support vector machine was 76.3%. This indicates that our proposed gazing properties can be used to quantitatively assess drowsiness.