Journal Search Engine
Search Advanced Search Adode Reader(link)
Download PDF Export Citaion korean bibliography PMC previewer
ISSN : 1598-7248 (Print)
ISSN : 2234-6473 (Online)
Industrial Engineering & Management Systems Vol.18 No.4 pp.748-760
DOI : https://doi.org/10.7232/iems.2019.18.4.748

Ocular Indicators as Fatigue Detection Instruments for Indonesian Drivers

Maya Arlini Puspasari*, Hardianto Iridiastadi, Iftikar Zahedi Sutalaksana, Ade Sjafruddin
Faculty of Industrial Technology, Institut Teknologi Bandung, Bandung, Indonesia
Faculty of Civil and Environmental Engineering, Institut Teknologi Bandung, Bandung, Indonesia
Corresponding Author, E-mail: mayaarlini@students.itb.ac.id
June 11, 2018 March 9, 2019 September 25, 2019

ABSTRACT


Fatigue is a well-known major cause of road accidents. Ocular indicators have been regarded as reliable indicators for measuring fatigue. However, results from previous investigations remain unclear about the performance of ocular parameters to detect fatigue in a real-time driving condition. This study was aimed at evaluating performance of several responsive ocular measures to detect fatigue during a simulated driving task. Thirteen participants drove a medium fidelity driving simulator continuously for 3 hours in high and low traffic density, after normal sleep duration (8 h) and sleep-deprived condition (4 h). Results from the present study showed that sleep deprivation substantially affects blink duration, percentage of eye closure (PERCLOS), microsleep, slow eye movement (SEM), and saccadic parameters. Traffic density, however, only had moderate effect toward ocular parameters. Among all ocular indicators, blink duration, PERCLOS, and saccadic PV demonstrated high accuracy, sensitivity, and specificity to detect fatigue. The present study suggests that blink duration has the highest performance to detect low-level fatigue and heavy fatigue, with a cut-off value of 285.17 ms and 512.31 ms, respectively, compared to other ocular indicators. The implications of this study are implementing a fatigue detection device based on blink duration, PERCLOS, and saccadic PV parameters.



초록


    1. INTRODUCTION

    1.1 Fatigue While Driving

    Road accidents are one of the main contributors to fatality. Statistics showed that approximately 1.25 million people died every year because of road accidents (World Health Organization, 2018). Furthermore, data from Indonesian Bureau of Statistics showed the number of accidents in Indonesia reached more than 100,000 cases, with the number of fatalities being more than 25,000 people (Indonesian Bureau of Statistics, 2016). Data from Indonesian National Transportation Safety Board also confirmed that 35% of fatalities occurred during accidents (Indonesian National Transportation Safety Board, 2016).

    Fatigue while driving remains as one of the most influential factors in road accidents. It is found that fatigue and sleepiness affect mental alertness, decelerate reaction time, reduce awareness, and impair judgment, thus leading to accidents (Balkin et al., 2000;Van Dongen and Maislin, 2003;Williamson et al., 2011). Fatigue has been estimated to be involved in at least 20% of road accidents (Zhang et al., 2013;Horne and Reyner, 1995). These data suggest the importance of fatigue related to road traffic accidents.

    Williamson et al. (2011) defined fatigue as a biological drive for recuperative rest, caused by time of day, homeostatic, and work-related factors. Phillips (2015) defined fatigue as a non-optimal psychophysiological condition caused by excessive exertion, whereas exertion consists of performance, sleep, and rest duration, circadian rhythm, and individual and environmental factors. Fatigue is also heavily affected by driver experience, vehicle and environmental conditions, and lack of government regulation (Zhang et al., 2016). Furthermore, fatigue induces vigilance loss (Desai and Haque, 2006), which leads to accidents. In summary, fatigue consists of physical and mental aspects that affect psychophysiological conditions of the human body, performance loss, and biological drive for rest, caused by time of day and homeostatic, work-related, individual and environmental factors.

    According to Williamson et al. (2011), May and Baldwin (2009), and Di Milia et al. (2011), several factors cause fatigue: sleep-related, task-related, and external factors. One of the prominent sleep-related factor is homeostatic, which is associated with reduction of sleep duration, sleep quality, and extension of time awake (Williamson et al., 2011). Several studies have demonstrated that less than 7 hours of sleep resulted in a significant increase of fatigue and sleepiness (Morad et al., 2009;Hanowski et al., 2007;Akerstedt et al., 2010;Abe et al., 2011). Sleep deprivation is a common phenomenon in the world, especially in Indonesia, where the majority of people have to get up very early (around 4:30 a.m.). This is due to cultural and religious needs. Furthermore, in certain month of the year, most Indonesians have to wake up earlier than 4:30 a.m. to participate in fasting, thus changing their sleep habit. Many Indonesians in large cities also spend longer time awake because of heavy traffic conditions.

    Fatigue is also influenced by task characteristics (May and Baldwin, 2009). Traffic density is one of the major problems related to task characteristics and associated with road driving (Gimeno et al., 2006;Cardona and Quevedo, 2014). In Indonesia, traffic density is an urgent problem. It is not uncommon for Indonesians to drive very early to their workplace, because of heavy traffic condition. In most large cities, the traffic causes 1.5 hours to be spent for a 20 km drive (the lowest point of vehicle speed approximately 5 km/hours), with most traffic experienced on toll roads (Kompas.com, 2014). This condition can influence performance of drivers; Shakouri et al. (2014) showed that lower traffic density causes 12% improvement in driving performance. However, studies that concentrate on effects of traffic density on fatigue are still limited, as most studies generally discussed driving task variation (Cardona and Quevedo, 2014;Gimeno et al., 2006) or driving environment (Thiffault and Bergeron, 2003;Larue et al., 2011).

    Fatigue can be evaluated by objective, subjective, and performance measurements. Objective measurements that were used in previous studies were electroencephalography (EEG) signal, ocular indicators, heart rate variability, and salivary amylase (Lal and Craig, 2001;Schleicher et al., 2008;Dawson et al., 2014). Subjective instruments such as Swedish Occupational Fatigue Inventory, Visual Analogue Scale, Karolinska Sleepiness Scale (KSS), Stanford Sleepiness Scale, Epsworth Sleepiness Scale (ESS) (Dawson et al., 2014;Ahsberg et al., 2000). Performance evaluations involve primary tasks (e.g., lane and steering deviation) and secondary tasks (e.g., a psychomotor vigilance task and a cognitive task) (Baulk et al., 2008).

    1.2 Ocular Indicators

    Ocular indicators are considered more promising to be determined as a fatigue parameter in real-time driving (Morad et al., 2009;Di Stasi et al., 2012). Ocular indicators have non-intrusive characteristics, involving involuntary movement and the complex neuron system; thus, more favored to measure fatigue (Morad et al., 2009;Di Stasi et al., 2012). Ocular indicators are categorized into saccade, pupil, blink, and eyelid closure (Abe et al., 2011). Eyelid closure consist of parameters related to eyelid movement, e.g., blink duration, PERCLOS (Percentage of Eye Closure), blink rate, microsleep, blink amplitude, blink velocity, and blink interval (Schleicher et al., 2008). On the other hand, saccadic parameters, slow eye movement (SEM), fixation duration, pupil diameter, and pupil constriction latency are often categorized as oculomotor parameters (Russo et al., 2003).

    Several investigations have discussed the effects of sleep deprivation toward ocular indicators (Russo et al., 2003;McClelland et al., 2010;Zils et al., 2005;Bocca and Denise, 2006). Zils et al. (2005) noted that peak velocity of saccade was significantly affected by sleep deprivation. Different results were reported by Russo et al. (2003) and McClelland et al. (2010). Their studies found that pupil diameter and pupil constriction latency were influenced by sleep deprivation. Another study by Bocca and Denise (2006) demonstrated that saccade duration was significantly affected by sleep deprivation. While the effect of sleep duration has been well-known, the effect of traffic density was not thoroughly investigated.

    Furthermore, studies regarding the performance of ocular indicators in detecting fatigue are still inconclusive. Schleicher et al. (2008) stated that blink duration, blink interval, delay of lid reopening, and lid closure speed as indicators that correlate most with subjective sleepiness. In line with Schleicher et al. (2008), the outcome of other studies identified eye closure speed and blink duration as parameters with the highest sensitivity and specificity (Akerstedt et al., 2010;Jackson et al., 2016). By contrast, Abe et al. (2011) found PERCLOS to be the most sensitive indicator of vigilance decrement. Works of Di Stasi et al. (2012) demonstrated that saccadic PV had a high reliability in fatigue measurement. These previous investigations had different results, and there was insufficient literature on the integration of eyelid closure parameters, saccadic parameters, and oculomotor parameters to generate the accurate, sensitive, and specific parameters to detect fatigue.

    1.3 Aim and Scope of the Study

    The present study was aimed at evaluating the performance of several ocular indicators that are responsive to fatigue during a simulated driving task, using sleep duration and traffic density as the main factors of fatigue. These two factors were used in the present study based on previous studies and the contextual problem of road drivers in Indonesia (Williamson et al., 2011). Furthermore, fatigue classification based on ocular indicators was also determined in the present study. The output of this study was to determine the accurate, sensitive, and specific parameters in ocular indicators that can be utilized to detect fatigue. Some contributions of the present study include incorporating traffic density as a fatigue-inducing factor and measuring ocular indicators that consist of eyelid closure parameters, saccadic parameters, and oculomotor parameters to generate accurate, sensitive, and specific indicators of fatigue.

    Practical implications of the present study were determining low-level fatigue (fatigue onset) and heavy fatigue experienced by drivers. Therefore, suggestions can be made for transportation companies or other stakeholders to implement a fatigue detection device based on ocular parameters. The device can then be used as one of instruments that increase the alertness of drivers, which will reduce the risk of road accidents.

    2. METHODS

    2.1 Participants

    Thirteen male subjects (mean age: 30.2 ± 4.97 years) were recruited. Each subject received monetary compensations of IDR 600,000 (US$50). One subject was excluded from this experiment due to illness after the second combination. The condition of participants was first screened out to ensure the homogeneity of data. Participants who are not heavy smokers (only smoke 1–2 cigarettes per day) were selected for this experiment. Most participants consumed tea on daily basis and occasionally consumed coffee (average consumption: 1.1 ± 0.65 cups). Another inclusion criterion for participants consisted of having experience in driving a car for at least two years, no visual correction, and a good health condition. They also reported normal sleep duration (6 hours/day on average) and absence of sleep disturbance.

    The participants were also screened out of their chronotype and daytime sleepiness level using Morningness– Eveningness Questionnaire (MEQ) and Epsworth Sleepiness Scale (ESS) (Abe et al., 2011;Johns, 1991). The chronotype of participants was tested by MEQ to determine whether the subjects categorized as morning or evening person (Horne and Östberg, 1976). An extreme morning person (score of MEQ: 70–86) and an extreme evening person (score of MEQ: 16–30) were excluded from this experiment (Abe et al., 2011). The daytime sleepiness level of participants was tested by ESS (Johns, 1991). Extremely sleepy people (score of ESS exceeding 10) were excluded from the present study (Abe et al., 2011).

    Participants were asked to refrain from smoking and consuming all substances that could affect their fatigue level 24 hours before the experiment, including coffee, tea, energy drinks, and medications (Schleicher et al., 2008;Akerstedt et al., 2010). Before the experiment began, all participants received detailed information about the experiment and gave written informed consent of their participation. The protocol of the present study has been approved by Institute of Research and Community Services of Bandung Institute of Technology as the Institutional Review Board.

    2.2 Procedure

    Before the test began, the heart rate and blood pressure of the participants were recorded to ensure their fitness level. Participants were given light meals provided by researchers before the driving session. Moreover, a training session was also conducted to familiarize participants with the driving simulator. Participants were instructed to drive a monotonous simulated driving for three hours with no specific scenario (free driving). The maximum speed required for each driver was 80 km/h. The experiment was from 8 a.m. to 11 a.m. to ensure the participants were in their best condition (Folkard and Lombardi, 2006).

    The task of the simulation was counterbalanced into four combinations that consisted of two main factors: sleep duration and traffic density. Sleep duration was divided into two levels: 8 hours of sleep and 4 hours of sleep. Traffic density was divided into two factors: low traffic and high traffic (determined by 20% and 80% in City Car Driving version 1.4, respectively). Participants were expected to do all four combinations in a 5-day interval to prevent the homeostatic effect. In a 4-hour sleep condition, one participant reported that he was unable to continue driving (after 2 hours of experiment); therefore, the experiment was rescheduled into the next three days.

    The participants were asked to control their sleep duration the day before the experiment with the activity tracker named Fitbit Charge HR (based on De Zambotti et al., 2016). If the participants’ sleep duration was not in the range of 3.5–4.5 hours (in a sleep-deprived condition) and 7.5-8.5 hours (in a normal sleep condition), the participants were asked to redo the experiment another day. Due to less control of participants regarding their sleep (only relying on the Fitbit device), the actual sleep duration of the participants varied from 3.75 ± 0.313 hours (for the sleep-deprived condition) to 7.45 ± 0.343 hours (for the normal sleep condition).

    2.3 Data Recording

    A medium fidelity driving simulator was used in the present study. The simulator comprised of three foot pedals, a steering wheel with force feedback, and a gear shift lever (G27 92, Logitech, US). A 55 inch HD screen was used as a projection of the driving environment for visual feedback. Driving simulator software (City Car Driving version 1.4, Multisoft, Russia) was used in this study as the driving environment (Figure 1). The driving environment selected for the present study was a highway with high and low traffic densities, with normal daylight weather, and a normal traffic behavior. A video camera was used to record facial expression and monitored by an experimenter sitting in the same room. An eye tracker (EyeLink II, SR Research, US) measured ocular indicators for the entire duration of the experiment. The frequency of line crossing and incidents were counted and divided into 20-minute time frames. Line crossing was counted as the frequency of a vehicle crosses the lane. Furthermore, incident frequency was defined as the number of crash experienced by the vehicle. Previous research noted about the relationship of fatigue with line crossing and incident frequency, e.g. Brookhuis et al. (2003), Philip et al. (2005), and Baulk et al. (2008). Brookhuis et al. (2003) stated that the number of line crossing is affected by sleep loss during driving in real roads. Similarly, Philip et al. (2005) stated that line crossing incident frequency significantly increased with extended wakefulness, which also confirmed by Baulk et al. (2008).

    Karolinska Sleepiness Scale (KSS) questionnaire was applied in this study as the subjective rating to validate the fatigue level. KSS has 9 scales which consists of very alert (scale 1), alert (scale 3), neither alert nor sleepy (scale 5), sleepy but not difficult staying awake (scale 7) and very sleepy, fighting sleep, difficulty staying awake (scale 9) (Pauly and Shankar, 2015;Kaida et al., 2006). Every 20 minutes, participants were asked to rate their subjective sleepiness using KSS questionnaire. The 20- minute time frame was determined by De Naurois et al. (2019), who stated that the KSS measurement should not be below the 15-minute interval as it would affect the alertness of participants.

    2.4 Data Analysis

    This study utilized 10 ocular parameters, namely blink duration, blink rate, percentage of eye closure (PERCLOS), microsleep, SEM, saccade amplitude, saccade duration, saccadic PV, pupil diameter, and fixation duration. Blink duration, PERCLOS, and microsleep were used in the present study because of their strong correlations with the sleep-deprived condition and time on task, as reported in previous studies (Akerstedt et al., 2010;Abe et al., 2011;Schleicher et al., 2008). Saccade amplitude, saccade duration, and saccadic PV were utilized because they showed significant results (p<0.050) toward driving duration in Di Stasi et al. (2012). Pupil diameter and fixation duration were proposed because of promising results by Russo et al. (2003) and Schleicher et al. (2018). SEM was adopted because of promising results by Shin et al. (2011).

    In the present study, an eye tracker was used to record the data with the sampling rate of 500 Hz. Blink duration was measured as the time from start of lid closure to the moment of lid reopening (Schleicher et al., 2008). Blink rate was defined as the number of blink per minute (frequency). Furthermore, PERCLOS was defined as the percentage of eyelid closure time (Dinges et al., 1998;Wierwille and Ellsworth, 1994). Eyelid closure was determined when diameters of two pupils could not be calculated (Abe et al., 2011). Moreover, microsleep was determined as the frequency of blink that had duration more than 500 ms (Sommer et al., 2009;Mahachandra et al., 2011). SEM was determined according to Shin et al. (2011), who divided saccade duration between 5º/s and 30º/s with observed time to obtain the percentage of SEM (Shin et al., 2011). Moreover, saccade was determined with a 1º minimum angle, a minimum velocity of 30º/s, and the duration had to be at least 4 ms. Saccade around blinks and fixation that were less than 100 ms were excluded from the analysis (Schleicher et al., 2008;Di Stasi et al., 2011;Di Stasi et al., 2013). Saccade amplitude was defined as the size of saccade, while saccade duration was measured as the time taken to complete the saccade, and saccadic PV was described as the highest velocity reached during the saccadic activity. Pupil diameter was defined as the size of the pupil in the eye, and fixation duration was measured as the time between two saccades when the pupil fixed on one object (Schleicher et al., 2008).

    Data of each participant were aggregated over a time window of 20 minutes (De Naurois et al., 2019) and present as mean and standard deviation. The data were categorized into nine segments of driving (3 hours). Performance indicators comprised of the frequency of line crossing and incident frequency. Subjective sleepiness scale was also measured by the KSS. Outliers of data were eliminated using Tukey’s Box Plot (Seo, 2002). This study applied five methods, which consisted of MANOVA, Friedman’s test, Spearman rho, binary logistics regression, and Receiver Operating Characteristics (ROC). Two-way Multivariate Analysis of Variance (MANOVA) with sleep duration (8 vs. 4 hours of sleep) and traffic density (20% and 80%) were performed on ocular indicators and performance measurements in order to determine whether there are any statistically significant differences between the means of groups. Subjective rating was analyzed by Friedman’s test for nonparametric data to determine the difference between each combination (Kaida et al., 2006). Spearman rho correlation was used for analyzing the relationships between parameters of ocular, performance, and subjective indicators (Schleicher et al., 2008). These three methods were used to determine the best predictors for fatigue. 10 ocular parameters will be ranked based on these three methods and the parameters that have the highest response will be analyzed further by binary logistics regression and ROC.

    Binary logistics regression was performed to classify the fatigue level, and Receiver Operating Characteristics (ROC) was applied to determine the cutoff point of each classification. Fatigue was classified into “alert condition,” “low-level fatigue,” and “heavy fatigue” (Table 1). The alert condition was defined when the KSS score fell into 1–5 (Pauly and Shankar, 2015). Low-level fatigue was defined when participants had a KSS score between 6 and 7. Furthermore, heavy fatigue was defined when participants had a KSS score between 8 and 9 and experienced line crossing at least once (Akerstedt et al., 2010;Liang et al., 2017).

    Youden index was used for cut-off point determination (formula 1). The highest point of the Youden index marked the cut-off point of each parameter (Hajian-Tilaki, 2013). The area under curve (AUC) in ROC marked the accuracy of the model, with a minimum point of 70% (Hajian-Tilaki, 2013).

    Youden Index = Sensitivity+Specificity–1
    (1)

    3. RESULTS

    3.1 Effects of Independent Variables (Sleep Duration and Traffic Density)

    3.1.1 Effects on Ocular Indicators

    Table 2 and Figure 2 summarized the effect of sleep duration and traffic density on ocular indicators. The effect of sleep duration was more prominent than that of traffic density, as shown by blink duration, PERCLOS, microsleep, saccadic parameters, and pupil diameter. In general, the sleep-deprived condition (4 hours) had a higher blink duration, PERCLOS, and microsleep than the normal sleep condition (8 hours) for approximately 157%, 90%, and 82%, respectively. The sleep-deprived condition was reported for having lower saccadic amplitude and saccadic PV for approximately 20% and 26%, respectively, higher than the normal sleep condition. The blink duration had strong effects on sleep duration, followed by PERCLOS, saccadic PV, microsleep, saccade duration, pupil diameter, and saccade amplitude. By contrast, fixation duration, SEM, and the blink rate showed no effects on sleep duration.

    Most variables showed moderate effects of traffic density. Higher traffic density influenced saccade amplitude, saccadic PV, SEM, and fixation duration (which increased) and pupil diameter (which decreased). The interaction effect of sleep duration and traffic density was shown in the blink rate, PERCLOS, saccade amplitude, SEM, and pupil diameter. The effect of high traffic density was clearly seen in the 8-hour sleep duration, which dramatically increased the blink rate, PERCLOS, pupil diameter. By contrast, the effect of high traffic density in the 4-hour sleep condition affected the saccade amplitude and saccadic PV (which increased) and SEM (which decreased).

    In Figure 2, blink duration, PERCLOS, and microsleep increased linearly with driving duration (for both 4 hours of sleep and 8 hours of sleep and high and low traffic densities). By contrast, saccadic parameters decreased when the driving duration increased. The effect of sleep duration was prominent, where 4-hour sleep affected the increase of eyelid closure parameters. By contrast, 4-hour sleep influenced the decrease of saccadic parameters. The effect of traffic density was moderate; however, the high traffic situation affected the increase of eyelid closure parameters, especially in 8-hour sleep.

    3.1.2 Effects on Subjective Ratings

    Generally, the KSS score of participants increased linearly during the driving period. KSS scores of participants were significantly affected by sleep duration (χ2 = 130.882, p = 0.000) and traffic density (χ2 = 14.041, p = 0.000). Participants who slept for 4 hours had higher KSS scores compared to those who slept for 8 hours. There were also combined effects regarding sleep duration and traffic density. In the 8-hour sleep condition, high traffic resulted in a higher KSS score than that due to low traffic. By contrast, in the 4-hour sleep condition, low traffic had higher KSS than the high traffic condition.

    3.1.3 Effects on Performance Measures

    Both line crossing and incident frequency increased in 4-hour sleep compared to the 8-hour sleep. Line crossing was significantly affected by sleep duration (p<0.001) and traffic density (p<0.001). Furthermore, the incident frequency was also significantly affected by sleep duration (p<0.001).

    3.2 Correlation Between Ocular, Performance, and Subjective Indicators

    The correlation between ocular indicators and other indicators (KSS, line crossing, and incident frequency) was computed using Spearman rank correlation (Table 3). The Spearman rank correlation was determined among KSS, line crossing, and incident frequency during a 20- minute time window and the statistical parameter (mean) of each ocular indicator.

    The results of Table 3 showed that blink duration, saccadic PV, PERCLOS, microsleep, and saccade amplitude consistently correlated with subjective sleepiness, line crossing, and the incident frequency. On average, the blink duration parameter had the highest Spearman’s rho, followed by saccadic PV, microsleep, PERCLOS, and saccade amplitude.

    3.3 Classification of Fatigue

    The classification of fatigue was conducted using binary logistics regression (Table 4). Results from Sections 3.1 and 3.2 generated the predictors for fatigue classification that consisted of blink duration, saccadic PV, PERCLOS, microsleep, and saccade amplitude. Based on aforementioned classification (Table 1), the univariate model of each predictor was compared to the multivariate model. Saccadic PV had the highest accuracy for predicting the data in cut-off point 1, and PERCLOS had the highest accuracy in cut-off point 2. Results indicated that the full model correctly predicted 74.3%–86.3% of the data with 0.369–0.370 R2. The full multivariate models added moderate R2 to the overall result. Generally, cut-off point 2 produced a better accuracy for each parameter.

    An analysis of the ROC curve was utilized to determine the performance of the predictive model (Table 5). The full model from cut-off points 1 and 2 generated AUC scores of 81.9% and 85.4%, respectively. Furthermore, the analysis of ROC was applied to each parameter to determine accuracy and cut-off point. Blink duration, PERCLOS, and saccadic PV consistently had AUC score above 70% from two classifications. Blink duration had the highest accuracy for separating between an alert and low-level fatigue condition and between low-level fatigue and heavy fatigue condition. The cut-off value of blink duration for the difference between alert and low-level fatigue was 285.17 ms, and the sensitivity and specificity of this cut-off value were 65.3% and 79.4%, respectively. The cut-off value of blink duration for the discrimination between low-level fatigue and heavy fatigue was 512.31 ms, and the sensitivity and specificity of this cut-off value were 62.1% and 84.2%, respectively.

    4. DISCUSSION

    4.1 Sleep Duration and Traffic Density

    The present study was aimed at determining the performance of ocular indicators that detect fatigue, by addressing the effect of sleep deprivation and traffic density in simulated driving. Sleep duration had pronounced effects on most indicators (ocular, subjective, and performance). The clearest effects were seen on KSS, line crossing, the incident frequency, saccadic PV, saccade amplitude, blink duration, PERCLOS, and microsleep. Roughly 56%-79% increase of line crossing and incident frequency was found in a sleep-deprived condition. It can be concluded that performance of driving was more affected by sleep duration than traffic density; this complemented findings of Gastaldi et al. (2014) and Baulk et al. (2008). Approximately a 57% increase of KSS was found in the sleep-deprived condition; this is coherent with the study of Williamson et al. (2011). The present study confirmed results in previous studies that found a strong effect of sleep deprivation on subjective sleepiness (Akerstedt et al., 2010;Abe et al., 2011). It verifies that experimental protocol was sufficient to induce subjective sleepiness of participants (Abe et al., 2011). Furthermore, the standard deviation of subjective sleepiness was lower than those of line crossing and incidents, indicating small variance of individual differences. It showed that individuals perceived sleepiness the same way (Akerstedt et al., 2010); thus, subjective sleepiness can be used as a reliable reference indicator of measuring sleepiness (Pauly and Shankar, 2015) as applied in the present study.

    The effect of sleep duration was also prominent on ocular indicators. Blink duration had a twofold increase in the sleep-deprived condition than in the normal sleep condition, in line with findings by Schleicher et al. (2008). By contrast, the blink rate parameter was not significantly affected by sleep duration. This is due to the nature of the blink rate, as it increases when subjects have early drowsiness but decrease when subjects suffer from heavier fatigue (Schleicher et al., 2008). The result was consistent with the studies of Schleicher et al. (2008) and Akerstedt et al. (2010), which indicated that blink duration was more sensitive in measuring fatigue than the blink rate. The sleep-deprived condition also affected both PERCLOS and microsleep parameters and thus coherent with studies of Abe et al. (2011). Sleep duration also significantly affected saccadic PV and saccade amplitude; this is consistent with a finding by Di Stasi et al. (2012) that sleep deprivation induced fatigue.

    On the contrary, the effect of traffic density was moderate. The present study is in agreement with Shakouri et al. (2014), who found that high traffic density increased mental workload (by saccadic parameters, SEM, fixation duration, and pupil diameter). However, the effect of traffic density was not as significant as that of sleep duration. High traffic density induced higher saccade amplitude, which indicated an increase in task complexity (Cardona and Quevedo, 2014). High traffic density was also consistent with the escalation of fixation duration; this correlates with the increase of cognitive activity (Schleicher et al., 2008). Traffic density had varied results toward saccadic PV, SEM, and pupil diameter. High traffic affected higher SEM and lower pupil diameter in the 8-hour sleep condition, which indicated sleepiness (Russo et al., 2003;Shin et al., 2011). On the contrary, saccadic PV increased by higher traffic density, especially in the 4- hour sleep condition, thus implying alertness (Di Stasi et al., 2012). It can be concluded that high traffic density induces high mental workload and has varied results in the sleep-deprived condition and the normal sleep condition. In the sleep-deprived condition, traffic density influences alertness, because of a higher workload that stimulates eye movement. By contrast, traffic density affects sleepiness in the normal sleep condition because of monotony. These findings reflected results about traffic density that were seldom discussed before.

    4.2 Ocular Indicators as Fatigue Detection Measurements

    The present study complements the work of Abe et al. (2011), who demonstrated PERCLOS as one of significant parameters that measure fatigue. In the present study, blink duration was regarded as the most sensitive and specific parameter of fatigue, followed by saccadic PV and PERCLOS. In this research, saccadic PV was more sensitive in measuring fatigue based on results from the ROC curve than other saccadic indices, consistent with the results of Di Stasi et al. (2012). Furthermore, saccade amplitude was also significantly affected by fatigue; this complements findings of Zils et al. (2005) and Bocca and Denise (2006).

    The correlations among ocular, subjective, and performance indicators, combined with the results from fatigue classifications, indicated that blink duration was the most sensitive indicator of fatigue. This finding was in line with Schleicher et al. (2008) and Akerstedt et al. (2010), who stated that eye closure indicators were one of the most sensitive measures for sleepiness, besides subjective sleepiness. The Spearman rank correlation showed somewhat strong correlation toward KSS indices; this was consistent with the study of Schleicher et al. (2008). As a result, the present study suggests blink duration, PERCLOS, and saccadic PV as the sensitive and specific parameters of fatigue detection. This in agreement with studies by Schleicher et al. (2008), Akerstedt et al. (2011), and Abe et al. (2011).

    4.3 Limitations

    Some limitations should be reflected in the present study. First, the present study had a medium fidelity driving simulator. According to Meuleners and Fraser (2015), results from the driving simulator were not significantly different from those of real driving, based on driving behavior (e.g., compliance to traffic signage and maximum driving speed). Furthermore, Davenne et al. (2012) stated that subjective fatigue did not significantly differ between the driving simulator and real driving, except the performance of drivers (e.g., line crossing). However, in the present study, we found an increment of sleepiness in daytime driving without sleep loss, indicating monotonous effects of the simulator itself (Akerstedt et al., 2010). Future research should include real driving to enhance the validity of the present study. Second, the sample size was small. Further studies with a larger sample are desirable to increase the reliability of ocular parameters as fatigue detectors. Third, it should be discussed further whether the cut-off value (285.17 ms and 512.31 ms) reported in the present study is practically suitable for detection of fatigue while driving. Fourth, the scenario developed in this study was not too rigid to control the drivers, e.g. participants can drive freely in their own style with speed restriction (80 km/h). The advantage of this condition is the situation will be similar to real driving, so this study can be generalized more closely to real driving situation. However, it implies a disadvantage such as the incident frequency will be higher than in real driving situation. Future research should include more control of participant as well as the scenario for simulator driving.

    4.4 Implications

    Some implications for future research and development are discussed. First, the results found a significant contribution of blink duration, saccadic PV, and PERCLOS in distinguishing stages of fatigue (the alert condition, low-level fatigue, and heavy fatigue). Those three parameters should be further clustered to be developed into a fatigue detection device, or one of the parameters be chosen as a fatigue detection instrument. This device can be used for detecting fatigue in a real-time driving condition and consequently reduce the risk of accidents.

    Second, the sensitivity and specificity of each parameter should also be discussed. The results designate the sensitivity and specificity estimation of 56%–82% and 62%–84%, respectively, for blink duration, saccadic PV, and PERCLOS parameters. The lower level of specificity indicates high frequency of false alarms, which is likely to lessen drivers’ confidence of the system. Drivers’ actions are likely to change after an alarm. Implications for real driving would be higher level of specificity to minimize false alarms, which would be close to 100%, with sensitivity above 50% to be accepted by the drivers (Akerstedt et al., 2010). In-depth research is needed to establish the level of sensitivity and specificity that practitioners in real driving conditions will accept.

    5. CONCLUSIONS

    The present study aimed to determine the performance of ocular indicators that detect fatigue, by addressing sleep duration and traffic density effects. The results showed that sleep deprivation substantially affected blink duration, saccadic parameters, PERCLOS, microsleep, and pupil diameter. By contrast, traffic density had a moderate effect toward saccade amplitude, saccadic PV, SEM, fixation duration, and pupil diameter. Furthermore, sleep duration influenced performance measures and subjective ratings more than traffic density did. Blink duration was regarded as the most accurate parameter to detect fatigue, as measured by correlation and the ROC curve, followed by saccadic PV and PERCLOS. In saccadic parameters, saccadic PV was proven to be more sensitive in measuring fatigue than other saccadic parameters. Thus, the present study suggests that blink duration, saccadic PV, and PERCLOS are accurate, sensitive, and specific parameters of fatigue while driving.

    ACKNOWLEDGMENT

    The present study was conducted with support from the research grant of the Directorate of Higher Education (DIKTI). The authors acknowledge Work System Engineering and Ergonomics Laboratory of the Bandung Institute of Technology for data processing and thank Ergonomics Centre Universitas Indonesia for providing the instruments for this experiment.

    Figure

    IEMS-18-4-748_F1.gif

    Instruments of this research

    IEMS-18-4-748_F2.gif

    Trends of ocular indicators.

    Table

    Prediction targets

    Summary of effects from MANOVA (F-values)

    Spearman rank correlation

    Summary of the predictive model (binary logistics regression)

    Receiver Operating Characteristics (ROC)

    REFERENCES

    1. Abe, T. , Nonomura, T. , Komada, Y. , Asaoka, S. , Sasai, T. , Ueno, A. , and Inoue, Y. (2011), Detecting deteriorated vigilance using percentage of eyelid closure time during behavioral maintenance of wakefulness tests, International Journal of Psychophysiology, 82(3), 269-274.
    2. Ahsberg, E. , Gamberale, F. , and Gustafsson, K. (2000), Perceived fatigue after mental work: An experimental evaluation of a fatigue inventory, Ergonomics, 43(2), 252-268.
    3. Akerstedt, T. , Ingre, M. , Kecklund, G. , Anund, A. , Sandberg, D. , Wahde, M. , Phillip, P. , and Kronberg, P. (2010), Reaction of sleepiness indicator to partial sleep deprivation, time of day and time on task in a driving simulator – the DROWSI project, Journal of Sleep Research, 19(2), 298-309.
    4. Balkin, T. , Thorne, D. , Sing, H. , Thomas, M. , Redmond, D. , Wesensten, N. , Russo, M. , Williamson, J. , Hall, S. , Belenky, G. (2000), Effects of sleep schedules on commercial motor vehicle driver performance, Report MC-00-133, National Technical Information Service, U.S. Department of Transportation, Springfield, VA.
    5. Baulk, S. D. , Biggs, S. N. , Reid, K. J. , van den Heuvel, C. J. , Dawson, D. (2008), Chasing the silver bullet: Measuring driver fatigue using simple and complex tasks, Accident Analysis and Prevention, 40(1), 396-402.
    6. Bocca, M. L. and Denise, P. (2006), Total sleep deprivation effect on disengagement of spatial attention as assessed by saccadic eye movements, Clinical Neurophysiology, 117(4), 894-899.
    7. Brookhuis, K. A. , De Waard, D. , and Fairclough, S. H. (2003), Criteria for driver impairment, Ergonomics, 46(5), 433-445.
    8. Cardona, G. and Quevedo, N. (2014), Blinking and driving: The influence of saccades and cognitive workload, Current Eye Research, 39(3), 239-244.
    9. Davenne, D. , Lericollais, R. , Sagaspe, P. , Taillard, J. , Gauthiera, A. , Espiéc, S. , and Philip, P. (2012), Reliability of simulator driving tool for evaluation of sleepiness, fatigue and driving performance, Accident Analysis and Prevention, 45, 677- 682.
    10. Dawson, D. , Searle, A. K. , and Paterson, J. L. (2014), Look before you (s)leep: Evaluating the use of fatigue detection technologies within a fatigue risk management system for the road transport industry, Sleep Medicine Reviews, 18(2), 141-152.
    11. de Naurois, C. J. , Bourdin, C. , Stratulat, A. , Diaz, E. , Vercher, J. L. (2019), Detection and Prediction of driver drowsiness using artificial neural network models, Accident Analysis and Prevention, 126, 95-104.
    12. Desai, A. V. and Haque, M. A. (2006), Vigilance monitoring for operator safety: A simulation study on highway driving, Journal of Safety Research, 37(2), 139-147.
    13. de Zambotti, M. , Baker, F. C. , Willoughby, A. R. , Godino, J. G. , Wing, D. , Patrick, K. , and Colrain, I. M. (2016), Measures of sleep and cardiac functioning during sleep using a multi-sensory commerciallyavailable wristband in adolescents, Physiology & Behavior, 158, 143-149.
    14. Dinges, D. F. , Mallis, M. , Maislin, G. , and dan Powell, J. W. (1998), Evaluation of techniques for ocular measurement as an index of fatigue and as the basis for alertness management, U.S. Department of Transportation, National Highway Traffic Safety Administration.
    15. Di Milia, L. , Smolensky, M. H. , Costa, G. , Howarth, H. D. , Ohayon, M. M. , and Philip, P. (2011), Demographic factors, fatigue, and driving accidents: An examination of the published literature, Accident Analysis and Prevention, 43(2), 516-532.
    16. Di Stasi, L.L. , Antoli, A. , and Cañas, J. J. (2011), Main sequence: An index for detecting mental workload variation in complex tasks, Applied Ergonomics, 42(6), 807-813.
    17. Di Stasi, L. L. , Renner, R. , Catena, A. , Cañas, J. J. , Velichkovsky, B. M. , and Pannasch, S. (2012), Towards a driver fatigue test based on the saccadic main sequence: A partial validation by subjective report data, Transportation Research Part C: Emerging Technologies, 21(1), 122-133.
    18. Di Stasi, L. L. , Catena, A. , Cañas, J. J. , Macknike, S. L. , Martinez-Conde, S. (2013), Saccadic velocity as an arousal index in naturalistic tasks, Neuroscience and Biobehavioral Reviews, 37(5), 968-975.
    19. Folkard, S. and Lombardi, D. A. (2006), Modelling the impact of the components of long work hours, American Journal of Industrial Medicine, 49(11), 953-963.
    20. Gastaldi, M. , Rossi, R. , and Gecchele, G. (2014), Effects of driver task-related fatigue on driving performance, Procedia - Social and Behavioral Sciences, 111, 955- 964.
    21. Gimeno, P. T. , Cerezuela, G. P. , and Montanes, M. C. (2006), On the concept and measurement of driver drowsiness, fatigue, and inattention: Implications for countermeasures, Int. J. Vehicle Design, 42(1/2), 67-86.
    22. Hanowski, R. J. , Hickman, J. , Fumero, M. C. , Olson, R. L. , Dingus, T. A. (2007), The sleep of commercial vehicle drivers under the 2003 hours-of-service regulations, Accident Analysis and Prevention, 39(6), 1140-1145.
    23. Hajian-Tilaki, K. (2013), Receiver operating characteristics (ROC) curve analysis for medical diagnostic test evaluation, Caspian J. Intern Med., 4(2), 627-635.
    24. Horne, J. A. and Reyner, L. A. (1995), Sleep-related vehicle accidents, Brit. Med. J., 310, 565-567.
    25. Horne J. A. and Östberg, O. (1976), A self-assessment questionnaire to determine morningness-eveningness in human circadian rhythms, Int J Chronobiol, 4(2), 97-110.
    26. Indonesian Bureau of Statistics (2016), Jumlah kecelakaan, korban mati, luka berat, luka ringan, dan kerugian materi yang diderita tahun 1992-2016, Available from: http://www.bps.go.id/linkTabelStatis/view/id/1415.
    27. Indonesian National Transportation Safety Board (2016), Data Investigasi Kecelakaan LLAJ 2010-2016, vailable from: http://knkt.dephub.go.id/knkt/ntsc_home/Media_Release/Media%20Release%20KNKT%202016/Media%20Release%202016%20-%20IK%20LLAJ%2020161130.pdf.
    28. Jackson, M. L. , Kennedy, G. A. , Clarke, C. , Gullo, M. , Swann, P. , Downey, A. L. , Hayley, A. C. , Pierce, R. J. , and Howard, M. E. (2016), The utility of automated measures of ocular metrics for detecting driver drowsiness during extended wakefulness, Accident Analysis and Prevention, 87, 127-133.
    29. Johns, M. W. (1991), A new method for measuring daytime sleepiness: The Epworth sleepiness scale, Sleep, 14(6), 540-546.
    30. Kaida, K. , Takahashi, M. , Åkerstedt, T. , Nakata, A. , Otsuka, Y. , Haratani, T. , and Fukasawa, K. (2006), Validation of the karolinksa sleepiness scale against performance and EEG variables, Clinical Neurophysiology, 117(7), 1574-1581.
    31. Kompas.com (2014), Jakarta Kian Macet, Kecepatan Kendaraan Hanya 5 km Per Jam, Available from: https://megapolitan.kompas.com/read/2014/06/04/1459211/Jakarta.Kian.Macet.Kecepatan.Kendaraan.Hanya.5.Km.Per.Jam.
    32. Lal, S. K. and Craig, A. (2001), A critical review of psychophysiology of driver fatigue, Biological Physiology, 55(3), 173-194.
    33. Larue, G. Rakotonirainy, A. , and Pettitt, A. N. (2011), Driving performance impairments due to hypovigilance on monotonous roads, Accident Analysis and Prevention, 43(6), 2037-2046.
    34. Liang, Y. , Horrey, W. J. , Howard, M. E. , Lee, M. L. , Anderson, C. , Shreeve, M. , O’Brien, C. S. , and Czeisler, C. A. (2017), Prediction of drowsiness events in night shift workers during morning driving, Accident Analysis and Prevention, 126, 105-114.
    35. Mahachandra, M. , Yassierli, Sutalaksana, I. Z. , and Suryadi, K. (2011), Sleepiness pattern of Indonesian professional driver based on subjective scale and eye closure activity, International Journal of Basic & Applied Sciences, 11(6), 87-96.
    36. May, J. and Baldwin, C. L. (2009), Driver fatigue: The importance of identifying causal factors of fatigue when considering detection and countermeasure technologies, Transportation Research Part F: Traffic Psychology and Behaviour, 12(3), 218-224.
    37. McClelland L. E. , Pilcher J. J. , and Moore D. D. (2010), Oculomotor measures as predictors of performance during sleep deprivation, Aviat Space Environ Med, 81(9), 833-842.
    38. Meuleners, L. and Fraser, M. (2015), A validation study of driving errors using a driving simulator, Transportation Research Part F: Traffic Psychology and Behaviour, 29, 14-21.
    39. Morad, Y. , Barkana, Y. , Zadok, D. , Hartstein, M. , Pras, E. , and Bar-Dayan, Y. (2009), Ocular parameters as an objective tool for the assessment of truck drivers fatigue, Accident Analysis and Prevention, 41(4), 856-860.
    40. Pauly, L. and Shankar, D. (2015), Detection of drowsiness based on HOG features and SVM classifiers, Proceedings of the IEEE International Conference on Computer Graphics, Vision, and Information Security (CGVIS), Kolkata, India.
    41. Philip, P. , Sagaspe, P. , Moore, N. , Taillard, J. , Charles, A. , Guilleminault, C. , and Bioulac, B. (2005), Fatigue, sleep restriction and driving performance, Accident Analysis and Prevention, 37(3), 473-478.
    42. Phillips, R. O. (2015), A review of definitions of fatigue – And a step towards a whole definition, Transportation Research Part F: Traffic Psychology and Behaviour, 29, 48-56.
    43. Russo, M. , Thomas, M. , Thorne, D. , Sing, H. , Redmond, D. , Rowland, L. , Johnson, D. , Hall, S. , Krichmar, J. , and Balkin, T. (2003), Oculomotor impairment during chronic partial sleep deprivation, Clinical Neurophysiology, 114(4), 723-736.
    44. Schleicher, R. , Galley, N. , Briest, S. , and Galley, L. (2008), Blinks and saccades as indicators of fatigue in sleepiness warnings: Looking tired?, Ergonomics, 51(7), 982-1010.
    45. Seo, S. (2002), A review and comparison of methods for detecting outliers in Univariate data sets, PhD dissertation, University of Pittsburgh, Pittsburgh.
    46. Shakouri, M. , Ikuma, L. H. , Aghazadeh, F. , Punniaraj, K. , and Ishak, S. (2014), Effects of work zone configurations and traffic density on performance variables and subjective workload, Accident Analysis & Prevention, 71, 166-176.
    47. Shin, D. , Sakai, H. , and Uchiyama, Y. (2011), Slow eye movement can prevent sleep-related accidents effectively in a simulated driving task, Journal of Sleep Research, 20(3), 416-424.
    48. Sommer, D. , Golz, M. , Schnupp, T. , Krajewski, J. , Trutschel, U. , and Edwards, D. (2009), A measure of strong driver fatigue, Proceedings of the Fifth International Driving Symposium on Human Factors in Driver Assessment, Training, and Vehicle Design, Big Sky, Montana, 9-15.
    49. Thiffault, P. and Bergeron, J. (2003), Monotony of road environment and driver fatigue: A simulator study, Accident Analysis and Prevention, 35(3), 381-391
    50. Van Dongen, H. and Maislin, G. (2003), The cumulative cost of additional wakefulness: Dose-response effects on neurobehavioral functions and sleep physiology from chronic sleep restriction and total sleep deprivation, Sleep, 26(2), 117-126.
    51. Wierwille, W. W. and Ellsworth, L. A. (1994), Evaluation of driver drowsiness by trained raters, Accident Analysis and Prevention, 26(5), 571-581.
    52. Williamson, A. , Lombardi, D. A. , Folkard, S. , Stutts, J. , Courtney, T. K. , and Connor, J. L. (2011), The link between fatigue and safety, Accident Analysis and Prevention, 43(2), 498-515.
    53. World Health Organization (2018), Road traffic injuries, cited 2018 Dec 7, Available from: http://www.who.int/mediacentre/factsheets/fs358/en/.
    54. Zhang, G. , Yau, K. , and Chen, G. (2013), Risk factors associated with traffic violations and accident severity in China, Accident Analysis and Prevention, 59, 18-25.
    55. Zhang, G. , Yau, K. W. , Zhang, X. , and Li, Y. (2016), Traffic accidents involving fatigue driving and their extent of casualties, Accident Analysis and Prevention, 87, 34-42.
    56. Zils, E. , Sprenger, A. , Heide, W. , Born, J. , and Gais, S. (2005), Differential effects of sleep deprivation on saccadic eye movement, Sleep, 28(9), 1109-1115.