Elsevier

Safety Science

Volume 119, November 2019, Pages 112-116
Safety Science

Continuous monitoring of visual distraction and drowsiness in shift-workers during naturalistic driving

https://doi.org/10.1016/j.ssci.2018.11.007Get rights and content

Highlights

  • Drowsiness and visual distraction examined in NDS of shift-workers.

  • Achieved continuous monitoring of real-world distracted driving behavioural signals.

  • Drivers significantly more likely to look toward their lap when drowsy.

  • Lap visual time-sharing sequences less frequent but longer duration on drowsy trips.

Abstract

Driver drowsiness is a significant public health problem and has previously been linked to an increase in drivers’ propensity to engage in visual distraction. This relationship however has yet to be examined under naturalistic driving conditions, where task demands may differ from lab-based experimental studies. This study aimed to examine the behavioural and physiological signals associated with visual distraction in real-world driving through a world-first application of a real-time driver monitoring system. Using a continuous driver monitoring system, shift-workers (N = 20) were observed on their commutes to and from work. Classifying off-road glances into glances to the driver lap and centre console regions of the vehicle revealed differences in the propensity for drivers to look away from the forward roadway toward these regions, with drivers significantly more likely to look toward their lap when drowsy. These glances were subsequently clustered and analysed in the context of visual time-sharing sequences. Our findings carry impact not only within the subjects of drowsiness and distraction but are also broadly applicable in the context of naturalistic driving methodology where real-time assessment of driver state can facilitate the analysis of large naturalistic datasets.

Introduction

Driver distraction is a significant public health problem, attributable to 20% of vehicle crashes in Australia (McEvoy et al., 2006). Defined as a diversion of attention toward competing activities other than required for safe driving, driver distraction is associated with a large array of impairments to driving performance in addition to crashes and near-crashes, including increased variance in speed and steering wheel angle, increased reaction time to perceived hazards, and decreased time headway (Boer et al., 2005, Briggs et al., 2011, Chen and Chiuhsiang, 2011, Engström et al., 2005, Horberry et al., 2006). Alarmingly, previous research has shown that drivers will willingly engage in distracting behaviours in spite of prior knowledge of the risks involved (McEvoy et al., 2006, Young and Lenné, 2010). Similarly, drivers’ decisions as to when they choose to engage in these behaviours are not necessarily linked to periods in a trip when it may be safer to do so, but rather are motivated by a range of external antecedents related more to the distracting behaviour than considerations of driving (Horrey and Lesch, 2008, Horrey and Lesch, 2009). The role of driver distraction in contributing to increased driving errors and safety critical outcomes continues to be consistently reported (Precht et al., 2017, Seppelt et al., 2017).

There is a clear need to better understand driver distraction as it occurs in real-world driving (with real-world risks and incentives) so that we may be better equipped to address the urgent issue of preventable death and injury on our roads. In particular, previous research has identified an increased propensity for drivers to engage in distracting behaviours as they become more drowsy (Anderson and Horne, 2013), as may be the case with shift-workers. This paper presents the initial findings on distraction from a naturalistic driving study (NDS) focusing on nurses working rotating night and day shifts.

Over the past decade, advances in sensing technology have facilitated a broad adoption of the (NDS) method to explore driver behaviour as it occurs in the real world (Eenink et al., 2014, Klauer et al., 2006, Van Schage et al., 2011). Large-scale NDS have now been conducted in North America (Victor et al., 2015), Europe (Eenink et al., 2014), Asia (Glaser et al., 2017), and Australasia (Williamson et al., 2015), with the aims of understanding driver behaviour in real-world settings where the constraints of participant recall of events or potential issues of ecological validity are removed. NDS has generated datasets of a breadth and depth not previously obtainable, providing researchers and practitioners a means to answer questions on topics as diverse as advanced driver assistance systems (ADAS) (Montgomery et al., 2014), child occupant safety (Arbogast et al., 2016), and distraction (Kuo et al., 2016).

The relative ease with which large quantities of data can be generated with NDS designs has generated a secondary problem, namely the intractable nature of large-scale data annotation and analysis of driver behaviour. Two main approaches have been implemented in previous research: brute-force manual annotation (with various statistical sampling techniques to reduce the problem space) (e.g. Stutts et al., 2005) and, more recently, machine learning and pattern recognition approaches which aim to replicate aspects of manual annotation (e.g. Kuo et al., 2014).

The former has yielded incredibly rich analyses of driver behaviour, providing the current gold standard on not only the range of behaviours that can be analysed but also the only means in which the context of those behaviours may be captured. The latter has yielded highly efficient analyses of target behaviours in which entire datasets can be readily processed, though at the obvious cost of context. Perhaps as testament to the infancy of current machine learning application in NDS, most implementations have ironically required non-trivial efforts in manual annotation to generate initial training data, and few have seen applications outside of the projects in which they were developed. A potential solution to this dilemma (at least in the context of analysing behaviours associated with driver distraction and drowsiness) is the implementation of existing, validated machine learning systems to NDS.

A key point of difference in the current study is the inclusion of real-time driver monitoring as a data collection method. Real-time driver monitoring presents significant advantages in this context in terms of supporting the collecting of a wider range of driver features and with greater accuracy while also removing the need for extensive manual annotation. Real-time driver-state monitoring has been shown to not only be technically feasible but also effective in positively changing behaviour (Fitzharris et al., 2017). The use of a research and development version of our automotive-grade driver monitoring system within an NDS design not only alleviates some of the data reduction challenges associated with NDS data, but allows for the continuous measurement of various behavioural signals.

Examination of driver states including distraction and drowsiness benefits greatly from an analysis of features including driver head orientation and gaze direction (Anderson and Horne, 2013, Klauer et al., 2006, Radwin et al., 2017), with glances to interior locations such as the centre console and driver lap commonly preceding crash events (Victor et al., 2015). Assessment of such driver states in research studies has historically been a significant challenge primarily because the analysis of these states has largely been reliant on the post-hoc analysis of video data. While this approach has been fruitful, it necessarily constrains the type of measures that can be collected (via video extraction), while also involving hundreds of hours of manual video coding prior to any data analysis.

Due to previous methodological challenges, the behavioural and physiological antecedents to driver distraction and drowsiness have yet to be examined under naturalistic driving conditions where task demands may differ from controlled experimental studies and where drivers may freely engage in additional secondary behaviours. This analysis aimed to examine continuous ocular markers of visual distraction under naturalistic conditions.

Section snippets

Participants

The current study observed shift-workers (N = 20) during their commutes to and from work over alternating periods of day and night shifts. Participants were recruited from Austin Health ICU nursing staff based on having sufficient transit duration between work and home, and a pattern of shifts likely to promote high levels of sleepiness (including a baseline period of day shifts).

The mean age of the sample was 33.15 years (SD = 8.81), comprising 13 female participants (65%). The majority of the

Results

Excluding trips of less than 5 min duration (on the basis that short-duration trips primarily involved changing the location where the vehicle was parked, or otherwise involved very short distances), the final dataset comprised 320 trips totalling 167.7 h.

Discussion

Despite substantial growth in research interest globally in naturalistic driving studies, significant challenges remain around data analysis. This study presents the first application of a continuous automated and real-time analysis of naturalistic driver-facing video. Using an NDS design, visual distraction was examined in a sample of shift-workers. Off-road glances were observed to increase significantly with drowsiness. Classification of off-road glances into glances to the driver lap and

Acknowledgements

This research is supported by the CRC for Alertness, Safety and Productivity. The authors acknowledge the valuable contributions of Grace Vincent, Julia Stone, Saranea Ganesan, Kaitlyn Crocker, Niamh McDonald, Jessica Papaleo, Matthew McLaren, Aaron Johnson, Marie-Antoinette Spina, Emma Giliberto, Trisha D'Lima, and Bronwyn Stevens. We also acknowledge the assistance from the Intensive Care Unit and Emergency Department at Austin Health, including specifically Simon Judkins, Yvonne Ballueer,

References (28)

  • J. Stutts et al.

    Driver's exposure to distractions in their natural driving environment

    Accid. Anal. Prev.

    (2005)
  • K. Young et al.

    Driver engagement in distracting activities and the strategies used to minimise risk

    Saf. Sci.

    (2010)
  • C. Ahlstrom et al.

    A generalized method to extract visual time-sharing sequences from naturalistic data

    IEEE Trans. Intell. Transp. Syst.

    (2017)
  • E.R. Boer et al.

    Steering entropy revisited

    Paper Presented at the Third International Driving Symposium on Human Factors in Driver Assessment

    (2005)
  • Cited by (20)

    • Understanding the drowsy driving crash patterns from correspondence regression analysis

      2023, Journal of Safety Research
      Citation Excerpt :

      In the latest national survey conducted by the American Automobile Association (AAA) in 2019, 24% of the standard probability sample of 3,511 respondents across the United States who were drivers aged 16 and older reported extreme drowsiness while driving during the last 30 days (AAA, 2020). There are examples in which survey questionnaire studies have been directed to identify specific areas of drowsy/fatigued driving (driving behaviors and attitudes, Vanlaar, Simpson, Mayhew, & Robertson, 2008; sleep habits, Sunwoo et al., 2017; lifestyle, Gnardellis, Tzamalouka, Papadakaki, & Chliaoutakis, 2008) and specific demographics (teen drivers, Hutchens, Senserrick, Jamieson, Romer, & Winston, 2008; truck drivers, Leechawengwongs, Leechawengwongs, Sukying, & Udomsubpayakul, 2006; shift-workers, Kuo et al., 2019; drivers with sleep apnea syndrome, Komada, Shiomi, Mishima, & Inoue, 2010; drivers in a particular state, McCartt, Ribner, Pack, & Hammer, 1996, etc.). An earlier literature review study compiled various issues of driver fatigue associated with traffic crashes from research in 2001 and prior years (RoSPA, 2001).

    • The EEG correlates and dangerous behavioral consequences of drowsy driving after a single night of mild sleep deprivation<sup>✰</sup>

      2022, Physiology and Behavior
      Citation Excerpt :

      However, the consequences of mild (e.g., a couple of hours) and acute (e.g., a single night) sleep loss, which is generally thought to be innocuous, are much less understood, despite this type of sleep loss being far more commonplace, widespread, and ecologically valid. However, recent research suggests that mild and acute sleep loss results in performance impairments [3,21,33], as well as negatively affecting emotional processing [53], visual attention [51], and behavioral preparedness and responding [33,72]. Thus, the gap in knowledge about mild, acute sleep loss is problematic, considering the direct implications for performance and safety in workplace settings (e.g., any job that requires sustained vigilance for monitoring monotonous tasks), academic settings (e.g., studying), and when driving (e.g., long-haul driving, commuting to/from work, etc.).

    • A data–information–knowledge cycle for modeling driving behavior

      2022, Transportation Research Part F: Traffic Psychology and Behaviour
    • EEG and behavioural correlates of mild sleep deprivation and vigilance

      2021, Clinical Neurophysiology
      Citation Excerpt :

      This is surprising, given that this is a commonplace form of sleep loss; one that many around the world reluctantly have to endure following the clock change when we “spring forward” from standard time to daylight savings time (Smith, 2016). This may, in part, be due to the fact that mild, acute sleep loss is generally thought to be innocuous, despite research suggesting that small amounts of sleep loss can negatively affect performance (Cote et al., 2009; Anderson and Horne, 2013), emotional processing (Lustig et al., 2018), as well as visual attention (Kuo et al., 2019), behavioral preparedness and responding (Stojanoski et al., 2019). Furthermore, compared to more extreme forms of sleep loss (i.e., sleep deprivation), mild and acute sleep restriction has much higher ecological validity, and has direct implications for any situation that requires continuous attention during a monotonous task (e.g., long highway drives, etc.).

    View all citing articles on Scopus
    View full text