The relative importance of real-time in-cab and external feedback in managing fatigue in real-world commercial transport operations

ABSTRACT

Objective: Real-time driver monitoring systems represent a solution to address key behavioral risks as they occur, particularly distraction and fatigue. The efficacy of these systems in real-world settings is largely unknown. This article has three objectives: (1) to document the incidence and duration of fatigue in real-world commercial truck-driving operations, (2) to determine the reduction, if any, in the incidence of fatigue episodes associated with providing feedback, and (3) to tease apart the relative contribution of in-cab warnings from 24/7 monitoring and feedback to employers.

Methods: Data collected from a commercially available in-vehicle camera-based driver monitoring system installed in a commercial truck fleet operating in Australia were analyzed. The real-time driver monitoring system makes continuous assessments of driver drowsiness based on eyelid position and other factors. Data were collected in a baseline period where no feedback was provided to drivers. Real-time feedback to drivers then occurred via in-cab auditory and haptic warnings, which were further enhanced by direct feedback by company management when fatigue events were detected by external 24/7 monitors. Fatigue incidence rates and their timing of occurrence across the three time periods were compared.

Results: Relative to no feedback being provided to drivers when fatigue events were detected, in-cab warnings resulted in a 66% reduction in fatigue events, with a 95% reduction achieved by the real-time provision of direct feedback in addition to in-cab warnings (p < 0.01). With feedback, fatigue events were shorter in duration a d occurred later in the trip, and fewer drivers had more than one verified fatigue event per trip.

Conclusions: That the provision of feedback to the company on driver fatigue events in real time provides greater benefit than feedback to the driver alone has implications for companies seeking to mitigate risks associated with fatigue. Having fewer fatigue events is likely a reflection of the device itself and the accompanying safety culture of the company in terms of how the information is used. Data were analysed on a per-truck trip basis, and the findings are indicative of fatigue events in a large-scale commercial transport fleet. Future research ought to account for individual driver performance, which was not possible with the available data in this retrospective analysis. Evidence that real-time driver monitoring feedback is effective in reducing fatigue events is invaluable in the development of fleet safety policies, and of future national policy and vehicle safety regulations. Implications for automotive driver monitoring are discussed.

KEYWORDS:

• Advanced driver assistance system (ADAS)
• driver monitoring system (DMS)
• auditory
• haptic warning
• fatigue

Introduction

Heavy vehicle safety is a subject of considerable interest due to the industry's importance to economic activity and also due to historically high rates of crash involvement (Bureau of Infrastructure, Transport and Regional Economics [BITRE] 2014; BITRE 2016a; BITRE 2016b; Button and Reggiani 2011; European Commission 2016a; Federal Motor Carrier Safety Administration [FMCSA] 2016a; Ministry of Transport 2014; Savy and Burnham 2013). With driver fatigue a contributing factor in up to 20% of crashes (Beanland et al. 2013; Connor et al. 2002; Horne and Reyner 1995), and crash risk being known to increase with reduced amounts of sleep and subjective judgements of fatigue (Bunn et al. 2005; Connor et al. 2002; Stutts et al. 2003), the mitigation of driver fatigue is a priority.

With long operating hours, fatigue is a significant issue in the commercial vehicle sector. This is exemplified by research reporting that nearly half of 593 long-haul truck drivers surveyed stated that they had fallen asleep while driving, with 25% having done so in the previous 12 months (McCartt et al. 2000). The link of fatigue to crashes is demonstrated by the finding that a truck driver falling asleep or being fatigued accounted for 46% of all driver impairment-related factors implicated in truck-involved fatality crashes (FMCSA 2016b).

Given these risks, and that drivers are known to be poor at predicting the onset of sleep (Howard et al. 2014) and/or face other pressures (including financial) to continue driving when arguably not safe to do so (Fletcher et al. 2005; Thompson and Stevenson 2014), real-time driver monitoring has strong appeal. The term “driver monitoring” is broad and includes systems that incorporate a forward-facing camera (Hickman and Hanowski 2011), those based on vehicle-based measures, or driver-facing camera-based approaches designed to objectively assess driver sleepiness. Notwithstanding differences in the aspect of driver performance measured, driver monitoring is increasingly important due to occupational health and safety and other regulatory regimes, and due to voluntary fleet safety programs that are now widely implemented following recognition that the vehicle is a workplace (Bidasca and Townsend 2016; FMCSA 2016a; Health and Safety Executive 2016; Newton et al. 2013; Occupational Safety and Health Administration [OSHA] 2016).

The objective measurement of driver fatigue has been the subject of considerable research and development (Dawson et al. 2014; Lenné and Jacobs 2016). Eye-closure measures were found to have promise nearly two decades ago (Dinges et al. 1998), with percent of eye closure (PERCLOS) being successfully used as the basis of fatigue measurement in the real-world short-haul context (Hanowski et al. 2003). Since that time, a number of propriety systems have emerged as after-market in-cab real-time systems, and the focus is now shifting to the factory-fitted installation of driver monitoring systems (DMS) in new vehicles as an advanced driver assistance system (ADAS). These DMS act to warn the driver when a safety-critical event, such as a fatigue episode, has been detected, and depending on the device settings, parties external to the vehicle can also receive warning information.

There are, however, few on-road studies examining the effectiveness of fatigue auditory warnings (Dawson et al. 2014), with some evidence that haptic warnings can be as effective as auditory feedback (Azmi 2012). To date, the effectiveness of multisensory real-time warning systems on mitigating fatigue episodes has not been explored in the commercial vehicle setting. Notwithstanding recent research (Hickman and Hanowski 2011; Law and Jones 2016; Lenné and Fitzharris 2016), there are few available data on the incidence of fatigue in real-world commercial vehicle operations and the efficacy of driver monitoring systems (DMS) in real-world conditions (Lenné and Fitzharris 2016).

The aim of this study was therefore to (1) document the incidence and duration of fatigue events in a commercial vehicle context; (2) determine the difference, if any, in the number of fatigue episodes detected without and then with real-time in-cab feedback, which was subsequently enhanced with direct feedback to the fleet operator; and (3) tease apart the relative contribution of these two types of feedback on the number, timing, and duration of fatigue events.

Methods

Study design

The study is a retrospective analysis of real-time fatigue event data collected via a DMS fitted to commercial vehicles in one company in the period 2011–2015. Three periods were defined: (1) “baseline,” where the DMS collected data but did not alert the driver or company (i.e., silent period); (2) “alarm-only,” referred to as intervention mode 1 (IM-1); and (3) “alarm plus feedback” (i.e., to driver and company), referred to as intervention mode 2 (IM-1) (Table 1).

Table 1. Study design and time periods under observation for the freight transport commercial vehicles.

Baseline (silent)	Intervention mode 1 (IM-1): driver feedback	Intervention mode (IM-2): driver and company feedback
July 21, 2011, to September 8, 2011	September 9, 2011, to January 31, 2014	February 1, 2014, to December 7, 2016

Setting and participants

Deidentified data from drivers employed by a large-scale commercial vehicle company that conducts short-, medium-, and long-haul freight transport in Australia were used.

Data capture

Description of the driver monitoring technology

The DMS fitted within the cabin of vehicles is comprised of a camera using a VGA (video graphics array) resolution 60-Hz global shutter image sensor and a pair of pulsed 850-nm infrared lights to obtain images of the face and eyes of drivers. The system is monocular and the system has operational capability in nearly all daytime and nighttime conditions. The driver-facing camera is located on the vehicle dashboard either directly in front of the driver, or to within a 30-degree angle to the left or right of the driver's forward view direction. The camera is positioned below the face, pitched up between 5 and 20 degrees.

Infrared illumination mitigates the effects of strong sunlight on the face and can penetrate most sunglasses, although some glass lenses and frames can interfere with eyelid measurements under particular conditions, while some sunglasses block the 850-nm wavelength. This interference can result in a lower degree of confidence in the event attributed by the system, and in instances where the eyelids cannot be tracked, these frames are filtered by the system.

Real-time event classification

To detect the fatigue events, and^{Table 2} not be confused by other events that emulate eyelid closure, the DMS uses machine-learned classifiers trained on a database of more than 500 verified drowsy drivers. These classifiers assess multiple aspects of the driver's behavior, including eyelid opening, eyelid shape, the pupil, and head pose. Events are stored locally and uploaded via wireless link to a central operations center for processing. The operations center is operated by the DMS vendor. All detected fatigue events are processed in real time by a trained expert and verified using defined heuristics. The expert observes a 3-s video where the driver's eyelids, head position, and historical information about the driver's speed, acceleration forces, and physical location are assessed. Events are classified as fatigue mitigation, drowsiness, or microsleeps, with a threshold of 1.5 s of eyes closed being required for classification as one of these three events.

Event notification

The DMS silent mode captures fatigue events that are verified but no feedback is provided. In the feedback mode (IM-1), when a fatigue event is detected the vehicle operator is given an auditory warning (“fatigue detected”) or an auditory tone (depending on vehicle), as well as a haptic warning consisting of vibration pulses issued at 1 Hz for 4 s delivered through the base of the driver's seat. Alerts were designed and tested to be easily perceptible in long-haul truck cabins. Warnings include all detected fatigue events, including those that are shortly after coded as “false positives” by the DMS vendor 24/7 operations center. This is designed to capture scenarios where the operator's head drops toward the lap, which may be a product of distraction or fatigue. This represents a conservative approach to event recognition and notification, thereby optimizing the safety of the vehicle operator.

In IM-2, feedback is provided to the driver (as described) and also to the company directly. Upon verifying a fatigue event, the transport company's fatigue management plan (FMP) is initiated. This involves the company's central dispatch center being notified of the fatigue event(s). By talking with their driver directly, the company may recommend the driver take a break or “swap out” a driver. Actions the company took were not available to the research team.

Data analysis

Deidentified data were extracted from the DMS vendors' global^{Table 3, Table 4} database. Data included date and time, location, vehicle speed, event type, truck identifier, the number of operational, moving, and stationary hours, and start–end dates and time of each trip. Driver details were not available to the DMS vendor or the research team.

Fatigue event rates and exposure measures (i.e., operational time, distance traveled) were calculated across the three time periods. Analysis of variance (ANOVA) tests via a regression model with robust standard errors were used to compare the mean event duration, distance traveled, and vehicle speed during each fatigue event across each group; post hoc contrasts were performed using a t-test with corrections for multiple comparisons (Mitchell 2015). Quantile regression was used to examine the difference, if any, in median event duration and distance traveled (Mitchell 2015). Analysis accounted for repeated trips undertaken by individual trucks.

To assess the differences, if any, in fatigue events across the period, a random-effects negative binomial regression (with beta effect) was used (Hilbe 2007; Vittinghoff et al. 2005). A standard Poisson regression model was not appropriate due to the overdispersion evident in the data, and the random effects model specifically accounts for the repeated observations from each truck, as trips in these trucks are not independent (due to route considerations, drivers, payload). The exposure variables used were the distance traveled and hours spent moving.

Direct comparison of driver fatigue events in IM-1 and IM-2 relative to the baseline period was performed with incidence rate ratios (IRRs) being presented. Kaplan–Meier nonparametric survival curves using the log rank test to compare the equality of fatigue event occurrence across the three periods was conducted (Cleeves et al. 2016; Hosmer and Lemeshow 1999).

All analyses were conducted in STATA/MP 12.1 (StataCorp 2011). Statistical significance was set at p ≤ 0.05. Ethics approval was obtained to analyze the deidentified data from the Monash University Human Research Ethics Committee.

Results

Descriptive analysis

A fatigue event was detected in 3.74% of trips in the baseline period, compared to 1.3% of trips in IM-1 and 0.18% of trips in IM-2 (Table 2). The majority of trips involved drivers experiencing one to three events, with fewer trips in IM-1 and IM-2 having four or more events compared to baseline (p ≤ 0.001). The reduction in the fatigue event rate is evident (Table 2), with 56 fatigue events per 1000 hours driven in the baseline period, compared to 17.87 in the IM-1 period and 1.3 in the IM-2 period. On a distance traveled basis, there were 1.47 fatigue events per 1000 km traveled (baseline), and this was lower in IM-1 and IM-2.

Table 2. Number and adjusted rate of fatigue events in the baseline and intervention periods.

Fatigue events detected	Baseline (silent) period	IM-1: Driver alarms only	IM-2: Driver alarms plus employer feedback
Number of trips	4539	161,541	530,075
Number of trips with a verified fatigue event	170	2108	971
Percent of trips with a fatigue event	3.74%	1.30%	0.18%
Total number of verified fatigue events	333	3948	1142
Number of fatigue events, by trip
1 event	65.9% (112)	50% (1055)	91.8% (891)
2–3 events	22.9% (39)	41.5% (874)	6.9% (67)
4–9 events	9.4% (16)	7.6% (161)	1% (10)
10 plus	1.8% (3)	0.9% (18)	0.3% (3)
Fatigue event rate
Per 1000 hours mobile	56.00	17.87	1.310
Per 1000 km traveled	1.47	0.30	0.020
Per 1000 miles traveled	2.37	0.49	0.030
Exposure parameters—distance and time moving (total)
Exposure—distance (km)	226,224.09	13,086,883.40	60,580,523.92
Exposure—distance (miles)	140,569.09	8,131,809.83	37,642,980.73
Exposure—time (h)	5946.57	220,911.13	870,376.06
Number of vehicles	16	142	342
Days in observation period	49	814	1040

Fatigue event duration and distance

Overall, the mean fatigue event duration was 2.33 s, the mean vehicle speed during these fatigue events was 61.4 km/h, and the mean distance covered was 29.74 m (Table 3). The mean fatigue event duration was shorter in IM-1 (M: 2.30, SD = 1.14) than the baseline period (M: 2.51, SD: 1.14) (p = 0.002), and marginally so compared to IM-2 (M: 2.35, SD: 2.49) (p = 0.5). IM-2 event duration was shorter than in baseline (p = 0.09) (−0.16). There were also differences in median fatigue event duration across each period (p ≤ 0.001). Despite shorter fatigue event times in IM-1 and IM-2, vehicles traveled a greater distance due to higher travel speeds (p = 0.05).

Table 3. Details of fatigue events in the three time periods.

	Baseline (silent) period	IM-1: Driver alarms only	IM-2: Driver alarms plus employer feedback
Duration of fatigue event
Mean (s) (SD)^a	2.51 (1.14)	2.30 (1.38)	2.35 (2.49)
Median (s)^a^,^b^,^c	2.03	1.93	1.85
Minimum–maximum (s)	1.62–11.42	1.57–48.47	1.37–59.30
Total time (min)	13.9	151.6	44.7
Percentiles (s)
75th	2.82	2.20	2.07
90th	3.94	3.43	2.83
95th	4.50	4.53	4.75
Vehicle speed during fatigue event
Mean (km/h)^a^,^b^,^c	52.09 (17.17)	60.76 (25.16)	66.34 (26.76)
Median (km/h)^b^,^c	56.00	61.00	75.00
Minimum–maximum (km/h)	15–82	11–106	15–102
Percentiles (km/h)	74.00	91.00	93.00
75th	78.30	94.00	97.00
90th	52.09 (17.17)	60.76 (25.16)	66.34 (26.76)
95th	56.00	61.00	75.00
Distance traveled during fatigue event
Mean (m)^b^,^c	36.65 (21.69)	38.69 (29.09)	44.25 (62.53)
Median (m)^b^,^c	32.00	35.00	39.00
Minimum–maximum (m)	8–180	7–891	7–1478
Percentiles (m)
75th	43.00	47.00	46.00
90th	65.20	61.00	57.00

^a IM-1 vs. baseline, p ≤ 0.05.

^b IM-2 vs. baseline, p ≤ 0.05.

^c IM-2 vs. IM-1, p ≤ 0.05.

Statistical modeling of fatigue event rates

Analysis of fatigue incidence rates

Adjusting for distance traveled and hours driven, the incidence of fatigue events was lower in IM-1 and IM-2 compared to baseline (Table 4). The incidence of fatigue events occurring in IM-1 was one-third of that in the baseline period, indicating a 66.2% reduction (IRR: 0.338, 95% confidence interval [CI]: 0.283–0.403, p ≤ 0.001). The implementation of 24/7 monitoring-center company feedback in IM-2 in addition to alarms saw a further reduction in fatigue events, with these occurring at a rate of 5.6% of the baseline period; this equates to a 94.4% reduction relative to the baseline period (IRR: 0.056, 95% CI: 0.046–0.068, p ≤ 0.001). In absolute terms, this represents an additional 28.2% reduction in the fatigue event incident rate, or 83.4% relative reduction (IRR: 0.166, 95% CI: 0.151–0.182, p ≤ 0.001). The distance adjusted IRRs indicate that the fatigue incidence rate was 68.8% lower in the IM-1 period and 94.9% lower in the IM-2 period compared to the baseline period, and the further reduction (26%_absolute or 83.7%_relative) between the IM-1 and IM-2 period was statistically significant (p ≤ 0.001).

Table 4. Fatigue event rate in the intervention period relative to the baseline period (Poisson regression analysis).

Fatigue event analysis	Referent (1.0)	IRR	95th% CI LCL–UCL	p
Rate per mobile time (h)
IM-1: Driver alarms only	Baseline	0.338	0.283–0.403	<0.001
IM-2: Driver alarms plus employer feedback	Baseline	0.166	0.151–0.182	<0.001
IM-2: Driver alarms plus employer feedback	IM-1: Driver alarms only	0.056	0.046–0.068	<0.001
Rate per distance traveled (km)
IM-1: Driver alarms only	Baseline	0.312	0.262–0.372	<0.001
IM-2: Driver alarms plus employer feedback	Baseline	0.163	0.148–0.179	<0.001
IM-2: Driver alarms plus employer feedback	IM-1: Driver alarms only	0.051	0.042–0.062	<0.001

Note. LCL, lower confidence limit; UCL, upper confidence limit.

The analysis just presented used all available data in the fleet for each of the three periods. As a secondary analysis, the data were limited to trips undertaken by the 16 vehicles represented in each of the three periods. The findings reflect the fleet analysis, which after adjustment for hours driven translates to a 68.8% reduction from baseline to IM-1 (IRR: 0.312, 95% CI: 0.263–0.383), a 97% reduction from baseline to IM-2 (IRR: 0.030, 95% CI: 0.019–0.477), and a 90.3% reduction from IM-1 to IM-2 (IRR: 0.0963, 95% CI: 0.063–0.147). Similarly, on a per distance rate, these reductions were (respectively), 65%, 95.8%, and 87.8% (IM-1 vs. baseline, IRR: 0.312, 95% CI: 0.263–0.383; IM-2 vs. baseline, IRR: 0.030, 95% CI: 0.019–0.477; IM-2 vs. IM-1, IRR: 0.096, 95% CI: 0.063–0.147).

Modeling of time to fatigue event

Drivers experienced a fatigue event in the baseline period significantly sooner, on average (mean: 510 min, 95% CI: 484.7 535.3) than did drivers in IM-1 (mean: 623.8 min, 95% CI: 612.9–634.7), and IM-2 (mean: 713.6 min, 95% CI: 712.8–714.6). The probability of drivers not having (Figure 1) and having a fatigue event (Figure 2) by time across the three periods differed (p ≤ 0.001). As an example, at 180 min, the probability of a driver having had a fatigue event in the baseline period was 8.5%, 2.8% in IM-1, and 0.3% in IM-2. Conversely, after 180 min, 91.5%, 97.1%, and 99.6% of drivers in the baseline, IM-1, and IM-2 periods, respectively, had not yet experienced a fatigue event.

Figure 1. Probability of remaining fatigue event free.

Figure 2. Probability of fatigue event.

Discussion

This study set out to examine the influence of a DMS using different feedback modes on the incidence of fatigue events, and to document the characteristics of these events in commercial transport operations.

In the absence of feedback, a verified fatigue event occurred in 3.74% of trips. Fatigue events occurred at a rate of 56 per 1000 moving hours or 2.37 per 1000 miles traveled. The mean duration of events was 2.51 s, which at a mean speed of 52 km/h meant vehicles traveled 36.65 m. This represents a significant road safety concern.

The provision of a fatigue alert to a driver reduced the incidence of fatigue events by 66.2%. Incorporating an additional layer of feedback via the driver's employer resulted in additional benefits, lifting this to a 94.4% reduction from the baseline rate. This finding is similar to earlier research that showed a similar 93.2% reduction in fatigue events using the same driver monitoring technology in South Africa, where the baseline fatigue incidence rate of 2.39 per 1000 miles traveled was remarkably similar to that of the present study (Lenné and Fitzharris 2016). The present study extends this earlier research by teasing apart the relative influence of IM-1 and IM-2, where it was found that the provision of direct feedback to the driver's employer resulted in an additional 28.2% benefit in absolute terms, or 84% in relative terms to the event rate in the driver alarm-only period.

The impact of the DMS extends into the duration and timing of events. In the feedback periods, fatigue events were of shorter duration, the first fatigue event occurred later in a trip, and fewer drivers had more than one fatigue event. These findings in particular could reflect the improved health of the driver through company programs or the company taking action to “swap out” the driver with another, although these aspects need to be explored in further research where detailed information about drivers is known (see Limitations section).

While near crashes, crashes, and other safety critical events such as lane position are not captured, the association between truck driver fatigue and these events is well known (FMSCA 2006; Klauer et al. 2005; Knipling and Wan 1994), as is the increased crash risk associated with eyes off road due to distraction (Hanowski et al. 2005; Klauer et al. 2014; Liang et al. 2012; Simons-Morton et al. 2014). Based on the findings reported here, systems designed to inform the driver directly, as well as operators of commercial fleets, with a view to modifying fatigue have significant merit.

A key question therefore is: What is the optimum ADAS configuration to address fatigue? This study has shown that 66% of the observed 94% reduction in fatigue events is achieved through the device human–machine interface (HMI) alone. Additional mechanisms—which in this case involved direct feedback to the company—are needed to achieve near elimination of fatigue events. Further research needs to address how these results can be achieved in the original equipment manufacturer (OEM) passenger fleets where “company monitoring” per se may not exist. The key challenge with nonfleet drivers is creating an incentive, or motivation, for drivers to take a break from driving when a fatigue event is detected.

Following the preceding point, the installation of the DMS into the commercial truck fleet was accompanied by a safety management program, including a fatigue management plan. Moreover, prior to the installation of the DMS, the transport company had very robust, sophisticated, and long-standing driver and fleet management plans, including global positioning system (GPS) monitoring of vehicles, speed management technologies, driver education, and health and safety plans. The DMS gave the company an opportunity to extend these existing strategies by using real-time, objective feedback on driver performance. The company was then able to counsel individual drivers on known risks and to optimize crews and rostering. We also note that the company introduced specific information on fatigue and health management in 2011, which coincided with the introduction of baseline DMS period; hence the separate effect of this on fatigue event rates is unknown. We note though that the reductions seen following the implementation of feedback were over and above any additions to fleet policy and practice from the baseline period, but may include the effects of any other (unspecified) changes in driver management that occurred during the feedback periods.

That the benefit achieved was higher with this additional level of feedback is unsurprising and is consistent with the broader safety literature. It is well known that motivations for, and the conduct of, safe behavior are largely influenced by the prevailing safety climate of the worker's organization (Zohar 2010). The most effective safety cultures are ones where there is a strong management commitment (Newnam et al. 2008; Swedler et al. 2015; Zohar et al. 2014) and where the policies complement, not compete against, the demands of the working role (Zohar 2010), even for workers operating remotely, including truck drivers (Sullman et al. 2016). Whether the findings reported here can be replicated by other commercial vehicle operators with a less well developed safety culture remains to be tested.

Limitations

The precise nature of the company's internal fatigue management plan in managing the drivers' health more broadly and in mitigating fatigue risks, while known to be strong, is not well understood. Clearly, this can play a role in the extent of reduction achieved in the number of fatigue events. It is reasonable to assert that for similar results to be achieved in other commercial vehicle settings, thought must be given to the way fatigue event information is used internally to manage fatigue risks, whether this be scheduling, pay methods, and/or driver health more broadly. Future research, perhaps using a prospective experimental design, ought to examine elements of fatigue management plans and the nature of device warnings themselves, to ensure they deliver optimal benefit to the driver and the company.

From an analytical perspective, the data have considerable strengths in that they were collected over a 5-year period (2011–2016) across 342 trucks for a combined distance traveled of 79.8 million km (45.9 million miles) across 1.097 million hours. Fatigue events were also verified by a trained observer in all three periods. We note, however, the difference in the number of trucks fitted with the DMS in each period as the progressive rollout across the company fleet occurred; this can explain differences in vehicle speeds across the periods, as there is greater diversity of truck type and operating routes. Crucially, the analysis using only the 16 vehicles involved in the baseline period was consistent with—even marginally better than—the fleet-wide analysis comparisons made when all vehicles in the IM-1 and IM-2 period were used. This provides confidence that the use of the baseline is appropriate and the fleet-wide effects are consistent with the smaller subgroup of trucks.

We also note the matter of potential rater bias in classifying fatigue events, but note that DMS monitoring staff are trained in event classification, which is governed by specified heuristics, and are subject to significant DMS vendor oversight, and all footage is accessible by companies with the DMS fitted. We further note commercial risks associated with systematic underreporting of fatigue events, particularly legal liability in the event of a critical safety events. We therefore consider any bias across intervention periods to be minimal.

As stated earlier, information about the truck driver in each trip was unknown. This is a consequence of the retrospective nature of the study using data from an operational commercial transport fleet, rather than being initiated as a dedicated research project. Importantly, the effect of the truck was statistically controlled for, which accounts for driver-, load-, and route-based effects; however these trucks may have been driven by different drivers across trips. From the perspective of a pure implementation trial in the real world, this limitation can also be viewed as a strength as it represents a “natural experiment.” Future work will seek to address this gap pertaining to lack of information about the individual drivers, at which time the effect of any breaks taken within each trip can be examined, as can the timing of subsequent fatigue events. This has particular relevance for work and rest hour requirements, which in Australia are mandated by the National Heavy Vehicle Regulator (NHVR 2017) and similar bodies such as the U.S. FMCSA.

The DMS used was inward focused and thus no information on roadway conditions or other changes in driver behavior—such as a lane departure events (Filtness et al. 2017)—that are known in the research literature to be associated with near crashes or crashes and that may be influenced by fatigue were available (Lenné and Jacobs, 2016). The integration of an outward-facing camera with the DMS implemented in the present study represents the next step in fatigue and distraction monitoring, and is the subject of a new Australian government initiative (Australian Government 2016). Linking these systems will provide the ability for greater integration of vehicle control systems to ensure road user safety. This would represent a significant advance in ADAS and has implications for the future semi-autonomous and autonomous vehicle, where resumption of control may need to occur in some scenarios.

In sum, this study has reported novel and important findings on the relative contribution of real-time in-cab alerts and additional company-level feedback in achieving reductions in fatigue events in commercial vehicle fleets. Given the global interest in driver state sensing, as seen in the NHTSA Drowsy Driving Research and Program Plan (NHTSA 2016), and the recent European Union (EU) report on regulating vehicle safety systems (European Commission 2016b), the findings of this study will be of interest to vehicle technology regulators and to operators of vehicle fleets. The findings support the installation of DMS and concurrent fatigue management strategies enabled by the information provided by the DMS itself across different vehicle platforms as a preventative road safety action. While highly effective in the commercial transport context due to command-and-control mechanisms, the efficacy of DMS within nonfleet vehicles needs to be established, particularly in light of more limited external feedback channels for the driver than is the case for fleet vehicles.

Acknowledgments

The authors thank Francis Cremen, Seeing Machines, for data extraction. Professor Mike Lenné joined Seeing Machines Ltd in the role of Chief Scientific Officer, Human Factors, in 2014. Seeing Machines funded this research via a research contract to Monash University Accident Research Centre. The study reports data collected from the Seeing Machines driver monitoring system as used in operational fleet settings. MF and ML conceived and developed the idea for the study, with MF leading the drafting of the article, and cleaning and conducting all statistical analysis on the data. AS, SL, and ML provided comments on the article and assisted with drafting the Introduction and Discussion sections in particular. All authors provided comment on the final version of the article. MF and ML are guarantors for the article.

Funding

This work was funded by Seeing Machines as a research contract with the Monash University Accident Research Centre.