Measuring visual distraction potential

One way to define driver distraction is to divide it into visual, cognitive, and manual distraction (Foley et al., 2013). However, visual inattention is the most hazardous form of inattention in traffic (e.g., Klauer et al., 2006). Visual inattention is also a form of inattention that can be operationalized and estimated with the eye-tracking technique. Hence, this dissertation is particularly focused on visual inattention. Therefore, only visual inattention, and further, visual distraction are discussed here.

Several lines of evidence suggest that there is an association between drivers’ off-road glances and accidents and near-accidents (e.g., Bálint et al., 2020;

Dingus et al., 2016). As a result, various authorities have published guidelines on how to assess drivers’ visual inattention caused by secondary in-car tasks (e.g., interacting with an application) for industrial testing purposes. For instance, the Alliance of Automobile Manufacturers (AAM, 2006), Japan Automobile Manufacturers (JAMA, 2004), and European Commission (EsOP, 2008) have provided glance durations and glance numbers that should not be exceeded while conducting secondary in-car tasks. Unfortunately, no guidelines were provided on how these glance durations and glance numbers should be exactly measured. The first one to do so, was the United States National Highway Traffic Safety Administration (NHTSA, 2013) which published guidelines in 2013 for measuring and assessing how distractive different in-car tasks are.

In these guidelines, distraction potential testing is conducted either using a visual occlusion method or in a driving simulator. In NHTSA’s (2013) visual occlusion method, participants complete in-car tasks in a series of 1.5-second glances in a stationary vehicle. To pass the test, the cumulative time of the glances should not exceed 12 seconds. The NHTSA’s (2103) occlusion method’s capability to measure in-car task’s visual distraction potential can be questioned since the method does not involve driving and, hence, is not described here in detail. The NHTSA’s (2013) visual occlusion method has been, however, used in previous studies to measure secondary tasks’ visual demand, see for instance Burnett et al.

(2011). Another testing method presented in the NHTSA guidelines (2013) utilizes a driving simulator. In the method, the testing of distraction potential is conducted in a driving simulator while driving on a straight four-lane road at 50 miles per hour, and following a lead vehicle, and performing secondary in-car tasks. According to the guidelines, testing should be performed with 24 randomly selected participants who are further divided into four groups of six, according to their age (18–24 years, 25–39 years, 40–54 years old, and older than 55 years). Three metrics are used to assess the tested in-car task: total glance time, mean glance duration, and the percentage of over 2-second glances. These metrics mean that (for 21 out of 24 participants):

1) the total glance time should not exceed 12 seconds when performing a task, 2) the mean glance time should be less than or equal to 2 seconds when

performing a task, and

26

3) the percentage of over 2-second glances should not exceed 15 % of the total number of in-car glances.

However, NHTSA’s (2013) distraction testing method has received criticism, for instance, for not taking into account the test participants’ individual glancing behaviors. This is significant since preceding research indicates that drivers have individual mean in-car glance durations that seem to be relatively constant across tasks (e.g., Broström et al., 2013, 2016; Donmez et al., 2010; Yang et al., 2021).

Based on the criticism, Broström et al. (2016) and Ljung Aust et al. (2015) tested how individual glancing behaviors affect the results of the distraction potential testing conducted following the NHTSA (2013) guidelines. They noticed that the results of the distraction potential testing were dependent on the driver sample.

This means that the same in-car task with a different driver sample could have had a different outcome in the distraction potential testing. This indicates that if the information on individual glancing behavior is neglected, the results of the distraction potential testing are greatly dependent on the driver sample – not necessarily on how distractive the task at hand is. Hence, the test result can even be false.

In addition, since the driving scenario in the NHTSA (2013) testing method is comprised of a straight four-lane road, another critical observation regarding the method is that it does not account for the visual demands of the driving scenario (e.g., Kujala et al., 2014). That is, the driving scenario in the NHTSA (2013) testing does not correspond sufficiently with the visual demands of real-life driving scenarios (e.g., Kujala et al., 2014; Large et al., 2015), for example, testing a navigation application is rather pointless on a straight road. This is significant since previous research (e.g., Risteska et al., 2021; Tivesten & Dozza, 2014; Tsimhoni & Green, 2001; Wierwille, 1993) has suggested that the visual demands of the driving scenario affect in-car glance durations. For instance, in the study by Large et al. (2015), off-road glances were longer in the NHTSA (2013) scenario than in the more complex scenario. Additionally, the visual demands of driving with different driving simulators, even in a similar scenario, may vary and this can also affect the results of distraction potential testing (e.g., Kujala et al., 2014). These findings suggest that, when conducting distraction potential testing, there is a need for information on how visually demanding certain situations are in the driving scenario. This information would provide a baseline for the accepted glancing behavior in that certain situation, and further, give instruments to assess if the driver is being attentive or not.

In order to respond to the neglects of individual glancing behaviors and visual demands of driving scenarios, Kujala and Mäkelä (2015) introduced a new distraction potential testing method. The new testing method is founded on the occlusion technique, which Senders et al. (1967) initially introduced. Note that this is different from NHTSA’s (2013) occlusion technique which does not include driving. In the original technique from Senders et al. (1967), the driver’s vision is occluded (i.e., driving blind) and when needed, the driver can see the forward road scene for 500 milliseconds at a time. During the occluded period, the time driven without visual information, is measured. Milgram (1987, p. 453) has

27

propounded that, with the occlusion technique, it is possible to "estimate the attentional demand, or information processing workload, imposed on a human monitor/controller of a (complex) system by recording the circumstances and rate at which he/she samples information from the system." Contrary to the original method, in the new testing method by Kujala and Mäkelä (2015), the distance driven during the occluded period is measured, not time. This distance is later called the occlusion distance. Occlusion distance stands for the driver’s preferred distance in meters that is driven during a period when there is no visual information available. Occlusion distance can also be seen as a measure of the driver’s situational spare visual capacity, following Ahlström et al.'s (2021) idea that drivers have a certain amount of time at their disposal to look away from the road scene ahead of them.

In the new distraction potential testing method by Kujala and Mäkelä (2015), the assessment of whether a tested task is too distractive is founded on 97 drivers’

occlusion distances (presented in Kujala, Mäkelä, et al., 2016) driven in simulated highways and suburban roads. These occlusion distances were measured and later mapped to the test routes. Each 1x1-meter route point in the map (see Kujala and Mäkelä, 2015) contains information on the median and 85^th percentile occlusion distances driven in that particular route point in the original experiment. When the same routes (as in the occlusion distance map) are used later in a distraction potential testing with a new participant sample, this information can be used for categorizing in-car glances as being appropriate or inappropriate glances based on both the distance driven during an in-car glance and the route point where that in-car glance starts.

If the glance is categorized as an appropriate glance, the distance driven during an in-car glance and the visual demands of that route point have been low enough for conducting a secondary in-car task – or a driver has spare visual capacity for conducting an in-car task, as Ahlström et al. (2021) and Kircher and Ahlström (2017) suggest. Low visual demand basically means there are no junctions or sharp road curviness. However, if the in-car glance is categorized as an inappropriate in-car glance, the in-car glance length has exceeded the occlusion distance of the 85th percentile of the original experiment’s driver sample (N = 97) on that particular route point. That is, the majority of the original experiment’s drivers preferred not to drive in that route point longer without visual information. This means that the in-car glance has been inappropriately long in relation to the visual demands of that given driving situation. These inappropriately long in-car glances are later called red in-car glances. In other words, a red in-car glance indicates the driver’s visual distraction, and the driver should have been looking at the forward road scene instead of the secondary in-car task on that route point. The idea behind the occlusion distance map is that it can determine the maximum acceptable duration of an in-car glance for each driving situation. This also means that the map provides a baseline for acceptable glancing behavior, which has the same basic idea to first assess attentive driving as in hindsight bias-free minimum attention requirements by Kircher and Ahlström (2017). It should also be noted that the driver can self-pace the

28

acceptable off-road glance duration by speed adjustment, meaning that drivers can regulate the time they drive without visual information while still complying with the acceptable occlusion distance threshold. This is also in line with Kircher and Ahlström’s (2017) idea of minimum attentional requirements regarding different, self-regulated sampling strategies to fulfill the minimum requirements for attentive driving.

Other than defining each route point’s visual demand, these original occlusion distances of 97 drivers (Kujala, Mäkelä, et al., 2016) are used for validating the new driver sample for the distraction potential testing. With comparing the tested participant sample’s occlusion distance distribution to the original occlusion distance distribution of 97 drivers, it is ensured that the new sample matches the original sample and contains drivers with different glancing behaviors – from those drivers who prefer only short occlusion distances to those drivers who prefer longer occlusion distances. A more detailed description of the method can be found in Kujala and Mäkelä (2015).

As argued earlier, there is a need for a more robust visual distraction