A Review of Occlusion as a Tool to Assess Attentional Demand in Driving

Objective The aim of this review is to identify how visual occlusion contributes to our understanding of attentional demand and spare visual capacity in driving and the strengths and limitations of the method. Background The occlusion technique was developed by John W. Senders to evaluate the attentional demand of driving. Despite its utility, it has been used infrequently in driver attention/inattention research. Method Visual occlusion studies in driving published between 1967 and 2020 were reviewed. The focus was on original studies in which the forward visual field was intermittently occluded while the participant was driving. Results Occlusion studies have shown that attentional demand varies across situations and drivers and have indicated environmental, situational, and inter-individual factors behind the variability. The occlusion technique complements eye tracking in being able to indicate the temporal requirements for and redundancy in visual information sampling. The proper selection of occlusion settings depends on the target of the research. Conclusion Although there are a number of occlusion studies looking at various aspects of attentional demand, we are still only beginning to understand how these demands vary, interact, and covary in naturalistic driving. Application The findings of this review have methodological and theoretical implications for human factors research and for the development of distraction monitoring and in-vehicle system testing. Distraction detection algorithms and testing guidelines should consider the variability in drivers’ situational and individual spare visual capacity.


INTRODUCTION
The visual occlusion technique was developed by Senders et al. (1967) to evaluate attentional demand in driving. By intermittently blocking (i.e., occluding) the driver's line of sight, for example with an occlusion visor or opaque glasses, the (visual) attentional demand can be estimated as the fraction of time that unoccluded. The occlusion technique can also be used to evaluate the analog but opposing concept of "spare visual capacity," which measures the fraction occluded ( ). These fractions can inform about the required (visual) information sampling frequency in driving. This understanding -ing and regulatory purposes.
Both attentional demand and spare visual capacity, as measured via visual occlusion, must be coupled with an assessment of successful driving performance. In the experiments of Senders et al. (1967), the drivers themselves determined what was to be considered as sucoccluded by default, meaning that they were essentially driving while blindfolded. However, as soon as the uncertainty about their own position in relation to the road became too high, the drivers could voluntarily unocclude their vision to recalibrate their mental model of the surroundings. The attentional demand of the situation was then estimated as the frequency of requested viewing instances. This is an example of a self-paced visual occlusion experiment. Senders et al. (1967) also did experiments with system-paced visual occlusion) to investigate 2023, Vol. 65(5) [792][793][794][795][796][797][798][799][800][801][802][803][804][805][806][807][808] how fast the drivers were willing to drive with intermittently occluded vision. The outcome of this series of experiments provided support for the assumption that the driver's attention needs to be only intermittently directed to the road and that the attentional demand varies with speed and road curvature.
As already indicated, there are several free parameters in the setup of a visual occlusion experiment (Lansdown et al., 2004), as can be seen in Figure 1 these settings lies in how much freedom the participant is given in controlling when and system-paced onset, the experimenter decides when, where, and for how long to occlude, including the special case of irrevocable occlusion. For safety reasons, this setting has mostly been used in simulators and on closed roads.
In system-paced occlusion, the location and duration of the occlusion are used as independent variables to assess how the withheld inforrelevant capacity (e.g., hazard perception). A self-paced onset, on the other hand, allows the driver to decide when to occlude or unocclude, depending on whether the default state of the apparatus is "unoccluded" or "occluded." With an unoccluded default state, which has mostly indicate with the activation of the occlusion when and where they do not need any visual information for the driving task at hand. An occluded default state requires the driver to activate access to visual information and has been more common in research done in simulators and on closed roads. The duration of the occluded period  Tables in Appendix. the occlusion duration is self-paced, drivers can choose freely when to terminate the occlusion depending on their level of uncertainty. In the self-paced onset setting, the choice to (un) occlude and, if self-paced as well, the occlusion time (or distance) are typically treated as dependent variables whose variation with external factors is of interest. The various settings reprevisual capacity.
With this review, we intend to assess how visual occlusion studies with various settings have contributed to the understanding of attentional demand and spare visual capacity in driving, thereby illustrating what the method can achieve while also pointing out its limitations. Based on this background, we then discuss how some of the weaknesses can be mitigated, for instance by triangulation with other methods. This can help future research go beyond the understanding of the attentional demands in driving.

METHOD
A literature search for the term "'visual occlusion' driving" was carried out using Google Scholar. The search found 2130 results (patents not included), which were sorted by scholar. google. com/ intl/ en/ scholar/ about. html). The inclusion criteria for the selected studies porally (as opposed to spatially) occluded (i.e., visually interrupted) while the participant was driving and (2) the studies had to assess the visual demands of driving. The search results were reviewed for title and excerpt of the articles, and for the abstract or full-text articles if inclusion criteria. Studies where the participant was watching images or videos of driving were excluded, as were papers where the participant was a passenger in a moving vehicle. Furthermore, a number of articles related to invehicle information system testing (e.g., Foley, 2008;Gelau et al., 2009;Lansdown et al., 2004) were excluded as these evaluations are usually carried out while standing still and the objective of this line of research is not to assess the visual demands of driving but rather to focus on the demands of in-car tasks. Relevant references in the review. In screening phase, 2046 search results were excluded. In total, 89 full-text articles were assessed for eligibility, of which inclusion criteria. As a result, 57 studies in 52 publications were included in the qualitative synthesis.

RESULTS
A summary of the 57 included studies reported in 52 publications is provided in Tables A1-A4 (Appendix). For each study, the main objective of the experiment, the number of participants, the driving environment (motorway, test track, or driving simulator), and the settings of the occlusion experiment are provided. The latter also includes the mechanisms used to operate the occlusion device, ranging from Figure  2) to simply closing one's eyes, and the occlusion area, which can consist of the entire visual based on the occlusion settings that are used (Figure 1), with system-paced studies in Table  A1, system-paced with irrevocable occlusion in Table A2, self-paced studies with an unoccluded default setting in Table A3, and self-paced studies with an occluded default setting in Table A4. The remainder of the results section provides an overview of what the reviewed visual occlusion studies have taught us about how to measure spare visual capacity in driving and what we have learned about it.

Benchmarking Spare Visual Capacity
Visual occlusions provide an estimate of spare visual capacity, but there are no obvious criteria stating if an observed occlusion frequency or duration indicates operation below, at, or above capacity. Anderson et al. (2000) assumed that self-paced occlusion is likely an underestimate of the actual spare capacity. Accordingly, many participants in Kircher et al. (2020), who used the self-paced setting, reported that they had occluded below their perceived maximum capacity, hence keeping a safety margin. An incident or collision caused by the driver during an occlusion would indicate that the minimum required information had not been sampled, but the absence of such occurrences is no guarantee To get a measure that is more indicative of true capacity in self-paced settings for a parttask of driving, such as steering, time-to-linecrossing (TLC) at the end of the occlusion (TLC end ) can be added to the occlusion time (OT), providing an estimate of the total time from occlusion onset until the vehicle leaves the lane . Whereas the OT + TLC end metric describes steering performance, similar integrative metrics can be used to describe thresholds of spare visual capacity in other dynamic part-tasks of driving (e.g., time-to-collision [TTC] in longitudinal control tasks; van der Horst, 1990). These metrics apply only to the lane-keeping and distance-keeping tasks in repeated measurements in controlled conditions. How to combine such separate models in more naturalistic driving situations with multiple concurrent demands is not known, but computational modeling and simulation of the interactions of the part-tasks' competing visual demands may be a way forward (Jokinen et al., 2020). The question is whether it will be possible to move from "capacity to stay on the road" in controlled environments to "minimum required attention" in more complex driving environments with multiple actors and competing demands.
Given the current lack of objective benchmarks, an important question is whether visual occlusion is a solid empiric correlate with respect to the construct of spare visual capacity. As reviewed in the next subsection, self-paced occlusion studies have shown systematic and expected variations in occlusion times based on the manipulations of factors such as speed, road curvature, road environment, lane width, -plexity indicate lower occlusion times across the studies. This is in line with the presumption that occlusion time has a correlation with spare visual capacity. Accordingly, systempaced studies have shown that drivers are able to maintain an acceptable level of performance while occluded for considerable time periods. An upper limit of spare visual capacity can be attained in a system-paced occlusion setting by gradually increasing the occlusion time until performance failures occur (e.g., Anderson et al., 2000). Studies with irrevocable occlusion have shown that drivers can refrain from crashing in a simulated motorway curve with , and on a straight road Figure 2. An example of a visual occlusion mechanism that is activated by a micro-switch placed on the Zwahlen & Balasubramanian, 1974).

Spare Visual Capacity Varies With Situational Demands
In spite of the lack of objective benchmarks, occlusion studies have shown direct evidence that there can be spare visual capacity in driving (Anderson et al., 2000;Kircher et al., 2020;Senders et al., 1967) and that occlusion times tional demands. The amount of available spare capacity depends on several factors. Under otherwise similar circumstances, drivers choose to occlude their vision for shorter percentages of time when driving at higher speeds (Courage et al., 2000;Farber & Gallagher, 1972;Mourant & Ge, 1997;Senders et al., 1967), with narrower lane widths (Courage et al., 2000;Farber & Gallagher, 1972;Mourant & Ge, 1997;Senders et al., 1967;van der Horst & Godthelp, 1989), when driving in sharper curves (Anderson et al., 2000;Backs et al., 2003;Courage et al., 2000;Senders et al., 1967;Tsimhoni & Green, 2001;Wooldridge et al., 1999) or on more complex road geometries (Easa & Ganguly, 2005;Easa & He, 2006). Many of the reviewed studies, especially in the system-paced category, attempt on one behavioral aspect, with a special focus on either longitudinal or lateral vehicle control. There is a risk that this approach misses interacwith multiple simultaneous requirements.
However, studies that investigated the tactical driving level (Michon, 1985) indicate that variations in more complex conditions also sion Liu et al., 2020;Mourant & Mourant, 1979;Steele & Gillespie, 2001).  found that maneuvers that require interaction fast lane, lead to fewer and shorter occlusions than driving in the slow lane. Motorway driving than urban or rural driving, in spite of higher speeds on the motorway (Liu et al., 2020; also occlusion distance: Kujala et al., 2016). In heterogeneous environments, drivers often choose to occlude their vision in relation to external circumstances, such as obstacles in the infrastrucbetween several such factors (Kircher et al., 2020, Liu et al., 2020Pekkanen et al., 2017Pekkanen et al., , 2018Steele & Gillespie, 2001). In general, the driving scenario or maneuver increases spare visual capacity.
With increasing levels of vehicle automation, driver assistance systems have also been found to reduce the need for visual sampling. For example, visual demand is reduced when driving with adaptive cruise control (Hoedemaker & Kopf, 2001) or with lane keeping assistance (de Vos & Godthelp, 1999;2005;Mars et al., 2014;Steele & Gillespie, 2001). The downside of this reduced visual demand is that the driver may be inclined to be visually and mentally distracted from the superenvironment by secondary activities. Only a handful of studies have assessed how additional demands of driving and drivers' ability to pre- Monk and Kidd (2008) found an improvement in lane tracking under cognitive load during occlusion, whereas hazard perception performance worsened during a concurrent visual task (Borowsky et al., 2015(Borowsky et al., , 2016Samuel & Fisher, 2015).

Inter-Individual Differences in Spare Visual Capacity
Already Senders et al. (1967) speculated on and observed with small sample sizes inter-in many occlusion studies, but the main evidence of large individual variations in spare visual capacity comes from self-paced studies. For instance, Liu et al. (2020) showed that spare visual capacity was highly dependent on unde-have also succeeded in revealing some of the occlusion behavior is age. Older drivers in motorway driving (Mourant & Mourant, 1979; ) and in curve driving with various curvatures (Tsimhoni & Green, 1999). One explanation could be that older drivers, on average, require more time to process the availin their visual search patterns (Shinar et al., 1978). In all these studies, some aged drivers performed at the same level as the younger drivers, indicating that aging does not lead average occlusion distance in self-paced occluwith age (Kujala et al., 2016), and neither has TTC assessments under occlusion (e.g., Kiefer et al., 2006).
A related factor is driving experience. When predicting occlusion times based on a supervisory control model, Blaauw et al. (1984) found that the predictions underestimated inexperienced drivers' actual occlusion times, but not suitable for inexperienced drivers. Based on the simple steering maneuvers while intermittently occluded, Cavallo et al. (1988) suggested that experienced drivers are more proactive when it comes to path control in curve driving, whereas novice drivers are more reactive. This is also use more complex and dynamic visual sampling strategies and have a better ability to estimate and predict the vehicle's position during occlusions (Chen, 2013). That experienced drivers are more capable of constructing useful predictions is also indicated by longer self-paced occlusion times for comparable levels of driving performance and more adaptation of occlusion times when occluded during driving-related multitasking and foggy or night-time driving conditions (Blaauw, 1984 made by Kujala et al. (2016) where the more experienced drivers were able to achieve more accurate lane-keeping performance than the least experienced drivers with similar occlusion distances. Senders et al. (1967) attributed interrate and the maximum level of tolerated uncertainty during occlusion, and also to how far ahead the driver samples information. The ability to make use of the sampled information and predict what is happening during the occlusion is yet another factor, which depends on the individual's ability to extract information during the preceding sampling (Shinar et al., 1978) and on its duration (Chen & Milgram, 2011).
preferred safety margins (Kujala et al., 2016;Pekkanen et al., 2017), in terms of how drivers try to maintain information redundancy during sampling (Milgram, 1983;Milgram et al., 1982) Pekkanen et al., 2017 at an individually comfortable level. Instead of overestimations, it seems that drivers tend to underestimate TLC , TTC (Kiefer et al., 2006), and distances (  2015) while occluded. The drivers in the study by  preferred to sample, on average, at about 40% of the available time before estimated lane crossing.

DISCUSSION
Visual occlusion in its varying forms has been used as a tool to establish the minimum visual information input necessary for driving in a foresighted and controlled manner. Occlusion studies have provided evidence of spare visual capacity in driving and support for open-loop (i.e., intermittent) control models of driving performance (e.g., . There is convincing evidence that spare visual capacity relates to what we can summarize as the predictability of the situation, which is dependent on a combination of factors like capabilities and maneuvering intentions. This qualitative literature review provided an literature, but meta-analysis-where applicable-would be the next logical step to acquire a more detailed quantitative understanding about to the disparity of experimental designs and targets of research (Tables in Appendix) and the fact that the data are often not reported or open access in a manner that would enable a meanif not impossible.

Limitations
gives access to insights about the availability of spare visual capacity and about the occasions when visual input is needed. However, as outlined in the benchmark section, the lack of criteria that determines the objectively available spare visual capacity is a fundamental challenge.
Visual occlusion can be seen as a coarse of the driver. In some studies, a small peripheral area was left intact (Borowsky et al., 2015(Borowsky et al., , 2016Kircher et al., 2020, Liu et al., 2020Pekkanen et al., 2017Pekkanen et al., , 2018Samuel & Fisher, 2015). This means that occlusion studies do not allow an are most crucial for the driver in each situation. A combination of occlusion with eye tracking can indicate the foveal targets in the unoccluded periods (e.g., Anderson et al., 2000;Borowsky et al., 2016;Kircher et al., 2020) but cannot determine the degree of importance of peripheral information. Partial occlusion (e.g., foveal vision only, peripheral vision only, or gazeoccluded based on real-time eye tracking) can used and what their relative importance is for part-tasks of driving, depending on how performance degrades or sampling changes if parts of Gordon, 1966;Wood & Troutbeck, 1994). Yet, removing altered sampling strategies. Therefore, a direct estimation of what information is normally sampled with foveal or peripheral vision may not be possible with partial occlusion.
Self-paced occlusion with an unoccluded default state resembles the situation where a driver chooses to execute an additional visual task while driving. However, in most studies, visual occlusion merely blanks out visual input, whereas additional visual tasks also require some mental focus but do not necessarily take away peripheral vision. Visual occlusion lets the driver focus on the latest impression of the scene, providing the possibility to make predictions about its likely development. Very few studies have investigated whether an additional task during visual occlusion hampers the driver's prediction abilities (Borowsky et al., 2015(Borowsky et al., , 2016Monk & Kidd, 2008;Samuel & Fisher, 2015). More studies on the topic are needed sus additional task execution that also involves mentally focusing on something else.
Visual occlusion is rather obtrusive and, especially in its self-paced version, can put mental load on the driver. For example, in some of 's (1967) experiments, the participants were given a break after 15 min because "the task of driving was an arduous one," and comments from participants in our own studies indicate that occlusion experiments are experienced as fatiguing. Also, blocking a driver's vision outside simulators can be perceived as ethically problematic (Anderson et al., 2000;Tsimhoni & Green, 1999), or as Senders et al. (1967 put it, "perhaps a little risky." This could be one reason behind the rareness of on-road studies employing the method. Using vehicles equipped with dual control and experienced safety drivers that are ready to intervene, as used in Kircher et al. (2020), is a relevant safety procedure in on-road simulator and on-road studies were outside the scope of the current review, but it should be noted that there might be variations in the outcomes depending on the ecological validity of the driving task.

Occlusion and Eye Tracking
Historically, studies using visual occlusion and studies using eye tracking have largely proceeded independently of each other, with the former mainly focusing on when visual information was not needed for driving and the latter categorizing what was sampled without really considering its necessity. As such, the two methods complement each other, and much can be gained by combining them. However, this is rarely done, and only a handful of studies have combined the techniques (see notes in Tables A1-A4, Appendix). Instead, eye tracking has gradually become the de facto standard in driver attention research. Mourant & Rockwell, 1970Rockwell et al., 1968). In these early days of eye-tracking research, the focus was on understanding the eye movement patterns of experienced, accident-free drivers. The importance of peripheral vision was still -Technological advancements not only led to a more frequent and widespread use of eye tracking, but also saw a shift in how the data were interpreted. It became more common to only categorizing them and labeling them as "relevant for driving" or not (Crundall et al., 2006;Garrison & Williams, 2013). This analysis of for its putative simplicity; however, it comes at the risk of neglecting what cannot be easily observed-namely, the information sampled via peripheral vision (Rosenholtz, 2016;Wolfe et al., 2019) and the determination of the necessary amount of information for a task (Kircher & Ahlström, 2017). Gradually, the advent of mobile eye trackers directed the focus of gets that drivers foveally focus upon, and to the conclusion that drivers are inattentive as soon as they glance at targets deemed "not relevant for driving" (Garrison, 2011) without considering any possibly available spare visual capacity.
As compared to occlusion, eye tracking actually needed for gathering the required information (Kircher et al., 2020). Eye trackgaze targets (e.g., pedestrian approaching a crosswalk), while occlusion makes sure that no redundant information is unjustly assumed to road). Capitalizing on this, it has been shown that there are redundant glances to the forward roadway in normal driving (Anderson et al., 2000;Kircher et al., 2020), but also necessary roadway (e.g., on mirrors; Kircher et al., 2020). Chen and Milgram (2013) argue that instead of gross metrics, the focus in occlusion research should be on the situational and individual variability of the information sampling. There are occlusion times in similar scenarios without explanation (e.g., Kujala et al., 2016). This also means the lack of objective benchmarks. However, we are not aware of a more objective method than occlusion for assessing spare visual capacity in dynamic tasks. This capacity always has a subjective component, even if the capacity can be argued to decrease (on average) with the increasing complexity or unpredictability of a scenario.

Minimum Required Attention and Future Work
For a deeper understanding of the minimum amount of information that needs to be sampled for attentive driving in real trafsingle-factor control models while attempting to preserve access to individual factors, we suggest combining the rather data-driven approach of visual occlusion with a theory In this context, visual occlusion, possibly in combination with other methods like eye tracking and think-aloud, could indicate how frequently the needed information is sampled and whether this is done foveally or peripherally. The Minimum Required Attention (MiRA) theory (Kircher & Ahlström, 2017) ing the attentional requirements, at least with respect to so-called static requirements, which tions. Self-paced occlusion frequency, duration, and location can be observed in relation to a systematic combination and variation of requirements, and system-paced occlusion of ments in dynamic situations, factors like trolling other factors, to assess the impact on occlusion possibilities. Computational modeling may prove to be useful for simulating the dynamic requirements and interactions of multiple demands (compare Jokinen et al., 2020). Equally important as knowing the situational targets is knowing how the situational information sampling frequencies of these targets. Existing self-paced occlusion studies have shown how drivers experience these frequencies, but more advanced tools are needed to evaluate the validity of these assessments in complex scenarios.

Implications for Human Factors Research and Practice
What is essential for minimum required information sampling in open-loop driving with various competing demands (compare example, Backs et al. (2003 and Green (1999, 2001), who operationalized it as unoccluded time divided by total time within a segment of interest. As a gross measure, similar to the percentage of unoccluded time per drive by Mourant and Ge (1997), it loses the situational information on the timing of the demands. Similarly, driver distraction guidelines (Young & Zhang, 2015) based on system-paced occlusion testing for in-car tasks neglect one of the most important factors of spare visual capacity, that is, the driver's ability to time the in-car glances in accordance with the variable visual demands of driving. As such, these methods seem to evaluate only the visual demands of the in-car task and not its compatibility with driving. The observed variability in spare visual capacity in driving casts doubts on general distraction monitoring and visual distraction testing of in-car tasks (e.g., Young & Zhang, 2015). When testing the distraction potential of in-car devices or tasks, the individual be considered and controlled in order to provide reliable results (Broström et al., 2016). Furthermore, the presented evidence suggests that acceptance of an in-car task to be conducted while driving cannot be judged only lated part-task of driving. On the other hand, the acceptance thresholds might be too strict if the driver's capacity to utilize peripheral vision in the task is not considered (e.g., as in lane keeping). Even more importantly, the thresholds might be too low if the visual and cognitive demands of the in-car task interfere with the demands of such part-tasks of driving that have not been evaluated (e.g., hazard perception; Borowsky et al., 2014).

CONCLUSIONS
The occlusion technique can complement eye tracking in studies on the attentional demand of driving by indicating the driver's spare visual capacity. Occlusion studies have shown that spare visual capacity varies across situations and drivers and have indicated environmental, situational, and inter-individual factors behind the variability. The level of understanding that can be achieved with the technique depends on the selection of the occlusion method (selfpaced vs. system-paced).
ological implications for human factors research and practical applications for the development of distraction monitoring and in-vehicle system testing. Distraction detection algorithms and testing guidelines need to consider the variability in situational and individual spare visual capacity. Oversimplifying the attentional demand of driving should be avoided in order to make valid and reliable conclusions on whether a driver is distracted or not. While there are a number of occlusion studies looking at various aspects of attentional demand, we are still only beginning to understand how the demands vary, interact and covary. Triangulation of various methods together with occlusion may be required for this inquiry.      Note. a Eye-tracking was used simultaneously with occlusion.

ACKNOWLEDGMENTS
This work was partially funded by Sweden's Innovation Agency (Grant 2019-05834).

KEY POINTS
Spare visual capacity varies with situation and driver. Combining eye tracking with the occlusion technique can enable the indication of requirements for and redundancy in visual information sampling. The appropriate occlusion setting depends on the target of the research. Distraction monitoring and testing needs to consider the variability in attentional demand. We are still only beginning to understand how attentional demands of driving vary, interact, and covary.