Glance Behavior Research Articles

Caught in the blind spot of a truck: A choice model on driver glance behavior towards cyclists at intersections

Vulnerable road users (VRUs) constitute an increasing proportion of the annual road fatalities across Europe. One of the crash types involved in these fatalities are blind spot crashes between trucks and bicyclists. Despite the presence of mandatory blind spot mirrors, truck drivers are often reported to have overlooked the presence of a bicyclist. This raises the question if and when truck drivers check their blind spot mirrors for the presence of bicyclists, and which factors contribute to such glance behavior. The current study presents the results of an analysis of naturalistic glance behavior by 39 truck drivers in 1,903 right-turning maneuvers at urban intersections, where in each maneuver there was a chance of crossing the path of a bicyclist. The descriptive analysis revealed that most often truck drivers did not cast a glance upon their blind spot mirrors as recommended by the examination guidelines. Furthermore, a choice model was developed with the main factors that have an impact on glance behavior. Drivers were more likely to glance with a priority regulation that allowed conflicts, with lower speed limits, with zebra crossings, without cyclist facilities, without a lead vehicle making the same maneuver, in presence of VRUs, without adverse sight conditions, in lower age groups, without certain non-driving related activities, when driving a truck with more direct vision on VRUs, and without a camera providing a view on the blind spot, and with less time between a standstill and starting the maneuver. Three factors did not significantly improve the choice model and were therefore left out, despite showing significant effects in bivariate tests: intersection layout (e.g., three vs. four legs), presence of advanced stopping lanes, and visual obstruction. Implications of the choice model are discussed for driver education (in terms of timely glances, reducing inattention, and hazard anticipation) and vehicle design (in terms of direct vision).

Driver Glance Behavior Modeling Based on Semi-Supervised Clustering and Piecewise Aggregate Representation

Glance behavior is significant because whether and how the driver is scanning and observing the driving scene is closely related to driving safety. This paper aims to improve the accuracy of glance behavior modeling and realize the spatiotemporal representation and visualization of glance behavior. Forty subjects were recruited to perform a freeway driving task using a driving simulator. The vehicle data were collected by the simulator. Drivers’ gaze points were collected by an eye tracker. The prior knowledge on gaze points obtained through a statistical analysis were provided for K-means (KM) to form a semi-supervised K-means (SSKM), which classifies gaze points into different fixation zones. The classification results were compared with the results of KM. Furthermore, a clustering center-based piecewise aggregate representation (CCPAR) was proposed to characterize glance behavior. Maneuvers identification was taken as a case to evaluate the proposed method. The k-nearest neighbour (KNN) based on the similarity of CCPAR identified driving maneuvers into lane-keeping, left lane change, and right lane change. The identification results were compared with the results of the Hidden Markov Model (HMM). The average classification accuracies of KM and SSKM were 55.28% and 94.75%, respectively. The accuracies of maneuvers identified by CCPAR-KNN and by HMM were 87.50% and 85.83%, respectively. The results indicate that SSKM and CCPAR are feasible for glance behavior modeling. SSKM eliminates the randomness of initial cluster center selection and improves the accuracy of gaze points classification. CCPAR is intuitive and convenient to describe and visualize the spatiotemporal characteristics of glance behavior.

Using Glance Behaviour to Inform the Design of Adaptive HMI for Partially Automated Vehicles

Partially automated vehicles present a large range of information to the driver in order to keep them in-the-loop and engaged with monitoring the vehicle’s actions. However, existing research shows that this causes cognitive overload and disengagement from the monitoring task. Adaptive Human Machine Interfaces (HMIs) are an emerging technology that might address this problem, by prioritising the information presented. To date, research aiming to define the driver’s glance fixation behaviour in a partially automated vehicle to contribute towards an adaptive interface is scarce. This study used a unique three-day longitudinal driving simulator study design to explore which information drivers in a partially automated vehicle require. Twenty-seven participants experienced nine partially automated driving simulations over three consecutive days. Nine information types, developed from standards, previous studies and industry collaboration, were displayed as discrete icons and presented on a surrogate in-vehicle display. Unique to the literature, this study showed that the recorded eye-tracking data demonstrated that usage of the information types changed with longitudinal driving simulator use. This study provides three key contributions: first, the longitudinal study design suggest that single exposure HMI evaluations may be limited in their assessment. Secondly, this study has methodologically shortlisted a list of nine information types that can be used in future studies to represent future partially automated vehicle interfaces. Finally, this is one of the first studies to characterise glance behaviour for partially automated vehicles. With this knowledge, this study contributes important design recommendations for the development of adaptive interfaces.

Looking out or Looking Away?—Exploring the Impact of Driving With a Passenger on Young Drivers’ Eye Glance Behavior

Objective To explore how passenger presence and the degree of association between young driver and passenger influences young drivers’ eye glance behavior when they are subjected to distraction. Background Young drivers (18–20 years old) are at an elevated crash risk when subjected to distraction. They are likely to be distracted even further when they drive with passengers. However, the eye glance behavior of these drivers when driving with passengers has not been explored. Method Eye glance data of 34 young drivers between the ages of 18 and 20 years were collected. Participants drove with and without a passenger while subjected to three distracting tasks (visual-manual, cognitive, or visual-cognitive) and driving scenarios that required driver attention. Results Visual-cognitive as well as visual-manual states of distraction result in higher mean and standard deviation of glance duration, along with higher number of glances away from road. Passenger presence is found to negatively influence young drivers’ eye glance behavior. The degree of association between the young driver and the passenger may help reduce the deviation of eye glances towards the task-related objects. Conclusion In addition to distraction, passengers have a negative influence on the eye glance behavior of young drivers. However, a high degree of association between driver and passenger may mitigate the negative impact of distraction on the eye glance behavior of young drivers. Application (non-theoretical works) This research may aid in the design of interventions that improve young drivers’ eye glance behavior when they drive with their peers.

Does LSTM outperform 4DDTW-KNN in lane change identification based on eye gaze data?

Lane change identification based on long short-term memory (LSTM) neural networks has received increasing attention. The most commonly known disadvantage of the approach is the “black box” nature. This study, to improve the interpretability, proposed a novel method for identifying lane changes. The philosophy was to measure the similarity of the glance behavior during the maneuver. The glance behavior in the same maneuver is more similar, while that in different maneuvers is different. First, a driving simulator was used to collect driving behavior data. The eye gaze time series were captured by the eye tracker. The median of the eye gaze sequence of 3 s before the initial moment of lane change was obtained. With the median as the center, the driver's visual field plane was divided into several grids with the side length of 600 pixels. A sliding space-time cuboid algorithm was proposed to extract the scanning path. Second, four-dimensional dynamic time warping (4DDTW) distance was used to measure the similarity of the scanning paths. Third, the K nearest neighbor (KNN) algorithm was used to classify the driving maneuver into the lane-keeping (LK), the right lane change (RLC), and the left lane change (LLC) based on the 4DDTW distance of the scan paths. LSTM was also used to classify the driving maneuver into LK, RLC, and LLC. The performances of 4DDTW-KNN and LSTM were compared. The accuracies of 4DDTW-KNN and LSTM were 86.50% and 86.33%, respectively. LSTM is not better than 4DDTW-KNN in lane change identification based on eye gaze data with a comprehensive consideration of the time efficiency, the accuracy and the interpretability.

Computational modeling of driver pre-crash brake response, with and without off-road glances: Parameterization using real-world crashes and near-crashes

When faced with an imminent collision threat, human vehicle drivers respond with braking in a manner which is stereotypical, yet modulated in complex ways by many factors, including the specific traffic situation and past driver eye movements. A computational model capturing these phenomena would have high applied value, for example in virtual vehicle safety testing methods, but existing models are either simplistic or not sufficiently validated. This paper extends an existing quantitative driver model for initiation and modulation of pre-crash brake response, to handle off-road glance behavior. The resulting models are fitted to time-series data from real-world naturalistic rear-end crashes and near-crashes. A stringent parameterization and model selection procedure is presented, based on particle swarm optimization and maximum likelihood estimation. A major contribution of this paper is the resulting first-ever fit of a computational model of human braking to real near-crash and crash behavior data. The model selection results also permit novel conclusions regarding behavior and accident causation: Firstly, the results indicate that drivers have partial visual looming perception during off-road glances; that is, evidence for braking is collected, albeit at a slower pace, while the driver is looking away from the forward roadway. Secondly, the results suggest that an important causation factor in crashes without off-road glances may be a reduced responsiveness to visual looming, possibly associated with cognitive driver state (e.g., drowsiness or erroneous driver expectations). It is also demonstrated that a model parameterized on less-critical data, such as near-crashes, may also accurately reproduce driver behavior in highly critical situations, such as crashes.

Evidence of inconsistent results using current eye tracking glance and visit analysis standards

Eye tracking glance analyses have been used to study the behavior of motor vehicle drivers for at least two decades and eye tracking is now emerging as an experimental protocol in construction research. Previous studies have identified relationships between driver glance behavior and performance, so existing motor vehicle driver glance analysis methods were applied to search for analogous relationships with construction craftworkers during a model assembly task. Eye tracking analysis software vendors have yet to adopt standardized eye tracking event detection algorithms, and researchers have historically reported insufficient details to render most eye tracking analyses reproducible. As a result, the purpose of the present work is to address the gap in knowledge in understanding the challenges encountered as a result of a lack of standardization. The methodology applied existing glance analysis methods to a study incorporating mobile eye tracking glasses, which made the experimental environment less controlled than a typical glance analysis study using a screen-based eye tracking system, such as a motor vehicle driver behavior study. A new noise filtering parameter, maximum off-stimulus fixations, was also introduced. The results show that the absence of standardized eye tracking parameters, such as minimum fixation duration, can lead to different statistical results. The study's primary contribution to the body of knowledge is that is identifies the need for standardization of eye tracking analyses to better ensure that future eye tracking studies be more consistent and reproducible.

A model for naturalistic glance behavior around Tesla Autopilot disengagements

We present a model for visual behavior that can simulate the glance pattern observed around driver-initiated, non-critical disengagements of Tesla's Autopilot (AP) in naturalistic highway driving. Drivers may become inattentive when using partially-automated driving systems. The safety effects associated with inattention are unknown until we have a quantitative reference on how visual behavior changes with automation. The model is based on glance data from 290 human initiated AP disengagement epochs. Glance duration and transition were modelled with Bayesian Generalized Linear Mixed models. The model replicates the observed glance pattern across drivers. The model's components show that off-road glances were longer with AP active than without and that their frequency characteristics changed. Driving-related off-road glances were less frequent with AP active than in manual driving, while non-driving related glances to the down/center-stack areas were the most frequent and the longest (22% of the glances exceeded 2 s). Little difference was found in on-road glance duration. Visual behavior patterns change before and after AP disengagement. Before disengagement, drivers looked less on road and focused more on non-driving related areas compared to after the transition to manual driving. The higher proportion of off-road glances before disengagement to manual driving were not compensated by longer glances ahead. The model can be used as a reference for safety assessment or to formulate design targets for driver management systems.

The Effect of Augmented Reality Cues on Glance Behavior and Driver-Initiated Takeover on SAE Level 2 Automated-Driving

In the present paper, we present a user study with an advanced-driver assistance system (ADAS) using augmented reality (AR) cues to highlight pedestrians and vehicles when approaching intersections of varying complexity. Our major goal is to understand the relationship between the presence and absence of AR, driver-initiated takeover rates and glance behavior when using a SAE Level 2 autonomous vehicle. Therefore, a user-study with eight participants on a medium-fidelity driving simulator was carried out. Overall, we found that AR cues can provide promising means to increase the system transparency, drivers’ situation awareness and trust in the system. Yet, we suggest that the dynamic glance allocation of attention during partially automated vehicles is still challenging for researchers as we still have much to understand and explore when AR cues become a distractor instead of an attention guider.

Advanced vehicle technology: Mapping mental model accuracy and system exposure to driver behavior

Advanced driver assistance systems have potential to increase safety and comfort for drivers; however, drivers need to understand the capabilities and limitations of these systems to use them appropriately. This study sought to explore how the quality (accuracy) of drivers’ mental models of adaptive cruise control (ACC) impacted their behavior and interactions while using the system. Seventy-eight participants drove in a high-fidelity driving simulator while operating an ACC system, in normal conditions and while interacting with the system interface. Participants with stronger (more accurate) mental models glanced to the road ahead more often during normal conditions early on compared to drivers were less accurate mental models; however, these differences diminished with increased system exposure. Glance behavior while interacting with the system and time to complete the interactions were less effected by the strength of the participant’s mental model. Results are discussed in the context of driver education and training.

Drivers Glance Like Lizards during Cell Phone Distraction in Assisted Driving

Head pose has been proposed as a surrogate for eye movement to predict areas of interest (AOIs) where drivers allocate their attention. However, head pose may disassociate with AOIs in glance behavior involving zero or subtle head movements, commonly known as “lizard” glance pattern. In contrast, “owl” glance pattern is used to describe glance behavior along with larger head movements. It remains unclear which glance pattern is prevalent during driver cell phone distraction and what are appropriate metrics to detect such distraction. To address this gap, we analyzed the gaze direction and head pose of 36 participants who completed an email-sorting task using a cell phone while driving a Tesla on the test track in Autopilot mode. The dispersion-threshold algorithm identified driver gaze fixations and synchronized them with head movements. The results showed that when using a cell phone either near the lap or behind the steering wheel, participants exhibited a dominant lizard-type glance pattern with minimal shift in head position. As a result, head pose alone may not provide sufficient information for cell phone distraction detection, and gaze metrics should be involved in enhancing this application.

Influence of Prior General Knowledge on Older Adults’ Takeover Performance and Attitude Toward Using Conditionally Automated Driving Systems

This study aims to investigate the influence of general knowledge, which denotes·driving automation’s taxonomy, definitions, function, driver role, and the request to intervene (RtI), on older adults’ takeover performance when using conditionally driving automation (DA), and to clarify the influence of knowledge on drivers’ attitudes toward DA. We introduced two types of DAs: full range and limited range. A driving simulator (DS) experiment was conducted to collect data, including driving behavior in taking over vehicle control and driver’s subjective evaluation of the DA. Data were collected from 36 elderly (mean age = 71.4±4.8 years) and 36 non-elderly (mean age = 40.8±9.5 years) participants. The results showed significant differences between the elderly and non-elderly, such that educating knowledge had a greater influence on the older adults, that is, instructing knowledge to the drivers contributed to a statistical increase in successful takeover rate from 0.66 to 0.80, but no effect was seen on either the response time or the maximum steering angular velocity. Furthermore, more frequent glance behavior of looking forward from a non-driving related task was observed in the educated groups. Self-rating scores of subjective evaluations revealed that older adults who were given the knowledge had a higher level of trust in and expectation from the DA, and more confidence in comprehending system functions. This study demonstrates the necessity of general knowledge instruction to enhance drivers’ positive attitudes toward DA.

Isolating the Effect of Off-Road Glance Duration on Driving Performance: An Exemplar Study Comparing HDD and HUD in Different Driving Scenarios.

We controlled participants' glance behavior while using head-down displays (HDDs) and head-up displays (HUDs) to isolate driving behavioral changes due to use of different display types across different driving environments. Recently, HUD technology has been incorporated into vehicles, allowing drivers to, in theory, gather display information without moving their eyes away from the road. Previous studies comparing the impact of HUDs with traditional displays on human performance show differences in both drivers' visual attention and driving performance. Yet no studies have isolated glance from driving behaviors, which limits our ability to understand the cause of these differences and resulting impact on display design. We developed a novel method to control visual attention in a driving simulator. Twenty experienced drivers sustained visual attention to in-vehicle HDDs and HUDs while driving in both a simple straight and empty roadway environment and a more realistic driving environment that included traffic and turns. In the realistic environment, but not the simpler environment, we found evidence of differing driving behaviors between display conditions, even though participants' glance behavior was similar. Thus, the assumption that visual attention can be evaluated in the same way for different types of vehicle displays may be inaccurate. Differences between driving environments bring the validity of testing HUDs using simplistic driving environments into question. As we move toward the integration of HUD user interfaces into vehicles, it is important that we develop new, sensitive assessment methods to ensure HUD interfaces are indeed safe for driving.

Pedestrian safety at roundabouts: Their crossing and glance behavior in the interaction with vehicular traffic

Smartphones are nowadays indispensable devices in daily life. Their rapid technological development makes it possible to have almost all necessary information on them, which facilitates their spread among all users. However, this extensive use has led to many distraction problems while performing other main tasks, such as activities on the road, which can also impact people's safety. Therefore, various experts focused their attention on issues related to drivers looking at their mobile phones; recently, researchers from different disciplines saw the need to deepen knowledge also on the phenomenon of vulnerable road users distracted by digital devices, especially when approaching signalized intersections.This study builds on and seeks to extend this area of research by analyzing the effects of digital distraction on pedestrians as they approach unsignalised intersections located on roundabout entrances and exits. The aim of the research is to understand the extent to which the task of checking social media affects pedestrian reaction and crossing times, as well as to identify which elements attract the most pedestrian attention. To achieve this goal, an eye-tracking study was designed, in which participants wearing eye-tracking glasses were asked to walk a predefined route, once checking their social media apps and once walking without distracting technological elements.The results showed an 84% increase in reaction time when using the phone, while only a slight rise in crossing time was found. The general conclusions about the most observed elements when walking are also consistent with the main findings of previous literature studies.

Determining the impact of augmented reality graphic spatial location and motion on driver behaviors

While researchers have explored benefits of adding augmented reality graphics to vehicle displays, the impact of graphic characteristics have not been well researched. In this paper, we consider the impact of augmented reality graphic spatial location and motion, as well as turn direction, traffic presence, and gender, on participant driving and glance behavior and preferences. Twenty-two participants navigated through a simulated environment while using four different graphics. We employed a novel glance allocation analysis to differentiate information likely gathered with each glace with more granularity. Fixed graphics generally resulted in less visual attention and more time scanning for hazards than animated graphics. Finally, the screen-fixed graphic was preferred by participants over all world-relative graphics, suggesting that graphic spatially integration into the world may not always be necessary in visually complex urban environments like those considered in this study.

Patterns in transitions of visual attention during baseline driving and during interaction with visual–manual and voice-based interfaces

Voice interfaces reduce visual demand compared with visual-manual interfaces, but the extent depends on design. This study compared visual demand during baseline driving with driving while using voice or manual inputs to place calls with Chevrolet MyLink, Volvo Sensus, or a smartphone. Mean glance duration and total eyes-off-road-time increased when using manual input compared with baseline driving; only eyes off road time increased with voice input. Confusion matrices developed with hidden Markov modelling characterise the similarity of glance sequences during baseline driving and while making phone calls. Glance sequences with the MyLink voice interface were misclassified as baseline driving more frequently than the other voice interfaces. Conversely, glance sequences with the Sensus and smartphone voice interfaces were more often misclassified as manual phone calling. Thus, the MyLink voice interface not only reduced the overall visual demand of placing calls, but produced glance patterns more similar to driving without another task. Practitioner Summary: The attention map and confusion matrix methodologies provide ways of characterising similarities and differences in glance behaviour across secondary task conditions, complementing traditional temporally based metrics (e.g. mean glance duration, long duration glances) while addressing some of the limitations of total-eyes-off-road-time (TEORT) for comparing secondary task behaviour to baseline driving.

Investigating the impact of driving automation systems on distracted driving behaviors

Driving automation systems (e.g., SAE Level 2) ultimately aim to enhance the comfort and safety of drivers. At present, these systems are able to control some portions of the driving task (e.g., braking, steering) for extended time periods, giving drivers the opportunity to disengage from the responsibilities associated with driving. In this study, data derived from two naturalistic driving studies involving automation-equipped vehicles were analyzed to evaluate driver behaviors with respect to driving automation system use, specifically distraction-related factors (i.e., secondary task engagement, eye-glance behavior, and drowsiness). The results indicate that when drivers had prior experience using driving automation systems, they were almost two times as likely to participate in distracted driving behaviors when the systems were active than during manual driving. Drivers with less experience and familiarity with driving automation systems were less likely to drive distracted when the systems were active; however, these drivers tended to be somewhat drowsy when driving with systems activated. The results provide important insights into different operational phases of driving automation system use (i.e., learning/unfamiliar vs experienced users), whereby experience results in overtrust and overreliance on the advanced technologies, which subsequently may negate some of the safety benefits of these systems. Thus, while the safety benefits of driving automation systems are evident, it is imperative to better understand the impact these advanced technologies may have on driver behavior and performance in order to evaluate and address any unintended consequences associated with system use.

Exploring the effectiveness of a digital voice assistant to maintain driver alertness in partially automated vehicles

Objective Vehicle automation shifts the driver's role from active operator to passive observer at the potential cost of degrading their alertness. This study investigated the role of an in-vehicle voice-based assistant (VA; conversing about traffic/road environment) to counter the disengaging and fatiguing effects of automation. Method Twenty-four participants undertook two drives– with and without VA in a partially automated vehicle. Participants were subsequently categorized into high and low participation groups (based on their proportion of vocal exchanges with VA). The effectiveness of VA was assessed based on driver alertness measured using Karolinska Sleepiness Scale (KSS), eye-based sleepiness indicators and glance behavior, NASA-TLX workload rating and time to gain motor readiness in response to take-over request and performance rating made by the drivers. Results Paired samples t-tests comparison of alertness measures across the two drives were conducted. Lower KSS rating, larger pupil diameter, higher glances (rear-mirror, roadside vehicles and signals in the drive with VA) and higher feedback ratings of VA indicated the efficiency of VA in improving driver alertness during automation. However, there was no significant difference in alertness or glance behavior between the driver groups (high and low-PR), although the time to resume steering control was significantly lower in the higher engagement group. Conclusion The study successfully demonstrated the advantages of using a voice assistant (VA) to counter these effects of passive fatigue, for example, by reducing the time to gain motor-readiness following a TOR. The findings show that despite the low engagement in spoken conversation, active listening also positively influenced driver alertness and awareness during the drive in an automated vehicle.

Adaptations in driver deceleration behaviour with automatic incident detection: A naturalistic driving study

Traffic congestion and crash rates can be reduced by introducing variable speed limits (VSLs) and automatic incident detection (AID) systems. Previous findings based on loop detector measurements have revealed that drivers reduce their speeds while approaching traffic congestion when the AID system is active. Notwithstanding these behavioural effects, most microscopic traffic flow models assessing the impact of VSLs do not describe driver response accurately.This study analyses the main factors that influence driver deceleration behaviour while approaching traffic congestion with and without VSLs. The Dutch VSL database was linked to the driver behaviour data collected in the UDRIVE naturalistic driving study. Driver engagement in secondary tasks and glance behaviour were extracted from the video data. Linear mixed-effects models predicting the characteristics of deceleration events were estimated.The results show that the maximum deceleration is high when approaching a slower leader, when driving at high speeds and short distance headways, and close to the beginning of traffic congestion. The minimum time headway is short when driving at high speeds and changing lanes. Certain drivers showed higher decelerations and shorter time headways than others. Controlled for these main factors, smaller maximum decelerations were found when the VSLs were present and visible, and when the gantries were within close proximity. These factors could be incorporated into microscopic traffic simulations to evaluate the impact of AID systems on traffic congestion more realistically. Further research is needed to clarify the link between engagement in secondary tasks, glance behaviour and deceleration behaviour.

The effect of driving demands on distraction engagement and glance behaviors: Results from naturalistic data

ObjectiveTo understand drivers’ engagement in distractions and their visual attention allocation under different driving demands. Background: Although distraction increases crash risk, drivers engage in distractions frequently with no negative consequences, likely in part due to their self-regulating behaviors. Prior research revealed a variety of self-regulating behaviors specifically related to cell-phone engagement, but very limited research has investigated whether and how driving demands affect engagement in distractions in general, particularly within a natural setting rather than in the simulator. Method: We used the Naturalistic Engagement in Secondary Tasks (NEST) dataset, a subset of SHRP2 data, to analyze secondary task engagement and off-path (not in direction of travel) eye glances. In addition to assessing their relation to environmental demand, we also considered driver age and chosen speed. Results: Higher visual difficulty environments (characterized as visually complex and/or with low visibility) were associated with a decreased likelihood of secondary task engagement as well as a decrease in off-path glances, particularly longer ones (>2s). Drivers 35 and older had lower rates of off-path glances compared to younger drivers. An increase in speed was associated with a decrease in the likelihood of task engagement in higher motor control difficulty environments (characterized as poor surface condition and/or curved road) but not in lower ones. Conclusion: Drivers appear to modulate their task engagement and off-path glances based on driving demands. However, given that inopportune short off-path glances can lead to crashes, interventions are still needed to help drivers better modulate their distraction engagement.