National Surface Transportation Safety Center for Excellence Reports (NSTSCE, VTTI)

Permanent URI for this collection

https://hdl.handle.net/10919/5529

http://www.vtti.vt.edu/national/nstsce/

Browse

Now showing 1 - 20 of 161

Equity in Transportation Safety
Robinson, Sarah; Medina, Alejandra; Gibbons, Ron; Kassing, Andrew; Myers, Bradley (National Surface Transportation Safety Center for Excellence, 2024-09-24)
Equity in transportation is a key issue for the Federal Highway Administration (FHWA), as well as state departments of transportation. Equitable transportation ensures safety for all road users across all modes of transportation for all communities. FHWA recommends the adoption and equitable application of a safe system approach to achieve Vision Zero objectives to eliminate traffic fatalities and severe injuries. A safe system fundamentally recognizes human error and accounts for it when designing systems and operations. Incorporating equity into roadway safety data is critical for conducting data-driven safety analysis. FHWA recommends collaboration with underserved communities through a process of collecting and analyzing data, engaging community representatives, implementing improvements, and evaluating impacts. Ensuring robust and accurate data is critical. State programs have worked to incorporate a wide variety of data into their crash models. Social and demographic data such as race, ethnicity, gender, age, education, employment status, income level, disability status, among many other variables, have been evaluated and demonstrated to be factors in the frequency of crashes. States have published mapping tools to visualize data trends and identify locations for targeted implementation efforts in conjunction with scoring metrics for evaluating proposed solutions.
Roadway Departure Events Using SHRP 2 NDS Data
Kassing, Andrew; Gibbons, Ronald B. (National Surface Transportation Safety Center for Excellence, 2024-09-16)
Roadway departures encompass a particularly dangerous subset of driving events during which a vehicle either crosses the centerline or edge line or otherwise leaves the lane of travel. Each year, roadway departure crashes account for roughly 50% of all fatal crashes reported to the National Highway Traffic Safety Administration. This study’s goal was to evaluate the factors contributing to roadway departure events. The data set used was naturalistic driving data collected during the Second Strategic Highway Research Program (SHRP 2). The full data set consisted of 28,937 driving events with information spanning 70 variables that characterized each event. For all events provided to the research team, reductionists categorized each as either a safety-critical or baseline event and reviewed their variable levels. Analyses determined that numerous driver behaviors and roadway environment elements influenced the odds and severity of roadway departure events. Overall, 80% of the adverse driver behavior categories were found to significantly increase the odds of roadway departure crashes at intersections. Drivers who were intoxicated, cut turns, took turns too widely, or were speeding were significantly overrepresented in roadway departure events. The prevalence data indicated that distracted driving was a very common behavior regardless of segment type or event outcome. More specifically, among the baseline events, drivers performing secondary tasks associated with distracted driving were very common. Throughout the baseline sample (i.e., no incidents), drivers were more likely to be distracted by tasks such as device usage, passenger interaction, personal hygiene, eating, smoking, etc., than performing no secondary tasks at all. Across all roadway departure events, adverse driver behavior was observed in 82% of incidents on tangent segments and 93% of incidents at intersections. Roadway environment changes, such as in pavement surface condition, were found to influence roadway departure frequency and severity. Analyses suggested that maintaining the skid resistance of roadway surfaces, even during inclement weather, may be essential to reducing the occurrence and severity of roadway departure events. Furthermore, roadway departures were overrepresented in incidents where sunlight, glare, headlamps, precipitation, vehicles, or infrastructure were obstructing the driver’s view. Researchers noted that visibility obstructions were significantly more common at intersections than tangent segments.
Risk Factors Re-evaluation with Bayesian Network Using SHRP 2 Data
Han, Shu; Guo, Feng (National Surface Transportation Safety Center for Excellence, 2024-09-11)
Traffic safety is a complex system influenced by numerous factors, including human behavior, road design, vehicle technology, and environmental conditions. Each of these factors can impact the safety of the transportation system in unique ways, and all factors could interact with each other in complex ways. The goal of this study was to evaluate the joint contribution of multiple risk factors to traffic safety by examining the interactions among different factors. This study considered 24 potential risk factors that reflect different perspectives in the analysis, including driver demographics, driving behavior, environmental conditions, road characteristics, traffic context, vehicle kinematics within a 5-second window of each event, and cell phone ban policies. There were two aspects to this study: first, it explored the relationships between traffic safety risk factors using unsupervised learning models with data from the Second Strategic Highway Research Program Naturalistic Driving Study. Second, with supervised learning models, the study developed a robust data-driven Bayesian network model, evaluated impacting risk factors, quantified their corresponding importance on driving risk, and consequently identified high-risk scenarios.
Development of a Naturalistic Observer-Based Rating of Near-Crash Severity in Naturalistic Driving Data
McClafferty, Julie A.; Walker, Stuart (National Surface Transportation Safety Center for Excellence, 2024-09-04)
The analysis of safety-critical events, including crashes and near-crashes, from naturalistic driving studies has proven extraordinarily useful in guiding transportation safety policies, transportation technology, and transportation infrastructure. Near-crashes, which are much more common than crashes, have the potential to answer many research questions. However, they are difficult to define, and their severity is difficult to rate. By definition, there is no impact to measure in a near-crash and therefore no injury or property damage to assess. Near-crashes cover a range of scenarios, and perceptions of severity can vary greatly depending on the person experiencing or perceiving them. From a research perspective, this variability makes near-crashes challenging. Severe near-crashes may be considered most similar to crashes and serve as better surrogates than less severe near-crashes, but less severe near-crashes are still very different from “normal” driving and are still relevant to policy, technology, and infrastructure development research. In this effort, an observer-based, naturalistic near-crash severity rating protocol was developed and tested to help researchers produce near-crash event data effectively and reduce associated variability. Goals included producing a protocol that can (1) produce consistent and meaningful ratings, (2) be incorporated effectively and efficiently into the standard primary event assessment, (3) be implemented by trained data reductionists with access to video and basic kinematic data charts, (4) be applied without complex models, calculations, or statistical modeling, and (5) mirror the existing crash severity scale in implementation and conceptualization. A key concept in this work was that of conflict urgency. There is no clear answer about how urgency can or should be observed or measured in naturalistic data, especially in non-crash scenarios. It is clear, however, that the concepts of collision imminence (a sense of conflict timing) and potential crash severity (related to possible damage and injury) are important factors. Thus, an additional goal was to achieve a balance between actual kinematics, predictive outcomes, and perceived subjective risks. Operational definitions, associated research protocols, and reference guides were developed for four levels of near-crash severity ranging from Critical Severity to Lower Severity. These are documented in the appendices. At their core, the definitions are based on objective metrics such as relative speed, time-to-collision, and type of conflict, but with room for subjective assessments. An iterative approach was used in the development of these definitions, and this included assessments to evaluate interrater reliability. Results indicated that reference materials and training improve interrater reliability.
Aerial Traffic
Viray, Reginald; Saffy, Joshua; Mollenhauer, Michael A. (National Surface Transportation Safety Center for Excellence, 2024-08-30)
This report documents a significant advancement in work-zone safety through the strategic integration of aerial drone technology and machine vision software. It summarizes the project’s phases: Technical Assessment and Procurement, System Integration and Validation, and Deployment Assessment. The Technical Assessment and Procurement phase led to the selection of Smartek ITS’s DataFromSky product for its unique real-time processing capabilities of aerial drone video, making it superior to other commercial offerings. The System Integration and Validation phase ensured that the video streams, whether real-time or recorded, were processed effectively for varying roadway scenarios, including work zones and intersection monitoring. Accompanying development work included a user-defined data interface with the capability to trigger intruding vehicle alarms. The Deployment Assessment phase confirmed the system’s precision, with object detection up to 150 meters and sub-500 millisecond latency in relaying data for real-time alerts. Despite some GPS data discrepancies due to wind-induced drone movements, the system showed promise in controlled and real-world environments. Overall, the project acquired and validated the system’s functionality, with successful tests on live and recorded video feeds, software video processing, and real-time data transmission, culminating in the development of a robust intruding vehicle alarm mechanism. The system demonstrated great potential for deployment across various Virginia Tech Transportation Institute research initiatives, setting a precedent for future work in enhancing work zone safety.
Highly Automated Ridesharing: Implications of Novel Seating Configurations and Seatbelt Use
Anderson, Gabrial T.; Radlbeck, Joshua (National Surface Transportation Safety Center for Excellence, 2024-08-29)
Upcoming novel vehicle designs, such as vehicles equipped with Level 4 (L4) driving automation features, are intended to be used as rideshare automated vehicles (RAVs). In vehicles without L4 automation features, drivers typically receive all regulated telltales, indicators, and alerts intended to encourage seatbelt use. In a vehicle with L4 features, a driver is not present, meaning these alerts need to be presented directly to the occupants. Understanding occupant behavior in a novel vehicle design can inform effective methods for encouraging seatbelt use. Thirty participants rode in an RAV on a closed test track. Participants rode in groups of three on a 5-minute route at speeds up to 15 mph. There were two benches with three seats per bench: a forward-facing row and rear-facing row. Human-machine interfaces were placed overhead for each seating position; these included a novel seatbelt reminder alert (SBR) that would chime if the passenger was unbuckled when the vehicle started to move. If participants remained unbuckled, the SBR would last 10 seconds before doubling in tempo until the participant in that seating position was buckled. A post ride survey was administered to capture participant opinions of their experience. Where appropriate, results from a previous proprietary study of 60 single riders were compared to the current study. Group riders were significantly more likely to buckle before vehicle movement than single riders. Group riders were more likely to sit in the rear-facing row of seats than the single riders. Across all participants, females were more likely to buckle before vehicle movement than males. For those participants who received the SBR alert (i.e., those who were not buckled before vehicle movement), the majority began buckling within the first ~7-seconds of SBR presentation. Results suggest riding context can impact seatbelt use. Although a majority of participants preferred to face forward, group dynamics forced some participants to sit backwards in all rides. Further research is needed to understand the impact of riding environment, other rider populations, and SBRs on buckling behavior in RAVs.
Investigation of ADAS/ADS Sensor and System Response to Rainfall Rate
Cowan, Jonathan B.; Stowe, Loren (National Surface Transportation Safety Center for Excellence, 2024-08-23)
Advanced driver assistance systems (ADAS) and automated driving systems (ADS) rely on a variety of sensors to detect objects in the driving environment. It is well known that rain has a negative effect on sensors, whether by distorting the inputs via water film on the sensor or attenuating the signals during transmission. However, there is little research under controlled and dynamic test conditions exploring how rainfall rate affects sensor performance. Understanding how precipitation may affect the sensor’s performance, in particular the detection and state estimation performance, is necessary for safe operation of the ADAS/ADS. This research strove to characterize how rainfall rate affects sensor performance and to provide insight into the effect it may have on overall system performance. The team selected a forward collision warning/automatic emergency braking scenario with a vehicle and surrogate vulnerable road user (VRU) targets. The research was conducted on the Virginia Smart Roads’s weather simulation test area, which can generate various simulated weather conditions, including rain, across a test range of 200 m. The selected sensors included camera, lidar, and radar, which are the primary sensing modalities used in ADAS and ADS. The rain rates during testing averaged 21 mm/h and 40 mm/h. Overall, the data backed up the expected trend that increasing rainfall rate worsens detection performance. The reduced detection probability was most prominent at longer ranges, thus reducing the effective range of the sensor. The lidars showed a general linear trend of 1% reduction in range per 1 mm/h of rainfall with some target type dependence. The long-range and short-range cameras show at least a 60% reduction in detection range at 40 mm/h. The object camera, which only detected the vehicle target, showed better performance with only a 20% reduction in range at 40 mm/h, which may be due to the underlying ADAS specific detection model. For vehicles, the radars typically showed a linear drop in detection range performance with an approximately 20% reduction in range at 40 mm/h rainfall rate. The VRU target showed a larger decrease in detection range compared to the vehicle target due likely to the smaller overall signal the VRU target returns.
Assessing the Impact of Disability on Drivers’ Equitable Use of Advanced Driver Assistance Systems (ADAS): A Literature Review
Stulce, Kelly E.; Antin, Jonathan F. (NSTSCE, 2024-08-22)
The growing prevalence of advanced driver assistance systems (ADAS) in the U.S. passenger fleet promises increased mobility and enhanced safety outcomes for all drivers, but particularly for disabled drivers, a group that comprises 11.9% of the driving population (U.S. Bureau of Labor Statistics, 2021). For ADAS to realize their full potential, stakeholders need to consider the difficulties associated with ADAS use by disabled drivers as well as the potential benefits. To support this reckoning, the authors reviewed the extant literature to discover emerging themes and to identify gaps in the literature. We then synthesized these results into a proposed road map for future work that addresses the challenges of using ADAS to enhance mobility and improve safety for all drivers, including those who are disabled. Our review of the literature reveals gaps that point the way forward for further work that will support the optimal implementation of ADAS to compensate for disability-induced driving performance deficits. Specifically, our gap analysis and research road map suggest that this work should begin with using subjective methodologies (e.g., focus groups, interviews, and surveys) to learn from the disabled driver community in a manner that centers these individuals. Such research should yield results that more authentically capture the experience (or lack thereof) of disabled individuals driving with and making use of ADAS. Additionally, longitudinal research is necessary to support extended observation of real-world ADAS use by disabled drivers across driving environments and their disability-related functional states, which are often transient
Pedestrian-Vehicle Interaction Data Dictionary and Analysis Protocol: Design, Development, and Pilot Application on Two Naturalistic Driving Databases
Mabry, J. Erin; Wotring, Brian; McClafferty, Julie A.; Soccolich, Susan A.; Boucher, Ben (2024-08-09)
Pedestrian conflicts with vehicles continue to be a serious problem in the United States. Unlike vehicle occupants, pedestrians do not have the protection of airbags, a steel structure surrounding them, or other vehicle safety technologies; their resultant vulnerability places them at higher risk of injury when involved in a traffic crash or conflict. Examining pedestrian behavior in a variety of settings and interaction severity levels supports research goals to improve pedestrian safety. The goals of this study were to: 1. Create an inclusive dictionary of video data analysis variables that details and describes interactions between pedestrians and motorized vehicles; and 2. Develop and pilot test a Pedestrian-Vehicle Interaction Data Reduction Protocol (PVIP) using existing naturalistic driving datasets. Implementing the PVIP confirmed that coding elements related to the interaction response between the pedestrian and vehicle from each perspective, and according to the three epoch stages (i.e., leading up to, during, and following the interaction), was critical for characterizing the entire interaction with consideration of all viewpoints. Pedestrian behaviors, locations, communication strategies, distractions, impairments, and glance behaviors were observed and coded at each stage of the epoch to account for behavioral, sensory-related, and positional changes of the pedestrian occurring over the course of the interaction that could impact the outcome. Similarly, coding the vehicle maneuver, driver behaviors, driving-related tasks, and glance behavior across the interaction epoch may be important elements to consider for pedestrian safety. Pedestrian location across the epoch was also an important variable in the pilot analyses. This study is the first of its kind to design, develop, and systematically apply a comprehensive, video-based, and pedestrian-centric data reduction protocol to NDS data to explore and describe interactions between pedestrians and vehicles for better understanding of pedestrian safety. The output of this project is a comprehensive and systematic PVIP that can be used to characterize pedestrian-vehicle interactions and behaviors. The protocol is organized so that researchers may select questions or groups of questions that are applicable for their specific research objectives in an à la carte fashion to create a focused protocol that fully explores a pedestrian-vehicle scenario using available data.
Creating a Dataset of Naturalistic Ambulance Driving: A Pilot Study of Two Ambulances
Valente, Jacob T.; Terranova, Paolo; Perez, Miguel A. (National Surface Transportation Safety Center for Excellence, 2024-08-02)
Motor vehicle collisions (MVCs) are an everyday occurrence in the United States. This pressing transportation and health care topic affects millions of citizens each year, and in many cases may result in fatality or lifelong injury complications. Despite best efforts, and notable success, to improve the frequency and severity of MVCs, these events are still a prevalent issue. In the wake of an MVC, crash occupants rely on emergency responders to quickly respond to the scene, control hazards, and administer necessary medical care. Efficiency within the emergency response event, to an MVC or some other medical care need, is contingent on a properly working transportation system, allowing emergency medical services (EMS) to travel to and from scenes both quickly and safely. Previous research has revealed that complex interactions with other road users not only hinders emergency response efficiency, but often results in hazardous and dangerous interactions on roadways. To capture these complex interactions from a firsthand perspective, this report details a naturalistic driving study that involved two ambulances and the subsequent dataset that was generated, which is the first of its kind. A custom data acquisition system was used to collect four external and three internal video perspectives on each vehicle, with continuous vehicle data that included vehicle speed, GPS location, and emergency system activation (i.e., emergency light or siren use). Following data collection, the dataset was summarized in the context of each participating agency, the consented drivers, trip type (emergent vs. non-emergent), trip duration, trip distance, and the time of day that the trip took place. The dataset was also processed through a map-matching algorithm that utilized the collected GPS data to provide additional context, including posted speed limit road classification. Finally, the dataset was subsampled to assess and interpret other road user behavior during emergent trips. The work outlined in this report serves as the foundation for additional research that could be leveraged from this dataset, as this dataset is intended to support the inquiry of future research questions within the scope of emergency vehicle operation and transportation. Additionally, some findings of this study and their implications apply beyond the scope of emergency MVC response and may be related more broadly to emergency response for all first responders and emergency events.
Fiber Sensing: Real-time, Long-distance Traffic Monitoring
Viray, Reg; Mollenhauer, Michael A.; Chen, Yuheng; Huang, Ming-Fang (2024-07-26)
The report presents the pilot testing of NEC’s distributed fiber optic sensing (DFOS) and artificial intelligence behavior detection technologies on the Virginia Smart Roads. The intent was to explore the viability of DFOS through existing fiber installations for traffic management and road safety, potentially superseding traditional methods like cameras or radar. Key use cases for DFOS include identifying wrong-way driving, traffic queues, emergency stops, and other non-standard driving behaviors, which are all critical for traffic management and safety. NEC collaborated with the Virginia Tech Transportation Institute to install fiber optics along a highway section conforming to Virginia Department of Transportation (VDOT) standards and executed orchestrated driving scenarios to test detection capabilities under various traffic and weather conditions. The results of these tests demonstrated that DFOS could effectively detect wrong-way driving across different traffic densities. However, challenges remain with vehicle count and speed accuracy in high-traffic situations. In addition, the DFOS hardware solutions would need to be deployable in roadside cabinets with limited environmental control to be viable for deployment across VDOT’s roadway system. The project underscores the strategic advantage of using existing network infrastructures for monitoring and suggests that while the NEC system shows promise in certain applications, further refinements are needed for handling complex traffic scenarios.
Evaluation of the Interaction of Driver Behavior Based on SHRP 2 NDS Data
Guo, Feng; Fang, Youjia (National Surface Transportation Safety Center for Excellence, 2024-07-25)
Driver behavior constitutes a significant factor contributing to traffic crashes. This study examines the relationship between three primary categories of driver behavior—driver distraction, driving errors, and driver impairment—utilizing data from the Second Strategic Highway Research Program Naturalistic Driving Study. The findings offer concrete evidence of interactions among these behaviors, indicating that the presence of one behavior significantly influences the likelihood of another and the occurrence of two high-risk behaviors amplifies crash risk. This study underscores the multifaceted nature of driver behavior and its profound impact on road safety. The findings can provide crucial information for driver education programs, safety countermeasure development, and advanced driving assistance systems.
Current Implementations of Over-the-Air Updates
Bowden, Zeb; Chandler, Jacob (National Surface Transportation Safety Center for Excellence, 2024-07-22)
Over-the-air (OTA) updates have become a significant aspect of the future development of connected vehicles. OEMs can address safety and security concerns, as well as provide a better customer experience, by delivering software improvements to vehicles faster and without requiring a visit to a service center. OTA updates will enable OEMs to address security, safety, and performance concerns much faster, at a lower cost, and allow delivery of updates in a manner more convenient to vehicle owners. From a cybersecurity perspective, this is important, as the number of security vulnerabilities in vehicles continues to grow (Sibros, 2022). Additionally, recalls are increasingly being addressed by software updates, potentially resolving recall issues faster. From March 2022–March 2023, 8.5 million vehicle units were recalled due to software-based or electrical components (Sibros, 2022; Vehicle Recalls, 2023). Advanced driver assistance systems and automated driving systems are continually being developed and improved. OTA updates are an enabling technology that allows improvements to these technologies to be realized on vehicles at a much more rapid pace than in the past. The objective of this report is to understand the capabilities, challenges, and characteristics of implementations of OTA updates in vehicles. This report summarizes and details key findings of OTA updates in current vehicles and focuses on the potential challenges in adopting this new advancement in technology.
Temporary Disability: Development of an Evaluation Framework
Werner, Alice; Perez, Miguel A. (National Surface Transportation Safety Center for Excellence, 2024-07-17)
The purpose of this work was to assess the impact of temporary disability on the ease and effectiveness of automobile driving tasks. The main objective of this work was to generate a framework to evaluate the extent to which a given temporary disability, caused by a given primary condition and associated with different symptoms, affects individual driving tasks at any given timepoint following the onset of the temporary disability. Injuries and medical conditions can limit driving ability—from chronic conditions to temporary conditions. Ultimately, the disability produced by these conditions may result in driving restrictions, reliance on compensatory driving techniques, or both. This can be particularly onerous for individuals that live in areas not well served by public transportation. In an ideal analysis of temporary disability, translating temporary disability into its disruptive impacts on components of the driving task (i.e., functional driving domains) should identify affected driving tasks, the degree of impact on driving tasks, and necessary/useful compensatory driving techniques and vehicle modifications, as well as inform return-to-drive guidelines that are not overly conservative. In reality, however, connections between temporary disability symptoms and their effects of specific driving tasks are either incomplete or completely nonexistent. Given this background, the current effort aimed to propose a preliminary framework and model for the systematic evaluation of effects of temporary disability on functional driving domains through a semi-hierarchical task analysis and identification of different elements of the temporary disability’s manifestation that should be considered in such a framework. A cascading model was developed that first associates a temporary disability event with specific temporary disabilities and then qualifies those disabilities into a set of affected capabilities. Finally, the affected capabilities are translated into impacts on specific driving tasks. The output from the model allows the user to understand the hindrance that a temporary disability event may create on someone’s driving so that reasonable return-to-drive guidelines may be generated and/or patients can self-regulate their driving to accommodate for temporarily diminished skills.
Trip Purpose Based on Naturalistic Driving Data
Costa, Rufina (National Surface Transportation Safety Center for Excellence, 2024-07-09)
Understanding trip purpose is important as it provides valuable information for driver behavior research, traffic management, and urban transportation planning. The primary challenge in analyzing trip purpose is upholding the constraints on personally identifiable information (PII) and ensuring confidentiality. Privacy protocols are particularly stringent for the Second Strategic Highway Research Program 2 (SHRP 2) regarding data at trip origins and destinations. Researchers have access to specific origin and destination data only within a secured enclave. Researchers receive only truncated trip data when it is exported, limiting their ability to identify specific origins and destinations that could indicate home or work-related trips. In this study, we successfully identified and labeled home and work-related trips without retaining any geographical information. This approach allows the use of the derived information in further analyses without depending on any PII, thus preserving individual privacy and complying with data protection regulations. The study utilized geospatial data, including latitude and longitude, as well as trip duration, Google API labels, driver demographic information, and trip date and time. To identify residential trips, we considered only those trips that the Google API labeled as residential destinations and that had an interim of at least 8 hours before the next trip. Geofencing was then employed across all destinations by each participant. The destination that was most frequently visited was identified as the resident’s home location. Work-related trips were identified for participants who reported full-time employment and had trips with a 2-hour or longer stay at destinations not classified as residential according to the Google API. Work-related trips were defined as those involving 16 consecutive visits to the same destination per month, covering at least 70% of the study time for the participant. Further analysis was conducted to link trips together, enabling the identification of round trips that either involved a single purpose or multiple destinations. Additionally, trips related to home and work that involved safety-critical events were specifically identified. Finally, trips that occurred 100 miles away from home were also analyzed, with a focus on examining the patterns based on the day of the week when each trip took place. The study used a novel method to identify home and work-related trips and successfully identified 1,546,798 home trip destinations from 2,937 drivers and 48,813 trips defined as work trips from 109 participants. These are now available for researchers as part of the SHRP 2 export, enhancing the scope for further academic and practical analyses.
Traffic Sign Characteristics for Machine Vision Safety Benefits
Kassing, Andrew; Gibbons, Ronald B.; Li, Eric; Palmer, Matthew; Hamen, Johann; Medina, Alejandra (National Surface Transportation Safety Center for Excellence, 2024-07-03)
Machine vision has become a central technology for the development of automated driving systems and advanced driver assistance systems. To support safe navigation, machine vision must be able to read and interpret roadway signs, which provide regulatory, warning, and guidance information for all road users. Complicating this task, transportation agencies use a large variety of signs, which can have significantly different shapes, sizes, contents, installation methods, and retroreflectivity levels. Additionally, many environmental factors, such as precipitation, fog, dew, and lighting, also affect the visibility and legibility of roadway signs. Understanding how environmental factors and sign conditions affect machine vision performance will be important for transportation agencies to maximize the technology’s safety benefits. Research began by conducting a literature review cataloguing current research concerning roadway sign and visual performance, vehicle vision systems, and sign significance for automated driving. Information and insight gained during the literature review process informed the design and system development of data collection systems. Field data collection was then performed over the course of 3 months in late spring to early summer in 2021. Simultaneously, sign data were harvested using Google Street View and mapped using ArcGIS. Data collected during the experimental trips were then reduced and carefully prepared for analysis. Researchers conducted a thorough data analysis, particularly looking at sign location, viewing distance, sign color, font size, sun position, and illumination, to assess the impact of many environmental and infrastructure factors on the legibility of sign characters. Results showed that blue and brown signage with white legend text provided the best chance of sign character legibility during the daytime; sign characters were easy to read during the day at all three experimental distances (200, 400, and 500 ft), with small characters becoming less legible as view distance increased; daytime legibility decreased as light levels decreased; sign images captured at nighttime illumination levels had poor legibility results; sign characters on overhead signage were found to be more legible and are expected to be identified at a higher rate by vehicle vision systems; and vehicle vision systems should use a high-quality camera capable of taking pictures at night without motion blur.
Real-time Risk Prediction Using Temporal Gaze Pattern and Evidence Accumulation
Yang, Gary; Kaskar, Omkar; Sarkar, Abhijit (National Surface Transportation Safety Center for Excellence, 2024-06-26)
Driver gaze information provides insight into the driver’s perception and decision-making process in a dynamic driving environment. It is important to learn whether driver gaze information immediately before a safety-critical event (SCE) can be used to assess crash risk, and if so, how early in the crash we can predict the potential crash risk. This requires a thorough understanding of the temporal pattern of the data. In this project, we explored multiple key research questions pertaining to driver gaze and its relation to SCEs. We first looked at how temporal gaze patterns in SCEs like crashes and near-crashes are different from baseline events. Then we extended this analysis to understand if the temporal gaze pattern can indicate the direction of the threat. Finally, we showed how contextual information can aid the understanding of gaze patterns. We used traffic density, positions of traffic actors, and vehicle speed as key factors that can influence the temporal nature of gaze and impact safety. This study introduces a real-time crash risk prediction framework leveraging deep learning models based on driver gaze and contextual data. The analysis utilizes the Second Strategic Highway Research Program (SHRP 2) Naturalistic Driving Study, preprocessing it to extract temporal driver gaze patterns and contextual data for two distinct subsets of data. Dataset 1 comprises one-dimensional temporal driver gaze patterns only (named “SHRP2_Gaze”). Dataset 2 includes additional contextual information such as vehicle speed, distance to lead vehicle, and level of service (named “SHRP2_Gaze_Context).” Dataset 1 was used to explore the feasibility of predicting crash risk solely based on driver gaze data. Dataset 2 was applied to assessing the potential for early crash warnings by analyzing both driver gaze patterns and contextual data. For SHRP2_Gaze, four deep learning models (long short-term memory [LSTM], one-dimensional convolutional neural network [1D CNN], 1D CNN+LSTM, and 1D CNN+LSTM+Attention) were trained and tested on various inputs (Input 1: 20 seconds from five zones; Input 2: different lengths from five zones; and Input 3: 20 seconds from 19 zones), considering different lengths and zones of driver gaze data. The 1D CNN with XGBoost exhibited superior performance in crash risk prediction but demonstrated challenges in distinguishing specific crash and near-crash events. This portion of the study emphasized the significance of the temporal gaze pattern length and zone selection for real-time crash risk prediction using driver gaze data only. With SHRP2_Gaze_Context, the 1D CNN model with XGBoost emerged as the top-performing model when utilizing 11 types of 20-second driver gaze and contextual data to predict crash, near-crash, and baseline events. Notably, vehicle speed proves highly correlated with event categories, showcasing the effectiveness of combining driver gaze and contextual data. The study also highlighted the model’s ability to adapt to early crash warning scenarios, leveraging different 2-second multivariate data without an increase in computing time. The deep learning models employed in this study outperform traditional statistical models in distinguishing features from driver gaze and contextual data. These findings hold significant implications for accurate and efficient real-time crash risk prediction and early crash warning systems. The study’s insights cater to public agencies, insurance companies, and fleets, providing valuable understanding of driver behavior and contextual information leading to diverse safety events.
Human-Machine Interface Review: A Comparison of Legacy and Touch-Based Center Stack Controls
Anderson, Gabrial T.; Antona-Makoshi, Jacobo; Klauer, Charlie (National Surface Transportation Safety Center for Excellence, 2024-04-19)
The current study investigated the effect of center stack design on driver distraction. Replacing physical center stack controls with touchscreens is an emerging trend in automotive design. This design decision requires a driver to take their eyes off the forward roadway to interact with a touchscreen center stack, as there is no tactile feedback like touching physical controls. Multiple resource theory (Wickens, 2004) suggests that performing dual tasks (i.e., driving and touchscreen interaction) that compete for similar resources (i.e., visual attention and manual input) can degrade performance on both tasks. It is important to understand the impact of touchscreen controls on driver distraction to ensure safe human-machine interface design. Data from legacy vehicles with physical center stack controls were extracted from the Second Strategic Highway Research Program, an NDS focusing on driver behavior over time in personal vehicles. Data from modern vehicles with touchscreen designs were extracted from the Virginia Connected Corridor 50 Elite Vehicle NDS and Virginia Tech Transportation Institute Level 2 NDS, both focusing on driver behavior in personal vehicles equipped with SAE Level 2 (L2) driving automation features. Twenty-second events that had a center stack interaction (CSI) and minimum speed of 35-mph or greater were selected from each dataset. For the modern vehicle dataset, L2 system status was coded for each event as L2 active or L2 inactive, and task type was coded as visual or visual-manual. The legacy vehicle dataset only had visual-manual CSIs. Driver distraction was defined as eye glances towards the center stack (eyes on center stack; EOCS) during the 20-second event. EOCS was split into total time, mean time, single longest glance, number of glances, and glances over 2 seconds in duration. Total time on task was recorded for the modern vehicles. Results suggest that CSIs with modern vehicle touchscreens have higher EOCS compared to legacy vehicle physical controls. Notably, these differences are even more pronounced when comparing visual-manual CSIs (e.g., adjusting climate control) across display type. Modern vehicle CSIs were also more likely to include glances over 2 seconds compared to legacy vehicle CSIs. Within the modern vehicle dataset, all EOCS metrics (except number of glances), time on task, and glances over 2 seconds were significantly higher when L2 systems were active versus inactive. Visual-manual CSIs were higher for all variables compared to visual CSIs. Glances over 2 seconds were more likely when L2 systems were active for all visual CSIs, but not for visual-manual CSIs. Touchscreen center stack designs are shown to be more distracting than legacy designs comprised of physical controls. When L2 systems are active, CSIs are more distracting than when L2 systems are inactive. Although display type has been shown to have a distracting effect, comparison of specific tasks (e.g., adjusting climate controls) is needed to represent true differences in driver distraction, as more complex tasks that are possible in modern vehicles versus legacy vehicles could contribute to the results of the current study.
VTTI L2 Naturalistic Driving Study: A Self-funded Effort to Capture L2 Feature Use Landscape
Hankey, Jonathan M.; Bedwell, Kaitlyn; Wiersma, Ethel; Stulce, Kelly E.; Perez, Miguel A. (2024-06-20)
The introduction of advanced driver assistance systems (ADAS) into the vehicle fleet continues to accelerate. In the past few years, that introduction has started to permeate the non-luxury vehicle segment, greatly increasing the availability of these technologies to a wide segment of the driving population. The implementation and capabilities of these systems, however, can vary widely across vehicle makes and models, which makes it imperative to have recent data that supports the study of driver adaptations in response to ADAS. While this data can take several forms, naturalistic driving data has proven to provide a flexible means of assessing real-world driver and system performance across a variety of domains and is well suited to understanding ADAS usage. The main objective of the VTTI L2 NDS data collection effort was to create a robust naturalistic driving dataset containing critical information about vehicles with ADAS. As ADAS continue to rapidly evolve and become more readily available in the vehicle fleet, it is essential to understand how these systems are being used and, in some instances misused, by drivers. This knowledge will facilitate the understanding of the safety, performance, and convenience benefits that these systems may offer drivers, along with unintended consequences from the use of these systems. The VTTI L2 NDS is available to help address a wide array of research questions that pertain to the usage of ADAS, along with traditional queries suited to NDS data, in a relatively modern fleet.
Truck Driver Compensation and Crash Risk
Guo, Feng; Xiang, Lanxin (National Surface Transportation Safety Center for Excellence, 2024-06-17)
Compensation methods might affect commercial drivers’ behavior and subsequently the possibility of being involved in a crash. This study investigated the relationship between various compensation strategies for commercial truck drivers and their associated risk of involvement in crashes. A nested case-control approach was adopted. Compensation data was sourced from the follow-up survey of the Commercial Driver Safety Risk Factors study. Upon a case-crash being identified, up to five consented non-crash drivers were contacted to complete a follow-up survey. The analysis employed standard logistic regression and two conditional logistic regression (CLR) models. The CLR models were stratified based on the date of the crash (reference date) and the years of commercial vehicle driving experience to control for temporal and experience-related factors that could confound the relationship between compensation methods and crash risk. To adjust for multiple comparison issues, Tukey’s tests were employed. The results indicate a significant difference in crash rate among driver compensation types. Drivers in PayPerMile-PayPerLoad are associated with a higher crash risk compared to PayPerHour, with odds ratios (ORs) of 3.512, 2.304, and 2.853, in three models, respectively (p < 0.05). Similarly, PayPerMile-PayPerTrip showed increased risk compared to PayPerHour, with ORs of 3.207, 2.260, and 2.548, respectively (p < 0.05). Tukey’s tests confirm significant differences among compensation methods, particularly between PayPerMile-PayPerLoad and PayPerHour, PayPerMile-PayPerLoad, and PayPerLoad, and within PayPerMile-PayPerLoad and PayPerMile. The study suggests a potential link between compensation methods and crash risk, highlighting the need for further research to pinpoint which compensation strategies significantly affect driving safety.

Browse

Recent Submissions