To improve safety, the SHRP2 Naturalistic Driving Study is collecting data on what happens when people crash, experience a near-crash -- or drive without incident.
Excessive speed, poor gap judgment, inept evasive action, in-attention, distraction -- drivers are human and make mistakes.
For a long time, the transportation safety community has known that driving behavior is a leading factor in roadway crashes. In the report, Tri-Level Study of the Causes of Traffic Accidents, released by the U.S. Department of Transportation (USDOT) in 1979, human factors are cited as the probable cause of 93 percent of crashes. But the gap remains wide between knowing about unsafe driving behavior and being able to do something about it through countermeasures such as engineering, enforcement, public information, and education.
"We still have a long way to go before we completely understand driver distraction, drowsy driving, and speeding behavior,” says Richard Compton, director of USDOT’s National Highway Traffic Safety Administration’s (NHTSA) Office of Behavioral Safety Research. "We definitely need a better understanding of why drivers underestimate the risks involved in these behaviors. Driver distraction, for example, entails different types of risks in terms of manual, visual, and cognitive forms of distraction.”
Reducing crashes by just 1 percent would prevent 330 deaths and save approximately $2 billion annually in medical expenses, according to a 2000 NHTSA report. In addition, crashes are a leading cause of traffic congestion. In 2011, congestion caused urban Americans to travel 5.5 billion hours more and to purchase an extra 2.9 billion gallons of fuel for a congestion cost of $121 billion, according to the 2012 Urban Mobility Report of the Texas A&M Transportation Institute. Today, roads are more congested than ever, cars have more bells and whistles, and drivers have cell phones and other distractions at their fingertips.
"The drivers of today are very different [from] the drivers of yesterday,” points out David Shinar, who serves on the Strategic Highway Research Program’s (SHRP2) Safety Technical Coordinating Committee and is a professor of human performance management at Ben-Gurion University of the Negev in Israel. "If the only thing we do is drive, we think we’re wasting time. We’ve gotten spoiled by all the available entertainment systems in the car.”
These factors certainly influence driving behavior -- but how exactly and to what degree? Having data that show what happens when people crash, when they experience a near-crash, and when they drive without incident could spur significant improvements in highway safety. The second SHRP2 Naturalistic Driving Study is aimed at doing just that.
Studying Driving Behavior
Established in 2006, SHRP2 aims to accelerate the renewal of the Nation’s highways; improve highway safety; advance reliable travel times; and provide highway capacity in support of U.S. economic, environmental, and social goals. The program is administered by the Transportation Research Board (TRB), a division of the National Academies, under a memorandum of understanding with the Federal Highway Administration (FHWA) and the American Association of State Highway and Transportation Officials.
SHRP2 is supporting research to evaluate the underlying causes of highway crashes and congestion and to address the role of driving performance and behavior in traffic safety. Most of the $67 million allocated for safety research is being spent on the Naturalistic Driving Study. According to TRB, this is the largest study of driving behavior ever conducted in terms of participants and data that will be generated. About 3,100 volunteer drivers at data collection sites in six States are participating. The drivers are allowing researchers to install cameras and sensors in their vehicles to deliver data captured in real time, making it available to scientists and engineers.
Data collection began in late 2010 and will end in November 2013. The data collected will exceed 4 petabytes, or twice the size of all the information contained in every academic research library in the United States in 1997. The study team expects the data to remain useful to transportation safety researchers and others for up to 30 years, providing an abundance of information regarding driving behavior, lane departures, and intersection activity.
"A long list of research topics has been identified, and interested researchers are eager to use these data to improve highway safety,” says Robert Skinner, TRB’s executive director. "They believe we can learn things we didn’t know before that will enable us to develop countermeasures, design vehicles better, and design more forgiving roadways. The challenge is to have the data available to them in formats that are understandable and usable.”
In collaboration with Battelle Memorial Institute and the University of Michigan Transportation Research Institute, Virginia Tech Transportation Institute (VTTI) designed the field study and defined requirements for a data acquisition system that collects kinematics (information about a vehicle’s motion, including velocity and acceleration data) and video of naturalistic (or real-world) driving behavior. Each of the 1,950 vehicles used for the study is equipped with this system, which will yield approximately 3,900 vehicle-years of data over 2 years. The system includes four video cameras, velocity and acceleration sensors, GPS, forward radar, an incident button, a light sensor, and a passive alcohol sensor. Machine vision tools to track lane fidelity and an eyes-forward monitor also are part of the system.
The VTTI researchers are storing and managing a databank that contains data on demographics, vehicle inventories, driver assessments, and crash investigations. To capture a range of geographic, road type and usage, State law, and weather factors, researchers are collecting data at locations in six States: Florida, Indiana, New York, North Carolina, Pennsylvania, and Washington. As the technical contractor for the study, VTTI is providing coordination and quality control, as well as oversight of the contractors that are gathering and storing data for future analysis projects.
"Data quality assurance is a big part of this,” explains VTTI research scientist Jonathan Antin. "We verify that the vehicles are collecting data and doing what they’re expected to do. Through an automated health check, the data acquisition system self-assesses certain aspects of its performance and functioning. It transmits that information by cellular network back to us, and we begin to interpret it.”
Participants and Trips
Age and gender were key criteria in recruiting participants in order to assemble a robust mix of young, middle-age, and senior drivers from both sexes. Volunteers took a series of assessments of driving-related skills and attributes, such as visual perception, visual-cognitive ability, psychomotor ability, physical ability, health and medication status, psychological factors, driving knowledge, and driver history. At enrollment, each volunteer in the study signed an informed consent. The institutional review boards of the National Academy of Sciences and other contracting agencies approved all activities required of the subjects.
Obtaining data about drivers, vehicles, and roads is necessary to answer questions about what influences the risk of being in a collision. Researchers need to see what drivers see and where they are looking. Key data collected include speed, distance from the car ahead, acceleration, steering and pedal action, seatbelt use, geographic location, and vehicle characteristics and performance. Onboard computers encrypt and store data obtained from the data acquisition system, then upload the data to a central repository.
Much of the study data consist of logged trips and road data. A trip begins just after a vehicle’s ignition is turned on and ends when it’s turned off. To enable data users to identify trips of interest, the researchers are organizing trips into summary files, which contain basic identifiers such as driver age and gender; vehicle make, model, and model year; maximum speed; highest deceleration; and time driving. As of February 2013, the study database included records on more than 3.4 million trips and about 250 crashes.
The Roadway Information Database
In addition to the trip database, the study includes a road database. Determining the relationship of roadway characteristics to crash risk and driver behavior requires detailed data about road grade, curvature, cross slope, lane and shoulder width, posted speed limits, medians, rumble strips, intersections, and other characteristics. Iowa State University’s Center for Transportation Research and Education (CTRE) is collecting these types of data.
Researchers at CTRE are creating a spatial database of roadway characteristics and features, as well as other supporting data -- including crash histories, traffic volumes, weather, work zones, and safety campaigns and laws -- that can be used to describe the context in which the participants drive. The CTRE researchers are populating the roadway information database with existing data acquired from State and local agencies and public and private sources.
The driving study has been operational at all six sites since spring 2011. From west to east, the sites are located in Seattle, WA; Bloomington, IN; Tampa, FL; Durham, NC; Buffalo, NY; and State College, PA.
The contractors selected to establish and operate the field data collection sites are responsible for installing equipment and conducting driver assessments, collecting and transmitting data, addressing problems, investigating crashes, and preparing periodic reports documenting the field study activities.
Analyzing the Data
The Naturalistic Driving Study is expected to produce data that researchers can use for at least 20 years. The following projects involve analyzing early data from the study.
Relationship between driver behavior and safety on curves: conducted by CTRE. Crash rates are 1.5 to 3 times higher on horizontal curves than on straight road sections. DOTs use various countermeasures to improve safety on curves, including installing signs and rumble strips to warn drivers about curves, delineating curves through chevrons and pavement markings, and minimizing the impact of road departures through paved shoulders and guardrails. But little information is available on how drivers respond to such roadway measures and why these roadway measures work or do not work. This CTRE study will use trip and roadway data from the Naturalistic Driving Study to examine how motorists interact with the roadway environment and what cues and measures are the most effective in influencing driver behavior. The study is expected to help highway departments implement more cost-effective measures to prevent or mitigate road-departure crashes on curves.
Driver inattention and crash risk: conducted by the SAFER Vehicle and Traffic Safety Centre at Chalmers University, Göteborg, Sweden. In 2009 distraction was involved in crashes causing 5,474 deaths and leading to 448,000 traffic injuries across the United States. Currently, measuring driver inattention and estimating the effect of inattention on crash risk remains a challenge. The SAFER study will use the Naturalistic Driving Study and roadway data to develop a measure of driver inattention based on observable driver actions, such as eye glances away from the road, and will estimate how driver inattention and the roadway environment combine to influence crash risk. The results will help establish guidelines for how long a driver can safely look away from the road and will help design invehicle technologies to measure driver inattention and warn inattentive drivers.
Evaluation of offset left-turn lanes: conducted by MRI Global, Kansas City, MO. Many intersections provide designated left-turn lanes where vehicles can wait apart from through-traffic lanes until it is safe to turn. However, vehicles waiting in standard left-turn lanes (in which the roadway’s centerline continues straight through the intersection) may have their view of oncoming through-traffic obstructed by vehicles in the opposing left-turn lane. One way to address this issue is to offset the left-turn lanes to the left, so vehicles waiting to turn are positioned to the left of the centerline of the opposite side of the intersection. Many highway designers have accepted these offset left-turn lanes in principle, but evidence of their effects on driver behavior or crashes remains inconclusive. The MRI Global study will use the Naturalistic Driving Study’s roadway data to analyze how driver left-turn behavior, such as gap acceptance, is influenced by intersection and traffic characteristics and especially by offset left-turn lanes. This will help DOT officials design intersections that balance construction and maintenance costs against crash risk.
At the same time, Fugro Roadware, a vendor of data collection solutions for roadway infrastructure, is collecting additional data on roads that the participants drive frequently. In total, the vendor will collect data on more than 2,000 centerline miles (3,219 kilometers) in each of the six study sites, totaling 12,500 centerline miles (20,117 kilometers), or 25,000 miles (40,234 kilometers) driven (the researchers are collecting the data in both directions).
SHRP2 will link the roadway data collected by CTRE to the naturalistic driving database to support analysis. Key aspects of the project include data integration, quality control and assurance, storage, retrieval, analysis, maintenance, reporting, and representation.
Data Processing and Analysis Phase
"Data collection is about half done, and about one-fourth of the data has been uploaded from the vehicles to the database at VTTI,” says SHRP2 safety chief program officer Ken Campbell. "Now our focus is shifting to data processing and data access. Issues [to be addressed] are the size of the database, over 4 petabytes, and maintaining the privacy assurances made to participants while providing access to data users.”
One of the benefits of the Naturalistic Driving Study will be the precrash, crash, and exposure (or baseline) data that it produces. Exposure data is obtained from normal, uneventful driving.
When the participants’ vehicles are running, the data acquisition system records continuously. Thus, the researchers will have access to crash and near-crash data, as well as baseline data representing samples of driving when no safety-related event occurred. A primary goal of the analysis is to determine the extent to which roadway design, traffic conditions, intersections, and advanced vehicle technology influence driving behavior and the risks inherent in these factors.
"Part of what we’re trying to do is develop confidence levels in the various data types we collect,” says Omar Smadi, director of CTRE’s Roadway Infrastructure Management & Operations Systems program. "We are working with the State DOTs [departments of transportation] in each area to get [their data] on the roadway and transportation infrastructure, and we’re working with VTTI to get GPS traces from participants to develop data collection maps.” In this context, GPS traces are the routes frequently driven by the study participants.
Professor Shinar adds: "In highway safety, we do not have a single, accepted theory of driver behavior. Thanks to the Tri-Level Study, we know that most crashes happen to normal people involved in normal activities but overwhelmed with a particular situation at the time of the crash, or just before it happens.”
For each crash, the records will contain a summary of what happened in the seconds before and after the event. The data is expected to show what the driver was doing that might have caused or contributed to the crash. After certain crashes (for example, those in which air bags deployed), researchers will conduct onsite crash investigations to gather additional data.
Road type, geometry, shoulders, safety infrastructure, signs, and pavement markings will be important factors in the study analysis. Researchers also will consider environmental variables such as traffic, lighting, and weather conditions.
"This information will support the development of new and improved countermeasures with greater effectiveness,” says Shinar.
The SHRP2 safety team will analyze the data to quantify the contribution of relevant driving, roadway, vehicle, and environmental factors, and will assess the implications of the findings in terms of potential countermeasures. Knowledge gained from these analyses, as well as those performed by other researchers, is expected to support public policy, rulemaking, infrastructure improvements, vehicle design, and other activities targeting crash reductions on the Nation’s roadways. Effective management of the data is essential for establishing systems and products to improve driving safety.
"We need to protect the human subjects in the research and ensure confidentiality, and at the same time we want researchers who have good reason to use the data to have a way of getting it,” says SHRP2 Director Ann Brach. "One of the decisions we made within TRB, SHRP2, and the SHRP2 Oversight Committee was to make sure we started funding the use of the data before we finished collecting it. This will enable us to demonstrate that it can be used and have some lessons learned so we can make improvements.”
SHRP2 is scheduled to end in March 2015, not quite 2 years after completion of the data collection. To develop options for the long-term stewardship and ownership of the data, FHWA is sponsoring a study with the John A. Volpe National Transportation Systems Center, a component of the USDOT’s Research and Innovative Technology Administration. The goals of the study are to protect the privacy of participants, make the data widely available for research purposes, ensure the sustainability of the data system for at least 20 years, and minimize the costs associated with supporting and maintaining a secure data system.
"Ownership of the data is the dominant issue underlying practically everything we’re discussing regarding long-term stewardship,” says Monique Evans, director of FHWA’s Office of Safety Research and Development at the Turner-Fairbank Highway Research Center. "Federal Government ownership brings with it major institutional requirements and standards that have to be met, covering privacy, security, record keeping, and other considerations. These requirements can be expensive and cumbersome, but are considered by many to be best practices.
"Currently TRB owns the data, and the Government has access rights to it. Our ultimate goal is to answer critical safety questions by ensuring that the research community has convenient and affordable access to the data while protecting the privacy of the study participants,” says Evans.
Privacy and Sharing
One of the major challenges of having such rich and comprehensive data will be figuring out how to make it publicly available while also protecting the privacy and identities of those who volunteered. Privacy protections were promised to participants, but their data will continue to be analyzed for decades after the study ends. Secure data facilities, data-sharing agreements, and approvals by institutional review boards will be used where appropriate.
Currently, all of the data from the study reside at VTTI. "For right now, the protocol for researchers who want to access the data requires a contract and funds to support data preparation, institutional review board approval, a data-sharing agreement, and SHRP2 approval for non-SHRP2 projects,” explains VTTI research scientist Suzie Lee. "Identifying video, which is video where participants are recognizable, and full GPS traces have to be viewed at VTTI’s secure data enclave, but nonidentifying data can be shipped out without further review.”
Data-sharing agreements are contracts between data stewards and research teams, specifying that data are to be shared and used in accordance with privacy promises made to the participants. Currently a 3- to 4-page document, the agreement requires researchers to present a scope of analysis, specify the dataset requested, secure institutional review board approval or proof of exemption for nonidentifying data, undergo human subjects training, and establish a timeframe for data retention.
"The human subjects training is typically an online course that can take between 45 to 90 minutes,” explains Lee. "It basically grounds you in the principles of human subjects’ protection and makes you aware of why we are protecting the privacy of the participants.”
All protections required for the rights and safety of participants in human subject research are in place, and the institutional review boards of the National Academy of Sciences and other contracting agencies have approved each step of the study design.
Using the Data in Projects
FHWA set aside $10 million out of SHRP2 funds for implementing projects to improve safety. As a way of targeting high-priority areas for safety research, FHWA developed roadmaps keyed to places where a large number of traffic fatalities are occurring. "We’re in the process of identifying projects from these roadmaps as well as new projects that might be higher risk or more advanced, or call for long-term studies in high-priority areas that could be addressed using SHRP2 safety data,” explains FHWA’s Evans.
Also, FHWA is pursuing projects in its Exploratory Advanced Research (EAR) Program to identify and develop tools to support the efficient use of massive amounts of data. One project is a 2-year study on automated extraction of video data to facilitate the analysis of large quantities of transportation research data and video.
Additional projects might result from a workshop on video analytics that was held at Turner-Fairbank on October 10–11, 2012. At the workshop, FHWA brought together professionals from academia, the private sector, and government to identify research needs and interests in advancing automation techniques that facilitate the efficient extraction of features in videos on driving behavior. "What you’re looking for depends on what you’re trying to study,” points out FHWA EAR Program Director David Kuehn. "So the questions are: What do you need to extract and what are the key behavioral or inattention problems you’re looking for? What are the differences in the algorithms you may need to identify?”
The EAR Program is interested in advancing technology for sensors, video, and signal processing that could result in transformative changes in the way researchers investigate human behavior and human-machine interaction. Kuehn expects the study on automated extraction of video data to demonstrate the effectiveness of advanced machine vision techniques applied to large and diverse datasets and enable the development of a comprehensive library of data processes and analysis tools. Information generated from the workshop could be useful in considering methods to save time by speeding up the process of examining particular data.
A National Discussion
The Naturalistic Driving Study marks the beginning of a national discussion that is likely to influence implementation of new standards and advanced technologies. Although it is too early to know the impact that the study’s data will have on safety, wireless communications for vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) applications have shown strong potential. For example, V2V systems that enable cars to talk to each other can warn drivers of dangerous road conditions, and intelligent intersections equipped with V2I technologies can advise drivers to slow down to avoid running a red light. Researchers are advancing tests of these applications under real-world, multimodal driving conditions to determine their effectiveness, safety, and ability to help reduce crashes. Might future V2V and V2I systems informed from the findings of the Naturalistic Driving Study serve to deter excessive speed, inattention, and other potentially dangerous driving behavior?
Future efforts and discussions aimed at ensuring that the study’s data are put to good use will continue to focus on instituting protocols and creating living documents that emphasize privacy protection and research intent. Questions of how to house and maintain the enormous database also will continue to be addressed. "The data will be housed in at least one secure facility -- Virginia Tech’s -- and possibly others,” says SHRP2 Director Brach. In the future, other facilities might hold and protect the data as well.
FHWA has initiated a feasibility study to establish a support enclave for safety data analysis at Turner-Fairbank. According to FHWA Director of Safety Research Evans, the enclave would serve to advance knowledge about the data and range of usage. "It would provide assistance to State DOTs regarding use of the data and opportunities for researchers to get hands-on training and experience with the data through internships, sabbaticals, and other avenues,” she says.
Researchers involved with data collection and analysis are discussing the feasibility of producing some additional categories of data files, featuring near-crash events, exposure samples, a small sample of trips, and specialized trip characteristics. Near-crash files would be assembled in a similar manner as crash files. For example, the files could contain events with extreme braking or steering maneuvers. Definitions for near-crashes are under development.
"We would like to get a better understanding of the relationship between crashes and near-crashes and critical maneuvers,” says NHTSA’s Compton. "These are typically the things that are not well understood, and not easily understood through post-crash investigations, where exposure and precrash driving behavior would make a tremendous difference in understanding how driver behavior contributes to crashes. I think this is one of the first things we’ll tackle with these data.”
Mark Fitzgerald is a senior writer at Woodward Communications and teaches writing at the University of Maryland, College Park. Before joining Woodward, he was the editor of several trade magazines and worked at the American Society of Civil Engineers. He has a B.A. in English from Franklin & Marshall College and an M.F.A. in creative writing from George Mason University.
For more information about the Naturalistic Driving Study, see www.shrp2nds.us. To read more about the SHRP2 safety research program, visit www.trb.org/SHRP2/safety or contact Mark Fitzgerald at 202–493–3995 or firstname.lastname@example.org.