Summary of Projects Funded By The Federal Highway Administration for The National Surface Transportation Safety Center for Excellence From July 2006 to June 2014

Fragility is estimated to be responsible for a major portion (60 to 95 percent) of the fatalities of older drivers and passengers.⁽³⁴⁾ Prior to this study, VTTI's DAS collected acceleration and gyro data at a temporal resolution of 10 Hz. While this was sufficient for the development of CNC detection algorithms and other related applications, it was not enough to model the biomechanical stress and injuries experienced during a vehicular crash.

Therefore, VTTI's Older Driver NDS would enhance the capabilities of the DAS to collect acceleration and gyro data with sufficient range, temporal resolution, and accuracy to facilitate subsequent biomechanics-injury analysis and modeling. That initial study produced no serious crashes to evaluate from a biomechanics perspective. However, the approach pioneered in that study has been applied to all of VTTI's subsequent NDSs, including SHRP2, the largest NDS of its kind ever conducted. Additionally, non-injurious exposure can be modeled to provide a set of data on tolerable events. This approach may benefit older drivers and likely many others as well, perhaps leading to countermeasures related to safety belt design, air bag design and deployment factors, seat factors, and cabin structures design.

Teenage drivers are four times more likely to be involved in a crash than any other age group. The reasons for these crash rates are largely unknown, but it is believed that if risky driving behaviors are reduced through providing feedback to the driver, crash rates will also fall.

The driver coach project evaluated the effectiveness of providing driving-related feedback to teens with the goal of reducing unsafe behaviors that lead to injury and fatal crashes. FHWA partially contributed funds, while VDOT, the National Institutes of Health, and the Toyota® Collaborative Safety Research Center contributed the additional funding. A total of 90 vehicles were instrumented with devices to record the driving performance and audio data of individuals who had received their learner's permit within the past 2 weeks. Driving performance feedback began a few weeks prior to licensure and continued for a period of 6 mo post licensure. During the seventh month, feedback was halted while researchers monitored the novice's driving performance to see if it remained constant or the driver reverted to unsafe behavior.

VTTI technicians developed a machine vision-based tool called the Mask that uses facial landmarks observed in naturalistic driving face video to determine the position of the driver's face and head as well as degree of rotation in space. While this tool cannot yet determine exact gaze location, some valuable information can be gleaned from its output in terms of breadth of head movement at intersections, gross metrics, or implications of eyes-off-road time.

This project developed the capability for use of the Mask in postprocessing of a naturalistic dataset. The Mask was effectively applied to the Older Driver and 40-Teen NDS databases and generated the expected head position and rotation data. Antin, Wotring, and Foley used this new approach to successfully evaluate data about breadth of head rotations at intersections from VTTI's Older Driver NDS.⁽³⁶⁾

The goal of this project was to evaluate two different training-based approaches to enhance senior driver safety. A total of 74 male and female licensed drivers aged 70—85 were recruited from the New River Valley area of Virginia to participate in an evaluation of a prototype in-vehicle implicit learning system designed by Toyota® engineers and a commercially available computer-based application from Posit Science®. A no-contact control group received no training.

Both training approaches emphasize expanding senior drivers' useful field of vision and otherwise strengthening other visual-cognitive functional abilities, such as speed of information processing and ability to visually track moving objects. Dependent measures include laboratory metrics (useful field of view, trail-making of parts A and B, and visualization of missing information), roadside object recognition distance, peripheral detection tasks, and public road driving performance. Data collection and analyses have been completed.

In a joint effort, VTTI researchers and their colleagues at the Monash University Accident Research Center in Melbourne, Australia, collected naturalistic driving data on older drivers. This project compared key aspects of the seniors' driving behavior across the two continents, specifically secondary task engagement. This type of engagement is associated with complexity of situation (i.e., less likely when vehicle is moving). Both the U.S. and Australian researchers noted differences in frequency of engagement with specific types of secondary activities such as cell phone use and adjustment of control panels, which could be attributed to differences in law enforcement priority. Yet general patterns were consistent across the continents and the findings supported previous research efforts.

Unfortunately, the study had several limitations, including a relatively small sample size and differences in vehicle familiarity; Australian drivers used an unfamiliar vehicle, whereas those in the United States used their personal vehicles. Although difficulties arose when aligning the two analyses post hoc, the results suggested a propensity for older drivers to self-regulate by reducing potentially distracting tasks at intersections when cognitive demand is high.

In this study, researchers evaluated the relationship between an older driver's fitness assessment profiles and their driving risk, represented primarily by CNC rate and secondarily by HGF event rate, all recorded during a naturalistic study of senior drivers. Due to the relatively small sample size of only 20 primary drivers in this pilot investigation, principal component analysis was used for dimension reduction and classification of the 60 total fitness profile metrics. Negative binomial regression models were employed to model the CNC and HGF events.

The results indicated that contrast sensitivity measures were significantly associated with CNC rate: the greater the sensitivity, the lower the CNC rate. In the HGF event analysis, the CNC rate was positively related to the HGF rate. The fitness metric contrast sensitivity was also related to the HGF event rate. In addition, two metrics related to metacognition (a measurement of one's perception of one's own cognitive status) were associated with HGF event rate. Higher HGF rates were associated with greater self-rating of cognitive status as well as greater disparities between that same self-rating and an objective metric of cognitive status.

The results of this study provided crucial information on the metrics and protocols, which could be applied by motor vehicle departments, physicians, occupational therapists, certified driving rehabilitation specialists, and others for whom determining seniors' fitness to drive is an important component of their work. Furthermore, these results can be further investigated and validated using the much larger database of senior driver data collected in the SHRP2 NDS.

The purpose of this project was to identify and develop the best procedures to evaluate driver coach trigger algorithms on the first 2 to 3 mo of driving in the 40-Teen Naturalistic Dataset. The 40-Teen study data provided the most accurate estimates because these drivers were within 2 weeks of receiving their provisional licenses and drove in the general geographic area as the driver coach participants. No other dataset, including the SHRP2 dataset, provided as precise estimates as the 40-Teen dataset.

Primary benefits of this venture included analysis and validation of driver coach triggers over an existing, complete dataset; increased ability to accurately estimate current and future project expenses (i.e., telemetry and labor); and establishment of a development framework to aid future algorithm recoding efforts.

Researchers gained information regarding the number of expected triggers, especially for the high-risk, moderate-risk, and low-risk teenage drivers. Results suggested that the mere presence of the MiniDAS coupled with emails to parents had a positive effect on novice driving performance because the number of triggered events was less than what was obtained during the NTDS.

Two methods of determining driver head pose positioning, both developed at VTTI prior to this contract, were compared to evaluate the feasibility of a real-time driver distraction trigger. The first method, colloquially referred to as "Driver Watchdog," used a support vector machine approach. The second approach drew from prior work completed on the Mask 2.0 gesture-sensing algorithm. The initial provision for both approaches required 80 percent or greater accuracy in detecting scenarios where the driver's face is outside of a ±15-degree cone of vision from the forward roadway. Both approaches were tested on a captured video feed to ensure they would be capable of running effectively within the constraints of the MiniDAS digital signal processor's power.

The Driver Watchdog algorithm requires a separate model for each individual driver to be created and stored on the MiniDAS platform. Therefore, the creation of these driver-specific models would require significant costs, in both human data reduction (labor) and database storage, and DAS package assignment capabilities. However, the Driver Watchdog algorithm shows promise as a means for identifying consented drivers in real-time, with identification and driver trip assignments taking place on the DAS. Therefore, future development will be directed toward this goal instead of driver distraction.

Upon review of the both approaches, it was determined that the MASK 2.0 algorithm was a more appropriate choice due to lower implementation costs. Further testing and refinement of the MASK 2.0 algorithm did not provide acceptable results for field deployment in the field for real-time monitoring and feedback. The algorithm did not achieve the necessary 80-percent accuracy mark for a ±15 degree cone of vision from the forward roadway.

Light-vehicle driver education programs containing content about heavy-vehicle operation may be helpful in reducing light-vehicle and heavy-vehicle interactions. However, the extent of current state curricula requirements, content, and perceived effectiveness of both public and private programs regarding heavy-vehicle operation and associated recommended procedures for light-vehicle driving is unclear. This project resulted in an online survey targeting instructors and administrators of State driver education programs to identify current curricula addressing heavy vehicles (or lack thereof) and perceived effectiveness.

At least two surveys were completed from each State and the District of Columbia. Survey results showed that while a large proportion of light-vehicle driver education programs include a component on how to safely share the road with heavy vehicles (91 percent), there is room for improvement in the content of these programs (82 percent perceived effectiveness). Qualitative analysis performed using the open-ended survey questions revealed areas of improvement.