How is the duration of distraction related to safety-critical events? Harnessing naturalistic driving data to explore the role of driving instability

INTRODUCTION

Due to a variety of secondary tasks performed by drivers, distracted driving has become a critical concern. At 50 mph, sending/reading a text for 5 seconds is equivalent to driving the length of a football field (360 ft) with eyes closed. A fundamental understanding of how distractions lead to crashes is needed to develop appropriate countermeasure strategies. A key question is whether distraction increases driving instability, which then further contributes to safety-critical events (SCEs).

METHODS

By harnessing newly available microscopic driving data and using the safe systems approach, a subsample of naturalistic driving study data were analyzed, collected through the second strategic highway research program. Rigorous path analysis (including Tobit and Ordered Probit regressions) is used to jointly model the instability in driving (using coefficient of variation of speed) and event outcomes (including baseline, near-crash, and crash). The marginal effects from the two models are used to compute direct, indirect, and total effects of distraction duration on SCEs.

RESULTS

Results indicate that a longer duration of distraction was positively but non-linearly associated with higher driving instability and higher chances of SCEs. Where, the chance of a crash and near-crash was higher by 34% and 40%, respectively, with a unit increase in driving instability. Based on the results, the chance of both SCEs significantly increases non-linearly with an increase in distraction duration beyond 3 seconds. For instance, the chance of a crash is 16% for a driver distracted for 3 seconds, which increases to 29% if a driver is distracted for 10 seconds.

CONCLUSIONS AND PRACTICAL APPLICATIONS

Using path analysis, the total effects of distraction duration on SCEs are even higher when its indirect effects on SCEs through driving instability are considered. Potential practical implications including traditional countermeasures (changes in roadway environments) and vehicle technologies are discussed in the paper.

Advancing investigation of automated vehicle crashes using text analytics of crash narratives and Bayesian analysis

Vehicle automation, manifested in self-driving cars, promises to provide safe mobility by reducing human errors. While the testing of automated vehicles (AVs) has improved their performance in recent years, automation technologies face challenges such as uncertainty of safety impacts in mixed traffic with human-driven vehicles. This study aims to examine the gaps in AV safety performance and identify what will be required on a preferential basis for AVs to guarantee an acceptable level of safety performance, especially in mixed traffic, by conducting a thorough analysis of crashes involving levels 2-3 AVs. Based on 260 AV collision reports from California from 2019 to 2021, this study extracts crash-related variables from crash records in a standardized form, crash locations, and, notably, crash narratives reported by AV manufacturers. This study untangles the complex interrelationships among pre-crash conditions, AV driving modes, crash types, and crash outcomes by applying a path-analytic framework with the frequentist and Bayesian approaches. Results show that 51.9 percent of crashes were rear-ends. Particularly, AVs become more vulnerable to rear-end collisions in the automated driving mode than in the conventional mode, given a crash. Additionally, the automated driving mode would not significantly affect the chance of a sideswipe collision, injury, or AV damage levels. Another interesting finding is that manual disengagement is more likely to happen when an AV interacts with a transit vehicle right before a crash occurs while having a negative relationship with injury crashes. Moreover, to reduce injury crashes, AVs would need more thorough testing to adapt to the critical roadway and infrastructure features such as intersections, ramps, and slip lanes; and roadway infrastructure would require improvements to support transportation automation. The risk factors identified in this study can be considered in AV safety assessment scenarios and future operations of mixed traffic. This study demonstrates that AV crash narrative data can be leveraged to improve knowledge of AV safety in mixed traffic.

Automated vehicle data pipeline for accident reconstruction: New insights from LiDAR, camera, and radar data

As automated vehicles are deployed across the world, it has become critically important to understand how these vehicles interact with each other, as well as with other conventional vehicles on the road. One such method to achieve a deeper understanding of the safety implications for Automated Vehicles (AVs) is to analyze instances where AVs were involved in crashes. Unfortunately, this poses a steep challenge to crash-scene investigators. It is virtually impossible to fully understand the factors that contributed to an AV involved crash without taking into account the vehicle's perception and decision making. Furthermore, there is a tremendous amount of data that could provide insight into these crashes that is currently unused, as it also requires a deep understanding of the sensors and data management of the vehicle. To alleviate these problems, we propose a data pipeline that takes raw data from all on-board AV sensors such as LiDAR, radar, cameras, IMU's, and GPS's. We process this data into visual results that can be analyzed by crash scene investigators with no underlying knowledge of the vehicle's perception system. To demonstrate the utility of this pipeline, we first analyze the latest information on AV crashes that have occurred in California and then select two crash scenarios that are analyzed in-depth using high-fidelity synthetic data generated from the automated vehicle simulator CARLA. The data visualization procedure is demonstrated on the real-world Kitti dataset by using the YOLO object detector and a monocular depth estimator called AdaBins. Depth from LIDAR is used as ground truth to calibrate and assess the effect of noise and errors in depth estimation. The visualization and data analysis from these scenarios clearly demonstrate the vast improvement in crash investigations that can be obtained from utilizing state-of-the-art sensing and perception systems used on AVs.

Exploring pathways from driving errors and violations to crashes: The role of instability in driving

This study explores how different driving errors, violations, and roadway environments contribute to safety-critical events through instability in driving speed. We harness a subsample (N = 9239) of the naturalistic driving study (NDS) data collected through the Second Strategic Highway Research Program (SHRP2). From a methodological standpoint, we use the safe systems approach relying on path analysis to jointly model outcomes. This accounts for the potential correlation between unobserved factors associated with both instability in driving speed and epoch (video stream) outcomes, i.e., baseline or event-free driving, near-crashes, and crashes. Tobit and ordered Probit regressions are estimated to model the coefficient of variation (COV) of speed and epoch outcomes, respectively. Results from the Tobit model indicate that driving errors and violations are associated with instability in the driving speed of the subject driver (COV of speed). The Probit model reveals that driving errors, violations, and instability in driving speed are associated with higher chances of crashes and near-crashes. Our key finding is that driving errors and violations not only induce event risk directly but also indirectly through instability in driving speed. For instance, recognition errors associate with higher crash risk by 6.78 % but this error is accompanied by instability in driving speed, which further increases event risk by 4.73 %, bringing the total increase in risk to 11.51 %. Moreover, significant correlations were found between unobserved factors reflected in the error terms of the two models. Ignoring such correlations can lead to inefficient parameter estimates. Based on the findings, practical implications are discussed, which can lead to effective countermeasures that effectively reduce crash risk.

Predicting intersection crash frequency using connected vehicle data: A framework for geographical random forest

Accurate crash frequency prediction is critical for proactive safety management. The emerging connected vehicles technology provides us with a wealth of vehicular motion data, which enables a better connection between crash frequency and driving behaviors. However, appropriately dealing with the spatial dependence of crash frequency and multitudinous driving features has been a difficult but critical challenge in the prediction process. To this end, this study aims to investigate a new Artificial Intelligence technique called Geographical Random Forest (GRF) that can address spatial heterogeneity and retain all potential predictors. By harnessing more than 2.2 billion high-resolution connected vehicle Basic Safety Message (BSM) observations from the Safety Pilot Model Deployment in Ann Arbor, MI, 30 indicators of driving volatility are extracted, including speed, longitudinal and lateral acceleration, and yaw rate. The developed GRF was implemented to predict rear-end crash frequency at intersections. The results show that: 1) rear-end crashes are more likely to happen at intersections connecting minor roads compared to major roads; 2) a higher number of hard acceleration and deceleration events beyond two standard deviations in the longitudinal direction is a leading indicator of rear-end crashes; 3) the optimal GRF significantly outperforms Global Random Forest, with a 9% lower test error and a substantially better fit; and 4) geographical visualization of variable importance highlights the presence of spatial non-stationarity. The proposed framework can proactively identify at-risk intersections and alert drivers when leading indicators of driving volatility tend to worsen.

Incorporating driving volatility measures in safety performance functions: Improving safety at signalized intersections

About 40 percent of motor vehicle crashes in the US are related to intersections. To deal with such crashes, Safety Performance Functions (SPFs) are vital elements of the predictive methods used in the Highway Safety Manual. The predictions of crash frequencies and potential reductions due to countermeasures are based on exposure and geometric variables. However, the role of driving behavior factors, e.g., hard accelerations and declarations at intersections, which can lead to crashes, are not explicitly treated in SPFs. One way to capture driving behavior is to harness connected vehicle data and quantify performance at intersections in terms of driving volatility measures, i.e., rapid changes in speed and acceleration. According to recent studies, driving volatility is typically associated with higher risk and safety-critical events and can serve as a surrogate for driving behavior. This study incorporates driving volatility measures in the development of SPFs for four-leg signalized intersections. The Safety Pilot Model Deployment (SPMD) data containing over 125 million Basic Safety Messages generated by over 2,800 connected vehicles are harnessed and linked with the crash, traffic, and geometric data belonging to 102 signalized intersections in Ann Arbor, Michigan. The results show that including driving volatility measures in SPFs can reduce model bias and significantly enhances the models' goodness-of-fit and predictive performance. Technically, the best results were obtained by applying Bayesian hierarchical Negative Binomial Models, which account for spatial correlation between signalized intersections. The results of this study have implications for practitioners and transportation agencies about incorporating driving behavior factors in the development of SPFs for greater accuracy and measures that can potentially reduce volatile driving.

Analyzing drivers' hazard recognition: Precursors to single-vehicle collisions

Extensive driver behavior and performance information provided by real-world video surveillance and sensor data in the SHRP2 Naturalistic Driving Study has enabled the examination of new layers and pathways leading to crash outcomes. We note that the prominence of hazards and the importance of recognizing them vary systematically across single vs. multi-vehicle crashes, and address a fundamental question about safety: why do around three-quarters of drivers involved in single-vehicle crashes not recognize, perceive, or react to the precipitating event (PE)? Using a path-analytic framework through marginal effects, this study investigates factors correlated to recognition of the PE in single-vehicle events, and how these correlations may act as crash precursors. Logit models, accounting for heterogeneity among events and drivers by estimating both fixed and random parameters, quantified correlations among key variables, given a crash or near-crash event (N = 543). The type of PE, roadway environment factors, and driving maneuvers heavily influenced recognition chances. Drivers had a harder time recognizing less conspicuous hazards (e.g. departing the travel way, decreased recognition chances by 48.29%), but seemed better at recognizing prominent hazards (e.g. vehicle losing control, increased recognition chances by 46.71%). In addition, drivers are less likely to recognize PEs when executing less involved driving maneuvers in more relaxed environments, such as daylight (decreased recognition chances by 16.00%), but are more adept in environments that already demand more attention. Recognition reduced the chances of a crash by 12.23%, so we found similar correlations with crash outcome. Future intelligent transportation systems may focus on increasing driver recognition of potential hazards by bringing attention to less conspicuous hazards and less involved driving environments and actions.

Integration of automated vehicles in mixed traffic: Evaluating changes in performance of following human-driven vehicles

The introduction of Automated Vehicles (AVs) into the transportation network is expected to improve system performance, but the impacts of AVs in mixed traffic streams have not been clearly studied. As AV's market penetration increases, the interactions between conventional vehicles and AVs are inevitable but by no means clear. This study aims to create new knowledge by quantifying the behavioral changes caused when conventional human-driven vehicles follow AVs and investigating the impact of these changes (if any) on safety and the environment. This study analyzes data obtained from a field experiment by Texas A&M University to evaluate the effects of AVs on the behavior of a following human-driver. The dataset is comprised of nine drivers that attempted to follow 5 speed-profiles, with two scenarios per profile. In scenario one, a human-driven vehicle follows an AV that implements a human driver speed profile (base). In scenario two, the human-driven vehicle follows an AV that executes an AV speed profile. In order to evaluate safety, these scenarios are compared using time-to-collision (TTC) and several other driving volatility measures. Likewise, fuel consumption and emissions are used to investigate environmental impacts. Overall, the results show that AVs in mixed traffic streams can induce behavioral changes in conventional vehicle drivers, with some beneficial effects on safety and the environment. On average, a driver that follows an AV exhibits lower driving volatility in terms of speed and acceleration, which represents more stable traffic flow behavior and lower crash risk. The analysis showed a remarkable improvement in TTC as a result of the notably better speed adjustments of the following vehicle (i.e., lower differences in speeds between the lead and following vehicles) in the second scenario. Furthermore, human-driven vehicles were found to consume less fuel and produce fewer emissions on average when following an AV.

Safety critical event prediction through unified analysis of driver and vehicle volatilities: Application of deep learning methods

Transportation safety is highly correlated with driving behavior, especially human error playing a key role in a large portion of crashes. Modern instrumentation and computational resources allow for the monitorization of driver, vehicle, and roadway/environment to extract leading indicators of crashes from multi-dimensional data streams. To quantify variations that are beyond normal in driver behavior and vehicle kinematics, the concept of volatility is applied. The study measures driver-vehicle volatilities using the naturalistic driving data. By integrating and fusing multiple real-time streams of data, i.e., driver distraction, vehicular movements and kinematics, and instability in driving, this study aims to predict occurrence of safety critical events and generate appropriate feedback to drivers and surrounding vehicles. The naturalistic driving data is used which contains 7566 normal driving events, and 1315 severe events (i.e., crash and near-crash), vehicle kinematics, and driver behavior collected from more than 3500 drivers. In order to capture the local dependency and volatility in time-series data 1D-Convolutional Neural Network (1D-CNN), Long Short-Term Memory (LSTM), and 1DCNN-LSTM are applied. Vehicle kinematics, driving volatility, and impaired driving (in terms of distraction) are used as the input parameters. The results reveal that the 1DCNN-LSTM model provides the best performance, with 95.45% accuracy and prediction of 73.4% of crashes with a precision of 95.67%. Additional features are extracted with the CNN layers and temporal dependency between observations is addressed, which helps the network learn driving patterns and volatile behavior. The model can be used to monitor driving behavior in real-time and provide warnings and alerts to drivers in low-level automated vehicles, reducing their crash risk.

Driving impairments and duration of distractions: Assessing crash risk by harnessing microscopic naturalistic driving data

Distracted and impaired driving is a key contributing factor in crashes, leading to about 35% of all transportation-related deaths in recent years. Along these lines, cognitive issues like inattentiveness can further increase the chances of crash involvement. Despite its prevalence and importance, little is known about how the duration of these distractions is associated with critical events, such as crashes or near-crashes. With new sensors and increasing computational resources, it is possible to monitor drivers, vehicle performance, and roadway features to extract useful information, e.g., eyes off the road, indicating distraction and inattention. Using high-resolution microscopic SHRP2 naturalistic driving data, this study conducts in-depth analysis of both impairments and distractions. The data has more than 2 million seconds of observations in 7394 baselines (no event), 1228 near-crashes, and 617 crashes. The event data was processed and linked with driver behavior and roadway factors. The intervals of distracted driving during the period of observation (15 seconds) were extracted; next, rigorous fixed and random parameter logistic regression models of crash/near-crash risk were estimated. The results reveal that alcohol and drug impairment is associated with a substantial increase in crash/near-crash event involvement of 34%, and the highest correlations with crash risk include duration of distraction through dialing on a cellphone, texting while driving, and reaching for an object. Using detailed pre-crash data from instrumented vehicles, the study contributes by quantifying crash risk vis-à-vis detailed driving impairment and information on secondary task involvement, and discusses the implications of the results.

Exploring the who, what, when, where, and why of automated vehicle disengagements

Automated vehicles are emerging on the transportation networks as manufacturers test their automated driving system (ADS) capabilities in complex real-world environments in testing operations like California's Autonomous Vehicle Tester Program. A more comprehensive understanding of the ADS safety performances can be established through the California Department of Motor Vehicle disengagement and crash reports. This study comprehensively examines the safety performances (159,840 disengagements, 124 crashes, and 3,669,472 automated vehicle miles traveled by the manufacturers) documented since the inauguration of the testing program. The reported disengagements were categorized as control discrepancy, environmental conditions and other road users, hardware and software discrepancy, perception discrepancy, planning discrepancy, and operator takeover. An applicable subset of disengagements was then used to identify and quantify the 5 W's of these safety-critical events: who (disengagement initiator), when (the maturity of the ADS), where (location of disengagement), and what/why (the facts causing the disengagement). The disengagement initiator, whether the ADS or human operator, is linked with contributing factors, such as the location, disengagement cause, and ADS testing maturity through a random parameter binary logit model that captured unobserved heterogeneity. Results reveal that compared to freeways and interstates, the ADS has a lower likelihood of initiating the disengagement on streets and roads compared to the human operator. Likewise, software and hardware, and planning discrepancies are associated with the ADS initiating the disengagement. As the ADS testing maturity advances in months, the probability of the disengagement being initiated by the ADS marginally increases when compared to human-initiated. Overall, the study contributes by understanding the factors associated with disengagements and exploring their implications for automated systems.

The role of pre-crash driving instability in contributing to crash intensity using naturalistic driving data

While the cost of crashes exceeds $1 Trillion a year in the U.S. alone, the availability of high-resolution naturalistic driving data provides an opportunity for researchers to conduct an in-depth analysis of crash contributing factors, and design appropriate interventions. Although police-reported crash data provides information on crashes, this study takes advantage of the SHRP2 Naturalistic Driving Study (NDS) which is a unique dataset that allows new insights due to detailed information on driver behavior in normal, pre-crash, and near-crash situations, in addition to trip and vehicle performance characteristics. This paper investigates the role of pre-crash driving instability, or driving volatility, in crash intensity (measured on a 4-point scale from a tire-strike to an injury crash) by analyzing microscopic vehicle kinematic data. NDS data are used to investigate not only the vehicle movements in space but also the instability of vehicles prior to the crash and their contribution to crash intensity using path analysis. A subset of the data containing 617 crash events with around 0.18 million temporal trajectories are analyzed. To quantify driving instability, microscopic variations or volatility in vehicular movements before a crash are analyzed. Specifically, nine measures of pre-crash driving volatility are calculated and used to explain crash intensity. While most of the measures are significantly correlated with crash intensity, substantial positive correlations are observed for two measures representing speed and deceleration volatilities. Modeling results of the fixed and random parameter probit models revealed that volatility is one of the leading factors increasing the probability of a severe crash. Additionally, the speed prior to a crash is highly correlated with intensity outcomes, as expected. Interestingly, distracted and aggressive driving are highly correlated with driving volatility and have substantial indirect effects on crash intensity. With volatile driving serving as a leading indicator of crash intensity, given the crashes analyzed in this study, early warnings and alerts for the subject vehicle driver and proximate vehicles can be helpful when volatile behavior is observed.

A geographically weighted regression to estimate the comprehensive cost of traffic crashes at a zonal level

Global road safety records demonstrate spatial variation of comprehensive cost of traffic crashes across countries. To the best of our knowledge, no study has explored the variation of this matter at a local geographical level. This study proposes a method to estimate the comprehensive crash cost at the zonal level by using person-injury cost. The current metric of road safety attributes safety to the location of the crash, which makes it challenging to assign the crash cost to home-location of the individuals who were involved in traffic crashes. To overcome this limitation, we defined Home-Based Approach crash frequency as the expected number of crashes by severity that road users who live in a certain geographic area have during a specified period. Using crash data from Tennessee, we assign those involved in traffic crashes to the census tract corresponding to their home address. The average Comprehensive Crash Cost at the Zonal Level (CCCAZ) for the period of the study was $18.2 million (2018 dollars). Poisson and Geographically Weighted Poisson Regression (GWPR) models were used to analyzing the data. The GWPR model was more suitable compared to the global model to address spatial heterogeneity. Findings indicate population of people over 60-years-old, the proportion of residents that use non-motorized transportation, household income, population density, household size, and metropolitan indicator have a negative association with CCCAZ. Alternatively, VMT, vehicle per capita, percent educated over 25-year-old, population under 16-year-old, and proportion of non-white races and individuals who use a motorcycle as their commute mode have a positive association with CCCAZ. Findings are discussed in line with road safety literature.

How instantaneous driving behavior contributes to crashes at intersections: Extracting useful information from connected vehicle message data

Connected and automated vehicles have enabled researchers to use big data for development of new metrics that can enhance transportation safety. Emergence of such a big data coupled with computational power of modern computers have enabled us to obtain deeper understanding of instantaneous driving behavior by applying the concept of "driving volatility" to quantify variations in driving behavior. This paper brings in a methodology to quantify variations in vehicular movements utilizing longitudinal and lateral volatilities and proactively studies the impact of instantaneous driving behavior on type of crashes at intersections. More than 125 million Basic Safety Message data transmitted between more than 2800 connected vehicles were analyzed and integrated with historical crash and road inventory data at 167 intersections in Ann Arbor, Michigan, USA. Given that driving volatility represents the vehicular movement and control, it is expected that erratic longitudinal/lateral movements increase the risk of crash. In order to capture variations in vehicle control and movement, we quantified and used 30 measures of driving volatility by using speed, longitudinal and lateral acceleration, and yaw-rate. Rigorous statistical models including fixed parameter, random parameter, and geographically weighted Poisson regressions were developed. The results revealed that controlling for intersection geometry and traffic exposure, and accounting unobserved factors, variations in longitudinal control of the vehicle (longitudinal volatility) are highly correlated with the frequency of rear-end crashes. Intersections with high variations in longitudinal movement are prone to have higher rear-end crash rate. Referring to sideswipe and angle crashes, along with speed and longitudinal volatility, lateral volatility is substantially correlated with the frequency of crashes. When it comes to head-on crashes, speed, longitudinal and lateral acceleration volatilities are highly associated with the frequency of crashes. Intersections with high lateral volatility have higher risk of head-on collisions due to the risk of deviation from the centerline leading to head-on crash. The developed methodology and volatility measures can be used to proactively identify hotspot intersections where the frequency of crashes is low, but the longitudinal/lateral driving volatility is high. The reason that drivers exhibit higher levels of driving volatility when passing these intersections can be analyzed to come up with potential countermeasures that could reduce volatility and, consequently, crash risk.