Individual driver risk assessment using naturalistic driving data

RiskFormer: Exploring the temporal associations between multi-type aberrant driving events and crash occurrence

With the development of driving behavior monitoring technologies, commercial transportation enterprises have leveraged aberrant driving event detection results for evaluating crash risk and triggering proactive interventions. The state-of-the-art applications were established based upon instant associations between events and crash occurrence, which assumed crash risk surged with aberrant events. Consequently, the generated crash risk monitoring results merely contain discrete abrupt changes, failing to depict the time-varying trend of crash risk and posing challenges for interventions. Given the multiple types of aberrant events and their various temporal combinations, the key to depict crash risk time-varying trend is the analysis of multi-type events' temporal coupling influence. Existing studies employed event frequency to model combined influence, lacking the capability to differentiate the temporal sequential characteristics of events. Hence, there is an urgent need to further explore multi-type events' temporal coupling influence on crash risk. In this study, the temporal associations between multi-type aberrant driving events and crash occurrence are explored. Specifically, a contrastive learning method, fusing prior domain knowledge and empirical data, was proposed to analyze the single event temporal influence on crash risk. After that, a novel Crash Risk Evaluation Transformer (RiskFormer) was developed. In the RiskFormer, a unified encoding method for different events, as well as a self-attention mechanism, were established to learn multi-type events' temporal coupling influence. Empirical data from online ride-hailing services were employed, and the modeling results unveiled three distinct time-varying patterns of crash risk, including decay, increasing, and increasing-decay pattern. Additionally, RiskFormer exhibited remarkable crash risk evaluation performance, demonstrating a 12.8% improvement in the Area Under Curve (AUC) score compared to the conventional instant-association-based model. Furthermore, the practical utility of RiskFormer was illustrated through a crash risk monitoring sample case. Finally, applications of the proposed methods and their further investigations have been discussed.

High-risk event prone driver identification considering driving behavior temporal covariate shift

Drivers who perform frequent high-risk events (e.g., hard braking maneuvers) pose a significant threat to traffic safety. Existing studies commonly estimated high-risk event occurrence probabilities based upon the assumption that data collected from different time periods are independent and identically distributed (referred to as i.i.d. assumption). Such approach ignored the issue of driving behavior temporal covariate shift, where the distributions of driving behavior factors vary over time. To fill the gap, this study targets at obtaining time-invariant driving behavior features and establishing their relationships with high-risk event occurrence probability. Specifically, a generalized modeling framework consisting of distribution characterization (DC) and distribution matching (DM) modules was proposed. The DC module split the whole dataset into several segments with the largest distribution gaps, while the DM module identified time-invariant driving behavior features through learning common knowledge among different segments. Then, gated recurrent unit (GRU) was employed to conduct time-invariant driving behavior feature mining for high-risk event occurrence probability estimation. Moreover, modified loss functions were introduced for imbalanced data learning caused by the rarity of high-risk events. The empirical analyses were conducted utilizing online ride-hailing services data. Experiment results showed that the proposed generalized modeling framework provided a 7.2% higher average precision compared to the traditional i.i.d. assumption based approach. The modified loss functions further improved the model performance by 3.8%. Finally, benefits for the driver management program improvement have been explored by a case study, demonstrating a 33.34% enhancement in the identification precision of high-risk event prone drivers.

How Can the Trust-Change Direction be Measured and Identified During Takeover Transitions in Conditionally Automated Driving? Using Physiological Responses and Takeover-Related Factors

OBJECTIVE

This paper proposes an objective method to measure and identify trust-change directions during takeover transitions (TTs) in conditionally automated vehicles (AVs).

BACKGROUND

Takeover requests (TORs) will be recurring events in conditionally automated driving that could undermine trust, and then lead to inappropriate reliance on conditionally AVs, such as misuse and disuse.

METHOD

34 drivers engaged in the non-driving-related task were involved in a sequence of takeover events in a driving simulator. The relationships and effects between drivers' physiological responses, takeover-related factors, and trust-change directions during TTs were explored by the combination of an unsupervised learning algorithm and statistical analyses. Furthermore, different typical machine learning methods were applied to establish recognition models of trust-change directions during TTs based on takeover-related factors and physiological parameters.

RESULT

Combining the change values in the subjective trust rating and monitoring behavior before and after takeover can reliably measure trust-change directions during TTs. The statistical analysis results showed that physiological parameters (i.e., skin conductance and heart rate) during TTs are negatively linked with the trust-change directions. And drivers were more likely to increase trust during TTs when they were in longer TOR lead time, with more takeover frequencies, and dealing with the stationary vehicle scenario. More importantly, the F1-score of the random forest (RF) model is nearly 77.3%.

CONCLUSION

The features investigated and the RF model developed can identify trust-change directions during TTs accurately.

APPLICATION

Those findings can provide additional support for developing trust monitoring systems to mitigate both drivers' overtrust and undertrust in conditionally AVs.

Unveiling pre-crash driving behavior common features based upon behavior entropy

Driving behavior is considered as the primary crash influencing factor, whereas studies claimed that over 90% crashes were attributed by behavior features. Therefore, unveil pre-crash driving behavior features is of great importance for crash prevention. Previous studies have established the correlations between features such as vehicle speed, speed variability, and the probability of crash occurrences, but these analyses have concluded inconsistent results. This is due to the varying operating characteristics among roadway facilities, where given the same driving behavior statistical features, the corresponding traffic states are not identical. In this study, a behavioral entropy index was proposed to address the abovementioned issue. First, through comparing the individual driving behavior with the group distribution, behavioral entropy index was calculated to quantify the abnormality of driving behavior. Then, crash classification models were established by comparing the behavioral entropy prior to crash events and normal driving conditions. The empirical analyses have been conducted based on 1,634,770 naturalistic driving trajectories and 1027 crash events. And models have been carried out for urban roadway sections, urban intersections, and highway sections separately. The results showed that utilizing the behavior entropy instead of the statistical features could enhance the crash classification accuracy by 11.3%. And common pre-crash features of increased behavioral entropy were identified. Moreover, the speed coefficient of variation (QCV) entropy was concluded as the most influencing factor, which can be used for real-time driving risk monitoring and enables individual-level hazard mitigation.

Pedestrian crash causation analysis and active safety system calibration

Over 20 % of global crash fatalities involve pedestrians, but pedestrian crash causation and pedestrian protection systems have not been thoroughly developed or reliably tested. To understand the causation characteristics of pedestrian crashes, 398 pedestrian crashes were extracted from the China in-depth accident study (CIDAS), and most of these crashes were aggregated into five scenarios. The two scenarios with the highest proportion of crashes were analyzed by the driving reliability and error analysis method (DREAM) to identify high-risk causation patterns. From these patterns, three main contributing factors were identified: 1) extremely environmental light disturbance; 2) distracted driving caused by drivers' own thoughts; 3) drivers violating pedestrian yield law. Based on these patterns and factors, a pedestrian protection system was designed. It consists of a forward vision sensor and radar to sense the environment and the three-stage autonomous emergency braking (AEB) algorithm to automatically avoid pedestrian collisions. Crash scenarios from CIDAS data were recreated in MATLAB Simulink to test the pedestrian protection system proposed in this study. This system was found to reduce pedestrian crashes by more than 90 %. The optimal parameters for three AEB stages were obtained, with decelerations of 0.2 g, 0.3 g, and 0.6 g. This study designed an active safety system based on causation analysis of the vehicle-pedestrian crashes and calibrated the AEB algorithm of it, thus providing reference and insight for further development of pedestrian protection systems.

Critical safety management driver identification based upon temporal variation characteristics of driving behavior

Identifying critical safety management drivers with high driver-level risks is essential for traffic safety improvement. Previous studies commonly evaluated driver-level risks based upon aggregated statistical characteristics (e.g., driving exposure and driving behavior), which were obtained from long-period driving monitoring data. However, given the great advancements of the connected vehicle and in-vehicle data instrumentation technologies, there has been a notable increase in the collection of short-period driving data, which has emerged as a prominent data source for analysis. In this data environment, traditionally employed aggregated behavior characteristics are unstable due to the time-varying feature of driving behavior coupled with insufficient data sampling periods. Thus, traditional modeling methods based upon aggregated statistical characteristics are no longer feasible. Instead of utilizing such unreliable statistical information to represent driver-level risks, this study employed temporal variation characteristics of driving behavior to identify critical safety management drivers in the short-period driving data environment. Specifically, the relationships between driving behavior temporal variation characteristics and individual crash occurrence probability were developed. To eliminate the impacts of drivers' driving behavior heterogeneity on model performance, "traffic entropy" index that could quantify the abnormal degrees of driving behavior was proposed. Deep learning models including convolutional neural network (CNN) and long short-term memory (LSTM) were employed to conduct the temporal variation feature mining. Empirical analyses were conducted using data obtained from online ride-hailing services. Experiment results showed that temporal variation characteristics based models outperformed traditional aggregated statistical characteristics based models. The area under the curve (AUC) index was improved by 4.1%. And the proposed traffic entropy index further enhanced the model performance by 5.3%. The best model achieved an AUC of 0.754, comparable to existing approaches utilizing long-period driving data. Finally, applications of the proposed method in driver management program development and its further investigations have been discussed.

Context-aware driver risk prediction with telematics data

Driving risk prediction is crucial for safety and risk mitigation. While traditional methods rely on demographic information for insurance pricing, they may not fully capture actual driving behavior. To address this, telematics data has gained popularity. This study focuses on using telematics data and contextual information (e.g., road type, daylight) to represent a driver's style through tensor representations. Drivers with similar behaviors are identified by clustering their representations, forming risk cohorts. Past at-fault traffic accidents and citations serve as partial risk labels. The relative magnitude of average records (per driver) for each cohort indicates their risk label, such as low or high risk, which can be transferred to drivers in a cohort. A classifier is then constructed using augmented risk labels and driving style representations to predict driving risk for new drivers. Real-world data from major US cities validates the effectiveness of this framework. The approach is practical for large-scale scenarios as the data can be obtained at scale. Its focus on driver-based risk prediction makes it applicable to industries like auto-insurance. Beyond personalized premiums, the framework empowers drivers to assess their driving behavior in various contexts, facilitating skill improvement over time.

Adaptive forward collision warning system for hazmat truck drivers: Considering differential driving behavior and risk levels

The risky driving behavior of hazmat truck drivers is a crucial factor in many severe traffic accidents. In-vehicle Advanced Driving Assistance Systems (ADAS), integrating vehicle active safety and driver assistance technology, has been installed into hazmat trucks aiming to reduce driving risks during emergencies. This paper presents an enhanced dynamic Forward Collision Warning (FCW) model tailored for hazmat truck drivers with different driving characteristics and risk levels. Our objective is to determine the optimal moment to alert drivers during risky situations. The novelty of our approach lies in analyzing the driver's response mechanism to the warning by considering their characteristics and real-time driving risk levels. We employ a multi-objective optimization method that integrates real-time driving risk, driver acceptance, and driving comfort to calculate the optimal warning time. Our findings indicate that the appropriate warning time is similar for all drivers under high-level risks, while significant differentiation exists for different driver categories under mid-level and low-level risks. Additionally, aggressive drivers tend to follow leading vehicles closely and exhibit lower deceleration intentions when faced with dangers compared to normal and cautious drivers. Our research outcomes enable the development of user profiles for hazmat truck drivers based on extensive historical driving records, facilitating the analysis of driver response differences to FCWs. This enhances driving safety and improves driver trust in ADAS systems.

Revision of the driver behavior questionnaire for Chinese drivers' aberrant driving behaviors using naturalistic driving data

The Manchester Driver Behavior Questionnaire (DBQ) is a widely used self-reported measure of aberrant driving behaviors. It provides a standardized way of evaluating drivers' safety awareness and motivation, but the effectiveness of the DBQ's application in different regions can be influenced by culture, social norms, and time period. Several studies have adjusted DBQ items to reflect driver behavior native to particular regions or times, but few have used objective measurements to make proper adjustments. A naturalistic driving study (NDS) provides vehicle kinematic data and in-vehicle videos that objectively capture actual driving behaviors. The gender, age, and driving experience characteristics of aberrant driving behaviors were analyzed, and, based on comparisons between the DBQ self-reported driving behaviors and those observed in the Shanghai, China, NDS, the existing items from the Manchester DBQ were subsequently adjusted. Sixty-two types of real-world aberrant driving behaviors were extracted from 490 valid crash and near crash events observed in the Shanghai NDS. Aberrant driving behavior rates were calculated for individual characteristics (gender, age, and driving experience), and factor rates were calculated based on the three DBQ factor types of violation, error, and lapse. Results revealed that (a) male drivers, drivers in their thirties, and those with three to five years of driving experience demonstrated higher rates of aberrant driving behaviors; and (b) there were weak correlations between observed NDS factor rates and self-reported DBQ scores, and only slight differences among drivers divided by factor rate level (e.g., high violation rate). The questionnaire calibrated for Chinese drivers includes 23 items. Five of the original 24 DBQ items were modified, eight were left unchanged, eleven were deleted, and ten field-observed combined behaviors were added. In addition to the importance of adjusting the DBQ for today's Chinese drivers, this study provides a method for objectively modifying DBQ items in the future in accord with observed driving behaviors in an NDS.

Application of naturalistic driving data: A systematic review and bibliometric analysis

The application of naturalistic driving data (NDD) has the potential to answer critical research questions in the area of driving behavior assessment, as well as the impact of exogenous and endogenous factors on driver safety. However, the presence of a large number of research domains and analysis foci makes a systematic review of NDD applications challenging in terms of information density and complexity. While previous research has focused on the execution of naturalistic driving studies and on specific analysis techniques, a multifaceted aggregation of NDD applications in Intelligent Transportation System (ITS) research is still unavailable. In spite of the current body of work being regularly updated with new findings, evolutionary nuances in this field remain relatively unknown. To address these deficits, the evolutionary trend of NDD applications was assessed using research performance analysis and science mapping. Subsequently, a systematic review was conducted using the keywords "naturalistic driving data" and "naturalistic driving study data". As a result, a set of 393 papers, Published between January 2002-March 2022, was thematically clustered based on the most common application areas utilizing NDD. the results highlighted the relationship between the most crucial research domains in ITS, where NDD had been incorporated, and application areas, modeling objectives, and analysis techniques involving naturalistic databases.

The Effect of Driving Style on Responses to Unexpected Vehicle Cyberattacks

Vehicle cybersecurity is a serious concern, as modern vehicles are vulnerable to cyberattacks. How drivers respond to situations induced by vehicle cyberattacks is safety critical. This paper sought to understand the effect of human drivers’ risky driving style on response behavior to unexpected vehicle cyberattacks. A driving simulator study was conducted wherein 32 participants experienced a series of simulated drives in which unexpected events caused by vehicle cyberattacks were presented. Participants’ response behavior was assessed by their change in velocity after the cybersecurity events occurred, their post-event acceleration, as well as time to first reaction. Risky driving style was portrayed by scores on the Driver Behavior Questionnaire (DBQ) and the Brief Sensation Seeking Scale (BSSS). Half of the participants also received training regarding vehicle cybersecurity before the experiment. Results suggest that when encountering certain cyberattack-induced unexpected events, whether one received training, driving scenario, participants’ gender, DBQ-Violation scores, together with their sensation seeking measured by disinhibition, had a significant impact on their response behavior. Although both the DBQ and sensation seeking have been constantly reported to be linked with risky and aberrant driving behavior, we found that drivers with higher sensation seeking tended to respond to unexpected driving situations induced by vehicle cyberattacks in a less risky and potentially safer manner. This study incorporates not only human factors into the safety research of vehicle cybersecurity, but also builds direct connections between drivers’ risky driving style, which may come from their inherent risk-taking tendency, to response behavior to vehicle cyberattacks.

Differences in frequency of occurrence, event characteristics, and pre-impact vehicle kinematics between crashes, near-crashes, and single vehicle conflicts in a large-scale naturalistic driving study

Objective: Motor vehicle crashes result in egregious personal injury, mortality, and economic cost but are relatively rare in naturalistic observations. There is, however, evidence of strong relationships between crashes and less severe (but more common) "surrogate" events (e.g., near-crashes). Despite this strong relationship, there can still be some important differences in findings when these surrogate events are investigated in lieu of, or combined with, crashes. Therefore, it is relevant to describe and quantify differences between crashes and crash-surrogate events. Consequently, the focus of this investigation was to establish how crashes and crash surrogate events in a large-scale naturalistic driving study compare in terms of frequency of occurrence, event characteristics, and pre-impact vehicle kinematics.Methods: Crashes, near-crashes, and single-vehicle conflicts (SVCs) derived from the Second Strategic Highway Research Program Naturalistic Driving Study were coded to summarize the environmental and contributing variables involved. The original coding for these events was downsized to the variables of interest, and those variables underwent recoding to simplify the coded options. Additional variables based on the kinematic characteristics for each event were also derived and analyzed. Multinomial logistic regression was used to assess the contributions of these different variables toward classification of an event as a crash, near-crash, or SVC.Results: The regression model comparing crashes with near-crashes and SVCs identified several variables that allowed differentiation between crashes and these surrogates, primarily the pre-incident maneuver of the subject vehicle and the evasive maneuver that was executed by the driver. Kinematic variables prior to event onset, however, were not predictive of event outcome.Conclusions: The results suggest that important differences exist between crashes and their near-crash surrogates, and between crashes and SVCs. These results, however, should not discourage the analysis of surrogate events, which still provide useful information in prevention and mitigation of crash circumstances. This investigation highlights how crashes are different from two types of surrogate events and provides information that may allow for more precise analysis of these surrogate events in the future.

Causation analysis of crashes and near crashes using naturalistic driving data

Understanding crash causation to the extent needed for applying countermeasures has always been a focus as well as a difficulty in the field of traffic safety. Previous research has been limited by insufficient crash data and analysis methods more suitable to single crashes. The use of crashes and near crashes (CNCs) and naturalistic driving studies can help solve the data problem, and use of pre-crash scenarios can identify the high-prevalence causes across multiple crashes of a given scenario. This study therefore proposes a two-stage crash causation analysis method based on pre-crash scenarios and a crash causation derivation framework that systematically categorizes and analyzes contributing factors. From the Shanghai Naturalistic Driving Study (SH-NDS), 536 CNCs were extracted, and were grouped into 23 different pre-crash scenarios based on the National Highway Traffic Safety Administration (NHTSA) pre-crash scenario typology. In-depth investigations were conducted, and CNCs sharing the same scenario were analyzed using the proposed framework, which identifies causation patterns based on the interaction of the framework's road user, vehicle, roadway infrastructure, and roadway environment subsystems. Through statistical analysis, the causation patterns and their contributing factors were compared for three common pre-crash scenarios of highest incidence: rear-end, lane change, and vehicle-pedalcyclist. Braking error in low-speed car following, following too closely, and non-driving-related distraction were important causes of rear-end scenarios. In lane change scenarios, the main causation patterns included illegal use of turn signals and dangerous lane changes as critical factors. Pedalcyclist scenarios were particularly impacted by visual obstructions, inadequate lanes for non-motorized vehicles, and pedalcyclists violating traffic regulations. Based on the identified causation patterns and their contributing factors, countermeasures for the three common scenarios are suggested, which provide support for safety improvement projects and the development of advanced driver assistance systems.

The influence of achievement goals on objective driving behavior

Investigating psychological characteristics through self-reported measures (e.g., anger, sensation seeking) and dynamic behaviors through objective measures (e.g., speed, 2D acceleration, GPS position etc.) may allow us to better understand the behavior of at-risk drivers. To assess drivers' motivation, the theoretical framework of achievement goals has been studied recently. These achievement goals can influence the decision-making and behaviors of individuals engaged in driving. The four achievement goals in driving are: seeking to improve or to drive as well as possible (mastery-approach), to outperform other drivers (performance-approach), to avoid driving badly (mastery-avoidance), and to avoid being the worst driver (performance-avoidance). Naturalistic Driving Studies (NDS) provide access to the objective measurements of data not accessible through self-reported measurements (i.e., speed, accelerations, GPS position). Three dynamic criteria have been developed to characterize the behavior of motorists objectively: driving events, time spent above acceleration thresholds (longitudinal and transversal), and the extent of dynamic demands. All these criteria have been measured in different road contexts (e.g., plain). The aim of this study was to examine the predictive role of the four achievement goals on these objective driving behaviors. 266 drivers (96 women, 117 men) took part in the study, and 4 242 482 km was recorded during 8 months. Simultaneously, they completed the Achievement Goals in Driving Questionnaire. The main results highlighted that mastery-approach goals negatively predicted hard braking and the extent of dynamic demands on plain and hilly roads. Mastery-approach goals seem to be the most protective goals in driving. Future research on the promotion of mastery-approach goals in driving may be able to modify the behavior of at-risk drivers.

A data-driven Bayesian network for probabilistic crash risk assessment of individual driver with traffic violation and crash records

In recent years, individual drivers' crash risk assessments have received much attention for identifying high-risk drivers. To this end, we propose a probabilistic assessment method of crash risks with a reproducible long-term dataset (i.e., traffic violations, license, and crash records). In developing this method, we used 7.75 million violations and crashes of 5.5 million individual drivers in Seoul, South Korea, from June 2013 to June 2017 (four years). The stochastic process of the Bayesian network (BN), whose structure is optimized by tabu-search, successfully evaluates individual drivers' crash and violation probability. In addition, the cluster analysis classifies drivers into five distinctive groups according to their estimated violation and crash probabilities. As a result, this study found that the estimated average crash rate within a cluster converges with the actual crash rate by the proposed framework without privacy issues. We also confirm that violation records and expected crash probability are strongly correlated, and there is a direct relationship between a driver's previous violations and crash record and the future at-fault crash. The proposed assessment method is valuable in developing proactive driver education programs and safety countermeasures, including adjusting the penalty system and developing user-based insurance by recognizing dangerous drivers and identifying their properties.

Predicting driving speed from psychological metrics in a virtual reality car driving simulation

Why do some people tend to drive faster than others? Personality characteristics such as the evaluation of risk to oneself or to others, impulsivity, adherence to norms, but also other personal factors such as gender, age, or driving experience all may play a role in determining how fast people drive. Since driving speed is a critical factor underlying accident prevalence, identifying the psychological metrics to predict individual driving speed is an important step that could aid in accident prevention. To investigate this issue, here, we used an immersive virtual reality driving simulation to analyze average driving speed. A total of 124 participants first took a comprehensive set of personality and background questionnaires and a behavioral risk-taking measure. In the virtual reality experiment, participants were required to navigate a difficult driving course in a minimally-restricted, non-urban setting in order to provide baseline results for speed selection. Importantly, we found that sensation seeking and gender significantly predicted the average driving speed, and that sensation seeking and age were able to predict the maximum driving speed.

Risk Levels Classification of Near-Crashes in Naturalistic Driving Data

Identifying dangerous events from driving behavior data has become a vital challenge in intelligent transportation systems. In this study, we compared machine and deep learning-based methods for classifying the risk levels of near-crashes. A dataset was built for the study by considering variables related to naturalistic driving, temporal data, participants, and road geometry, among others. Hierarchical clustering was applied to categorize the near-crashes into several risk levels based on high-risk driving variables. The adaptive lasso variable model was adopted to reduce factors and select significant driving risk factors. In addition, several machine and deep learning models were used to compare near-crash classification performance by training the models and examining the model with testing data. The results showed that the deep learning models outperformed the machine learning and statistical models in terms of classification performance. The LSTM model achieved the highest performance in terms of all evaluation metrics compared with the state-of-the-art models (accuracy = 96%, recall = 0.93, precision = 0.88, and F1-measure = 0.91). The LSTM model can improve the classification accuracy and prediction of most near-crash events and reduce false near-crash classification. The finding of this study can benefit transportation safety in predicting and classifying driving risk. It can provide useful suggestions for reducing the incidence of critical events and forward road crashes.

Safety assessment of trucks based on GPS and in-vehicle monitoring data

Increasingly, drivers are choosing to buy usage-based automobile insurance (UBI). Manage-how-you-drive (MHYD) insurance, a new type of UBI, incorporates active safety management to monitor driver behavior and issue warnings as needed. While researchers have introduced telematics data into automobile insurance pricing, the specific effect of in-vehicle active safety management on driver risk assessment has been neglected, especially for truck drivers, whose crashes have more serious consequences. This study uses telematics and in-vehicle monitoring features to examine the key factors underlying large commercial truck crashes, and quantifies the effect of these factors on crash risk. Data from 2,185 trucks in Shanghai, China, were collected for a total of 105,786 trips and 465,555 in-vehicle warnings to investigate three types of factors affecting risk: travel characteristics, driving behavior, and in-vehicle warnings. A zero-inflated Poisson (ZIP) regression model was built, and a ZIP model without the warning variables as well as a basic Poisson model with warnings were considered for comparison. It was found that the ZIP model considering in-vehicle warning information performed significantly better than the other models. The standardized regression coefficient method was used to identify the most important variables. In-vehicle yawn and smoking warnings had significantly more association with the number of crashes than did the travel characteristics and driving behavior variables, though freeway distance traveled, average freeway speed, percentage of trips on sunny days, and percentage of trips at night also correlated significantly with crash risk. These results can provide a reference for UBI insurance professionals considering in-vehicle active safety management, as well as support freight companies in drafting appropriate working regulations.

Global lessons learned from naturalistic driving studies to advance traffic safety and operation research: A systematic review

The state of practice of investigating traffic safety and operation is primarily based on traditional data sources, such as spot sensors, loop detectors, and historical crash data. Recently, researchers have utilized transportation data from instrumented vehicles, driving simulators, and microsimulation modeling. However, these data sources might not represent the actual driving environment at a trajectory level and might introduce bias due to their experimental control. The shortcomings of these data sources can be overcome via Naturalistic Driving Studies (NDSs) considering the fact that NDS provides detailed real-time driving data that would help investigate the safety and operational impacts of human behavior along with other factors related to weather, traffic, and roadway geometry in a naturalistic setting. With the enormous potential of the NDS data, this study leveraged the Preferred Reporting Items for Systematic Reviews and Meta-Analyzes (PRISMA) approach to shortlist the most relevant naturalistic studies out of 2304 initial studies around the world with a focus on traffic safety and operation over the past fifteen years (2005-2020). A total of 117 studies were systematically reviewed, which were grouped into seven relevant topics, including driver behavior and performance, crash/near-crash causation, driver distraction, pedestrian/bicycle safety, intersection/traffic signal related studies, detection and prediction using NDSs data, based on their frequency of appearance in the keywords of these studies. The proper deployment of Connected and Autonomous Vehicles (CAV) require an appropriate level of human behavior integration, especially at the intimal stages where both CAV and human-driven vehicles will interact and share the same roadways in a mixed traffic environment. In order to integrate the heterogeneous nature of human behavior through behavior cloning approach, real-time trajectory-level NDS data is essential. The insights from this study revealed that NDSs could be effectively leveraged to perfect the behavior cloning to facilitate rapid and safe implementation of CAV.

Assessing safety critical driving patterns of heavy passenger vehicle drivers using instrumented vehicle data - An unsupervised approach

Assessing the individual's driving profile and identifying the at-fault behaviors contributes to road safety, riding comfort, and driver assistance systems. This study proposes a framework to identify aggressive driving patterns in longitudinal control using real-time driving profiles of heavy passenger vehicle (HPV) drivers. The main objective is to detect and quantify the instantaneous driving decisions and classify the identified maneuvers (acceleration, braking) using unsupervised machine learning techniques without any prior-ground truth. To this end, total 8295 acceleration events, and 7151 braking events, were extracted from 142 driving profiles collected using high-resolution (10 Hz) GPS instrumentation. The principal component analysis was conducted on a multi-dimensional feature set, followed by a two-stage k-means clustering on the reduced feature subspace. The results showed that 86.5% of accelerations and 65.3% of braking maneuvers were characterized as non-aggressive, indicating safe or base-line driving behavior. However, 13.5% of accelerations and 34.7% of braking maneuvers were featured to be aggressive, indicative of the actual risky behaviors. Further analysis demonstrated the heterogeneity in drivers' trip-level frequency of aggressive maneuvers and highlighted the need for a continuous driving assessment. The study also revealed that the thresholds derived from the obtained clusters featuring the aggressive accelerations (+0.3 to +0.48 g) and aggressive braking (-0.42 to -0.27 g) maneuvers were beyond the acceptable limits of passenger safety and comfort. The insights from the study aids in developing driver assistance systems for personalized feedback provision and improve driver behavior.

Quantifying the effect of roadway, driver, vehicle, and location characteristics on the frequency of longitudinal and lateral accelerations

The purpose of this study is to understand and quantify the simultaneous effects of roadway speed category, driver age, driver gender, vehicle class, and location on the rates of longitudinal and lateral acceleration epochs. The rate of usual as well as harsh acceleration epochs are used to extract insights on driving risk and driver comfort preferences. However, an analysis of acceleration rates at multiple thresholds incorporating various effects while using a large-scale and diverse dataset is missing. This analysis will fill this research gap. Data from the 2nd Strategic Highway Research Program Naturalistic Driving Study (SHRP2 NDS) was used for this analysis. The rate of occurrence of acceleration epochs was modeled using negative binomial distribution based generalized linear mixed effect models. Roadway speed category, driver age, driver gender, vehicle class, and location were used as the fixed effects and the driver identifier was used as the random effect. Incidence rate ratios were then calculated to compare subcategories of each fixed effect. Roadway speed category has the strongest effect on longitudinal and lateral accelerations of all magnitudes. Acceleration epoch rates consistently decrease as the roadway speed category increases. The difference in the rates depends on the threshold and is up to three orders of magnitude. Driver age is another significant factor with clear trends for longitudinal and lateral acceleration epochs. Younger and older drivers experience higher rates of longitudinal accelerations and decelerations. However, the rate of lateral accelerations consistently decreases with age. Vehicle class also has a significant effect on the rate of harsh accelerations with minivans consistently experiencing lower rates.

Analysis of factors influencing aggressive driver behavior and crash involvement

OBJECTIVE

Aggressive driver behavior is one of the major contributing factors to road crashes. However, the relationship between aggressive driver behavior and crash risk is scarcely explored. The present study focused on quantifying the effect of aggressive driver behavior on crash probability.

METHOD AND DATA SOURCES

A sample of 405 Indian drivers were analyzed to model the aggressive driver behavior using self-reported measures. Generalized linear models were developed to quantify the effects of independent variables such as age, gender, personality traits (e.g., driving anger, physical aggression, hostility), and individual predilections to commit violations (e.g., excessive speeding and frequent risky overtaking) on aggressive driver behavior and crash probabilities.

RESULTS

K-means clustering technique was applied to the Aggressive Driving Scale (ADS) scores to cluster the drivers into three groups (aggressive, normal, and cautious). Gender was significantly correlated with aggressive driver behavior. Compared to female drivers, male drivers were 2.57 times more likely to engage in aggressive driving. Driver's age was negatively correlated with aggressive driving. With one-year increment in driver's age, the tendency of a driver to engage in aggressive driving was reduced by 26%. In addition, the likelihood of being engaged in aggressive driving was increased by 2.98 times and 2.15 times for the drivers who engage in excessive speeding and frequent risky overtaking, respectively. Driver's personality traits were significantly correlated with aggressive drivers. The crash involvement model showed that aggressive drivers were 2.79 times more likely to be involved in road crashes than cautious drivers. Further, married drivers were 2.17 times less likely to be involved in crashes, whereas for professional drivers the crash involvement probability was increased by 75%.

CONCLUSIONS

The results revealed that in addition to age and gender personality traits were significant predictors of driving aggression. Further, the driver's marital status was negatively correlated with the crash involvement and professional drivers were more likely to be involved in crashes than nonprofessional drivers. The study findings can be used in identifying specific risk-prone drivers to provide safety measures via in-vehicle Advanced Driver Assistance Systems (ADAS).

Developing personalised braking and steering thresholds for driver support systems from SHRP2 NDS data

Examining the relationships between the factors associated with the crash development enabled the realisation of driver support systems aiming to proactively avert and control crash causation at various points within the crash sequence. Developing such systems requires new insights in personalised pre-crash driver behaviour with respect to braking and steering to develop crash prevention strategies. Therefore, the current study utilises Strategic Highway Research Program 2 Naturalistic Driving Studies (SHRP2 NDS) data to investigate personalised steering and braking thresholds by examining the last stage of a crash sequence. More specifically, this paper carried out an in-depth examination of braking and steering manoeuvres observed in the final 30 s prior to safety critical events. Two algorithms were developed to extract braking and steering events by examining deceleration and yaw rate and another developed and applied to determine the sequence of the manoeuvres. Based on the analysis, thresholds for detecting emerging situations were recommended. The investigation of driver behaviour before the safety critical events, provides valuable insights into the transition from normal driving to safety critical scenarios. The results indicate that 20% of the drivers did not react to the impending event suggesting that they were not aware of the imminent safety critical situation. Future development of Advanced Driver Assistance Systems (ADAS) can focus on individual drivers' needs with tailored activation thresholds. The developed algorithms can facilitate driver behaviour and safety analysis for NDS while the thresholds recommended could be exploited for the design of new driver support systems.

Driver influence on vehicle trajectory prediction

Drivers continually interact with other road users and use information from the road environment to make decisions to control their vehicle. A clear understanding of different parameters impacting this interaction can provide us with a new design approach for a more effective driver assistance system - a personalised trajectory prediction system. This paper highlights the influential factors on trajectory prediction system performance by (i) identifying driver behaviours impacting the trajectory prediction system; and (ii) analysing other contributing factors such as traffic density, secondary task, gender and age group. To explore the most influential contributing factors, we first train an interaction-aware trajectory prediction system using time-series data derived from the Second Strategic Highway Research Program (SHRP2) Naturalistic Driving Study (NDS). Prediction error is then analysed based on driver characteristics such as driver profile which is subjectively measured through self-reported questions, and driving performance which is based on evaluation of time-series information such as speed, acceleration, jerk, time, and space headway. The results show that prediction error significantly increased in the scenarios where the driver engaged in risky behaviour. Analysis shows that trajectory prediction system performance is also affected by factors such as traffic density, engagement in secondary tasks, driver gender and age group. We show that the driver profile, which is subjectively measured using self-reported questionnaires, is not as significant as the driving performance information, which is objectively measured and extracted during each specific driving scenario.

Decision-adjusted driver risk predictive models using kinematics information

Accurate prediction of driving risk is challenging due to the rarity of crashes and individual driver heterogeneity. One promising direction of tackling this challenge is to take advantage of telematics data, increasingly available from connected vehicle technology, to obtain dense risk predictors. In this work, we propose a decision-adjusted framework to develop optimal driver risk prediction models using telematics-based driving behavior information. We apply the proposed framework to identify the optimal threshold values for elevated longitudinal acceleration (ACC), deceleration (DEC), lateral acceleration (LAT), and other model parameters for predicting driver risk. The Second Strategic Highway Research Program (SHRP 2) naturalistic driving data were used with the decision rule of identifying the top 1% to 20% of the riskiest drivers. The results show that the decision-adjusted model improves prediction precision by 6.3% to 26.1% compared to a baseline model using non-telematics predictors. The proposed model is superior to models based on a receiver operating characteristic curve criterion, with 5.3% and 31.8% improvement in prediction precision. The results confirm that the optimal thresholds for ACC, DEC and LAT are sensitive to the decision rules, especially when predicting a small percentage of high-risk drivers. This study demonstrates the value of kinematic driving behavior in crash risk prediction and the necessity for a systematic approach for extracting prediction features. The proposed method can benefit broad applications, including fleet safety management, use-based insurance, driver behavior intervention, as well as connected-vehicle safety technology development.

An integrated methodology for real-time driving risk status prediction using naturalistic driving data

Real-time driving risk status prediction is critical for developing proactive traffic intervention strategies and enhance driving safety. However, the optimal observation time window length and prediction time window length, which should be the prerequisite for the timeliness and accuracy of real-time driving risk status prediction model, have been rarely explored in previous studies. In this study, a methodology which integrates driving risk status identification, rolling time window-based feature extraction, real-time driving risk status prediction and driving risk influencing factors analysis was proposed to accurately evaluate and predict real-time driving risk status. The methodology was tested based on 1,440 car-following events from Shanghai Naturalistic Driving Study. Results show that four driving risk statuses (safe, low-risk, median-risk and high-risk) are most appropriate to establish risk labelling criteria. In addition, results from driving risk status prediction show that when the observation time window length is 0.5 s, the accuracy rate of predicting medium-risk or high-risk status occurring in the next 0.7 s is higher than 85 % using multi-layer perceptron model. Meanwhile, the results from the analysis of influencing factors show that the input variables related to the risk status score higher in the ranking of feature importance. A part from that, speed difference, headway distance, speed and acceleration are still important in predicting driving risk status. The proposed methods in this paper can be applied in connected and autonomous vehicle (CAV) to reduce driver cognitive workload and hence improve driving safety fed with naturalistic driving data collected using in-vehicle systems.

Temporal analysis of driving efficiency using smartphone data

This paper attempts to shed light on the temporal evolution of driving safety efficiency with the aim to acquire insights useful for both driving behavior and road safety improvement. Data exploited herein are collected from a sophisticated platform that uses smartphone device sensors during a naturalistic driving experiment, at which the driving behavior from a sample of two hundred (200) drivers during 7-months is continuously recorded in real time. The main driving behavior analytics taken into consideration for the driving assessment include distance travelled, acceleration, braking, speed and smartphone usage. The analysis is performed using statistical, optimization and machine learning techniques. The driver's safety efficiency index is estimated both in total and in several consecutive time windows to allow for the investigation of safety efficiency evolution in time. Initial data analysis results to the most critical components of microscopic driving behaviour evolution, which are used as inputs in the k-means algorithm to perform the clustering analysis. The main driving characteristics of each cluster are identified and lead to the conclusion that there are three main driving groups of the a) moderate drivers, b) unstable drivers and c) cautious drivers.

The mediating effect of driver characteristics on risky driving behaviors moderated by gender, and the classification model of driver's driving risk

High-risk drivers are more likely to be involved in traffic accidents, and the driving risk level of drivers could be affected by many potential factors, such as demographics and personality traits. Based on the Structural Equation Model (SEM), this study involves a sample of 3150 drivers from the Strategic Highway Research Program 2 (SHRP 2), to explore the relationships among drivers' demographic characteristics (gender, age, and cumulative driving years), sensation seeking, risk perception, and risky driving behaviors. More specifically, the mediation model of driver characteristics on risky driving behaviors moderated by gender is constructed by the SEM. The results show that the effects of driving experience on risky driving behaviors are partially mediated by sensation seeking and risk perception for male drivers, while those are completely mediated by sensation seeking and risk perception for female drivers. Moreover, the development trend of risky driving behavior engagements declines greater with the growing of driving experience for female drivers than male drivers. Finally, a classification model of the driver's driving risk is proposed by the Random Forest classifier, in which the driving risk level of the driver evaluated by the crash and near-crash rate could be classified through the driver's self-reported demographics, sensation seeking, risk perception, and risky driving behaviors. The classification accuracy achieves up to 90 percent, which offers an alternative approach to identifying potential high-risk drivers to reduce property losses, injuries, and death caused by traffic accidents.

Fusing crash data and surrogate safety measures for safety assessment: Development of a structural equation model with conditional autoregressive spatial effect and random parameters

Most existing efforts to assess safety performance require sufficient crash data, which generally takes a few years to collect and suffers from certain limitations (such as long data collection time, under-reporting issue and so on). Alternatively, the surrogate safety measure (SSMs) based approach that can assess traffic safety by capturing the more frequent "near-crash" situations have been developed, but it is criticized for the potential sampling and measurement errors. This study proposes a new safety performance measure-Risk Status (RS), by fusing crash data and SSMs. Real-world connected vehicle data collected in the Safety Pilot Model Deployment (SPMD) project in Ann Arbor, Michigan is used to extract SSMs. With RS treated as a latent variable, a structural equation model with conditional autoregressive spatial effect and corridor-level random parameters is developed to model the interrelationship among RS, crash frequency, risk identified by SSMs, and contributing factors. The modeling results confirm the proposed interrelationship and the necessity to account for both spatial autocorrelation and unobserved heterogeneity. RS can integrate both crash frequency and SSMs together while controlling for observed and unobserved factors. RS is found to be a more reliable criterion for safety assessment in an implementation case of hotspot identification.

Analyzing driver behavior under naturalistic driving conditions: A review

For a decade, researchers working in the area of road safety have started exploring the use of driving behavior data for a better understanding of the causes related to road accidents. A review of the literature reveals the excellent potential of naturalistic driving studies carried out by collecting vehicle performance data and driver behavior data during normal, impaired, and safety-critical situations. An in-depth understanding of driver behavior helps analyze and implement pre-crash safety measures - the development of enforcement policies, infrastructure design, and intelligent vehicle safety systems. The present paper attempts to review the naturalistic driving studies that have been undertaken so far. The paper begins with an overview of different methods for collecting unobtrusive driver behavior data during their day to day trip, followed by a discussion of various factors affecting driving behavior and their influence on vehicle performance parameters. The paper also discusses the strategies mentioned in the literature for improving driving behavior using naturalistic driving studies to enhance road safety. Some of the major findings of this review suggest that i) driver behavior is a major cause in the majority of the road accidents ii) drivers generally reduce their speed and increases headway as a compensatory measure to reduce the workload imposed during distracting activity and adverse weather conditions iii) mobile phone has emerged as a potential device for collecting naturalistic driving data and, iv) improvement in driving behavior can be achieved by providing feedback to the drivers about their driving behavior. This can be done by implementing usage-based insurance schemes such as pay as you drive (PAYD), pay how you drive (PHYD), and manage how you drive (MHYD). While a considerable amount of research has been done to analyze driving behavior under naturalistic conditions, some areas which are yet to be explored are highlighted in the present paper.

Risk factors associated with truck-involved fatal crash severity: Analyzing their impact for different groups of truck drivers

INTRODUCTION

Fatal crashes that include at least one fatality of an occupant within 30 days of the crash cause large numbers of injured persons and property losses, especially when a truck is involved.

METHOD

To better understand the underlying effects of truck-driver-related characteristics in fatal crashes, a five-year (from 2012 to 2016) dataset from the Fatality Analysis Reporting System (FARS) was used for analysis. Based on demographic attributes, driving violation behavior, crash histories, and conviction records of truck drivers, a latent class clustering analysis was applied to classify truck drivers into three groups, namely, ''middle-aged and elderly drivers with low risk of driving violations and high historical crash records," ''drivers with high risk of driving violations and high historical crash records," and ''middle-aged drivers with no driving violations and conviction records." Next, equivalent fatalities were used to scale fatal crash severities into three levels. Subsequently, a partial proportional odds (PPO) model for each driver group was developed to identify the risk factors associated with the crash severity. Results' Conclusions: The model estimation results showed that the risk factors, as well as their impacts on different driver groups, were different. Adverse weather conditions, rural areas, curved alignments, tractor-trailer units, heavier weights and various collision manners were significantly associated with the crash severities in all driver groups, whereas driving violation behaviors such as driving under the influence of alcohol or drugs, fatigue, or carelessness were significantly associated with the high-risk group only, and fewer risk factors and minor marginal effects were identified for the low-risk groups. Practical Applications: Corresponding countermeasures for specific truck driver groups are proposed. And drivers with high risk of driving violations and high historical crash records should be more concerned.

Driving safety assessment for ride-hailing drivers

Ride-hailing services, which have become increasingly prevalent in the last decade, provide an efficient travel mode by matching drivers and travelers via smartphone apps. Ride-hailing services enable millions of non-traditional taxi drivers to provide travel services, but may also raise safety concerns due to heterogeneity in the driver population. This study evaluated crash risk factors for ride-hailing drivers, including driving history and ride-hailing operational characteristics, using a sample of 189,815 drivers. We utilized the Poisson generalized additive model to accommodate for the potential nonlinear relationship between crash rate and risk factors. Results showed that crash history, the percentage of long-shift bookings, driving distance, operations during peak hours, years of being a ride-hailing driver, and passenger rating were significantly associated with crash risk. Several factors showed nonlinear relationships with crash risk. We adopted the SHapley Additive exPlanation (SHAP) method to assess and visualize the impact of each risk factor. The results indicated that passenger average rating, total driving distance, and crash history were the leading contributing factors. The findings of this study provide critical information for the development of safety countermeasures, driver education programs, as well as safety regulations for the ride-hailing industry.

The use of machine learning improves the assessment of drug-induced driving behaviour

RATIONALE

Car-driving performance is negatively affected by the intake of alcohol, tranquillizers, sedatives and sleep deprivation. Although several studies have shown that the standard deviation of the lateral position on the road (SDLP) is sensitive to drug-induced changes in simulated and real driving performance tests, this parameter alone might not fully assess and quantify deviant or unsafe driving.

OBJECTIVE

Using machine learning we investigated if including multiple simulator-derived parameters, rather than the SDLP alone would provide a more accurate assessment of the effect of substances affecting driving performance. We specifically analysed the effects of alcohol and alprazolam.

METHODS

The data used in the present study were collected during a previous study on driving effects of alcohol and alprazolam in 24 healthy subjects (12 M, 12 F, mean age 26 years, range 20-43 years). Various driving features, such as speed and steering variations, were quantified and the influence of administration of alcohol or alprazolam was assessed to assist in designing a predictive model for abnormal driving behaviour.

RESULTS

Adding additional features besides the SDLP increased the model performance for prediction of drug-induced abnormal driving behaviour (from an accuracy of 65 %-83 % after alprazolam intake and from 50 % to 76 % after alcohol ingestion). Driving behaviour influenced by alcohol and alprazolam was characterised by different feature importance, indicating that the two interventions influenced driving behaviour in a different way.

CONCLUSION

Machine learning using multiple driving features in addition to the state-of-the-art SDLP improves the assessment of drug-induced abnormal driving behaviour. The created models may facilitate quantitative description of abnormal driving behaviour in the development and application of psychopharmacological medicines. Our models require further validation using similar and unknown interventions.

Personalized Subjective Driving Risk: Analysis and Prediction

Subjective risk assessment is an important technology for enhancing driving safety, because an individual adjusts his/her driving behavior according to his/her own subjective perception of risk. This study presents a novel framework for modeling personalized subjective driving risk during expressway lane changes. The objectives of this study are twofold: (i) to use ego vehicle driving signals and surrounding vehicle locations in a data-driven and explainable approach to identify the possible influential factors of subjective risk while driving and (ii) to predict the specific individual’s subjective risk level just before a lane change. We propose the personalized subjective driving risk model, a combined framework that uses a random forest-based method optimized by genetic algorithms to analyze the influential risk factors, and uses a bidirectional long short term memory to predict subjective risk. The results demonstrate that our framework can extract individual differences of subjective risk factors, and that the identification of individualized risk factors leads to better modeling of personalized subjective driving risk.

Investigating the Significant Individual Historical Factors of Driving Risk Using Hierarchical Clustering Analysis and Quasi-Poisson Regression Model

Driving risk varies substantially according to many factors related to the driven vehicle, environmental conditions, and drivers. This study explores the contributing historical factors of driving risk with hierarchical clustering analysis and the quasi-Poisson regression model. The dataset of the study was collected from two sources: naturalistic driving experiments and self-reports. The drivers who participated in the naturalistic driving experiment were categorized into four risk groups according to their near-crash frequency with the hierarchical clustering method. Moreover, a quasi-Poisson model was used to identify the essential factors of individual driving risk. The findings of this study indicated that historical driving factors have substantial impacts on individual risk of drivers. These factors include the total number of miles driven, the driver’s age, the number of illegal parking (past three years), the number of over-speeding (past three years) and passing red lights (past three years). The outcome of the study can help transportation officials, educators, and researchers to consider the influencing factors on individual driving risk and can give insights and provide suggestions to improve driving safety.

Risk Assessment of Commercial dangerous -goods truck drivers using geo-location data: A case study in China

The primary objective of this study is to understand the relationship between driving risk of commercial dangerous-goods truck (CDT) and exposure factors and find a way to evaluate the risk of specific transportation environment, such as specific transportation route. Due to increasing transportation demand and potential threat to public, commercial dangerous goods transportation (CDGT) has drawn attention from decision makers and researchers within governmental and non-governmental safety organization. However, there are few studies focusing on driving risk assessment of commercial dangerous-goods truck by environmental factors. In this paper we employ survival analysis methods to analyze the impact of risk exposure factors on non-accident mileage of commercial dangerous-good truck and assess risk level of specific driving environment. Using raw location data from six transportation companies in China, we derive a set of 17 risk exposure factors that we use for model parameters estimation. The survival model and hazard model were estimated using the Weibull distribution as the baseline distribution. The results show that four factors - weather, traffic flow, travel time and average velocity have a significant impact on the non-accident mileage of driver in this company, and the assessment results of survival function and hazard function are robust to the different levels of testing data. The employment time has some effect on the results but does not result in a significant difference in most cases, and the task stability has little impact on the results. The findings of this study should be useful for decision makers and transportation companies to better risk assessment of CDT.

Risk factors of road accident severity and the development of a new system for prevention: New insights from China

Road accident fatalities and accident severity costs have become top priorities and concerns for Chinese policymakers. Understanding the principal factors that explain accident severity is considered to be the first step towards the adequate design of an accident prevention strategy. In this paper, we examine the contribution of various types of factors (vehicle, driver and others) in explaining accident severity in China. Unlike previous studies, the analysis gives a particular focus on fatal accidents. Using a large sample of 405,177 observations for 4-wheeled vehicles in the year 2017 and various statistical and econometrics approaches (e.g., OLS, quantile regression and extreme value theory), the results show that the factors explaining the severity of accidents differs significantly between normal and extreme severity accidents, e.g. across quantiles. Interestingly, we find that the gender factor is only significant for fatal accidents. In particular, the analysis shows that male drivers have an increased likelihood of extreme risk taking. On the basis of these empirical findings, a new ratemaking approach that aims to improve road safety and prevention is discussed and proposed.

A Review of Data Analytic Applications in Road Traffic Safety. Part 1: Descriptive and Predictive Modeling

This part of the review aims to reduce the start-up burden of data collection and descriptive analytics for statistical modeling and route optimization of risk associated with motor vehicles. From a data-driven bibliometric analysis, we show that the literature is divided into two disparate research streams: (a) predictive or explanatory models that attempt to understand and quantify crash risk based on different driving conditions, and (b) optimization techniques that focus on minimizing crash risk through route/path-selection and rest-break scheduling. Translation of research outcomes between these two streams is limited. To overcome this issue, we present publicly available high-quality data sources (different study designs, outcome variables, and predictor variables) and descriptive analytic techniques (data summarization, visualization, and dimension reduction) that can be used to achieve safer-routing and provide code to facilitate data collection/exploration by practitioners/researchers. Then, we review the statistical and machine learning models used for crash risk modeling. We show that (near) real-time crash risk is rarely considered, which might explain why the optimization models (reviewed in Part 2) have not capitalized on the research outcomes from the first stream.

Socio-demographic and experience factors affecting drivers' runoff risk along horizontal curves of two-lane rural highway

INTRODUCTION

An improper driving strategy is one of the causative factors for a high probability of runoff and overturning crashes along the horizontal curves of two-lane highways. The socio-demographic and driving experience factors of a driver do influence driving strategy. Hence, this paper explored the effect of these factors on the driver's runoff risk along the horizontal curves.

METHOD

The driving performance data of 48 drivers along 52 horizontal curves was recorded in a fixed-base driving simulator. The driving performance index was estimated from the weighted lateral acceleration profile of each driver along a horizontal curve. It was clustered and compared with the actual runoff events observed during the experiment. It yielded high, moderate, and low-risk clusters. Using cross-tabulation, each risk cluster was compared with the socio-demographic and experience factors. Further, generalized mixed logistic regression models were developed to predict the high-risk and high to moderate risk events.

RESULTS

The age and experience of drivers are the influencing factors for runoff crash. The high-risk event percentage for mid-age drivers decreases with an increase in driving experience. For younger drivers, it increases initially but decreases afterwards. The generalized mixed logistic regression models identified young drivers with mid and high experience and mid-age drivers with low-experience as the high-risk groups.

CONCLUSIONS

The proposed index parameter is effective in identifying the risk associated with horizontal curves. Driver training program focusing on the horizontal curve negotiation skills and graduated driver licensing could help the high-risk groups. Practical applications: The proposed index parameter can evaluate driving behavior at the horizontal curves. Driving behavior of high-risk groups could be considered in highway geometric design. Motor-vehicle agencies, advanced driver assistance systems manufacturers, and insurance agencies can use proposed index parameter to identify the high-risk drivers for their perusal.

Driving risk assessment using driving behavior data under continuous tunnel environment

Objective: Driving behavior is the key feature for determining the nature of traffic stream qualities and reflecting the risk of operating environments. However, evaluating the driving risk accurately and practically in continuous tunnels (tunnels with a space more than 250 m and less than 1000 m) still faces severe challenges due to the complex driving conditions. The objective of this study is to predict the driving risk indicators and determine different risk levels.Methods: The naturalistic driving system equipped with a road environment and driving behavior data acquisition system combined with the fixed-point test method was used for data collection in 130 tunnels on four highways. A traditional AASHTO braking model and convex hull algorithm were adopted to predict the critical safety speed and the critical time headway of each risk feature point in tunnels. According to the risk constraints under free-flow, car-following and lane-changing conditions, the average traffic flow risk index (TFRI) representing six risk levels and the safety threshold of the corresponding risk indicators were determined.Results: The findings of this study revealed that the critical safety speed at nighttime is slower than in other daytime conditions in continuous tunnels. The time headway slightly changes under 90 km/h. As the speed continues to increase, speed has a significant influence on the critical time headway. The only reliable interaction involved the different adverse weather conditions on the mean critical safety speed in the continuous tunnels (short plus long) (F = 9.730, p＜0.05) and single long tunnels (F = 12.365, p＜0.05).Conclusions: It can be concluded that driving behaviors significantly vary in different tunnel risk feature points and the combined effect of high speed and luminance variation may result in high driving risk. The performance validation indicted that the risk assessment level determined by the proposed approach is consistent with the real safety situations. The study provides an effective and generally acceptable method for identifying driving risk criteria that can also be applied for traffic management and safety countermeasures with a view to possible implementation in continuous tunnels.

Modeling of time-dependent safety performance using anonymized and aggregated smartphone-based dangerous driving event data

Safety performance functions (SPFs) are generally used to relate exposure to the expected number of crashes aggregated over a long time (e.g. a year) by holding all other risk factors constant, and to identify hotspots that have excessive crashes regardless of different time periods. However, it is highly likely that the relationships of exposure, risk factors and crash occurrence can vary across different times of day. This study aims to establish time-dependent SPFs for urban roads by using large-scale dangerous driving event data captured by smartphones in different times of day. Multivariate conditional autoregressive (MVCAR) models are developed to jointly account for spatial and temporal dependence of crash observations. Results of two-sample Kolmogorov-Smirnov tests affirm the heterogeneity of the safety effects of dangerous driving events in different time periods. Time-dependent hotspots are identified using potential for safety improvement (PSI) metric. The assumption here is that due to the change of traffic conditions and environment across different times of day, safety hotspots for different time periods should be different from each other. According to the results of Wilcoxon signed-rank tests, hotspots identified by times of day are found to be mostly different from each other. The findings of this study provide insights into temporal effects of risk factors and can support the development of time-dependent safety countermeasures. Besides, this study also shows the potential of leveraging anonymized and aggregated dangerous driving data to assess traffic safety issues.

Adjusting finite sample bias in traffic safety modeling

Poisson and negative binomial regression models are fundamental statistical analysis tools for traffic safety evaluation. The regression parameter estimation could suffer from the finite sample bias when event frequency is low, which is commonly observed in safety research as crashes are rare events. In this study, we apply a bias-correction procedure to the parameter estimation of Poisson and NB regression models. We provide a general bias-correction formulation and illustrate the finite sample bias through a special scenario with a single binary explanatory variable. Several factors affecting the magnitude of bias are identified, including the number of crashes and the balance of the crash counts within strata of a categorical explanatory variable. Simulations are conducted to examine the properties of the bias-corrected coefficient estimators. The results show that the bias-corrected estimators generally provide less bias and smaller variance. The effect is especially pronounced when the crash count in one stratum is between 5 and 50. We apply the proposed method to a case study of infrastructure safety evaluation. Three scenarios were evaluated, all crashes collected in three years, and two hypothetical situations, where crash information was collected for "half-year" and "quarter-year" periods. The case-study results confirm that the magnitude of bias correction is larger for smaller crash counts. This paper demonstrates the finite sample bias associated with the small number of crashes and suggests bias adjustment can provide more accurate estimation when evaluating the impacts of crash risk factors.

Assessing the impact of sleep time on truck driver performance using a recurrent event model

Driver fatigue is a major safety concern for commercial truck drivers and is directly related to the total hours of sleep prior to a working shift. To evaluate changes in driving performance over a long on-duty driving period, we propose a mixed Poisson process recurrent-event model with time-varying coefficients. We use data from 96 commercial truck drivers whose trucks were instrumented with an advanced in situ data acquisition system. The driving performance is measured by unintentional lane deviation events, a known performance deterioration related to fatigue. Driver sleep time and other activities are extracted from a detailed activity register. The time-varying coefficients are used to model the baseline intensity and difference among three cohorts of shifts in which the driver slept less than 7 hours, between 7 to 9 hours, and more than 9 hours prior to driving. We use the penalized B-splines approach to model the time-varying coefficients and an expectation-maximization algorithm with embedded penalized quasi-likelihood approximation for parameter estimation. Simulation studies show that the proposed model fits low and high event rate data well. The results show a significantly higher intensity after 8 hours of on-duty driving for shifts with less than 7 hours of sleep prior to work. The study also shows drivers tend to self-adjust sleep duration, total driving hours, and breaks. This study provides crucial insight into the impact of sleep time on driving performance for commercial truck drivers and highlights the on-road safety implications of insufficient sleep and breaks while driving.

Assessing the relationship between self-reported driving behaviors and driver risk using a naturalistic driving study

The Manchester Driver Behavior Questionnaire (DBQ) identifies risky driving behaviors resulting from psychological mechanisms. Investigating the relationships between these behaviors and drivers' crash risk can provide a better understanding of the personal factors contributing to the incidence of crashes, allowing the more effective development of safety education and road management countermeasures and interventions. The objectives of this study are therefore: 1) to determine the extent to which driver involvement in both crashes and near crashes (CNCs) is related to self-reported driving behaviors, and 2) to assess the relationship between each type of risky behavior and individual driver CNC risk. Driver and crash data were acquired from the Shanghai Naturalistic Driving Study and included 45 males and 12 females, participants with the mean age of 38.7. A K-mean cluster method was adopted to classify participants into three CNC groups of high-, moderate- and low-risk drivers. Drivers completed the DBQ to self-evaluate the frequency during their daily driving of the questionnaire's 24 risky behaviors. Principal component analysis of the 24 items led to a five-component structure including aggressive violations, ordinary violations, lapses, inattention errors, and inexperience errors. Two logistic regression models were developed to investigate the correlation between the five DBQ components and drivers' CNC levels. Conclusions are as follows: 1) high-risk drivers were significantly more likely to have engaged in inattention errors (e.g., missing a "yield" sign) and ordinary violations (e.g., running a red light) than the other drivers, and, 2) aggressive violations (e.g., racing against others) and ordinary violations were positively related to the probability of being a high- or moderate-risk driver.

Exploring the effects of critical driving situations on driver perception time (PT) using SHRP2 naturalistic driving study data

Driver PT is critical when a driver faces an imminent crash risk and needs to determine what evasive maneuvers to execute. Therefore, it is of utmost importance to study how PT varies across different critical driving situations. PT refers to the time drivers need to recognize the nature and significance of external stimuli. Driver PT is critical when he or she faces a potentially hazardous driving situation, and must determine what action(s) or evasive maneuver(s) to execute. Although past research has identified many factors associated with PT, little research has been done on the effects of critical driving situations on PT, let alone in a real-world driving environment. Naturalistic driving study (NDS) data provides an unprecedented opportunity to look into PT prior to the occurrence of safety-related events. This study seeks to shed light on how critical driving situations influence driver PT, as well as how the driving environment and driver behavior affect PT during real-world driving by utilizing the Second Strategic Highway Research Program (SHRP2) NDS data. An NDS consists of two primary features that distinguish it from retrospective approaches: vehicles are equipped with video camera technologies that observe the driver and the road ahead of the vehicle continuously while driving, and drivers are asked to drive as they normally would. To best study PT while minimizing the effects of confounding factors, this study focused on a total of 1417 rear-end crashes and near crashes. It was found that critical driving situations, the driving environment, and driver behavior are all influential factors in explaining the variation of PT among different drivers. The longest PTs are during critical driving situations where the vehicle ahead is stop-and-go, which can be as long as 2.84 s while controlling for the effects of driving environment and driver behavior factors, compared to other types of driving situations such as a vehicle ahead decelerating or lane changing.

Naturalistic Driving Study Investigating Self-Regulation Behavior in Early Alzheimer's Disease: A Pilot Study

BACKGROUND

Because cognitive processes decline in the earliest stages of Alzheimer's disease (AD), the driving abilities are often affected. The naturalistic driving approach is relevant to study the driving habits and behaviors in normal or critical situations in a familiar environment of participants.

OBJECTIVE

This pilot study analyzed in-car video recordings of naturalistic driving in patients with early-stage AD and in healthy controls, with a special focus on tactical self-regulation behavior.

METHODS

Twenty patients with early-stage AD (Diagnosis and Statistical Manual of Mental Disorders, Fourth Edition [DSM-IV] criteria), and 21 healthy older adults were included in the study. Data collection equipment was installed in their personal vehicles. Two expert psychologists assessed driving performance using a specially designed Naturalistic Driving Assessment Scale (NaDAS), paying particular attention to tactical self-regulation behavior, and they recorded all critical safety events.

RESULTS

Poorer driving performance was observed among AD drivers: their tactical self-regulation behavior was of lower quality. AD patients had also twice as many critical events as healthy drivers and three times more "unaware" critical events.

CONCLUSION

This pilot study used a naturalistic approach to accurately show that AD drivers have poorer tactical self-regulation behavior than healthy older drivers. Future deployment of assistance systems in vehicles should specifically target tactical self-regulation components.

A Deep Learning Framework for Driving Behavior Identification on In-Vehicle CAN-BUS Sensor Data

Human driving behaviors are personalized and unique, and the automobile fingerprint of drivers could be helpful to automatically identify different driving behaviors and further be applied in fields such as auto-theft systems. Current research suggests that in-vehicle Controller Area Network-BUS (CAN-BUS) data can be used as an effective representation of driving behavior for recognizing different drivers. However, it is difficult to capture complex temporal features of driving behaviors in traditional methods. This paper proposes an end-to-end deep learning framework by fusing convolutional neural networks and recurrent neural networks with an attention mechanism, which is more suitable for time series CAN-BUS sensor data. The proposed method can automatically learn features of driving behaviors and model temporal features without professional knowledge in features modeling. Moreover, the method can capture salient structure features of high-dimensional sensor data and explore the correlations among multi-sensor data for rich feature representations of driving behaviors. Experimental results show that the proposed framework performs well in the real world driving behavior identification task, outperforming the state-of-the-art methods.