Revisiting the hybrid approach of anomaly detection and extreme value theory for estimating pedestrian crashes using traffic conflicts obtained from artificial intelligence-based video analytics.

Pedestrians represent a group of vulnerable road users who are at a higher risk of sustaining severe injuries than other road users. As such, proactively assessing pedestrian crash risks is of paramount importance. Recently, extreme value theory models have been employed for proactively assessing crash risks from traffic conflicts, whereby the underpinning of these models are two sampling approaches, namely block maxima and peak over threshold. Earlier studies reported poor accuracy and large uncertainty of these models, which has been largely attributed to limited sample size. Another fundamental reason for such poor performance could be the improper selection of traffic conflict extremes due to the lack of an efficient sampling mechanism. To test this hypothesis and demonstrate the effect of sampling technique on extreme value theory models, this study aims to develop hybrid models whereby unconventional sampling techniques were used to select the extreme vehicle-pedestrian conflicts that were then modelled using extreme value distributions to estimate the crash risk. Unconventional sampling techniques refer to unsupervised machine learning-based anomaly detection techniques. In particular, Isolation forest and minimum covariance determinant techniques were used to identify extreme vehicle-pedestrian conflicts characterised by post encroachment time as the traffic conflict measure. Video data was collected for four weekdays (6 am-6 pm) from three four-legged intersections in Brisbane, Australia and processed using artificial intelligence-based video analytics. Results indicate that mean crash estimates of hybrid models were much closer to observed crashes with narrower confidence intervals as compared with traditional extreme value models. The findings of this study demonstrate the suitability of machine learning-based anomaly detection techniques to augment the performance of existing extreme value models for estimating pedestrian crashes from traffic conflicts. These findings are envisaged to further explore the possibility of utilising more advanced machine learning models for traffic conflict techniques.

A bi-level framework for real-time crash risk forecasting using artificial intelligence-based video analytics.

This study proposes a bi-level framework for real-time crash risk forecasting (RTCF) for signalised intersections, leveraging the temporal dependency among crash risks of contiguous time slices. At the first level of RTCF, a non-stationary generalised extreme value (GEV) model is developed to estimate the rear-end crash risk in real time (i.e., at a signal cycle level). Artificial intelligence techniques, like YOLO and DeepSort were used to extract traffic conflicts and time-varying covariates from traffic movement videos at three signalised intersections in Queensland, Australia. The estimated crash frequency from the non-stationary GEV model is compared against the historical crashes for the study locations (serving as ground truth), and the results indicate a close match between the estimated and observed crashes. Notably, the estimated mean crashes lie within the confidence intervals of observed crashes, further demonstrating the accuracy of the extreme value model. At the second level of RTCF, the estimated signal cycle crash risk is fed to a recurrent neural network to predict the crash risk of the subsequent signal cycles. Results reveal that the model can reasonably estimate crash risk for the next 20-25 min. The RTCF framework provides new pathways for proactive safety management at signalised intersections.

A Bayesian extreme value theory modelling framework to assess corridor-wide pedestrian safety using autonomous vehicle sensor data.

Pedestrians are a vulnerable road user group, and their crashes are generally spread across the network rather than in a concentrated location. As such, understanding and modelling pedestrian crash risk at a corridor level becomes paramount. Studies on pedestrian crash risks, particularly with the traffic conflict data, are limited to single or multiple but scattered intersections. A lack of proper modelling techniques and the difficulties in capturing pedestrian interaction at the network or corridor level are two main challenges in this regard. With autonomous vehicles trialled on public roads generating massive (and unprecedented) datasets, utilising such rich information for corridor-wide safety analysis is somewhat limited where it appears to be most relevant. This study proposes an extreme value theory modelling framework to estimate corridor-wide pedestrian crash risk using autonomous vehicle sensor/probe data. Two types of models were developed in the Bayesian framework, including the block maxima sampling-based model corresponding to Generalised Extreme Value distribution and the peak over threshold sampling-based model corresponding to Generalised Pareto distribution. The proposed framework was applied to autonomous vehicle data from Argoverse-a Ford Motors subsidiary. This autonomous vehicle fleet of Agro AI (owner of Argoverse dataset) is equipped with two 64 beams synchronised LiDAR sensors, a cluster of seven high-resolution cameras, and a pair of stereo-vison high-resolution cameras to capture surrounding road users' information within a range of 200 meters. A subset of the Argoverse dataset, focussing on an arterial corridor in Miami, USA, was used to extract pedestrian and vehicle trajectories. From these trajectories, vehicle-pedestrian conflicts were identified and measured using post encroachment time. The non-stationarity of extremes was captured by vehicle volume, pedestrian volume, average vehicle speed, and average pedestrian speed in the extreme value model. Both block maxima and peak over threshold sampling-based models were found to provide a reasonable estimate of historical pedestrian crash frequencies. Notably, the block maxima sampling-based model was more accurate than the peak over threshold sampling-based model based on mean crash estimates and confidence intervals. This study demonstrates the potential of using autonomous vehicle sensor data for network-level safety, enabling an efficient identification of pedestrian crash risk zones in a transport network.

Examining the performance of engineering treatment evaluation methodologies using the hypothetical treatment and actual treatment settings.

The selection of treatment evaluation methodology is paramount in determining reliable crash modification factors (CMFs) for engineering treatments. A lack of ground truth makes it cumbersome to examine the performance of treatment evaluation methodologies. In addition, a sound methodological framework is critical for evaluating the performances of treatment evaluation methodologies. In addressing these challenges, this study proposed a framework for assessing treatment evaluation methodologies by hypothetical treatments with known ground truth and actual real-world treatments. In particular, this study examined three before-after treatment evaluation approaches: 1) Empirical Bayes, 2) Simulation-based Empirical Bayes, and 3) Full Bayes methods. In addition, this study examined the Cross-Sectional treatment evaluation methodology. The methodological framework utilized five datasets of hypothetical treatment with known ground truth based on the hotspot identification method and a real-world dataset of wide centerline treatment on two-lane, two-way rural highways in Queensland, Australia. Results showed that all the methods could identify the ground truth of hypothetical treatments, but the Full Bayes approach better predicts the known ground truth compared to Empirical Bayes, Simulation-based Empirical Bayes, and Cross-Sectional methods. The Full Bayes approach was also found to provide the most precise estimate for real-world wide centerline treatment along rural highways compared to other methods. Moreover, the current study highlighted that the Cross-Sectional method offers a viable estimate of treatment effectiveness in case the before-period data is limited.

A hierarchical multinomial logit model to examine the effects of signal strategies on right-turn crash injury severity at signalised intersections.

The severity of right-turn crashes (or left-turn crashes for the roads in the US) at signalised intersections tends to be high because of the relatively high conflicting speeds and angle of impact. However, right-turn crash injury severity at signalised intersections was not sufficiently studied. In particular, the effects of signal control strategies on crash injury severity are not known. This study developed crash injury severity models for right-turn crashes at signalised intersections with a novel approach of linking crashes with signal strategies which enabled assessing the effects of signal strategies on crash injury severity. The study provided a comprehensive understanding of the impacts of signal strategies, intersection geometry and traffic factors on crash injury severity of right-turn crashes at signalised intersections. Crash injury severity models were estimated with crash data from 221 signalised intersections in Queensland from 2012 to 2018. To address the hierarchical structure of crash data, two-level hierarchical Multinomial Logit models were applied, hypothesising that the first level includes individual crash characteristics while the second level includes intersection characteristics. The applied hierarchical model accounts for the correlation among crashes within intersections. Results showed that crashes during Lagging right-turn and Diamond overlap turns are likely to be more severe than other signal strategies at intersections, with the Lagging right-turn signal being the most hazardous. The results also illustrate that the probability of severe injuries increases with the number of conflicting lanes, whereas the corresponding probability decreases with the occupancy of the conflicting lane.

Effects of design consistency on run-off-road crashes: An application of a Random Parameters Negative Binomial Lindley model.

Run-off-road crashes are one of the most common crash types, especially in rural roadway environments contributing significantly to fatalities and severe injuries. These crashes are complex and multi-dimensional events, and factors like road geometry, driver behaviour, traffic characteristics and roadside features contribute to their occurrence, separately or interactively. Sudden changes in road geometry, in particular, can influence driver behaviour, and therefore, in developing a micro-level crash risk model for run-off-road crashes, one of the challenges is incorporating the effects of driver behaviour (disaggregated information) that may arise from the variations in road geometry (aggregated information). This study aims to examine the interaction between road geometry and driver behaviour through a set of measures for design consistency on two-lane rural roads. Multiple data sources, including crash data for 2014-18, traffic data, probe speed data and roadway geometric data, for twenty-three highways in Queensland, Australia, have been fused for this study. Seventeen types of design consistency measures with regard to alignment consistency, operating speed consistency and driving dynamics are tested. A run-off-road crash risk model is estimated by employing the Random Parameters Negative Binomial Lindley regression framework, which accounts for excess zeros in the crash counts and captures the effects of unobserved heterogeneity in the parameter estimates. Results indicate that the geometric design consistency capturing the interaction between driver behaviour and operational factors better predicts run-off-road crashes along rural highways. In addition, roadside attributes like clear zone width, infrastructures, terrain, and roadway remoteness also contribute to run-off-road crashes. The findings of the study provide a comprehensive understanding of the influence of variations in roadway geometry on driver behaviour and run-off-road crashes along rural highways.

Modelling braking behaviour of distracted young drivers in car-following interactions: A grouped random parameters duration model with heterogeneity-in-means.

Braking is an important characteristic of driving behaviour that has a direct relationship with rear-end collisions in a car-following task. Braking becomes more crucial when drivers' cognitive workload increases because of using mobile phones whilst driving. This study, therefore, investigates and compares the effects of using mobile phones whilst driving on braking behaviour. Thirty-two young licenced drivers, evenly split by gender, faced a safety-critical event, that is, leader's hard braking, in a car-following situation. Each participant drove the CARRS-Q Advanced Driving Simulator and was required to respond to a braking event in the simulated environment in three phone conditions: baseline (no phone conversation), handheld, and hands-free. A random parameters duration modelling approach is employed to (i) model drivers' braking (or deceleration) times using a parametric survival model, (ii) capture unobserved heterogeneity associated with braking times, and (iii) account for repeated experiment design. The model identifies the handheld phone condition as a random parameter whilst vehicle dynamics variables, hands-free phone condition, and driver-specific variables are found as fixed parameters. The model suggests that most distracted drivers (in the handheld condition) reduce their initial speeds more slowly than undistracted drivers, reflecting their delayed initial braking that may lead to abrupt braking to avoid a rear-end collision. Further, another group of distracted drivers exhibits faster braking (in the handheld condition), recognising the risk associated with mobile phone usage and delayed initial braking. Provisional licence holders are found to be slower in reducing their initial speeds than open licence holders, indicating their risk-taking behaviour because of their less experience and more sensitivity to mobile phone distraction. Overall, mobile phone distraction appears to impair the braking behaviour of young drivers, which poses significant safety concerns for traffic streams.

A Hierarchical Multinomial Logit model to examine the effects of signal strategies on right-turn crash risks by crash movement configuration.

Crash risk models relying on total crash counts are limited in their ability to extract meaningful insights regarding the context of crashes and to identify effective remedial measures. In addition to the typical classification of collisions noted in the literature (e.g., angle, head-on and rear-end), crashes can also be categorised according to vehicle movement configurations (Definitions for Coding Accidents or DCA codes in Australia). This classification presents an opportunity to extract useful insights into road traffic collision causes and contributing factors that are highly contextual. With this aim, this study develops crash-type models by DCA crash movement, with a focus on right-turn crashes (equivalent to left-turn crashes for right-hand traffic) at signalised intersections using a novel approach for linking crashes with signal control strategies. The modelling approach with contextual data enables quantification of the effect of signal control strategies on right-turn crashes, offering potentially unique and novel insights into right-turn crash causes and contributing factors. Crash-type models are estimated with the crash data of 218 signalised intersections in Queensland from 2012 to 2018. Multilevel (Hierarchical) Multinomial Logit Models with random intercepts are employed to capture the hierarchical influence of factors on crashes and unobserved heterogeneities. These models capture upper-level influences on crashes from intersection characteristics and lower-level influences from individual crash characteristics. The models specified in this way account for the correlation among crashes within intersections and influences on crashes across spatial scales. The model results reveal that the probabilities of the opposite approach crash type are significantly higher than the same direction and adjacent approach crash types for all right-turn signal control strategies at intersections except the split approach, for which the opposite is true. The results also suggest that the number of right-turning lanes and occupancy in conflicting lanes are positively associated with the likelihood of crashes for the same direction crash type.

A data fusion approach of police-hospital linked data to examine injury severity of motor vehicle crashes.

Injury severity studies typically rely on police-reported crash data to examine risk factors associated with traffic injuries. The police crash database includes essential information on roadways, crashes and driver-vehicle characteristics but may not contain accurate and sufficient information on traffic injuries. Despite sizable efforts on injury severity modelling, very few studies have employed hospital records to classify injury severities accurately. As such, the inferences drawn from the police-recorded injury severity classifications may be questionable. This study investigates factors affecting road traffic injuries of motor vehicle crashes in two approaches (1) police-reported injury severity data and (2) a data fusion approach linking police and hospital records. Data from 2015 to 2019 were collected from the Abu Dhabi Traffic Police Department and linked with hospital records by the Department of Health, Abu Dhabi. A total of 6,333 casualty crashes were categorised into non-severe, severe, and fatal crashes following police-reported data and non-hospitalised, hospitalised and fatal crashes based on the police-hospital linked data. The state-of-the-art random thresholds random parameters hierarchical ordered Probit models were then employed to examine the differences in factors affecting crash-injury severities between police-reported and police-hospital linked data. While there are similarities between these two approaches, there are numerous notable differences in injury severity factors. For instance, head-on collisions are associated with high crash-injury severities in the model with police-hospital linked data, but they tend to show low injury severities in the model with police-reported data. In addition, the police-reported approach identifies that crashes occurred in remote areas and angle collisions are associated with low injury severities, which is not intuitive. These findings highlight that modelling the misclassified injury severity in police crash data may lead to wrong estimations and misleading inferences. Instead, the data fusion approach of police-hospital linked data provides critical and accurate insights into road traffic injuries and is a valuable approach for understanding traffic injuries.

Leading pedestrian intervals - Yay or Nay? A Before-After evaluation of multiple conflict types using an enhanced Non-Stationary framework integrating quantile regression into Bayesian hierarchical extreme value analysis.

A pedestrian was estimated to be killed every 85 min and injured every 7 min on US roads in 2019. Targeted safety treatments are particularly required at urban intersections where pedestrians regularly conflict with turning vehicles. Leading Pedestrian Intervals (LPIs) are an innovative, low-cost treatment where the pedestrian and vehicle usage of the potential conflict area (a crosswalk) is staggered in time to give the pedestrians a head start of a few seconds and reduce the "element of surprise" for right-turning vehicles. The effectiveness of LPI treatment on pedestrian safety is mixed, and most importantly, its effect on vehicle-vehicle conflicts is unknown. This study investigates the before-after effects of LPI treatments on vehicle-pedestrian and vehicle-vehicle crash risk by applying traffic conflict techniques. In particular, this study has developed a quantile regression technique within the extreme value model to estimate and compare crash risks before and after the installation of the LPI treatment. The before-after traffic movement video data (504 h in total) were collected from three signalized intersections in the City of Bellevue, Washington. The recorded movements were analyzed using Microsoft's proprietary computer vision platform, Edge Video Service, and Advanced Mobility Analytics Group's cloud-based SMART SafetyTM platform to automatedly extract traffic conflicts by analyzing road user trajectories. The treatment effect was measured using a Bayesian hierarchical extreme value model with the peak-over threshold approach. For the extreme value model, a Bayesian quantile regression analysis was conducted to estimate the conflict thresholds corresponding to a high (95th) quantile. Odds ratios were estimated for both conflict types using untreated crossing as a control group. Results indicate that the LPI treatment reduces the crash risk of pedestrians as measured by the reduction in extreme vehicle-pedestrian conflicts by about 42%. The LPI treatment has also been found not to negatively affect rear-end conflicts along the approaches leading to the LPI-treated pedestrian crossing at the signalized intersections. The findings of this study further emphasize the effectiveness of video analytics in proactive safety evaluations of engineering treatments.

A before-after evaluation of protected right-turn signal phasings by applying Empirical Bayes and Full Bayes approaches with heterogenous count data models.

Right-turn crashes (or left-turn crashes for the US or similar countries) represent over 40 % of signalized intersection crashes in Queensland, Australia. Protected right-turn phasings are a widely used countermeasure for right-turn crashes, but the research findings on their effects across different crash types and intersection types are not consistent. Methodologically, the Empirical Bayes and Full Bayes techniques are generally applied for before-after evaluations, but the inclusion of heterogeneous models within these techniques has not been considered much. Addressing these research gaps, the objective of this study is to evaluate the effectiveness of protected right-turn signal phasings at signalized intersections employing heterogeneous count data models with the Empirical Bayes and Full Bayes techniques. In particular, the Empirical Bayes approach based on random parameters Poisson-Gamma models (simulation-based Empirical Bayes), and the Full Bayes approach based on random parameters Poisson-Lognormal intervention models (simulation-based Full Bayes) are applied. A total of 69 Cross intersections (with ten treated sites) and 47 T intersections (with six treated sites) from Southeast Queensland in Australia were included in the analysis to estimate the effects of protected right-turn signal phasings on various crash types. Results show that the change of signal phasing from a permissive right-turn phasing to the protected right-turn phasing at cross and T intersections reduces about 87 % and 91 % of right-turn crashes, respectively. In addition, the effect of protected right-turn phasings on rear-end crashes was not significant. The heterogenous count data models significantly address extra Poisson variation, leading to efficient safety estimates in both simulation-based Empirical Bayes and simulation-based Full Bayes approaches. This study demonstrates the importance of accounting for unobserved heterogeneity for the before-after evaluation of engineering countermeasures.

Influence of segmentation approaches on the before-after evaluation of engineering treatments: A hypothetical treatment approach.

The segmentation of highways is a fundamental step in estimating crash frequency models and conducting a before-after evaluation of engineering treatments, but the effects of segmentation approaches on the engineering treatment evaluations are not known very well. This study examined the effects of segmentation approaches on the before-after evaluation of engineering treatments. In particular, this study evaluated four segmentation approaches by applying the Empirical Bayes technique to a dataset for which the ground truth was known. Four segmentation approaches included Highway Safety Manual (HSM), Fixed (kilometre post), Fisher's, and K-means segmentation. This study utilized a 440 km stretch of rural two-lane two-way highway in Queensland, Australia, to prepare a dataset with known ground truth. The treatment under evaluation was a hypothetical treatment, which should yield a crash modification factor (CMF) of 1. For assigning hypothetical treatment, a total of fifteen datasets were prepared, including ten datasets based on the random assignment and five datasets based on the hotspot identification method. Following the before-after evaluation using the Empirical Bayes technique, the results showed that HSM and Fixed segmentation approaches predict the ground truth in both dataset types. From random assignment datasets, the estimated CMFs using HSM, Fixed, Fisher's, and K-means segmentation approaches deviated from the true CMF (i.e., 1) by 2.32 %, 5.30 %, 6.08 %, and 8.62 %, respectively. In the case of hotspots, the corresponding deviations of CMFs were 8.57 %, 9.37 %, 28.84 %, and 35.43 %, respectively. Overall, HSM segmentation best identified the actual treatment effect, followed by the Fixed segmentation. If the variables to define homogeneity for HSM segmentation are limited, then Fixed segmentation can yield reliable crash modification factors from the before-after treatment evaluations than the crash-based segmentation approaches.

Transferability of multivariate extreme value models for safety assessment by applying artificial intelligence-based video analytics.

Traffic conflict techniques represent the state-of-the-art for road safety assessments. However, the lack of research on transferability of conflict-based crash risk models, which refers to applying the developed crash risk estimation models to a set of external sites, can reduce their appeal for large-scale traffic safety evaluations. Therefore, this study investigates the transferability of multivariate peak-over threshold models for estimating crash frequency-by-severity. In particular, the study proposes two transferability approaches: (i) an uncalibrated approach involving a direct application of the uncalibrated base model to the target sites and (ii) a threshold calibration approach involving calibration of conflict thresholds of the conflict indicators. In the latter approach, the conflict thresholds of the Modified Time-To-Collision (MTTC) and Delta-V indicators were calibrated using local data from the target sites. Finally, the two transferability approaches were compared with a complete re-estimation approach where all the model parameters were estimated using local data. All three approaches were tested for a target set of signalized intersections in Southeast Queensland, Australia. Traffic movements at the target intersections were observed using video cameras for two days (12 h each day). The road user trajectories and rear-end conflicts were extracted using an automated artificial intelligence-based algorithm utilizing state-of-the-art Computer Vision methods. The base models developed in an earlier study were then transferred to the target sites using the two transferability approaches and the local data from the target sites. Results show that the threshold calibration approach provides the most accurate and precise predictions of crash frequency-by-severity for target sites. Thus, for peak-over threshold models, the threshold parameter is the most important, and its calibration improves the performance of the base models. The complete re-estimation of models for individual target sites yields inferior fits and less precise crash estimates than the two transferability approaches since they utilize fewer traffic conflict extremes in their development than the larger dataset utilized in base model development. Therefore, the study results can significantly advance the applicability of traffic conflict models for crash risk estimation at transport facilities.

A simulation-based empirical bayes approach: Incorporating unobserved heterogeneity in the before-after evaluation of engineering treatments.

The Empirical Bayes approach for before-after evaluation methodology utilizing the negative binomial model does not account well for unobserved heterogeneity. Building on the Empirical Bayes approach, the objective of this study was to propose a framework to accommodate unobserved heterogeneity in before-after countermeasure evaluation. In particular, this study has proposed a simulation-based Empirical Bayes approach by applying the panel random parameters negative binomial model with parameterized overdispersion (PRNB-PO) to evaluate the effectiveness of engineering treatments. The proposed framework has been tested for the wide centerline treatment (WCLT) on rural two-lane two-way highways in Australia. The empirical analysis included 511 km of WCLT treated highways in a before-after evaluation within a time period of 2010 - 2018 and 430 km of reference sites in Queensland, Australia. The PRNB-PO models outperformed the traditional negative binomial models in terms of goodness-of-fit and prediction performance for total injury crashes, and fatal and serious injury (FSI) crashes. The simulation-based Empirical Bayes approach using the PRNB-PO model resulted in more precise estimates of crash modification factors than the standard Empirical Bayes approach. The WCLT is found to result in significant reductions in total injury crashes by 28.21% (95% confidence interval (CI) = 22.92 - 33.50%), FSI crashes by 13.90% (95% CI = 6.99 - 20.81%), and head-on crashes by 25.45% (95% CI = 14.87 - 36.03%). Overall, WCLT is an effective engineering treatment and should be considered a low-cost countermeasure on rural two-lane two-way highways.

Temporal instability assessment of injury severities of motor vehicle drivers at give-way controlled unsignalized intersections: A random parameters approach with heterogeneity in means and variances.

Unsignalized intersections are highly susceptible to traffic crashes compared to signalized ones. By taking into account temporal stability and unobserved heterogeneity, this study investigates the determinants of the injury severity of drivers involved in crashes at unsignalized intersections controlled by give-way (yield) signs. Mixed logit models with three approaches were employed, namely random parameters, random parameters with heterogeneity in means, and random parameters with heterogeneity in means and variances. The investigation covered four years (2015-2018) of motor vehicle crashes in South Australia, and the injury severity was categorized into no injury, minor injury, and severe injury. Log-likelihood ratio tests revealed that there is a significant temporal instability in the four years of crashes. Thus, each year was considered separately to avoid any potential erroneous conclusions and unreliable countermeasures. The study found 28 indicator variables were temporally unstable over the four years of crashes, such as driver gender, time of the crash, rear-end involvement, sideswipes, right-angle crash type, vehicle movement at crash time, and crash time. Whereas several variables were stable over the same period, for example, crashes within metropolitan areas were temporally stable over four years, crashes in dry pavement condition were temporally stable over three consecutive years. Four factors have temporal stability over two consecutive years: alcohol involvement crashes, hitting fixed objects, hitting cyclists, and crashes during winter. Overall, mixed logit models with heterogeneity in means and with/without variance performed better than the standard one. It is recommended that temporal instability be considered in order to avoid any potential inconsistent countermeasures.

A systematic mapping review of surrogate safety assessment using traffic conflict techniques.

Safety assessment of road sections and networks have historically relied on police-reported crash data. These data have several noteworthy and significant shortcomings, including under-reporting, subjectivism, post hoc assessment of crash causes and contributing factors, limited behavioural information, and omitted potential important crash-related factors resulting in an omitted variable bias. Moreover, crashes are relatively rare events and require long observation periods to justify expenditures. The rarity of crashes leads to a moral dilemma-we must wait for sufficient crashes to accrue at a site-some involving injuries and even death-to then justify improvements to prevent crashes. The more quickly the profession can end its reliance on crashes to assess road safety, the better. Surrogate safety assessment methodologies, in contrast, are proactive in design, do not rely on crashes, and require shorter observation timeframes in which to formulate reliable safety assessments. Although surrogate safety assessment methodologies have been developed and assessed over the past 50 years, an overarching and unifying framework does not exist to date. A unifying framework will help to contextualize the role of various methodological developments and begin a productive discussion in the literature about how the various pieces do or should fit together to understand road user risk better. This paper aims to fill this gap by thoroughly mapping traffic conflicts and surrogate safety methodologies. A total of 549 studies were meticulously reviewed to achieve this aim of developing a unifying framework. The resulting framework provides a consolidated and up-to-date summary of surrogate safety assessment methodologies and conflict measures and metrics. Further work is needed to advance surrogate safety methodologies. Critical research needs to include identifying a comprehensive and reliable set of surrogate measures for risk assessment, establishing rigorous relationships between conflicts and crashes, developing ways to capture road user behaviours into surrogate-based safety assessment, and integrating crash severity measures into risk estimation.

A home-based approach to understanding seatbelt use in single-occupant vehicles in Tennessee: Application of a latent class binary logit model.

Although the enforcement of seatbelt use is considered to be an effective strategy in reducing road injuries and fatalities, lack of seatbelt use still accounts for a substantial proportion of fatal crashes in Tennessee, United States. This problem has raised the need to better understand factors influencing seatbelt use. These factors may arise from spatial/temporal characteristics of a driving location, type of vehicle, demographic and socioeconomic attributes of the vehicle occupants, driver behaviours, attitudes, and social norms. However, the above factors may not have the same effects on seatbelt use across different individuals. In addition, the behavioural factors are usually difficult to measure and may not always be readily available. Meanwhile, residential locations of vehicle occupants have been shown to be associated with their behavioural patterns and thus may serve as a proxy for behavioural factors. However, the suitability of geographic and residential locations of vehicle occupants to understand the seatbelt use behaviour is not known to date. This study aims to fill the above gaps by incorporating the residential location characteristics of vehicle occupants in addition to their demographics and crash characteristics into their seatbelt use while accounting for the varying effects of these factors on individual seatbelt use choices. To achieve this goal, empirical data are collected for vehicular crashes in Tennessee, United States, and the home addresses of vehicle occupants at the time of the crash are geocoded and linked with the census tract information. The resulting data is then used as explanatory variables in a latent class binary logit model to investigate the determinants of vehicle occupants' seatbelt use at the time of the crash. The latent class specification is employed to capture the unobserved heterogeneity in data. Results show that Tennessean drivers belong to two general categories-conformist and eccentric-with gender, vehicle type, and income per capita determining the likelihood of these categories. Overall, male drivers, younger drivers, and drivers who have consumed drugs are less likely to wear a seatbelt, whereas drivers who come from areas with higher population density, travel time, and income per capita are more likely to wear a seatbelt. In addition, driving during the day and in rainy weather are associated with an increased likelihood of seatbelt use. The findings of this study will help developing effective policies to increase seatbelt use rate and improve safety.

The impact of the connected environment on driving behavior and safety: A driving simulator study.

The connected environment provides surrounding traffic information to drivers via different driving aids that are expected to improve driving behavior and assist in avoiding safety-critical events. These driving aids include speed advisory, car-following assistance, lane-changing support, and advanced information about possible unseen hazards, among many others. While various studies have attempted to examine the effectiveness of different driving aids discretely, it is still vague how drivers perform when they are exposed to a connected environment with vehicle-to-vehicle and vehicle-to-infrastructure communication capabilities. As such, the objective of this study is to examine the effects of the connected environment on driving behavior and safety. To achieve this aim, an innovative driving simulator experiment was designed to mimic a connected environment using the CARRS-Q Advanced Driving Simulator. Two types of driving aids were disseminated in the connected environment: continuous and event-based information. Seventy-eight participants with diverse backgrounds drove the simulator in four driving conditions: baseline (without driving aids), perfect communication (uninterrupted supply of driving aids), communication delay (driving aids are delayed), and communication loss (intermittent loss of driving aids). Various key driving behavior indicators were analyzed and compared across various routine driving tasks such as car-following, lane-changing, interactions with traffic lights, and giving way to pedestrians at pedestrian crossings. Results suggest that drivers in the perfect communication scenario maintain a longer time-to-collision during car-following, a longer time-to-collision to pedestrian, a lower deceleration to avoid a crash during lane-changing, and a lower propensity of yellow light running. Overall, drivers in the connected environment are found to make informed (thus better) decisions towards safe driving.

Applying a joint model of crash count and crash severity to identify road segments with high risk of fatal and serious injury crashes.

Both crash count and severity are thought to quantify crash risk at defined transport network locations (e.g. intersections, a particulate section of highway, etc.). Crash count is a measure of the likelihood of occurring a potential harmful event, whereas crash severity is a measure of the societal impact and harm to the society. As the majority of safety improvement programs are focused on preventing fatal and serious injury crashes, identification of high-risk sites-or blackspots-should ideally account for both severity and frequency of crashes. Past research efforts to incorporate crash severity into the identification of high-risk sites include multivariate crash count models, equivalent property damage only models and two-stage mixed models. These models, however, often require suitable distributional assumptions for computational efficiency, neglect the ordinal nature of crash severity, and are inadequate for capturing unobserved heterogeneity arising from possible correlations between crash counts of different severity levels. These limitations can ultimately lead to inefficient allocation of resources and misidentification of sites with high risk of fatal and serious injury crashes. Moreover, the implication of these models in blackspot identification is an important, unanswered question. While a joint econometric model of crash count and crash severity has the flexibility to account for the limitations mentioned previously, its ability to identify high-risk sites also needs to be examined. This study aims to fill this research gap by employing the joint model for blackspot identification. Using data from state-controlled roads in Queensland, Australia, a new risk score is developed based on predicted crash counts by severity, weighted by the cost ratio of severity levels. This weighted risk score is then used for identifying road segments with high risk of fatal and injury crashes. Results show that the joint model of crash count and crash severity has substantially improved prediction accuracy compared to the traditional count models. The correlation between crash counts of different severity levels captures the unobserved heterogeneity caused by the extra-variation in total crash counts and moderates the parameters in the joint model. In comparison with the traditional approaches, the proposed weighted risk score approach with the joint model of crash count and crash severity leads to the identification of a higher number of fatal and serious injury crashes in the top ranked sites flagged for safety improvements.

Incorporating behavioral variables into crash count prediction by severity: A multivariate multiple risk source approach.

The frequency and severity of traffic crashes have commonly been used as indicators of crash risk on transport networks. Comprehensive modeling of crash risk should account for both frequency and injury severity-capturing both the extent and intensity of transport risk for designing effective safety improvement programs. Previous research has revealed that crashes are correlated across severity categories because of the combined influence of risk factors, observed or unobserved. Moreover, crashes are the outcomes of a multitude of factors related to roadway design, traffic operations, pavement conditions, driver behavior, human factors, and environmental characteristics, or in more general terms: factors reflect both engineering and non-engineering risk sources. Perhaps not surprisingly, engineering risk sources have dominated the list of variables in the mainstream modeling of crashes whereas non-engineering sources, in particular, behavioral factors, are crucially omitted. It is plausible to assume that crash contributing factors from the same risk source affect crashes in a similar manner, but their influences vary across different risk sources. Conventional crash frequency modeling hypothesizes that the total crash count at any roadway site is well-approximated by a single risk source to which several explanatory variables contribute collaboratively. The conventional formulation is not capable of accounting for variations between risk sources; therefore, is unable to discriminate distinct impacts between engineering variables and non-engineering variables. To address this shortcoming, this study contributes to the development of multivariate multiple risk source regression, a robust modeling technique to model crash frequency and severity simultaneously. The multivariate multiple risk source regression method applied in this study can effectively capture the correlation between severity levels of crash counts while identifyinging the varying effects of crash contributing factors originated from distinct sources. Using crashes on Wisconsin rural two-lane highways, two risk sources - engineering and behavioral - were employed to develop proposed models. The modeling results were compared with a single equation negative binomial (NB) model, and a univariate multiple risk source model. The results show that the multivariate multiple risk source model significantly outperforms the other models in terms of statistical fit across several measures. The study demonstrates a unique approach to explicitly incorporating behavioral factors into crash prediction models while taking crash severity into consideration. More importantly, the parameter estimates provide more insight into the distinct sources of crash risk, which can be used to further inform safety practitioners and guide roadway improvement programs.