Crash Features Research Articles

Exploration of the Characteristics of Elderly-Driver-Involved Single-Vehicle Hit-Fixed-Object Crashes in Pennsylvania, USA

With the acceleration of population aging, the elderly driving safety issue is increasingly prominent. Method: With the crash data of Pennsylvania from 2010 to 2019, this study exclusively discusses features of single-vehicle hit-fixed-object crashes (SVHFOCs), one of the most common and deadliest crash types for elderly drivers. Results: Firstly, we demonstrate that elderly drivers are much more likely to be injured and killed than young drivers in SVHFOCs by checking crash consequences. The descriptive analysis indicates that elderly drivers have very different crash features from young drivers. They are found to drive with more caution in many aspects, such as more low-speed local travels, fewer illegal behaviors, fewer nighttime travels, etc. Then, a logistic regression model is built to find the factors significantly influencing the severity of SVHFOCs from driver, vehicle, roadway, and environment. The estimation results indicate that female sex, not wearing a seatbelt, DUI, rural area, and SUV involvement tend to be associated with more severe SVHFOCs. Additionally, illumination, weather, and road type could also significantly affect crash severity. Especially, SVHFOCs in adverse weather, in dark conditions, and at intersections are found to be less severe, which implies that elderly drivers might drive more carefully in complex environments. Practical Applications: These findings are expected to provide new insights for agencies in formulating customized measures to prevent elderly drivers from being involved in SVHFOCs.

Quantifying the heterogeneity impact of risk factors on regional bicycle crash frequency: A hybrid approach of clustering and random parameter model

The existence of internal and external heterogeneity has been established by numerous studies across various fields, including transportation and safety analysis. The findings from these studies underscore the complexity of crash data and the multifaceted nature of risk factors involved in accidents. However, most studies consider the effects of unobserved heterogeneity from one perspective −- either within clusters (internal) or between clusters (external) −- and do not investigate the biases from both simultaneously on crash frequency analysis. To fill this gap, this study proposes a hybrid approach combining latent class cluster analysis with the random parameter negative binomial regression model (LCA-RPNB) to explore the association between risk factors and bicycle crash frequency. First, the bicycle crash data is categorized into three clusters using LCA based on crash features such as gender, trip purposes, weather, and light conditions. Then, the separated crash frequency models for different clusters and the overall model are developed based on RPNB using regional factors of crash locations as independent variables and the crash frequency of different clusters respectively as dependent variables. The hybrid approach enables a comprehensive examination of internal and external heterogeneities among bicycle crash frequency factors simultaneously. Results suggest that the proposed hybrid approach exhibits superior fitting and predictive performance compared to the model only considers the effects of unobserved heterogeneity from one perspective with the lower values of Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). This approach can help policymakers and urban planners to design more effective safety interventions by understanding the distinct needs of different bicyclist clusters and the specific factors that contribute to crash risk in each group.

Learning from Moped Crash Data: Identifying Risk Factors Contributing to the Severity of Injuries Sustained by Moped Riders

Mopeds are light vehicles that provide solutions for short-distance travel, traffic congestion, and environmental pollution issues. However, because of the limited physical protection they offer, moped riders are more likely to sustain severe injuries in a crash than vehicle passengers. This study aims to investigate the features and patterns associated with recent moped crashes and analyzed the severity of injuries sustained by riders and related risk factors in 1,657 moped crashes in Michigan during a five-year period (2017–2021). A multinomial logit model was built to examine the association between crash injury severity levels and 13 variables from three main categories: crash features; rider behaviors/cognitive status; and environmental conditions. The descriptive statistics showed that 79% of injuries sustained by moped riders in crashes were minor, and the injured riders tended to be younger, male, and not wearing helmets. The findings indicate a significant correlation between the severity of injuries sustained by riders and several factors: the type of crash (whether single, head-on, or angle collision); the age of the riders (with older riders being more susceptible), visibility conditions (especially during nighttime); and helmet usage. Notably, single and head-on crashes were the primary causes of injuries. These results highlight the importance of heightened vigilance among moped riders, particularly when navigating poorly maintained surfaces or riding under low-light conditions. It is imperative that riders consistently use helmets to minimize risks. Additionally, setting guidelines or restrictions on helmet use, rider age, riding hours, and permissible areas in which to ride could further contribute to improving overall safety for moped riders.

Determination of functional scenarios for intersection collisions

IntroductionThe National Highway Traffic Safety Administration (NHTSA) estimated that in 2019, intersection crashes accounted for $179 billion of economic damages and $639 billion in societal damages. Intersection advanced driver assist systems (I-ADASs) and automated driving systems (ADS) are designed and have been actively deployed to avoid or mitigate these intersection crash scenarios. Given the indeterminate parameter space for describing collision scenarios, evaluators, and designers are all challenged with condensing the possible intersection crash configurations into digestible, executable conditions for scenario-based simulation testing. The objective of this study is to identify functional intersection crash configurations for I-ADAS and ADS safety evaluation. MethodsReal-world intersection crash characteristics are important considerations for scenario testing as these features can directly correlate to or influence causality, controllability, and potential injury severity. To identify functional intersection crash types, similar crash scenarios were grouped together by identified critical features using an unsupervised decision tree model. A key advantage of this approach was that the implemented cluster crash scenarios would be understandable and interpretable by users. Unsupervised decision trees work by generating uniformly distributed synthetic data with features from real data and classifying all the data as real or synthetic. Long, non-diverging branches were manually pruned to reduce overfitting and improve model performance. Feature importance values were computed based on how effective a given variable grouped the crashes together. Data sourcesThis analysis selected intersection cases that only involved two vehicles from the Crash Investigation Sampling System (CISS) spanning 2017 to 2020. Crash features such as road geometry, intersection signal, and vehicle configuration were important to consider for scenario generation. CISS contained the traffic device, device functionality, vehicle intended pre-event movement, road alignment, road profile, trafficway flow, number of lanes, and crash type for each crash case. Intersection geometry, intersecting road angle, each vehicles’ legal moves, and the presence of a two-way-left-turn-lane (TWLTL), channelized roads, bike lanes, crosswalks, street parking, slip lanes, and visual obstructions were manually recorded from the scene diagram. ResultsThe tree identified 44 functional intersection crash configurations after pruning. These clusters have five main sections: Straight-crossing path (SCP) crashes at 4-legged intersections, Left-Turn-Across-Path/Opposite Direction (LTAP/OD) crashes at 4-legged intersections, other crash types at 4-legged intersections, roundabout and multileg intersections, and 3-legged intersection crashes. The features that best split the data were TWLTL, lane travel direction violation, and traffic control device functionality. The largest cluster was SCP crashes at 4-legged, undivided intersections where the traffic control device was working and both vehicles did not violate the direction of their lane of travel. This cluster was adjacent to 32 vehicles in similar SCP crashes except a vehicle performed an unexpected maneuver based on their lane position. ConclusionThese 44 identified crash configurations could be useful in bolstering the robustness of I-ADAS and ADS intersection scenario testing as they are a compact representation of all the police reported intersection crashes where a vehicle was towed. Future studies could generate logical scenarios with distributions of initial conditions and behaviors from these clusters that could be used to evaluate an I-ADAS or ADS.

Identifying typical pre-crash scenarios based on in-depth crash data with deep embedded clustering for autonomous vehicle safety testing

Choosing appropriate scenarios is critical for autonomous vehicles (AVs) safety testing. Real-world crash scenarios can be used as critical scenarios to test the safety performance of AVs. As one of the dominant types of traffic crashes, the car to powered-two-wheelers (PTWs) crash results in a higher possibility of fatality than ordinary car-to-car crashes. Generally, typical testing scenarios are chosen according to the subjective understanding of the safety experts with limited static features of crashes (e.g., geometric features, weather). This study introduced a novel method to identify typical car-to-PTWs crash scenarios based on real-world crashes with dynamic pre-crash features investigated from the China In-depth Mobility Safety Study-Traffic Accident (CIMSS-TA) database. First, we present crash data collection and construction methods of the CIMSS-TA database to construct testing scenarios. Second, the stacked autoencoder methods are used to learn and obtain embedded features from the high-dimensional data. Third, the extracted features are clustered using k-means clustering algorithm, and then the clustering results are interpreted. Six typical car-to-PTWs scenarios are obtained with the proposed processes. This study introduces a typical high-risk scenario construction method based on deep embedded clustering. Unlike existing researches, the proposed method eliminates the negative impacts of manually selecting clustering variables and provides a more detailed scenario description. As a result, the typical scenarios obtained from AV testing are more robust.

Identifying the heterogeneous effects of road characteristics on Motorcycle-Involved crash severities

This study identifies the heterogeneous effects of road characteristics on motorcycle-involved crash severities by using the data of 11,611 crashes in London from 2017 to 2020. Latent class clustering (LCC) is employed to identify the typical classes and features of motorcycle-involved crashes, considering the driver, vehicle, and environmental characteristics. The partial proportion odds (PPO) model is then applied to discuss the heterogeneous effects of road characteristics on motorcycle-involved crash severities for each class. Parameter evaluations and marginal effects are given to interpret the results better and discuss the effects of road characteristics. Results suggest that four latent classes developed by LCC are more insightful and effective in identifying the heterogeneous effects of road characteristics on motorcycle-involved crash severity. Differences between the four classes do exist, confirming that significant heterogeneity exists both within and between the four classes. Many road characteristics are found to contribute to crash severities, such as junction controls, road obstacles, restricted lanes, and speed limits. Notably, junctions with proper controls can greatly alleviate serious injury, especially junctions with gesture control (marginal effects for fatal injury − 35.33%). Road obstacles have significant impacts on aggregating serious injury, and facility-related obstacles (marginal effects for serious injury + 19.94%) are more severe than non-facility obstacles (marginal effects for serious injury + 13.37%). These findings offer critical policy implications to enhance motorcycle-involved crash safety by improving road characteristics.

Analysis of Expert System Implementation in Computer Damage Diagnosis with Forward Chaining Method

The necessity for computerization is currently growing quickly because computers are now necessary for all needs connected to business and daily life. Therefore, users' ability to quickly and easily access information seems to be hindered by maintenance and repairs. An expert system is a computer-based method that uses facts, knowledge, and reasoning to solve issues that are often best handled by a subject-matter expert. Expert systems are created for expertise that is close to human capabilities in a given sector. A computer expert currently needs a lot of time to diagnose computer damage, and even technicians frequently put off their work in order to come up with fixes. Forward chaining was used in the construction of this system. In order to derive conclusions, forward chaining is employed to test the given factors against the recorded rules in the system. As a result, this expert system was developed to assist users in dealing with the damage and early maintenance that frequently affect computer systems. using computers in everyday tasks.If we are familiar with the Forward Chaining method used by the expert system to trace computer damage, the known crash features to address frequent crashes on that PC, and the application in object-oriented programming languages such as Visual Basic 6.0, we can determine where the damage is located.

Far from reality, or somehow accurate? Social beliefs and perceptions about traffic crashes in the Dominican Republic.

Despite the considerable effort made during the last decades, emerging countries are still among the highest road safety concerns because they still account for most of the deaths caused by traffic crashes. Various studies suggest that one of the factors involved in this negative outcome could be road safety. However, this issue remains pending to be addressed in most emerging countries, including the Dominican Republic. This study aimed to assess the beliefs and perceptions of Dominicans regarding some key road risky-related issues and to discuss them in the light of objective data. For this cross-sectional study, the responses by a full sample of 1,260 Dominicans (50.1% men, 49.9% women) with a mean age of 39.4 years participating in a set of surveys conducted across the country, were used. Although Dominicans (especially women) seem to attribute high importance to road crashes, there is a low perceived likelihood of getting involved in a traffic crash. As for subjective versus objective data comparisons, perceived crash features and objective crash report data considerably match. However, the numbers largely differ in terms of crash frequency and importance and relevance given to road crashes, and their consequences. Further, perceptions of traffic violations and lack of law enforcement were pertinent predictors of the degree of relevance attributed to traffic crashes. Overall, the results of this study suggest that, despite a relative awareness of their actual traffic crash features, Dominicans systematically underestimate the causes, frequency, and consequences of these crashes, including yearly fatality rates. These outcomes suggest the need to strengthen road safety awareness and beliefs in further road safety actions and policymaking in the region.

Cannabis presentations to the emergency department after MVC in the era of legalization for recreational use

Introduction: The objectives of this study were to examine cannabis and alcohol use among injured patients presenting to emergency departments (ED) in cannabis-legal states to capture an expanded profile of cannabis use and evaluate differences in motor-vehicle collision (MVC) characteristics among those using cannabis alone and in combination with alcohol. Methods: This was a cross-sectional study of ED visits by drivers in MVC who presented to one of three study sites. Event-related and usual drug and alcohol use information were obtained using a detailed interviewer-administered computerized questionnaire. We also obtained data from blood and breathalyzer tests and the electronic medical record. We examined frequency and types of acute and past-year cannabis and alcohol use and crash mechanisms and characteristics. Our primary method of determining substance use was self-report; we used biosamples secondarily. Results: Eight percent of drivers reported cannabis use in the 8 h prior to MVC, alone or in combination with alcohol; however, a higher proportion (18%) were positive by biosample. High-risk crash features were common in MVCs associated with cannabis, as they were for alcohol use and co-use of cannabis and alcohol; however, patients injured seriously enough to require admission were less likely to report cannabis use (7% vs. 9%) and more likely to report alcohol use (16% vs. 10%). Conclusions: Cannabis use was common among patients presenting after MVC in this sample of cannabis-legal states. Practical Applications: Differences between self-report and biosample data for cannabis and alcohol use were significant and supports the need to use both means of assessing acute use.

Classifying features of freeway crashes using machine learning

The identification and control of features influencing traffic crashes have become a concern in safety and human factor engineering. The decision tree algorithm can play an effective role in the identification of these features by discovering hidden patterns in traffic crash data. The present study was conducted to identify the most important features influencing the prediction of the types of crashes on a freeway in Iran using the decision tree algorithm. Reports were collected on 2039 traffic crashes on the freeway during 2015–2017. Each crash report form included 35 features, each of which had more sub-features. To optimize the features, the participle swarm optimization (PSO) approach was adopted. The remaining features from the optimization process were analyzed based on the decision tree algorithm in MATLAB. Applying PSO, 13 features were introduced to the decision tree algorithm. Based on the calculations of the algorithm, a decision tree model was constructed. The model demonstrated that seven features generated the highest separation for determining crash types on the freeway, including the complete cause of the crash, road surface conditions, the geometry of the crash location, weather conditions, time of the crash, the human component, and helmet/safety belt. The overall prediction accuracy of the model was found to be 72.21%. The prediction accuracy of the model was also separately calculated for different types of crashes. The prediction accuracy results were 93.62%, 64.20%, and 65.47% for fatal, injury, and damage crashes, respectively. This study can be regarded as the intersection of crash data, data mining, safety sciences. The results can be used to develop strategies and control/supervision plans in order to reduce crash and injury rates.

Exploring traffic safety climate with driving condition and driving behaviour: a random parameter structural equation model approach

Abstract This study aimed to explore traffic safety climate by quantifying driving conditions and driving behaviour. To achieve the objective, the random parameter structural equation model was proposed so that driver action and driving condition can address the safety climate by integrating crash features, vehicle profiles, roadway conditions and environment conditions. The geo-localized crash open data of Las Vegas metropolitan area were collected from 2014 to 2016, including 27 arterials with 16 827 injury samples. By quantifying the driving conditions and driving actions, the random parameter structural equation model was built up with measurement variables and latent variables. Results revealed that the random parameter structural equation model can address traffic safety climate quantitatively, while driving conditions and driving actions were quantified and reflected by vehicles, road environment and crash features correspondingly. The findings provide potential insights for practitioners and policy makers to improve the driving environment and traffic safety culture.

Crash severity analysis of vulnerable road users using machine learning.

Road crash fatality is a universal problem of the transportation system. A massive death toll caused annually due to road crash incidents, and among them, vulnerable road users (VRU) are endangered with high crash severity. This paper focuses on employing machine learning-based classification approaches for modelling injury severity of vulnerable road users-pedestrian, bicyclist, and motorcyclist. Specifically, this study aims to analyse critical features associated with different VRU groups-for pedestrian, bicyclist, motorcyclist and all VRU groups together. The critical factor of crash severity outcomes for these VRU groups is estimated in identifying the similarities and differences across different important features associated with different VRU groups. The crash data for the study is sourced from the state of Queensland in Australia for the years 2013 through 2019. The supervised machine learning algorithms considered for the empirical analysis includes the K-Nearest Neighbour (KNN), Support Vector Machine (SVM) and Random Forest (RF). In these models, 17 distinct road crash parameters are considered as input features to train models, which originate from road user characteristics, weather and environment, vehicle and driver condition, period, road characteristics and regions, traffic, and speed jurisdiction. These classification models are separately trained and tested for individual and unified VRU to assess crash severity levels. Afterwards, model performances are compared with each other to justify the best classifier where Random Forest classification models for all VRU modes are found to be comparatively robust in test accuracy: (motorcyclist: 72.30%, bicyclist: 64.45%, pedestrian: 67.23%, unified VRU: 68.57%). Based on the Random Forest model, the road crash features are ranked and compared according to their impact on crash severity classification. Furthermore, a model-based partial dependency of each road crash parameters on the severity levels is plotted and compared for each individual and unified VRU. This clarifies the tendency of road crash parameters to vary with different VRU crash severity. Based on the outcome of the comparative analysis, motorcyclists are found to be more likely exposed to higher crash severity, followed by pedestrians and bicyclists.

A comparative analysis of factors influencing the injury severity of daytime and nighttime crashes on a mountainous expressway in China

This study focuses on analyzing the injury severity of crashes occurring at different times of the day and providing insights into crash prevention countermeasures. The data contained 1521 police-reported crashes that occurred on a mountainous expressway between 2012 and 2017, which were divided into two subsamples: the daytime crash sample and the nighttime crash sample. The statistical plausibility of developing separate models for the two subsamples was verified by the likelihood ratio test. Two random thresholds random parameters hierarchical ordered probit models were independently conducted to analyze the relationship between injury severity and risk factors. The results showed that the contributing factors to injury severity were different between the two subsamples. The three factors most likely to cause serious crashes during the daytime were ‘not keep a safe distance’, ‘drunk driving’ and ‘adverse weather’, while during the nighttime, these three factors were ‘fatigued driving’, ‘adverse weather’ and ‘drunk driving’. The same factors can cause more serious consequences during the nighttime than that during the daytime. After fully identifying the contributing factors related to injury severity, and in particular distinguishing the features of daytime and nighttime crashes, some useful suggestions were proposed to mitigate crash injury severity.

A probabilistic approach to online classification of bicycle crashes

When a bicycle crash takes place, it is paramount for an emergency center to recognize the physical state of the cyclist as early as possible. However, an injured bicyclist may be incapable of making a phone call to the emergency center. In this study, we propose an online approach to classify bicycle crashes based on signals from an onboard inertial measurement unit (IMU), which can be used as a trigger function for an automatic emergency system. For this purpose, we define several bicycle crash features according to the kinematic properties of bicycle accidents. The input signals (variables) influencing the individual crash features are determined by the ANOVA method (analysis of variance). With the determined input signals, probabilistic models for each crash feature are trained on the base of logit models and 20,000 km naturalistic driving data including 20 real crashes. In addition, further crash and corner case data has been collected for the model training. A decision tree describing all probabilistic crash features is used to classify different kinematic events and crash scenarios. A series of driving tests with a crash-dummy and a crash-car are performed to verify the proposed crash classification approach, showing a sensitivity of 96.8%, a specificity of 99.6% and an accuracy of 99.5% of the trained model.

Comparison of motor vehicle-involved e-scooter and bicycle crashes using standardized crash typology

Introduction: The market share of e-scooters in the United States has proliferated in cities: 86 million trips were made on shared e-scooters in 2019, a more than 100% increase compared to 2018. However, the interaction of e-scooters with other road users and infrastructure remains uncertain. Method: This study scrutinized 52 e-scooter and 79 bicycle police-reported crashes in Nashville, Tennessee, from April 2018 to April 2020 from the Tennessee Integrated Traffic Analysis Network (TITAN) database. We used descriptive analysis and a recent prototype version of the Pedestrian and Bicycle Crash Analysis Tool (PBCAT) to classify crashes based on the locations of the crashes relative to roadway segments or intersections, as well as the maneuver of the motor vehicle and e-scooter/bicycle relative to the motor vehicle. Results: Two crash typologies can explain the majority of e-scooter crashes, while bicycle crashes are distributed over several crash typologies. Additionally, 1 in 10 e-scooter- and bicycle-motor vehicle crashes leads to the injury or fatality of the e-scooter rider or bicyclist. Furthermore, we noted statistically significant differences in spatial and temporal distribution, demographics, lighting conditions, and crash distance from home for e-scooter and bicycle crashes. Conclusions: The police crash report provides a comprehensive picture of e-scooter safety complementing existing literature. We found that e-scooter crash characteristics do not fully overlap with features of bicycle crashes. Practical Implications: A generalized engineering, education, and enforcement treatment to reduce and prevent e-scooter and bicycle crashes, injuries, and fatalities might not result in equal outcomes for each mode. More rigorous enforcement could be implemented to deter e-scooters riders under the age of 18 years and e-scooter safety campaigns could target female riders.

GENERAL FEATURES OF ROAD CRASHES IN CLUJ COUNTY, ROMANIA. SPATIALITY AND CAUSALITY

Road crashes have become a serious issue, and their negative impact, both socially and economically, has been subject to policies and programmes worldwide as well as a research topic for numerous studies in various fields. The present study aims to identify and demonstrate the persistence of certain features related to the occurrence and location of road crashes in Cluj County, Romania, which is a territory recording a constant high road crash incidence in the last decade. We used descriptive statistics to illustrate the key features of road crashes occurring in urban and rural areas, by road type, within the administrative territory of Cluj County, Romania. The analysis was focused on four main aspects: causes, effects in terms of persons injured and deaths, occurrence by road type, and location within and outside urban areas. The years 2019, 2009, and 2018 were considered as reference moments for the values recorded for all indicators in the analysis. Results showed a general trend of decrease in road crash incidence in 2019 compared to 2009, which was also confirmed by the absolute and relative increase in the period 2018-2019. The most significant decrease was found in the number of deaths, especially in the case of road cras h e soccurring on national roads and urban streets, where the incidence is still quite high. However, when ranked considering the number of crashes caused, we noted the persistence of certain categories of triggering factors for the high incidence of road crashes related to both drivers and pedestrians. Road safety is very much related to the behaviour of all participants in traffic and not so much to the road infrastructure and quality, although road capacity may be a triggering factor for drivers’ behaviour. Further measures are needed to enhance road safety and meet the European target of halving the road crash number and fatalities until 2020 and in the next decade.

A Motorcyclist-Injury Severity Analysis: A Comparison of Single-, Two-, and Multi-Vehicle Crashes Using Latent Class Ordered Probit Model

Motorcycle crashes increasingly become a high proportion of the overall motorized vehicle fatalities. However, limited research has been conducted to compare the injury severity of single-, two- and multi-vehicle crashes involving a motorcycle. This study aims to investigate the effects of rider characteristics, road conditions, pre-crash situations, and crash features on motorcycle severities with respect to different numbers of vehicles involved. The crash data used was obtained through a comprehensive Motorcycle Crash Causation Study (MCCS) by the Federal Highway Administration. An anatomic injury severity indicator, the New Injury Severity Score (NISS), is utilized to calculate a total score as the sum of squared the abbreviated injury scale scores of each of the rider’s three most severe injuries. A hybrid approach integrating Latent Class Clustering (LCC) and Ordered Probit (OP) models was used to uncover the unobserved heterogeneity and to explore the major factors which significantly affect the injury severities resulting from single-, two- and multi-vehicle crashes involving a motorcycle. The results show that the significant differences in severity exist between different numbers of vehicles involved. More importantly, they also indicate dividing motorcycle crashes into homogeneous classes before modelling helps to discover insightful information. Pre-speed of the motorcycle is found to be a main factor associated with serious and critical injuries in most types of crashes. Findings of the study provide specific and insightful countermeasures targeting at the contributing factors of motorcycle crashes.

Crash report data analysis for creating scenario-wise, spatio-temporal attention guidance to support computer vision-based perception of fatal crash risks

Reducing traffic fatal crashes has been an important mission of transportation. With the rapid development of sensor and Artificial Intelligence (AI) technologies, the computer vision (CV)-based crash anticipation in the near-crash phase is receiving growing attention. The ability to perceive fatal crash risks in an early stage is of paramount importance as well because it can improve the reliability of crash anticipation. Yet this task is challenging because it requires establishing a relationship between the driving scene information that CV can recognize and the fatal crash features that CV will not get until the crash occurrence. Image data with the annotation for directly training a reliable AI model for the early visual perception of fatal crash risks are not abundant. The Fatality Analysis Reporting System (FARS) contains big data on fatal crashes, which is a reliable data source for finding fatal crash clusters and discovering their distribution patterns to tell the association between driving scene characteristics and fatal crash features. To enhance CV's ability to perceive fatal crash risks earlier, this paper develops a data analytics model from fatal crash report data, which is named scenario-wise, spatio-temporal attention guidance. First, the paper identifies five descriptive variables that are sparse and thus allow for decomposing the 5-year (2013–2017) fatal crash dataset to develop scenario-wise attention guidance. Then, an exploratory analysis of location- and time-related descriptive variables suggests dividing fatal crashes into spatially defined groups. A group's temporal distribution pattern is an indicator of the similarity of fatal crashes in the group. Hierarchical clustering and K-means clustering further merge the spatially defined groups into six clusters according to the similarity of their temporal patterns. After that, association rule mining discovers the statistical relationship between the temporal information of driving scenes with fatal crash features, such as the first harmful event and the manner of collisions, for each cluster. The paper illustrates how the developed attention guidance supports the design and implementation of a preliminary CV model that can identify agents of a possibility to involve in fatal crashes from their environmental and context information.

Analysis of Factors Affecting the Severity of Automated Vehicle Crashes Using XGBoost Model Combining POI Data

The research and development of autonomous vehicle (AV) technology have been gaining ground globally. However, a few studies have performed an in-depth exploration of the contributing factors of crashes involving AVs. This study aims to predict the severity of crashes involving AVs and analyze the effects of the different factors on crash severity. Crash data were obtained from the AV-related crash reports presented to the California Department of Motor Vehicles in 2019 and included 75 uninjured and 18 injured accident cases. The points-of-interest (POI) data were collected from Google Map Application Programming Interface (API). Descriptive statistics analysis was applied to examine the features of crashes involving AVs in terms of collision type, crash severity, vehicle movement preceding the collision, and degree of vehicle damage. To compare the classification performance of different classifiers, we use two different classification models: eXtreme Gradient Boosting (XGBoost) and Classification and Regression Tree (CART). The result shows that the XGBoost model performs better in identifying the injured crashes involving AVs. Compared with the original XGBoost model, the recall and G-mean of the XGBoost model combining POI data improved by 100% and 11.1%, respectively. The main features that contribute to the severity of crashes include weather, degree of vehicle damage, accident location, and collision type. The results indicate that crash severity significantly increases if the AVs collided at an intersection under extreme weather conditions (e.g., fog and snow). Moreover, an accident resulting in injuries also had a higher probability of occurring in areas where land-use patterns are highly diverse. The knowledge gained from this research could ultimately contribute to assessing and improving the safety performance of the current AVs.

A New Video-Based Crash Detection Method: Balancing Speed and Accuracy Using a Feature Fusion Deep Learning Framework

Quick and accurate crash detection is important for saving lives and improved traffic incident management. In this paper, a feature fusion-based deep learning framework was developed for video-based urban traffic crash detection task, aiming at achieving a balance between detection speed and accuracy with limited computing resource. In this framework, a residual neural network (ResNet) combined with attention modules was proposed to extract crash-related appearance features from urban traffic videos (i.e., a crash appearance feature extractor), which were further fed to a spatiotemporal feature fusion model, Conv-LSTM (Convolutional Long Short-Term Memory), to simultaneously capture appearance (static) and motion (dynamic) crash features. The proposed model was trained by a set of video clips covering 330 crash and 342 noncrash events. In general, the proposed model achieved an accuracy of 87.78% on the testing dataset and an acceptable detection speed (FPS > 30 with GTX 1060). Thanks to the attention module, the proposed model can capture the localized appearance features (e.g., vehicle damage and pedestrian fallen-off) of crashes better than conventional convolutional neural networks. The Conv-LSTM module outperformed conventional LSTM in terms of capturing motion features of crashes, such as the roadway congestion and pedestrians gathering after crashes. Compared to traditional motion-based crash detection model, the proposed model achieved higher detection accuracy. Moreover, it could detect crashes much faster than other feature fusion-based models (e.g., C3D). The results show that the proposed model is a promising video-based urban traffic crash detection algorithm that could be used in practice in the future.