Improving traffic accident severity prediction using MobileNet transfer learning model and SHAP XAI technique.

Traffic accidents remain a leading cause of fatalities, injuries, and significant disruptions on highways. Comprehending the contributing factors to these occurrences is paramount in enhancing safety on road networks. Recent studies have demonstrated the utility of predictive modeling in gaining insights into the factors that precipitate accidents. However, there has been a dearth of focus on explaining the inner workings of complex machine learning and deep learning models and the manner in which various features influence accident prediction models. As a result, there is a risk that these models may be seen as black boxes, and their findings may not be fully trusted by stakeholders. The main objective of this study is to create predictive models using various transfer learning techniques and to provide insights into the most impactful factors using Shapley values. To predict the severity of injuries in accidents, Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Residual Networks (ResNet), EfficientNetB4, InceptionV3, Extreme Inception (Xception), and MobileNet are employed. Among the models, the MobileNet showed the highest results with 98.17% accuracy. Additionally, by understanding how different features affect accident prediction models, researchers can gain a deeper understanding of the factors that contribute to accidents and develop more effective interventions to prevent them.

Advancing urban traffic accident forecasting through sparse spatio-temporal dynamic learning.

Traffic accidents have emerged as one of the most public health safety matters, raising concerns from both the public and urban administrators. The ability to accurately predict traffic accident not only supports the governmental decision-making in advance but also enhances public confidence in safety measures. However, the efficacy of traditional spatio-temporal prediction models are compromised by the skewed distributions and sparse labeling of accident data. To this end, we propose a Sparse Spatio-Temporal Dynamic Hypergraph Learning (SST-DHL) framework that captures higher-order dependencies in sparse traffic accidents by combining hypergraph learning and self-supervised learning. The SST-DHL model incorporates a multi-view spatiotemporal convolution block to capture local correlations and semantics of traffic accidents, a cross-regional dynamic hypergraph learning model to identify global spatiotemporal dependencies, and a two-supervised self-learning paradigm to capture both local and global spatiotemporal patterns. Through experimentation on New York City and London accident datasets, we demonstrate that our proposed SST-DHL exhibits significant improvements compared to optimal baseline models at different sparsity levels. Additionally, it offers enhanced interpretability of results by elucidating complex spatio-temporal dependencies among various traffic accident instances. Our study demonstrates the effectiveness of the SST-DHL framework in accurately predicting traffic accidents, thereby enhancing public safety and trust.

Implementation of a realistic artificial data generator for crash data generation.

In this paper, a framework is outlined to generate realistic artificial data (RAD) as a tool for comparing different models developed for safety analysis. The primary focus of transportation safety analysis is on identifying and quantifying the influence of factors contributing to traffic crash occurrence and its consequences. The current framework of comparing model structures using only observed data has limitations. With observed data, it is not possible to know how well the models mimic the true relationship between the dependent and independent variables. Further, real datasets do not allow researchers to evaluate the model performance for different levels of complexity of the dataset. RAD offers an innovative framework to address these limitations. Hence, we propose a RAD generation framework embedded with heterogeneous causal structures that generates crash data by considering crash occurrence as a trip level event impacted by trip level factors, demographics, roadway and vehicle attributes. Within our RAD generator we employ three specific modules: (a) disaggregate trip information generation, (b) crash data generation and (c) crash data aggregation. For disaggregate trip information generation, we employ a daily activity-travel realization for an urban region generated from an established activity-based model for the Chicago region. We use this data of more than 2 million daily trips to generate a subset of trips with crash data. For trips with crashes crash location, crash type, driver/vehicle characteristics, and crash severity. The daily RAD generation process is repeated for generating crash records at yearly or multi-year resolution. The crash databases generated can be employed to compare frequency models, severity models, crash type and various other dimensions by facility type - possibly establishing a universal benchmarking system for alternative model frameworks in safety literature.

Does the priority of ambulance guarantee no delay? a MIPSSTW model of emergency vehicle routing optimization considering complex traffic conditions for highway incidents.

Globally, traffic accidents on the highway network contribute significantly to a high fatality rate, drawing considerable attention from health institutions. The efficiency of transportation plays a vital role in mitigating the severe consequences of these incidents. This study delves into the issues of emergency vehicles experiencing delays despite having priority. Therefore, we construct mixed-integer linear programming with semi-soft time windows (MIPSSTW) model for optimizing emergency vehicle routing in highway incidents. We analyze the time-varying and complex traffic situations and respectively propose corresponding estimation approaches for the travel time of road segments, intersections on the urban road network, and ramp-weave sections on the highway network. Furthermore, we developed a modified cuckoo search(MCS) algorithm to solve this combinatorial problem. Optimization strategies of Lévy flight and dynamic inertial weight strategy are introduced to strengthen the exploration capability and the diversity of solution space of the CS algorithm. Computational experiments based on the Chinese emergency medical system data are designed to validate the efficacy and effectiveness of the MIPSSTW model and MCS algorithm. The results show that our works succeed in searching for high-quality solutions for emergency vehicle routing problems and enhance the efficacy of strategic decision-making processes in the realm of incident management and emergency response systems.

A method to enhance drivers' hazard perception at night based on "knowledge-attitude-practice" theory.

During nighttime driving, the inherent challenges of low-illuminance conditions often lead to an increased crash rate and higher fatalities by impairing drivers' ability to recognize imminent hazards. While the severity of this issue is widely recognized, a significant research void exists with regard to strategies to enhance hazard perception under such circumstances. To address this lacuna, our study examined the potential of an intervention grounded in the knowledge-attitude-practice (KAP) framework to bolster nighttime hazard detection among drivers. We engaged a cohort of sixty drivers split randomly into an intervention group (undergoing specialized training) and a control group and employed a holistic assessment that combined eye movement analytics, physiological response monitoring, and driving performance evaluations during simulated scenarios pre- and post-intervention. The data showed that the KAP-centric intervention honed drivers' visual search techniques during nighttime driving, allowing them to confront potential threats with reduced physiological tension and ensuring more adept vehicle handling. These compelling findings support the integration of this methodology in driver training curricula and present an innovative strategy to enhance road safety during nighttime journeys.

Statistical estimation of fatal and serious injuries saved by iRAP protocols in 74 countries.

OBJECTIVE: Road traffic crashes cause 1.19 million deaths and millions more injuries annually. The persistently high burden has drawn attention from national and international stakeholders worldwide. Unsafe road infrastructure is one of the major risk factors for traffic safety, particularly in low- and middle-income countries. METHODS: Aiming to eliminate high-risk roads in all countries, the International Road Assessment Programme (iRAP) developed a robust and evidence-based approach to support country transportation agencies. RESULTS: Thus far, the iRAP protocols have been used to collect 1.8 million kilometers of Crash Risk Mapping and 1.5 million kilometers of Star Rating and FSI estimations in 128 countries. Deploying an observational before-and-after (or pre-post) study design, this report estimated the fatal and series injuries (FSI) saved through use of the iRAP protocols. The study is based on 441,753 kilometers of assessed roads from 1,039 projects in 74 countries. Our results show that the implementation of iRAP's proposed countermeasures saves about 159,936 FSI annually. Throughout the lifetime of the implemented countermeasures, a total of 3.2 million FSI could be saved. CONCLUSION: While quantifying the success of the iRAP protocols, our results suggest an opportunity to save many millions more lives on the roads through expanding iRAP implementation to more regions and countries.

A systematic review of the use of in-vehicle telematics in monitoring driving behaviours.

BACKGROUND: Road traffic deaths are increasing globally, and preventable driving behaviours are a significant cause of these deaths. In-vehicle telematics has been seen as technology that can improve driving behaviour. The technology has been adopted by many insurance companies to track the behaviours of their consumers. This systematic review presents a summary of the ways that in-vehicle telematics has been modelled and analysed. METHODOLOGY: Electronic searches were conducted on Scopus and Web of Science. Studies were only included if they had a sample size of 10 or more participants, collected their data over at least multiple days, and were published during or after 2010. 45 relevant papers were included in the review. 27 of these articles received a rating of "good" in the quality assessment. RESULTS: We found a divide in the literature regarding the use of in-vehicle telematics. Some articles were interested in the utility of in-vehicle telematics for insurance purposes, while others were interested in determining the influence that in-vehicle telematics has on driving behaviour. Machine learning analyses were the most common forms of analysis seen throughout the review, being especially common in articles with insurance-based outcomes. Acceleration, braking, and speed were the most common variables identified in the review. CONCLUSION: We recommend that future studies provide the demographical information of their sample so that the influence of in-vehicle telematics on the driving behaviours of different groups can be understood. It is also recommended that future studies use multi-level models to account for the hierarchical structure of the telematics data. This hierarchical structure refers to the individual trips for each driver.

Exploring the associations of demographics and scale measures with cognitive driving behavior among older drivers in China.

Age-related changes and frailty are reasons for the high proportion of older drivers in certain types of crashes, such as giving right of way at intersections and turning left. The identified crash causes include the driver's demographics, driving style, cognitive function, and mental workload. This study aimed to explore the associations of demographics and scale measures with cognitive driving behavior. Thirty-nine drivers, consisting of twenty younger drivers (18-60 years old) and nineteen older drivers (above 60 years old), participated in driving simulation experiments after completing scale tests. The selected scale measures included the demographic questionnaire, Multidimensional Driving Style Inventory (MDSI-C), Mini-Mental State Examination (MMSE), Trail Making Test Part A (TMT-A) and Part B (TMT-B), and the National Aeronautics and Space Administration Task Load Index (NASA-TLX) for obtaining subjective information from drivers. Driving scenarios were developed based on the driving characteristics of older adults to investigate age-related driving ability. The driving behavior parameters included reaction time, lateral stability, and driving speed, corresponding to reaction, perception, and execution. Three stepwise regression models showed that NASA-TLX, the interaction between age and driving experience, and the interaction between age and TMT-A significantly explained 53.3 % of reaction time variance; TMT-A, risk driving style, anxiety driving style, and gender significantly explained 53.5 % of lateral stability variance; TMT-A, NASA-TLX, and MMSE significantly explained 60.6 % of driving speed variance. Subsequently, the impact of four age-related predictor variables on driving behavior was further discussed. It is worth noting that a rich driving experience may compensate for driving performance. However cognitive impairment impairs this compensation. Driving behavior is influenced by a combination of various factors. Age, as a physiological indicator, is not sufficient to be a strong predictive factor for lateral stability and driving speed. The results provide a reference for traffic safety management departments to streamline driving suitability test procedures and propose targeted training methods for older drivers.

A method for generating case-specific vehicle models from a single-view vehicle image for accurate pedestrian injury reconstructions.

Developing vehicle finite element (FE) models that match real accident-involved vehicles is challenging. This is related to the intricate variety of geometric features and components. The current study proposes a novel method to efficiently and accurately generate case-specific buck models for car-to-pedestrian simulations. To achieve this, we implemented the vehicle side-view images to detect the horizontal position and roundness of two wheels to rectify distortions and deviations and then extracted the mid-section profiles for comparative calculations against baseline vehicle models to obtain the transformation matrices. Based on the generic buck model which consists of six key components and corresponding matrices, the case-specific buck model was generated semi-automatically based on the transformation metrics. Utilizing this image-based method, a total of 12 vehicle models representing four vehicle categories including family car (FCR), Roadster (RDS), small Sport Utility Vehicle (SUV), and large SUV were generated for car-to-pedestrian collision FE simulations in this study. The pedestrian head trajectories, total contact forces, head injury criterion (HIC), and brain injury criterion (BrIC) were analyzed comparatively. We found that, even within the same vehicle category and initial conditions, the variation in wrap around distance (WAD) spans 84-165 mm, in HIC ranges from 98 to 336, and in BrIC fluctuates between 1.25 and 1.46. These findings highlight the significant influence of vehicle frontal shape and underscore the necessity of using case-specific vehicle models in crash simulations. The proposed method provides a new approach for further vehicle structure optimization aiming at reducing pedestrian head injury and increasing traffic safety.

Assessing the collision risk of mixed lane-changing traffic in the urban inter-tunnel weaving section using extreme value theory.

Urban inter-tunnel weaving (UIW) sections are characterized by short lengths and frequent lane-changing behaviors in the area, commonly used for fast through traffic. These features increase the likelihood of collisions, however, collision risk assessment in this area has been inadequate. The aim of this study was to evaluate the potential collision risk of urban inter-tunnel weaving (UIW) sections in mixed lane-changing traffic conditions in morning rush hours, utilizing surrogate safety measures. The investigation involved the collection of trajectory data via an unmanned aerial vehicle (UAV). Time to collision (TTC) and extended time to collision (ETTC) were chosen as surrogate safety indicators. The estimation of collision risk was conducted using Extreme Value Theory (EVT) by means ofsurrogate safety indicators. It was found that the threshold of TTC and ETTC in this area was 1.25 s. Furthermore, a comprehensive evaluation of collision risks associated with various vehicle types was performed, revealing an inverse relationship between thecollisions riskof vehicles in mixed traffic and their size. It was worth noting that while heavy vehicles exhibit a lower collision risk, they resulted in the highest energy loss and inflicted greater harm in the event of a collision. By an examination of the distribution features pertaining to conflict types during the operation of heavy vehicles, it showed that the highest likelihood of conflict with heavy vehicles occurred when adjacent lanes are involved. Consequently, the implementation of assisted driving technology for heavy vehicles was imperative in order to mitigate the risk associated with side collisions.

Examining road safety impacts of Green Light Optimal Speed Advisory (GLOSA) system.

Mobility and environmental benefits of Green Light Optimal Speed Advisory (GLOSA) systems have been reported by many previous research studies, however, there is insufficient knowledge on the safety implications of such an application. For safe deployment of GLOSA system, it is most critical to identify and address potential safety issues in the design process. It can be argued that implementation of GLOSA system can improve safety by reducing traffic conflicts associated with the interrupted traffic flow at signalised intersections. However, more research findings are needed from field and simulation based studies to evaluate the impacts on safety under a variety of real-world scenarios. As part of the LEVITATE (Societal Level Impacts of Connected and Automated Vehicles) project under European Union's Horizon 2020 Programme, the main objective of this study is to examine the safety impacts of GLOSA under mixed traffic compositions with varying market penetration rates (MPR) of connected and automated vehicles (CAVs). A calibrated and validated microsimulation model (developed in Aimsun) of the greater Manchester area was used for this study where three signalised intersections in a corridor were identified for implementing GLOSA system. An improved algorithm was developed by identifying the potential issues/limitations in some of the GLOSA algorithms found in literature. Behaviours of CAVs were modelled based on the findings of a comprehensive literature review. Safety analysis was performed through processing the simulated vehicular trajectories in the surrogate safety assessment model (SSAM) by the Federal Highway Administration (FHWA). The surrogate safety assessment results showed small improvement in safety with the GLOSA implementation at multiple intersections in the test network only at low MPR (20%) scenarios of CAVs, as compared to the respective without GLOSA scenarios. No or rather slightly lower improvement in safety was observed with GLOSA implementation under mixed fleet scenarios with 40 % or higher 1st Generation or 2nd Generation CAVs, as compared to the respective scenarios without GLOSA. The implementation of GLOSA system was also found to have some impact on the traffic conflict types (although not consistent across all MPR scenarios), where rear-end conflicts were found to decrease while a slight increase was observed in lane-change conflicts.

Enhancing intersection safety in autonomous traffic: A grid-based approach with risk quantification.

Existing studies on autonomous intersection management (AIM) primarily focus on traffic efficiency, often overlooking the overall intersection safety, where conflict separation is simplified and traffic conflicts are inadequately assessed. In this paper, we introduce a calculation method for the grid-based Post Encroachment Time (PET) and the total kinetic energy change before and after collisions. The improved grid-based PET metric provides a more accurate estimation of collision probability, and the total kinetic energy change serves as a precise measure of collision severity. Consequently, we establish the Grid-Based Conflict Index (GBCI) to systematically quantify collision risks between vehicles at an autonomous intersection. Then, we propose a traffic-safety-based AIM model aimed at minimizing the weighted sum of total delay and conflict risk at the intersection. This entails the optimization of entry time and trajectory for each vehicle within the intersection, achieving traffic control that prioritizes overall intersection safety. Our results demonstrate that GBCI effectively assesses conflict risks within the intersection, and the proposed AIM model significantly reduces conflict risks between vehicles and enhances traffic safety while ensuring intersection efficiency.

Improving Child Passenger Safety Through Proper Restraint System Use.

Nurses could help close caregivers' knowledge gap.

Development and classification of autonomous vehicle's ambiguous driving scenario.

Human drivers are gradually being replaced by highly automated driving systems, and this trend is expected to persist. The response of autonomous vehicles to Ambiguous Driving Scenarios (ADS) is crucial for legal and safety reasons. Our research focuses on establishing a robust framework for developing ADS in autonomous vehicles and classifying them based on AV user perceptions. To achieve this, we conducted extensive literature reviews, in-depth interviews with industry experts, a comprehensive questionnaire survey, and factor analysis. We created 28 diverse ambiguous driving scenarios and examined 548 AV users' perspectives on moral, ethical, legal, utility, and safety aspects. Based on the results, we grouped ADS, with all of them having the highest user perception of safety. We classified these scenarios where autonomous vehicles yield to others as moral, bottleneck scenarios as ethical, cross-over scenarios as legal, and scenarios where vehicles come to a halt as utility-related. Additionally, this study is expected to make a valuable contribution to the field of self-driving cars by presenting new perspectives on policy and algorithm development, aiming to improve the safety and convenience of autonomous driving.

Heterogeneity in crash data: A cross comparison between ordered probability model and its variant using crashes at suburban type arterial.

Transportation researchers have long been using the statistical analysis of traffic crash data to create a proactive awareness of traffic safety, make important decisions about the design of vehicles and highways, and develop and implement safe preventive strategies to improve safety. Despite significant progress toward maintaining and analyzing traffic crash data, researchers still encounter several challenges and methodological barriers when conducting statistical analysis. One of these challenges is dealing with the issue of unobserved heterogeneity in crash data. This study uses state-of-the-art methodologies to model the injury severity of traffic crashes that occurred on a specific road segment, namely, a suburban-type road (STR), simultaneously addressing issues related to unobserved heterogeneity in data. Multiple heterogeneity ordered probit models are evaluated against Ohio crash data from the Highway Safety Information System (HSIS). The findings reveal the heterogeneous nature of some variables, such as the nighttime indicator, and demonstrate the distinctive feature of each model to capture the effect of unobserved heterogeneity in analyzing data with such variables. Furthermore, the result helps comprehend the contextual scenarios of crashes at STRs and formulate practical plans to lower the severity of such crashes.

Lateral safety evaluation of autonomous trucks in overtaking maneuvers and adverse weather: Exploratory comparison of different lateral control modes.

As the detrimental impact of the commonly recommended centered driving mode for autonomous trucks on road longevity is gaining attention, more lateral control modes are being proposed to enhance road sustainability. However, there is currently a lack of research on the lateral safety analysis of autonomous trucks with different lateral control modes, especially in complex driving scenarios (such as overtaking) and adverse weather conditions. Therefore, this study developed a safety assessment framework to comparatively analyze the risk probability differences in lateral accidents during overtaking maneuvers by autonomous trucks with different lateral control modes under adverse weather conditions. Based on aerodynamics and vehicle dynamics simulations to capture the multifactorial influences on truck lateral deviation, the results are used for model validation and training. In the reliability approach, Support Vector Machine Regression (SVR) is introduced to establish the SVR response surface model with optimal predictive performance, and combined with Monte Carlo simulations for safety assessment, quantifying safety indices. The results indicate that trucks being overtaken during overtaking maneuvers are more prone to lateral accidents under crosswind influences. The overall impact of lateral control modes on the lateral safety trends is minor. Compared to other lateral control modes, following the centered zero-drift mode is generally safer. However, in conditions of low wind speeds (below 20 km/h) or on highly slippery road surfaces (road friction coefficient below 0.1), autonomous trucks following a uniform distribution mode can better maintain a low-risk level. This study provides crucial insights for future considerations integrating road longevity and truck safety in a collaborative manner, and the proposed methodology has broad applications.

Driver Drowsiness Multi-Method Detection for Vehicles with Autonomous Driving Functions.

The article outlines various approaches to developing a fuzzy decision algorithm designed for monitoring and issuing warnings about driver drowsiness. This algorithm is based on analyzing EOG (electrooculography) signals and eye state images with the aim of preventing accidents. The drowsiness warning system comprises key components that learn about, analyze and make decisions regarding the driver's alertness status. The outcomes of this analysis can then trigger warnings if the driver is identified as being in a drowsy state. Driver drowsiness is characterized by a gradual decline in attention to the road and traffic, diminishing driving skills and an increase in reaction time, all contributing to a higher risk of accidents. In cases where the driver does not respond to the warnings, the ADAS (advanced driver assistance systems) system should intervene, assuming control of the vehicle's commands.

What can we learn from the AV crashes? - An association rule analysis for identifying the contributing risky factors.

The objective of this study is to explore the contributing risky factors to Autonomous Vehicle (AV) crashes and their interdependencies. AV crash data between 2015 and 2023 were collected from the autonomous vehicle collision report published by California Department of Motor Vehicles (DMV). AV crashes were categorized into four types based on vehicle damage. AV crashes features including crash location and time, driving mode, vehicle movements, crash type and vehicle damage, traffic conditions, and among others were used as potential risk factors. Association Rule Mining methods (ARM) were utilized to identify sets of contributing risky factors that often occur together in AV crashes. Several association rules suggest that AV crashes result from complex interactions between road factors, vehicle factors, and environmental conditions. No damage and minor crashes are more likely affected by the road features and traffic conditions. In contrast, the movements of vehicles are more sensitive to severe AV crashes. Improper vehicle operations could increase the probability of severe AV crashes. In addition, results suggest that adverse weather conditions could increase the damage of AV crashes. AV interactions with roadside infrastructure or vulnerable road users on wet road surfaces during the night could potentially lead to significant loss of life and property. Furthermore, the safety effects of vehicle mode on the different AV crash damage are revealed. In some contexts, the autonomous driving mode can mitigate the risk of crash damages compared with conventional driving mode. The findings of this study should be indicative of policy measures and engineering countermeasures that improve the safety and efficiency of AV on the road, ultimately improving road transportation's overall safety and reliability.

Crash risk estimation of Heavy Commercial vehicles on horizontal curves in mountainous terrain using proactive safety method.

Heavy commercial vehicles (HCVs) face elevated crash risks in mountainous terrains due to the challenging topography and intricate geometry, posing a significant challenge for transportation agencies in mitigating these risks. While safety studies in such terrains traditionally rely on historical crash data, the inherent issues associated with crash data have led to a shift towards proactive safety studies using surrogate safety measures (SSM) in recent years. However, the scarcity of accurate microscopic data related to HCV drivers has limited the application of proactive safety studies in mountainous terrains. This study addresses this gap by employing an SSM known as anticipated collision time (ACT) to explore the impact of horizontal curves on the crash risk of HCVs in mountainous terrain. To perform the crash risk analysis, a collection of videos was gathered from horizontal curves in the mountainous terrain along the Guwahati-Shillong bypass in the Northeastern region of India. Subsequently, trajectories were extracted from these videos using semi-automated image processing software. Traffic conflicts were identified using ACT, and the crash risk was estimated through the Peak-Over Threshold (POT) approach of the Extreme Value Theory (EVT). The findings indicate that Run-Off-Road (ROR) traffic events happen more frequently on or near the horizontal curves falling in mountainous terrain. However, the frequency of severe ROR traffic events is lower, indicating the lower propensity for such collisions on the selected curves. The threshold for the safety margin of ROR traffic events involving HCVs was 2 s. The study revealed that stationary models exhibit an overestimation of crash frequency (0, 6) compared to the observed crash frequency (0, 2). Consequently, non-stationary crash risk models were developed, incorporating road geometry and the braking and yaw rates of HCVs as covariates. The results demonstrate that the estimated confidence bounds (1, 2) align with the observed crash frequency (0, 2), emphasizing the applicability of POT models for safety analysis in mountainous terrains in India. The study identified curve radius, length of the approach tangent, and the distance between the center points of horizontal and vertical curves as influential factors affecting the Run-Off-Road (ROR) crash risk of HCVs. Notably, sharp curves with radii less than 200 m or more are associated with a significantly higher crash risk. Additionally, an increased distance between the midpoints of horizontal and vertical curves beyond 1 m was found to escalate the ROR crash risk of HCVs. To mitigate these risks, it is recommended to reduce the length of the approach tangent to prevent high-speed travel on sharp curves. Furthermore, proper signage should be strategically placed to warn drivers and avert potential hazards.

Real-time combined safety-mobility assessment using self-driving vehicles collected data.

The study presents a real-time safety and mobility assessment approach using data generated by autonomous vehicles (AVs). The proposed safety assessment method uses Bayesian hierarchical spatial random parameter extreme value model (BHSRP), which can handle the limited availability and uneven distribution of conflict data and accounts for unobserved spatial heterogeneity. The approach estimates two real-time safety metrics: the risk of crash (RC) and return level (RL), using Time-To-Collision (TTC) as conflict indicator. Additionally, a Risk Exposure (RE) index was developed to reflect the risk of an individual vehicle to travel through a corridor. In parallel, the mobility of corridor were assessed based on the highway Capacity manual methodology using real-time traffic data (Highway Capacity Manual, 2010). The study used a 440-hour AVs' dataset of a corridor in Palo Alto, California. After normalizing for each LOS representation in the dataset, LOS E was identified as the most hazardous operating condition with the highest average crash risk. However, the time spent under different operating condition would affect the safety of individual vehicles traveling through a road facility (i.e., vehicle's exposure time). Accounting for exposure time, the vehicle has the highest chance of encountering an extremely risky driving condition at intersections and segments under LOS D and E, respectively.