Accid Anal Prev [JOURNAL]
Sun
200 papers
Beyond the norm: Identifying rare and high-risk pedestrian crash patterns using unsupervised learning.
Bayati Z, Khattak AJ
Accid Anal Prev · 2026 Apr · PMID 41548419 · Publisher ↗
Pedestrian safety remains a major concern, with fatalities rising despite infrastructure and safety improvements. To make meaningful progress, efforts should focus more intensely on reducing the most dangerous and fatal...
Pedestrian safety remains a major concern, with fatalities rising despite infrastructure and safety improvements. To make meaningful progress, efforts should focus more intensely on reducing the most dangerous and fatal cases, given the growing importance of conventional and automated vehicle safety in shaping crash outcomes. This study introduces a composite unsupervised edge case detection framework that combines Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction with Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN). Each crash receives a composite score based on its cluster membership uncertainty and its distance from the core of typical crash patterns in the UMAP space. Based on these scores, crashes are classified into three interpretive layers: Core, Moderate Edge, and Strong Edge. Core cases represent common patterns, while Strong Edge cases reflect rare and complex situations. The framework is applied to 10,108 police-reported crashes from North Carolina coded with the Pedestrian and Bicycle Crash Analysis Tool (PBCAT), a relatively clean database of pedestrian crashes. Crash severity and contextual characteristics were compared across the three layers. Strong Edge crashes were substantially more severe, with 36.6% resulting in fatal injuries compared to 8.1% in the Core group. These high-risk cases often occurred in rural areas, under poor lighting conditions, in non-intersection locations, and involved behaviors such as unusual circumstances or crossing expressways. The findings show that the built environment and crash type influence pedestrian crash patterns. The edge case framework helps detect rare, high-risk crashes often missed by traditional methods, supporting targeted safety efforts.
Inferring the structure of pedestrian flows at a transportation hub.
Jia X, Feliciani C, Murakami H … +5 more , Tanida S, Chen L, Yue H, Yanagisawa D, Nishinari K
Accid Anal Prev · 2026 Apr · PMID 41547102 · Publisher ↗
In transportation hubs, pedestrian flows form complex network structures, leading to serious congestion at peak hours. Understanding their dynamics is crucial to managing safe and efficient transportation. Although many...
In transportation hubs, pedestrian flows form complex network structures, leading to serious congestion at peak hours. Understanding their dynamics is crucial to managing safe and efficient transportation. Although many experimental and theoretical studies have investigated pedestrian interactions at the microscopic level, computational models that account for pedestrians' macroscopic origin and destination (OD) demands and mesoscopic route choices in large walking facilities are rare and lack empirical validation. In other words, pedestrians' decision-making at strategic (macroscopic) and tactical (mesoscopic) levels, other than the operational (microscopic) level, has remained largely unexplored. Here, we propose an integrated Strategic-Tactical-Operational model for transportation hub (STO-Hub model), and validate it using 0.87 million pedestrian trajectories collected over three days by means of 11 LiDAR sensors at JR Shinjuku station in Japan. Based on an abstracted graph of the main concourse with directed links between different platform entrances and gates, we employ the gravity model at the strategic layer to estimate time-varying OD demand, a logit route-choice model at the tactical layer to capture route choice behavior, and an agent-based model to reproduce interactions with the surrounding environment and pedestrians. The STO-Hub model accurately reconstructs OD demand and route-choice behavior, achieving high agreement with directed flow counts, and the simulation delineates local congested areas evident in the sensing data. By estimating OD demand and route splits and by reproducing local interactions at any selected section, the STO-Hub model captures pedestrian dynamics across all three levels, including at congested locations. We further propose a STO-Hub framework that integrates sensing, the STO-Hub model, and management plans, providing a practical 10-min-resolution basis for OD-informed pedestrian guidance and control in transportation hubs. The study fills a gap in strategic modeling and management for large transportation hubs and supports congestion prevention, improved safety, and higher operational efficiency.
A graph-based spatio-temporal framework for predicting safety-critical pedestrian-vehicle interactions at unsignalized crosswalks.
Muduli K, Ghosh I, Ukkusuri SV
Accid Anal Prev · 2026 Apr · PMID 41547101 · Publisher ↗
Pedestrian safety remains a critical global concern, especially in countries like India, where unsignalized crossings with limited traffic control contribute to high pedestrian fatality rates. This study proposes a novel...
Pedestrian safety remains a critical global concern, especially in countries like India, where unsignalized crossings with limited traffic control contribute to high pedestrian fatality rates. This study proposes a novel graph-based framework for analyzing pedestrian-vehicle interactions, advancing beyond traditional indicator-based, trajectory prediction, and pairwise modeling approaches. Indicator-based methods (e.g., PET, TTC) are retrospective and fail to capture evolving dynamics. Trajectory prediction models suffer from error accumulation, while pairwise models like LSTM or Transformer architectures are limited to two-agent interactions, restricting scalability and scene comprehension. In contrast, the proposed framework constructs holistic, scene-level multi-relational graphs, representing pedestrians and vehicles as interconnected nodes, with explicit modeling of pedestrian-pedestrian, vehicle-vehicle, and pedestrian-vehicle interactions. Unlike existing approaches that treat each pedestrian-vehicle pair in isolation, the proposed method represents all road users present in the scene as nodes in a dynamic spatio-temporal graph, enabling the model to learn not only direct pedestrian-vehicle relationships but also indirect influences mediated by surrounding pedestrians and vehicles. This design eliminates the need to pre-select interaction pairs, simplifying real-world deployment and improving scalability in dense traffic scenarios. Disentangled Multi-Scale Aggregation (DMSA) captures group behavior by focusing on contextually relevant agents, while a temporal CNN backbone models both short- and long-range dependencies efficiently. Empirical evaluations demonstrate the superior performance of the proposed model, which achieved a test accuracy of 90.6%, F1-score of 0.906, precision of 0.927, recall of 0.886, specificity of 0.928, and an AUC of 0.950, outperforming widely used baselines from the literature, such as GRU, LSTM, and Transformer-MLP, that have been applied in pedestrian interaction modeling tasks. Ablation studies confirmed the importance of Multi-Relational Adjacency Matrices (MRAM) and DMSA in improving accuracy and reducing false positives. By modeling scene-level dynamics, the framework enables context-aware prediction of critical events, supporting proactive conflict warning systems.
DrowsyDG-Phys: Generalizable driver drowsiness estimation in conditional automated vehicles using physiological signals.
Wang J, Li W, Wang Z … +4 more , Ayas S, Donmez B, He D, Wu K
Accid Anal Prev · 2026 Apr · PMID 41547100 · Publisher ↗
Driver drowsiness is one of the leading causes of crashes, injuries, and fatalities on the road. Traditional drowsiness detection models relied on manually extracted physiological features processed through machine learn...
Driver drowsiness is one of the leading causes of crashes, injuries, and fatalities on the road. Traditional drowsiness detection models relied on manually extracted physiological features processed through machine learning algorithms. However, these methods lacked flexibility and robustness across diverse real-world conditions. Although recent advances in deep learning have improved detection accuracy through automated feature extraction based on larger learnable parameter space, the generalization of existing models is still limited due to domain shifts. In this study, we proposed DrowsyDG-Phys, a novel domain generalization (DG) framework for driver drowsiness detection using three physiological signals (i.e., electrocardiogram, electrodermal activity, and respiration signals) that can be measured by in-vehicle or wearable sensors. Our approach introduced a backbone network for explicit time and frequency domain feature learning. In addition, our approach integrated three novel loss functions: a prior knowledge-based contrastive regularization for robustness, a feature centralization loss to promote generalization in heterogeneities, and a novel loss function to align drowsiness assessment criteria. Finally, we established a multi-source DG benchmark and evaluated our model on three existing datasets and a self-collected dataset involving 60 participants in a simulated SAE Level-3 driving scenario. Our proposed DrowsyDG-Phys achieves 78.5% accuracy on the DG protocol, as well as 88.4% accuracy on the cross-subject protocol. Experimental results demonstrated that DrowsyDG-Phys outperformed baseline methods, and improved generalization and robustness of physiological signal-based drowsiness monitoring.
Cooperative or competitive? Resolving social dilemmas in autonomous vehicles through evolutionary game theory.
Li R, Liu Y, Sun J … +1 more , Tian Y
Accid Anal Prev · 2026 Apr · PMID 41547099 · Publisher ↗
With the large-scale deployment of autonomous vehicles (AVs), AV-human-driven vehicle (HV) interactions are increasingly common. AVs face a social dilemma: competitive behavior raises ethical and public acceptance concer...
With the large-scale deployment of autonomous vehicles (AVs), AV-human-driven vehicle (HV) interactions are increasingly common. AVs face a social dilemma: competitive behavior raises ethical and public acceptance concerns, whereas cooperative behavior can invite exploitation and degrade efficiency. We use Evolutionary Game Theory (EGT) to model long-run adaptation between AVs and HVs and quantify agent sociality via a data-calibrated Social Value Orientation (SVO) metric. After calibrating HV social preferences from unprotected left-turn trajectories, we incorporate HV heterogeneity into a two-population EGT with cooperative and competitive types. SVO-informed rewards are used to construct payoff matrices for replicator analyses to identify evolutionarily stable strategies (ESS). Experiments show that AV policies with moderate egoism mitigate the social dilemma and tend to achieve population-level dominance in both roles (left-turning and straight-going), whereas overly cooperative policies are evolutionarily unstable. Moreover, AVs benefit from opponent-aware, dynamically adjustable sociality to accommodate diverse HV preferences. To test the theory, we run agent-based imitation simulations. Sensitivity analyses indicate that AV advantages are hard to observe at low market penetration but become pronounced as penetration approaches about 50%, after which convergence accelerates. Overall, the framework clarifies when and why AV sociality preferences succeed over time, offering actionable guidance for designing adaptive, socially compatible AV decision policies in mixed traffic.
ROAR: Robust accident recognition and anticipation for autonomous driving.
Liu X, Guan Y, Liao H … +2 more , He Z, Li Z
Accid Anal Prev · 2026 Apr · PMID 41547098 · Publisher ↗
Accurate accident anticipation is essential for enhancing the safety of autonomous vehicles (AVs). However, existing methods often assume ideal conditions, overlooking challenges such as sensor failures, environmental di...
Accurate accident anticipation is essential for enhancing the safety of autonomous vehicles (AVs). However, existing methods often assume ideal conditions, overlooking challenges such as sensor failures, environmental disturbances, and data imperfections, which can significantly degrade prediction accuracy. Additionally, previous models have not adequately addressed the considerable variability in driver behavior and accident rates across different vehicle types. To overcome these limitations, this study introduces ROAR, a novel approach for accident detection and prediction. ROAR combines Discrete Wavelet Transform (DWT), a self-adaptive object-aware module, and dynamic focal loss to tackle these challenges. The DWT effectively extracts features from noisy and incomplete data, while the object-aware module improves accident prediction by focusing on high-risk vehicles and modeling the spatial-temporal relationships among traffic agents. Moreover, dynamic focal loss mitigates the impact of class imbalance between positive and negative samples. Evaluated on three widely used datasets - Dashcam Accident Dataset (DAD), Car Crash Dataset (CCD), and AnAn Accident Detection (A3D) - our model consistently outperforms existing baselines in key metrics such as Average Precision (AP) and mean Time-to-Accident (mTTA). These results demonstrate the model's robustness in real-world conditions, particularly in handling sensor degradation, environmental noise, and imbalanced data distributions. This work offers a promising solution for reliable and accurate accident anticipation in complex traffic environments.
Enhancing vision-based traffic crash detection performance consistency across day-night scenes: A depth-aware and domain-adaptive network.
Yang Y, Chen X, Wang J … +3 more , Dong Y, Qie K, Yuan Z
Accid Anal Prev · 2026 Apr · PMID 41544332 · Publisher ↗
Closed-circuit television (CCTV)-based traffic video crash detection systems require stable and consistent cross-scene performance to support all-day crash response and rescue efficiency. However, due to the substantial...
Closed-circuit television (CCTV)-based traffic video crash detection systems require stable and consistent cross-scene performance to support all-day crash response and rescue efficiency. However, due to the substantial domain discrepancies between daytime and nighttime scenes-particularly in illumination and imaging quality-traffic crash detection still suffers from significant performance degradation when transferred across heterogeneous lighting conditions. To address this issue, this research proposed a depth-aware and domain-adaptive network built upon the Visual State Space Model (VSSM) to achieve robust crash detection across heterogeneous lighting environments. The proposed model employed a two-stream architecture that integrated appearance, motion and 3D depth information, in which the depth enhancement module captured fine-grained spatial geometry to provide complementary structural constraints, while the domain adaptation constraint effectively mitigated domain shift, thereby improving the overall robustness and reliability of crash detection. Experimental results demonstrated that the proposed model achieved a recall of 96.043 %, a miss rate of only 2.507 %, and an F1-score of 97.003 %, significantly outperforming several widely used baseline models. Ablation experiments further confirmed the critical roles of optical flow representation, 3D depth features, and the dual-level domain adaptation mechanism in enhancing spatiotemporal consistency. Moreover, the model required only 0.623 GFLOPs and achieved a real-time inference speed of 118 frames per second (FPS), demonstrating high computational efficiency. The proposed framework effectively mitigates the performance discrepancy between daytime and nighttime crash detection, and its high inference speed can contribute to faster emergency response and reduced casualty risk, offering a practical foundation for developing stable and transferable intelligent traffic safety monitoring systems.
The role of ADHD in aggressive driving behavior among young adult drivers: effects of traffic aggressiveness and roadway environments.
Duany JM, Mouloua M, Hancock PA
Accid Anal Prev · 2026 Apr · PMID 41539044 · Publisher ↗
This study examined the effects of Attention-Deficit Hyperactivity Disorder (ADHD), traffic aggressiveness, and roadway environment on driving behavior. Fifty-seven participants (26 ADHD, 31 non-ADHD; M = 20.75, SD = 5.1...
This study examined the effects of Attention-Deficit Hyperactivity Disorder (ADHD), traffic aggressiveness, and roadway environment on driving behavior. Fifty-seven participants (26 ADHD, 31 non-ADHD; M = 20.75, SD = 5.19; 33 males, 24 females) completed questionnaires related to driving behavior. Participants then completed a series of simulated aggressive and non-aggressive drives in both city and freeway environments. Prior to the experimental drives, all participants completed a baseline control drive. Driving performance metrics (i.e., steering angle, acceleration pressure, brake pressure, and speed) and mental workload were recorded across all simulated drives. It was hypothesized that ADHD diagnosis, traffic aggressiveness, and roadway environment would each affect driving performance. Results showed that drivers with ADHD exhibited higher driving speed, while traffic aggressiveness and roadway environment exerted significant effects on steering angle and braking. Notably, ADHD drivers exhibited lower HRV (RMSSD), and NASA-TLX scores tended to be higher under aggressive city driving. The implications of these results for driver assessment, traffic safety, and public health are discussed.
A doubly robust estimation framework to quantify potential bias in linked crash-EMS-trauma data with multi-cohort overlap.
Karimi S, Kluger R
Accid Anal Prev · 2026 Apr · PMID 41520570 · Publisher ↗
Reliable estimation of injury severity is essential for informing trauma care, evaluating crash interventions, and guiding EMS resource allocation; however, analyses based on linked administrative datasets are often comp...
Reliable estimation of injury severity is essential for informing trauma care, evaluating crash interventions, and guiding EMS resource allocation; however, analyses based on linked administrative datasets are often compromised by incomplete linkage and selection bias. This study employs a doubly robust estimation framework to address potential bias in injury severity estimation when integrating multiple datasets. Using Augmented Inverse Probability Weighting (AIPW), we adjust for selection bias introduced by incomplete linkage while improving robustness to misspecification in either the selection or outcome model. Using data from a multi-source linkage of crash, EMS, and trauma records, we estimate the Injury Severity Score (ISS) under three approaches: naïve complete-case analysis, inverse probability weighting (IPW), and AIPW. The naïve approach yielded a mean ISS of 13.52, while both IPW (10.86) and AIPW (10.93) provided adjusted estimates accounting for selection. Subgroup analyses revealed substantial differences in effect size and direction between models. For instance, the impact of male gender on ISS was estimated at 3.98 in AIPW versus 2.22 in naïve analysis. Similarly, secondary collisions and frontage-road crashes showed ISS increases exceeding 10 points under AIPW, compared to considerably lower naïve estimates. Several protective factors, including airbag deployment and crash setting, also demonstrated stronger effects when adjusted for bias. Our results demonstrate that traditional analyses of linked data may underestimate or misstate key risk and protective associations. The proposed AIPW framework offers a practical, statistically rigorous solution for producing population-level inferences in injury severity research using linked administrative data.
Exploring novel surrogate safety indicators measuring conflict riskiness and severity: a case study in Sacramento, United States.
Chen Y, Bian Y, Yuan Q … +5 more , King M, He J, Cao X, Ruan X, Zheng Y
Accid Anal Prev · 2026 Apr · PMID 41518779 · Publisher ↗
Determining appropriate traffic conflict indicators is essential for accurately conducting road safety evaluations. However, existing studies often fail to comprehensively address key factors such as the presence or abse...
Determining appropriate traffic conflict indicators is essential for accurately conducting road safety evaluations. However, existing studies often fail to comprehensively address key factors such as the presence or absence of collision courses, the riskiness before and after the conflict point, and the mobility of the conflict point when selecting conflict riskiness indicators. Moreover, current conflict severity indicators are based on fully inelastic collision theory, which lacks sufficient modeling accuracy. This study delves into the characterization of collision and crossing courses in both angle and straight-line conflicts, providing a comprehensive analysis of the risks faced by road users both before reaching a conflict point and after one has passed it. Based on this analysis, a new combination scheme of conflict riskiness indicators is proposed. A two-dimensional, six-degrees-of-freedom potential collision model is then developed based on the theory of partially elastic collisions, and a new indicator, Extended Delta-E, is introduced. Finally, correlation analysis based on crash and conflict data is performed to compare the proposed indicators with traditional riskiness and severity indicators, evaluating their accuracy in road safety evaluation. The results demonstrate that the conflict rates and severity rates derived from the proposed indicators exhibit the strongest correlation with actual crash rates and crash severity rates, respectively, compared to other indicators. The conflict riskiness and severity indicators proposed in this study offer a more nuanced characterization of conflict events, and can serve as highly accurate alternative measures for crash-based road safety evaluations.
Street vitality and traffic risk: a multiscale analysis of Barcelona and Warsaw.
Galaktionova A, Istrate AL, Tamagusko T … +1 more , Carroll P
Accid Anal Prev · 2026 Apr · PMID 41518778 · Publisher ↗
This study examines the relationship between street vitality and traffic crash risk in Barcelona and Warsaw using street-level spatial regression models and group-based temporal clustering. Street vitality, operationalis...
This study examines the relationship between street vitality and traffic crash risk in Barcelona and Warsaw using street-level spatial regression models and group-based temporal clustering. Street vitality, operationalised as functional density and computer-vision-derived streetscape safety scores, shows a positive association with crash frequency across the street network in both cities, supporting the hypothesis that lively streets increase exposure and possibilities for conflict. However, street vitality explains only part of this risk dynamic: street length remains the strongest predictor overall, while building density produces mixed effects. In hotspot models, street vitality effects weaken substantially; functional density becomes insignificant, and visual safety effects diminish, suggesting that once crash concentrations form, risk is shaped more by localised design and behavioural factors than by land-use intensity. These findings underscore the importance of combining system-wide and site-specific perspectives in street safety planning, highlighting the need for design interventions that reconcile lively public spaces with traffic safety.
Injury severity analysis of e-bike crashes: An age-stratified study of riders aged 40 and above.
Jia J, Yue H, Yang S … +2 more , Jia X, Qiu Y
Accid Anal Prev · 2026 Apr · PMID 41518777 · Publisher ↗
As electric bikes (e-bikes) gain popularity, traffic safety concerns have intensified, particularly for riders aged 40 and above, who face heightened risks due to declining physiological capabilities. However, research a...
As electric bikes (e-bikes) gain popularity, traffic safety concerns have intensified, particularly for riders aged 40 and above, who face heightened risks due to declining physiological capabilities. However, research analyzing crash injury severity factors for this demographic remains limited. This study examined 2452 e-bike crashes involving riders aged 40 and above in Jiaozhou, China, divided into three groups: 40-50 years, 50-60 years, and 60 years and above. A hybrid methodological framework combining the eXtreme Gradient Boosting (XGBoost) algorithm with Shapley Additive exPlanations (SHAP) and a Random Parameters Binary Logit model with Heterogeneity in Means (RPBL-HM) was constructed. Results showed that rural areas, primary/secondary roads, and holidays increase severe injury likelihood across all riders aged 40 and above. Each age group exhibited distinct risk patterns. The 40-50 age group showed higher severe injury probability with sub-zero temperatures and truck-involved crashes. The 50-60 age group faced elevated risks during nighttime, dawn, rainy or snowy weather, sub-zero temperatures, unhealthy air quality, and weekday nights. The 60 and above age group demonstrated higher risks when riders were farmers, unhealthy air quality, off-peak hours, motorcycle/truck involvement, rural autumn, and autumn crashes involving trucks. These findings provide evidence for developing age-targeted traffic safety interventions, offering significant implications for improving e-bike safety among elderly riders in an increasingly aging society.
Evaluating underground space of rail transit hub evacuation under fire scenarios: Virtual reality meets agent-based simulation.
Yuan Z, Lin S, Mao Z … +4 more , Xie C, Tong Y, Li Y, Jia H
Accid Anal Prev · 2026 Apr · PMID 41512487 · Publisher ↗
The rapid growth of urban rail transit has improved transportation efficiency but presents significant safety challenges, particularly in high-density transit hubs where aging infrastructure, overcrowding, and extreme ev...
The rapid growth of urban rail transit has improved transportation efficiency but presents significant safety challenges, particularly in high-density transit hubs where aging infrastructure, overcrowding, and extreme events converge. Fires in such hubs create critical evacuation bottlenecks due to enclosed spaces and complex rescue environments. Existing methods of evacuation research lack the realism and adaptability required to address these dynamic scenarios, particularly for optimizing Selective Door Opening (SDO) strategies. To address these gaps, we developed a high-fidelity virtual environment replicating emergency scenarios in underground transit tunnels. A VR experiment was conducted to collect participants' evacuation trajectories and eye-tracking data under different SDO strategies. Behavioral mechanisms, movement dynamics, gaze patterns and evacuation speeds in different areas were analyzed, with bivariate Gaussian distributions fitted to describe visual perception regions of pedestrians. Features derived from the foregoing VR experiment informed agent-based simulations, enabling the quantitative evaluation of SDO performance across diverse emergency scenarios. The results of the VR experiment indicate that during evacuation, pedestrians exhibit significant differences in evacuation speed and gaze patterns across various spatial regions. In addition, pedestrians show notable variation in perception scale for different types of AOIs. Furthermore, the evaluation of SDO using agent-based simulation reveals that the effectiveness of SDO is influenced by the scenario issues, with passenger volume and the weighting of safety factors dynamically impacting the selection of the optimal SDO.
Uncertainty-aware spatiotemporal interaction learning for pre-conflict risk evolution with a risk-increase prior.
Zhao C, Li M, Liu J … +3 more , Zhang Z, Niu S, Song D
Accid Anal Prev · 2026 Apr · PMID 41494432 · Publisher ↗
Quantifying real-time conflict risk and revealing its evolution are of great importance for enhancing vehicle active safety. Recent studies estimate dynamic risk via conflict probability, yet annotation still relies on t...
Quantifying real-time conflict risk and revealing its evolution are of great importance for enhancing vehicle active safety. Recent studies estimate dynamic risk via conflict probability, yet annotation still relies on threshold based static views and uncertainty is only partially modeled, which limits the assessment of a model's ability to learn conflict evolution. Addressing this gap, we posit a prior hypothesis of increasing pre-conflict risk and develop a risk quantification model that integrates driver control inputs and multi vehicle spatiotemporal interactions with explicit uncertainty outputs. The model is evaluated for accuracy and stability of risk perception, parameter sensitivity, and capacity for pattern learning. Experiments show that, relative to Time To Collision (TTC), Deceleration Rate to Avoid a Crash (DRAC), Proportion of Stopping Distance (PSD), Anticipated Collision Time (ACT),and Emergency Index (EI), the proposed model achieves stronger risk discrimination. On the test set of 15 conflict events used in this study, the proposed model detects elevated conflict risk on average 1.15 s before the conflict point. In four representative scenarios, including car following, ego lane change, unobstructed cut in and cut in under occluded view, the proposed model yields a lower false alarm rate than TTC and, on average, perceives rising conflict risk 1.44 s before the conflict point. Uncertainty analysis indicates lower uncertainty during the rising risk phase, enabling reliable capture of risk evolution. Sensitivity results support the expressiveness of the proposed hypothesis and reveal a common regularity across scenarios, where risk begins to increase approximately 4-6 s before conflict. The results establish a pre conflict risk modeling paradigm that jointly estimates risk and its confidence, supports calibration and transfer across scenarios, and provides an operational basis for proactive safety assessment.
Safety-oriented passenger flow control at a congested metro hub: A microscopic approach.
Zhang J, Shi D, Yang W … +1 more , Ma J
Accid Anal Prev · 2026 Apr · PMID 41494431 · Publisher ↗
Massive passenger influxes at a metro hub pose significant risks. While previous studies have primarily focused on service-oriented passenger flow control from a macroscopic perspective, they have lacked detailed mechani...
Massive passenger influxes at a metro hub pose significant risks. While previous studies have primarily focused on service-oriented passenger flow control from a macroscopic perspective, they have lacked detailed mechanisms to mitigate crowd-related risks from a microscopic viewpoint. To address this gap, this study develops a safety-oriented passenger flow control method based on a comprehensive microscopic pedestrian simulation model for metro hubs. The simulation model operates on two levels: the tactical level, which determines route choices for each pedestrian, and the operational level, which models pedestrian movement using the Social Force Model (SFM). We then introduce a dynamic entry ticket gate service time delay strategy to regulate pedestrian flow and mitigate crowding risks. The strategy leverages Model Predictive Control (MPC) integrated with a Kalman filter to enable real-time decision-making in a stochastic environment. The MPC's predictive model is a linear representation derived from input-output data generated by our simulation. The control objective, defined as crowd danger, represents the state-of-the-art understanding of pedestrian dynamics. Computational experiments demonstrate the effectiveness of this approach in stabilizing the crowd danger metric around a predefined target level. Furthermore, the results reveal two important management insights: (1) Increasing the crowd danger reference value intensifies fluctuations in crowd behavior, as shown by our control performance analysis and correlation analysis, highlighting the need for a lower reference value to maintain stable control. (2) While this study focuses on crowd safety in the station hall, the proposed approach also improves platform service levels under various train headways and passenger densities in carriages. Thus, this safety-oriented strategy can be integrated into broader passenger flow management efforts to enhance both safety and service efficiency.
Interpretable analysis of risk factors for intercity bus operation safety on plateau roads using the SHAP method.
Yun M, Zheng H, Lu P
Accid Anal Prev · 2026 Mar · PMID 41483662 · Publisher ↗
Identifying and analyzing factors that influence traffic risk is crucial for reducing crash frequency and severity. This is particularly true for intercity buses operating on high-altitude plateaus, where unique environm...
Identifying and analyzing factors that influence traffic risk is crucial for reducing crash frequency and severity. This is particularly true for intercity buses operating on high-altitude plateaus, where unique environmental and topographical conditions heighten operational risks. However, quantitative research on how these factors influence traffic operation risk in such contexts, especially across freeways, general highways, and urban roads, remains insufficient. This study uses two months of GPS trajectory data from 10 intercity bus routes in Qinghai province, China to address this gap. We assessed traffic operation risks at discrete road segments using three indicators: Coefficient of Speed Variation (CSV), Severity of Rapid Acceleration (SRA), and Severity of Rapid Deceleration (SRD). A Random Forest regression model, interpreted by the SHAP method, was employed to explore the complex effects of influencing factors. The results quantify the unique risks of the plateau environment, showing that driving instability surges at high altitudes, and the impact is significantly amplified by high speeds. The analysis further highlights the operational risks of intercity buses, identifying a clear fatigue threshold where continuous driving leads to more frequent aggressive maneuvers. Critically, the research differentiates these risks by road type, demonstrating a shift in dominant safety factors, as road geometry poses a greater threat on general highways than on freeways. These findings provide quantitative evidence for developing targeted safety interventions that address the unique operational risks of plateau roadways.
Time-to-event crash severity prediction at highway-rail grade crossings with monotonic neural networks.
Keramati A, Lu P, Ren YH
Accid Anal Prev · 2026 Mar · PMID 41483661 · Publisher ↗
Despite advances in highway-rail grade crossing (HRGC) safety, including widespread use of active control devices, crashes at these intersections still lead to severe outcomes. Conventional crash prediction models often...
Despite advances in highway-rail grade crossing (HRGC) safety, including widespread use of active control devices, crashes at these intersections still lead to severe outcomes. Conventional crash prediction models often fail to capture severity-level dependencies, rely on assumption-driven and computationally intensive methods, and overlook links between severity and time between events. This study introduces a predictive framework based on positive monotonic neural networks (MNNs) for modeling time-to-crash outcomes with severity at HRGCs. Considering the relative newness of the time-to-crash paradigm in HRGC safety, the Neural Fine-Gray model is adopted as a core MNN implementation to estimate severity-specific crash likelihoods. This approach eliminates the numerical integration required in traditional time-to-event models, substantially reducing computational burden and accelerating training for large datasets. The framework naturally handles imbalanced HRGC data by treating event-free records as right-censored, avoiding the resampling required in traditional machine-learning approaches. To examine severity-level dependencies-an aspect largely overlooked in the literature-four MNN architectures are developed and evaluated. Using a 29-year North Dakota HRGC dataset, results show trade-offs between predictive accuracy and computational efficiency. The cause-specific MNN performs best for medium- and long-term horizons, whereas the multi-head MNN converges faster and excels at short horizons. Moreover, benchmarking against traditional time-to-event models-cause-specific Cox and Fine-Gray-shows modest calibration gains and 2%-50% stronger discrimination, reflecting the alignment between MNNs and the nonlinear, high-dimensional HRGC data. The framework also enhances interpretability by revealing paradoxical effects, including the "adding flashing lights paradox" and the "adding stop signs paradox."
Loudness of In-Vehicle auditory Warnings: Sustained counteraction of task-related driver fatigue and individual differences in alertness maintenance.
Wu M, Wang X, Filtness A … +3 more , Yue L, Huang Z, Jiao Y
Accid Anal Prev · 2026 Mar · PMID 41483660 · Publisher ↗
In-vehicle fatigue warnings alert drivers when signs of fatigue are detected. These warnings can temporarily restore alertness, helping drivers respond before a crash occurs. They play a key role in reducing fatigue-rela...
In-vehicle fatigue warnings alert drivers when signs of fatigue are detected. These warnings can temporarily restore alertness, helping drivers respond before a crash occurs. They play a key role in reducing fatigue-related crash risk. However, the optimal loudness and effective duration of such warnings are not yet well understood. This study aims to analyze the effectiveness of auditory warnings from two perspectives: first, by employing the Kaplan-Meier method and the Random Survival Forest model to analyze alertness maintenance time across different warning loudness levels and to identify influencing factors. And second, by applying a Bayesian Cox proportional hazards model with random intercepts and slopes to examine individual differences in baseline alertness maintenance risk and sensitivity to 50 dB auditory warnings within the driver population. The results indicate that auditory warnings heighten drivers' awareness of fatigue and extend the duration over which drivers maintain alertness following a warning. For mild task-related fatigue, one to three warnings produced an incremental benefit compared with no warning, as indicated by longer alertness maintenance. Compared to novice drivers, experienced drivers exhibited a higher baseline risk of reduced vigilance, greater sensitivity to the 50 dB warning, and a lower risk of post-warning fatigue. Overall, this study provides practical guidance for fatigue warning design and accounts for individual differences in alertness maintenance risk. It also introduces a new perspective for evaluating warning effectiveness based on alertness maintenance time.
Uncovering latent structures of crash typology in narcotic-involved fatal crashes for safe system interventions.
Tusti AG, Chakraborty R, Chowdhury TI … +3 more , Islam MM, Mimi MS, Das S
Accid Anal Prev · 2026 Mar · PMID 41483659 · Publisher ↗
Narcotic-impaired driving increases the risk of fatal crashes, yet existing studies rarely provide narcotic-specific crash typologies that link driver impairment to roadway, traffic, and environmental conditions. This ga...
Narcotic-impaired driving increases the risk of fatal crashes, yet existing studies rarely provide narcotic-specific crash typologies that link driver impairment to roadway, traffic, and environmental conditions. This gap limits the design of Safe System interventions that can proactively address the most common high-risk configurations. Using Fatality Analysis Reporting System data from 2018 to 2022, this study applies Cramér's V statistic for variable selection and Cluster Correspondence Analysis (CCA) to explore unsupervised crash typologies and latent patterns of narcotics-involved fatal crashes. CCA biplot coordinates group crashes into four clusters: high-speed lane changes on uncontrolled arterials, run-off-road impacts with rollovers, nighttime pedestrian or cyclist strikes on unlit roads, and moderate-speed angle crashes at signalized intersections. Results show that speed and lateral control failures dominate the first two clusters, narcotic-induced sensory and cognitive deficits under low visibility drive the third, and decision-making errors during turn phases characterize the fourth. Key factors such as posted speed limit, lighting condition, and driver age exert cluster-specific influences on incapacitating and fatal injury outcomes. These findings underscore the inadequacy of appropriate countermeasures and point to Safe System-aligned interventions, including dynamic speed management, enhanced roadside clear zones, targeted lighting upgrades, and intersection control strategies.
Enhancing collaborative perception through multi-scale contextual information integration.
Ma M, Hong D, Zhang H … +3 more , Miao Z, Wang W, Zhao G
Accid Anal Prev · 2026 Mar · PMID 41478138 · Publisher ↗
In autonomous driving, perception systems face challenges from dynamic environments such as occlusions, changing lighting, and unpredictable traffic. These conditions make it hard to capture fine local details and broad...
In autonomous driving, perception systems face challenges from dynamic environments such as occlusions, changing lighting, and unpredictable traffic. These conditions make it hard to capture fine local details and broad context, while real-time operation demands high efficiency and low communication cost. In this paper, we propose a multi-scale contextual information integration (MSCI) framework designed to enhance collaborative perception. The method employs multi-scale adaptive attention augmentation to focus on relevant features across spatial scales, capturing fine-grained details and wider contextual cues to improve perception accuracy. The context-aware perception enhancement module then combines local and global information. It refines features to keep perception robust and stable in changing or challenging environments. Finally, the GRU-based dynamic booster embeds a motion-aware mechanism into the recurrent unit. This strengthens temporal modeling for sequential data and improves real-time decision making. Experimental results demonstrate that the proposed method achieves notable improvements in both detection accuracy and communication efficiency.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe