Accid Anal Prev [JOURNAL]
Sun
200 papers
A Bayesian spatial bivariate multinomial logit model for joint analysis of crash type and injury severity at signalized intersections.
Ma J, Ma X, Su Z … +2 more , Jia Y, Wang C
Accid Anal Prev · 2026 Sep · PMID 42172803 · Publisher ↗
Crash type and injury severity at signalized intersections may be shaped by overlapping risk mechanisms and spatially clustered influences, yet they are often modeled as separate outcomes. This study applies a Bayesian s...
Crash type and injury severity at signalized intersections may be shaped by overlapping risk mechanisms and spatially clustered influences, yet they are often modeled as separate outcomes. This study applies a Bayesian spatial joint bivariate random-parameters logit framework to jointly analyze crash type and binary injury severity at signalized intersections using Texas crash records from 2022 to 2024. Crash type is represented by rear-end, sideswipe, and left-turn/angle crashes, while injury severity is modeled as injury versus no injury. The framework integrates three components: cross-equation dependence through correlated intersection-level latent effects, unobserved heterogeneity through random parameters, and residual spatial dependence through conditional autoregressive effects across neighboring intersections. Model comparison based on WAIC and DIC shows that the spatial joint random-parameters specification consistently outperforms independent and non-spatial alternatives. The results indicate that unobserved intersection-level factors jointly influence crash-type composition and injury outcomes, with injury severity more strongly aligned with latent risk associated with left-turn/angle crashes than with sideswipe crashes. Several covariates, including driver age, adverse roadway or weather conditions, multi-vehicle involvement, and distraction-related factors, show distinct associations across crash types and injury severity. These findings support a more integrated and spatially informed interpretation of signalized-intersection safety and provide evidence for mechanism-oriented diagnosis and targeted intervention prioritization.
Revealing the divergences between LLM-simulated and human takeover decision-making in ADS-equipped HGV operations.
Xu Z, Xi Y, Hua J … +2 more , Cen X, Zheng N
Accid Anal Prev · 2026 Sep · PMID 42150399 · Publisher ↗
Global adoption of Level-3 (L3) autonomous driving systems (ADS) in commercial heavy goods vehicles (HGVs) remains limited, as concerns about reliability and safety foster trust miscalibration and reinforce experience-de...
Global adoption of Level-3 (L3) autonomous driving systems (ADS) in commercial heavy goods vehicles (HGVs) remains limited, as concerns about reliability and safety foster trust miscalibration and reinforce experience-dependent risk heuristics. This study examines human-ADS interaction in automated HGV operations and investigates whether large language models (LLMs) can generate takeover decisions that resemble human behavior. Human-in-the-loop experiments were conducted within the fully compiled CARLA simulation platform featuring a large-scale road environment (30 × 30 km) with realistic background traffic flow dynamics. Human reactions, particularly takeover behavior during ADS operations, were systematically examined from compliance rates, decision latency, intervention strategies, and crash outcomes across freeway merging, freeway navigating, and urban driving scenarios. The same scenarios were presented to LLM-agent-equipped HGVs, where the agents were prompted and contextualized as scalable surrogates for "truck drivers", and their performance was then recorded and compared with human counterparts. Results showed that (i) crash rates dropped substantially from human participants (12.0 %) to LLM-agent-controlled ADS (2.8 %-4.6 %); (ii) whereas LLM agents relied exclusively on deceleration and steering for safety responses, human participants also frequently used acceleration as a preferred intervention strategy; (iii) participants with 6-20 years of driving experience intervened more pre-emptively and exhibited below 50 % compliance with ADS operation. These findings suggest that human takeover decisions are shaped by experience-based risk heuristics and exhibit substantial heterogeneity across drivers. They also indicate that current LLMs are not yet reliable proxies for human supervisory behavior during the autonomous driving. The methodological framework developed in this investigation offers a replicable approach for researchers examining human-ADS interaction across transportation domains, and the empirical findings inform appropriate boundaries for LLM application in safety-critical human factors research.
Analysis of the factors affecting the evolution of the severity stages of traffic conflicts in long tunnels.
Zhang Y, Qin W, Xiao F … +3 more , Wei P, Hu Z, Meng X
Accid Anal Prev · 2026 Sep · PMID 42150398 · Publisher ↗
Rear-end conflicts in long freeway tunnels are safety-critical yet difficult to study using crash data alone, because crashes are rare and tunnels impose constrained sight distance and abrupt environmental changes that c...
Rear-end conflicts in long freeway tunnels are safety-critical yet difficult to study using crash data alone, because crashes are rare and tunnels impose constrained sight distance and abrupt environmental changes that can accelerate risk escalation. This study examines how conflict severity evolves as vehicles approach a critical interaction point and whether the underlying mechanisms vary across stages. We analyze trajectory-based rear-end conflicts extracted from continuous video monitoring in two long tunnels on Guangxi expressways and reconstruct each conflict at four spacing-to-leader stages (80, 60, 40, and 20 m). Methodologically, we first conduct horizontal stage-wise estimation using ordered logit, random-parameters ordered logit, and a heterogeneous-means extension. We then perform sensitivity analyses under alternative outcome definitions, balancing schemes, and stage definitions. Finally, we implement a longitudinal Bayesian hierarchical ordered-logit model to link coefficients across stages and decompose them into an overall stage trend and stage-specific deviations. Results show that heterogeneity-aware models outperform the basic ordered logit, and that the effects of relative-motion indicators and traffic-state factors change as the conflict point nears, revealing distinct upstream and near-conflict sensitivities. The sensitivity analyses indicate that the principal stage-wise mechanisms remain stable across reasonable modeling choices. The Bayesian results further show that common variables do not evolve uniformly across stages: some effects are governed mainly by smooth cross-stage trends, whereas others retain clearer local departures, thereby distinguishing structurally stable evolution from stage-specific adjustment. Overall, the proposed horizontal-longitudinal framework provides an interpretable basis for distance-aware tunnel warning logic and stage-sensitive safety management in long tunnels.
Crash root-cause identification via trace-rewarded causation chain reasoning large language model.
Xie N, Huang J, Zhou X … +1 more , Yu R
Accid Anal Prev · 2026 Aug · PMID 42140126 · Publisher ↗
Road traffic crash is one of the top leading causes of death worldwide. To support the deployment of modern safety improvements such as Highly Automated Driving (HAD), collision analysis requires to reveal crash formatio...
Road traffic crash is one of the top leading causes of death worldwide. To support the deployment of modern safety improvements such as Highly Automated Driving (HAD), collision analysis requires to reveal crash formation and find the root-causes. This study explores Large Language Model (LLM)-based techniques to reconstruct the crash causation chain, and reveal the root-causes underlying crash occurrence. By designing trace-reward functions, a domain reasoning model base upon DeepSeek-R1-Distill-Qwen-1.5B is constructed to identify the root-causes. Specifically, the trace-rewards are constructed from accuracy in recognizing crash types and entities, accuracy in extracting crash-related behaviors, and alignment degree of behavior and root-cause. Monte Carlo Tree Search (MCTS) is then employed to broaden the exploration of potential root-causes and build inference paths. Finally, Group Relative Policy Optimization (GRPO) is applied to identify the optimal inference traces through training the model. Empirical analyses are conducted using Multi-Modal Accident Video Understanding (MM-AU) dataset. The results show that the proposed method raises the Micro Accuracy of root-cause identification from 0.427 to 0.870 and improves the Macro Recall from 0.389 to 0.852. This demonstrates that the proposed method effectively enhances the benchmark model's ability to understand the process of crash formation. Based on the identified root-causes, further analyses and discussions are conducted, providing effective support for traffic safety management and the development of preventive strategies.
Does acceptance of driver monitoring systems differ across countries? A latent profile analysis approach.
Öztürk İ, Madigan R, Lehtonen E … +3 more , Aittoniemi E, Lee YM, Merat N
Accid Anal Prev · 2026 Aug · PMID 42134196 · Publisher ↗
Driver Monitoring Systems (DMS) are poised to become a standard safety feature in new vehicles, with regulations in regions such as the European Union mandating their inclusion. Yet little is known about how privacy and...
Driver Monitoring Systems (DMS) are poised to become a standard safety feature in new vehicles, with regulations in regions such as the European Union mandating their inclusion. Yet little is known about how privacy and security concerns influence their acceptance. This study investigated attitudes towards DMS among 9025 licensed drivers from nine countries (Germany, Spain, France, Japan, Poland, Sweden, United Kingdom, United States, and China). Participants rated their acceptance, concerns (including data collection, secondary use, and perceived insecurity), and behavioural intention to use DMS. Through Latent Profile Analysis, five distinct user profiles were identified: Enthusiasts, Somewhat enthusiastic, Moderate, Sceptical, and Resistant, each exhibiting systematic differences in acceptance, concerns, and behavioural intentions. Cross-national comparisons revealed significant cultural variations, with Enthusiasts and Somewhat enthusiastic profiles being more prevalent in China, whereas Resistant and Sceptical profiles were disproportionately represented in France, United Kingdom, and Sweden. Notably, age and gender effects were significant, as older drivers and women were more likely to belong to the resistant profile. These findings underscore the necessity for targeted interventions, transparent data handling policies, and culturally tailored communication strategies to enhance user acceptance of DMS and the widespread adoption as part of the broader transition towards automated and connected driving systems.
The safety impacts of pedestrian delay at signalized intersections.
Cabe R, Jaeger M, Knezevich R … +1 more , Tsai J
Accid Anal Prev · 2026 Aug · PMID 42127591 · Publisher ↗
Pedestrian delay is widely associated with non-compliance and risky crossing behavior, yet few studies have linked delay directly to pedestrian crash outcomes. Using automated traffic signal performance measure (ATSPM) d...
Pedestrian delay is widely associated with non-compliance and risky crossing behavior, yet few studies have linked delay directly to pedestrian crash outcomes. Using automated traffic signal performance measure (ATSPM) data and pedestrian crashes from more than 3,000 signalized intersections in Georgia, this study integrates probability-based, count-based, and economic models to quantify how delay influences pedestrian safety. A Wilcoxon signed-rank test, which controlled for roadway scale and speed limit by comparing high-delay and low-delay groups within geometrics blocks, confirmed that pedestrian delay is a significant factor associated with thru-movement and left-turn crashes (p < 0.01). Following this, multivariate logistic and negative binomial regression models were used to isolate the effects of delay while controlling for roadway geometry and pedestrian exposure (measured by the number of pedestrian push button activations). These results were integrated into an economic cost model to provide agencies with a data-driven tool for quantifying the safety and societal benefits of signal timing optimizations. Based on the analysis, for a 30 s increase in average pedestrian delay, the odds of a thru-movement vs pedestrian crash increase by 23%, and the odds of a left-turn crash increase by 39%. While delay, roadway geometry, and actuation frequency drive the likelihood of a crash occurrence, speed remains the primary determinant of severity; intersections with speed limits above 35 mph experience a 41.2% increase in the odds of a fatal or serious injury (FSI) outcome. Together, these probability, count, and cost frameworks provide transportation agencies with a unified, data-driven approach for identifying high-risk intersections and quantifying the safety and economic benefits of reducing pedestrian delay within Vision Zero and Safe System frameworks.
LLM-enhanced causal graph learning for real-time crash risk prediction.
Li C, Huang H, Zhou H … +3 more , Li F, Zhou R, Zhong L
Accid Anal Prev · 2026 Aug · PMID 42127590 · Publisher ↗
Accurate and reliable risk prediction is crucial for preventing traffic crashes and improving road safety. However, existing approaches often struggle to balance predictive performance with causal interpretability. To br...
Accurate and reliable risk prediction is crucial for preventing traffic crashes and improving road safety. However, existing approaches often struggle to balance predictive performance with causal interpretability. To bridge this gap, we propose a closed-loop framework that integrates causal discovery, semantic enhancement, and spatio-temporal prediction. First, transfer entropy extracts event-specific causal relationships from dangerous driving scenarios to construct initial causal graphs. These graphs are then refined using a GPT-2-based language model fine-tuned with Low-Rank Adaptation (LoRA), which employs a graph enhancement strategy to reinforce salient causal pathways. Finally, the semantically enhanced graphs are fused with temporal driving features using a spatio-temporal model that combines Graph Attention Networks (GAT) and Long Short-Term Memory (LSTM) networks for real-time risk prediction. Evaluated on car-following scenarios from the highD dataset, our framework achieves an accuracy of 0.956, F1-score of 0.872, and AUC of 0.985, significantly outperforming traditional baselines. Ablation studies further confirm that the GPT-2-based causal enhancement with LoRA fine-tuning is a significant contributor to the model's predictive accuracy. These findings indicate that our framework offers a powerful solution for enhancing both precision and interpretability in collision risk prediction, holding strong potential for deployment in real-time traffic management systems.
Safety-to-trust dynamics: can automated vehicles navigate yellow-light dilemma zones from the driver's perspective?
Wang S, Jiang J, Li Z … +5 more , Liu Z, Zhao W, Li P, Zhang G, Ou S
Accid Anal Prev · 2026 Aug · PMID 42127589 · Publisher ↗
At signalized intersections, the yellow-light Dilemma Zone (DZ) is the roadway segment within which, at the onset of yellow, drivers cannot clearly determine whether stopping before the stop line or proceeding through th...
At signalized intersections, the yellow-light Dilemma Zone (DZ) is the roadway segment within which, at the onset of yellow, drivers cannot clearly determine whether stopping before the stop line or proceeding through the intersection is the safer maneuver, thereby elevating the risk of rear-end and right-angle crashes. Existing research primarily relies on signal-based or Connected Vehicle strategies (e.g., broadcasting signal phase/countdown information to drivers) to mitigate DZ-related safety risks, but these approaches often compromise intersection efficiency or impose additional cognitive demand on drivers. Recent advances in automated driving suggest a potential alternative, yet a clear gap remains: existing research has not conclusively quantified whether automated driving improves safety in DZ scenarios, particularly across Levels 3 and 4 automation, nor has it clarified whether such safety improvements transform into greater drivers' trust during the transitional period in which human supervision and possible intervention are still required. To address this gap, this study develops a safety-to-trust framework to examine whether higher levels of automation improve objective safety in yellow-light DZ scenarios and whether these safety improvements constitute the mechanism through which automation shapes drivers' trust. A driving simulator study was conducted by recruiting 52 participants to drive through a signalized intersection at the onset of yellow under Levels 0, 3, and 4, respectively. Driving behavior and its related safety performance were collected and analyzed. A linear mixed-effects model and a logistic regression model were applied to evaluate the effects of automation level on minimum time-to-collision (TTC) and traffic conflict, respectively, thereby assessing the safety benefits of different automation levels. Structural equation modeling (SEM) was further used to examine whether safety performance mediated the relationship between automation level and drivers' trust. Results suggest that, under DZ conditions, higher automation levels significantly improve safety. SEM further reveals that increased automation improves safety, which in turn elevates drivers' trust in automated driving, highlighting safety as the key mediating linkage between automation level and trust. Together, these findings quantify the safety benefits of automated driving systems in yellow-light DZ and clarify how those benefits shape trust, thereby providing an integrated basis for informing Automated Vehicle (AV) deployment and human-machine interface strategies at urban signalized intersections.
Influences of illuminated speed limit markings on driving behaviors-A driving simulation study.
Li Z, Wang Y, Zhao W … +2 more , Sze NN, Ma W
Accid Anal Prev · 2026 Aug · PMID 42127588 · Publisher ↗
Roadside speed limit signs and conventional painted markings are widely used, but their effectiveness is often compromised under low-visibility conditions, leading to insufficient speed awareness and compliance. To addre...
Roadside speed limit signs and conventional painted markings are widely used, but their effectiveness is often compromised under low-visibility conditions, leading to insufficient speed awareness and compliance. To address this issue, this study introduces illuminated speed limit markings of two different sizes and compares their effects on driving behavior with conventional signs under different conditions. Using linear mixed models (LMM) and generalized linear mixed models (GLMM), we examined the effects of driver characteristics, environmental factors, and sign types on five key driving performance indicators, moderating effects of weather and time. Results show that illuminated markings significantly outperform conventional signs under low-visibility conditions: small illuminated markings reduce speed variation and improve compliance in foggy weather, while large ones excel at nighttime. Both sizes of illuminated markings induce more pronounced deceleration tendencies and smoother speed adjustments. These findings highlight the potential of illuminated markings to enhance traffic safety, especially under low-visibility conditions. Future field trials are needed to validate their real-world effectiveness and practical feasibility.
Hysteresis effect of comprehensive driving risk in highway tunnel-group: A naturalistic field test study.
Meng Y, Zhang J, Zhang Y … +6 more , Ren C, Deng Y, Zhang X, Li B, Liu Z, Liu T
Accid Anal Prev · 2026 Aug · PMID 42119163 · Publisher ↗
Driving through tunnel-group sections in mountainous highway involves frequent abrupt changes in the driving environment, which can easily induce fluctuations in driver workload and elevate driving risk. To investigate t...
Driving through tunnel-group sections in mountainous highway involves frequent abrupt changes in the driving environment, which can easily induce fluctuations in driver workload and elevate driving risk. To investigate the variation patterns of driving risk in tunnel-group sections, this study utilized naturalistic field test data and focused on drivers' heart rate variations, pupil characteristics, and driving behaviors. An evaluation index system was established covering aggressive driving behavior, driving workload, and speed-variability-related risk indicators. All indicators were normalized, and a hybrid weighting scheme was adopted by integrating the CRITIC method with the entropy weight method. The comprehensive risk index for each sub-section of the tunnel group was then calculated using a linear weighted approach. The results indicate that driving risk is more sensitive in tunnel entrance and exit sections. As the number of traversed tunnels increases, the overall tunnel driving risk shows a decreasing trend, demonstrating a pronounced hysteresis effect. Moreover, the hysteresis intensity varies significantly with longitudinal spacing length, i.e. the longer the longitudinal spacing, the greater the decrease scale of driving risk for each tunnel within the tunnel group, indicating a stronger hysteresis intensity. Specifically, the decrease scale of the maximum comprehensive driving risk index increased from 0.06 and 0.08 in short-spacing tunnel groups to 0.09and 0.10 in medium-spacing tunnel groups, and further to 0.10and 0.11 in long-spacing tunnel groups. The comprehensive driving risk index developed can effectively characterize the risk evolution and key influencing factors in continuous tunnel groups. The identified risk evolution patterns provide a basis for safety optimization of longitudinal spacing in tunnel groups and can guide the rational deployment of lighting and visual guidance facilities, thereby improving overall driving safety.
Driving behavior and safety performance in spiral tunnel groups: A naturalistic data analysis.
He S, Du Z, Wong YD … +2 more , Wu P, Ding X
Accid Anal Prev · 2026 Aug · PMID 42114247 · Publisher ↗
The characteristics of naturalistic driving along freeway tunnels were examined to better understand driving operations and safety performance along a series of closely-spaced spiral tunnel environment. Thirty-four valid...
The characteristics of naturalistic driving along freeway tunnels were examined to better understand driving operations and safety performance along a series of closely-spaced spiral tunnel environment. Thirty-four valid participants completed runs in both travel directions of spiral tunnel groups over 25 km. Driving behavior was quantified using speed and speed consistency, longitudinal acceleration, lateral lane position, time-to-line crossing, steering wheel angle, and steering entropy. These measures were treated as safety performance indicators and were integrated using the rank-sum ratio method to construct a composite safety performance index for comprehensive evaluation and comparison. The results showed that: (1) Clear directional differences in driving operations. In the downhill direction, vehicles operated at higher speeds with poorer speed consistency and a stronger tendency to speed. In the uphill direction, acceleration and deceleration fluctuated more, indicating greater longitudinal control effort. (2) Safety risk was significantly higher in the downhill direction. The composite safety performance index was markedly worse, accompanied by more frequent speeding and larger dispersion of lateral lane position, reflecting reduced control stability. (3) Pronounced section-dependent degradation was observed. Connector tangent sections exhibited the poorest safety performance. The tunnel entrance zone also performed significantly worse than the interior middle zone and the tunnel exit zone. (4) Longitudinal grade dominated horizontal curve direction in explaining safety performance variation. The effect associated with uphill versus downhill direction was substantially larger than that associated with left versus right turning direction, indicating that grade-driven speed regulation and longitudinal load were primary contributors to elevated risk in spiral tunnel groups. These findings pinpoint high-risk directions and critical sections and support data-driven improvements in geometric design and safety management for freeway spiral tunnel groups.
Traffic risk spatial-temporal evolution mechanism and dynamic prediction approach from the vehicle-to-vehicle interaction perspective.
Du Z, Lyu N, Wen J
Accid Anal Prev · 2026 Aug · PMID 42114246 · Publisher ↗
In intelligent connected vehicles, lane-changing decisions and trajectory planning often prioritize local safety, neglecting the broader impact of vehicle interactions on traffic flow. However, interaction patterns betwe...
In intelligent connected vehicles, lane-changing decisions and trajectory planning often prioritize local safety, neglecting the broader impact of vehicle interactions on traffic flow. However, interaction patterns between vehicles significantly affect both lane-change risk and traffic dynamics. Moreover, existing models often overlook spatial dependencies in highly interactive environments. This study addresses these gaps by proposing a classification of four interaction patterns based on vehicle acceleration and inter-vehicle gap trends. Using the TOD trajectory dataset, we analyze the dynamic evolution of traffic risk across these patterns and introduce a spatiotemporally-aware risk prediction model. A Spatial Durbin Model captures the spatial influence of neighboring vehicles on ego-vehicle risk, while a Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) dynamically adjusts transition probabilities to forecast risk levels and vehicle motion states. This integrated approach enables short-term, macroscopic traffic flow risk forecasting. Results show that aligned interaction intentions lead to slower risk dissipation and higher risk levels. Compared to the GMM-HMM and Recurrent Neural Network (RNN) models that do not account for spatial effects, the proposed model achieves a prediction accuracy of 85.5% in a 5-second prediction window, which is 13.5% and 5.0% higher, respectively. Although it is slightly lower than the 86.5% achieved by the Temporal Fusion Transformer (TFT), the computational time is reduced by 1.814 s, demonstrating better real-time performance. The model shows good prediction accuracy across four interaction modes: steady-state interaction, strong competitive interaction, strong cooperative interaction, and weak cooperative interaction. The accuracy rates for predicting the risk of subsequent vehicles within 5 s are 87.8%, 76.4%, 84.7%, and 81.9%, respectively. This study provides valuable insights into the mechanisms through which different vehicle interaction patterns influence traffic flow, and offers essential guidance and optimization directions for lane-change trajectory planning in autonomous driving systems.
Accounting for under-reporting in wildlife-vehicle collision hotspot identification using copulas and Bayesian model averaging.
Moeinaddini A, Zhang T, D'Agostino C … +2 more , Xie Y, Zou Y
Accid Anal Prev · 2026 Aug · PMID 42114245 · Publisher ↗
Wildlife-vehicle collisions (WVCs) pose significant risks to highway safety and wildlife populations, leading to considerable economic losses. However, efforts to systematically characterize and mitigate WVCs are often c...
Wildlife-vehicle collisions (WVCs) pose significant risks to highway safety and wildlife populations, leading to considerable economic losses. However, efforts to systematically characterize and mitigate WVCs are often constrained by substantial under-reporting in official crash record databases. This study integrates police-reported crash data with carcass removal records from 7041 road segments across Washington State, USA, to provide a more accurate picture of true WVC occurrence and risk factors. Building on prior copula-based approaches, we propose a novel hybrid copula-based framework to jointly model reported WVC frequencies and under-reporting probabilities, incorporating Bayesian Model Averaging (BMA). The hybrid models link a binary logit under-reporting with the count submodel using Gaussian and Student-t copulas to capture dependence. Key findings reveal that under-reporting is less likely near wolverine habitat, on state routes, in low ecological value areas, and on multi-lane roads, but more likely near white-tailed deer habitat and in segments of highest ecological value. Reported WVC frequency increases with Annual Average Daily Traffic (AADT), wider shoulders, proximity to large mammal habitat, and higher speed limits, while it decreases with higher truck percentages and more lanes. Empirical Bayes (EB) hotspot identification (HSID) shows copula-based models outperform independent specifications, capturing 2.8-4.4% more true collisions at selective thresholds (2-5%), with further gains from BMA. Residual diagnostics further confirm the framework's robustness. Hybrid BMA-copula specifications substantially reduce heteroscedasticity and alleviate remaining endogeneity bias compared to Gaussian, no-copula. These results provide transportation agencies with clearer guidance for prioritizing mitigation measures, such as wildlife crossings, fencing, and speed management, on high-risk segments, particularly two-lane roads in ecologically sensitive areas near large mammal habitats, while accounting for systematic under-reporting biases linked to specific species and roadway types.
Traffic conflict characteristics and evolution mechanism in freeway weaving segments with varying spacing.
Jin X, Yanling L, Zhiwei C … +1 more , Tao W
Accid Anal Prev · 2026 Aug · PMID 42107220 · Publisher ↗
With the rapid expansion of China's freeway network and the increasing density of interchange clusters, reduced interchange spacing has emerged as a critical challenge for weaving section safety. To systematically charac...
With the rapid expansion of China's freeway network and the increasing density of interchange clusters, reduced interchange spacing has emerged as a critical challenge for weaving section safety. To systematically characterize how spacing affects the distribution, severity, and duration of traffic conflicts in weaving areas, this study collected aerial video data from four weaving sections with spacing of 400 m, 600 m, 850 m, and 1400 m along the Changhu Freeway in Dongguan. Using the Data From Sky video analysis platform for trajectory extraction and time-to-collision (TTC)-based surrogate safety analysis, the spatiotemporal evolution of traffic conflicts under varying spacing conditions was investigated. The research demonstrates that spacing is a critical factor influencing conflict risk. Reduced spacing exacerbates risk through spatial compression effects, manifested by a leftward shift of the primary TTC distribution peak, reduced collision avoidance time, and a significant increase in both the proportion of critical conflicts and TET duration. In contrast, longer spacing (850 m and 1400 m) provides necessary buffer space, optimizes conflict distribution, and reduces severity. The weaving flow ratio shows a positive correlation with the unit conflict rate. The 850 m scenario exhibits the highest conflict rate due to its combination of high traffic volume and high weaving flow ratio; however, conflict severity remains primarily regulated by spacing. Conflicts between heavy vehicles (L-L) present the highest risk, with a mean TTC of only 2.3 s and a critical conflict proportion of 18.53%, while light vehicle combinations (S-S) perform optimally. Vehicle movement intentions also significantly influence conflict characteristics: combinations with shared intentions (S-S, F-F, H-H) exhibit elevated risks due to high trajectory overlap, and their conflict numbers show an increasing trend under longer spacing. In contrast, heterogeneous vehicle combinations (S-F, S-H) are less affected by spacing variations. Based on these findings, a minimum spacing of 850 m is recommended as a practical safety threshold for weaving section design, and vehicle-type-specific management measures are advised for constrained sections below this benchmark. The findings of this study can provide a theoretical foundation and practical guidance for the planning, design, and management of weaving sections in freeway interchanges.
A novel method for determining the trajectory of vulnerable road users from video recordings and its practical implications.
Janoška Z, Bílová M, Bíl M
Accid Anal Prev · 2026 Aug · PMID 42105696 · Publisher ↗
Accurate trajectory reconstruction is essential for reliable surrogate safety analysis, yet most computer-vision pipelines still represent road users using the geometric centre of a bounding box-a simplification that int...
Accurate trajectory reconstruction is essential for reliable surrogate safety analysis, yet most computer-vision pipelines still represent road users using the geometric centre of a bounding box-a simplification that introduces systematic spatial bias. We propose a novel trajectory-estimation method that computes a representative point derived from object geometry and motion dynamics, yielding a more realistic approximation of the true occupied space. The method was evaluated on 64 h of video data covering 3 traffic environments and 223 intersections of car-pedestrian trajectories. Using Post-Encroachment Time (PET) as the primary performance indicator, we show that the proposed method produces PET values that differ from the bounding-box-centre approach by 0.69 s on average, with discrepancies exceeding 1 s in 16.7 % of all conflict events. These deviations systematically shift the classification of conflict severity, leading to up to 3.6 % of events being re-assigned to a different PET-based risk category. The results demonstrate that conventional trajectory representations may substantially distort surrogate safety metrics. Beyond safety applications, the method enables more precise estimation of the spatial footprint of different road-user types, supporting advanced analyses of street-space allocation and utilisation.
UIRAM: An intention-uncertainty-based risk assessment framework for interactive traffic scenarios.
Wang C, Xiong C, Hu M … +4 more , Liu Y, Ma C, Zhu J, Liu Q
Accid Anal Prev · 2026 Aug · PMID 42102654 · Publisher ↗
The growing heterogeneity of driving behaviors makes accurate identification and quantitative assessment of imminent collisions a cornerstone of safety assurance. Yet existing risk detection and scoring methods often gen...
The growing heterogeneity of driving behaviors makes accurate identification and quantitative assessment of imminent collisions a cornerstone of safety assurance. Yet existing risk detection and scoring methods often generalize poorly across complex interactive scenarios and insufficiently account for uncertainty in surrounding agents' future behavior. To address this issue, we propose UIRAM, an intention-uncertainty-based risk assessment framework for interactive traffic participants. UIRAM first rapidly selects collision-relevant candidates using a simplified two-dimensional Gaussian conflict domain, then predicts for each participant an uncertainty-aware Gaussian trajectory distribution through an intention prediction network that fuses environmental context and interaction dynamics. Based on these predictions, the framework estimates collision probability and further combines it with consequence severity to derive an overall risk score, using the maximum per-agent risk as the scenario-level indicator. Experiments on the FLUID dataset show that UIRAM achieves up to a 0.22-point AUC improvement over representative baselines while completing per-vehicle uncertainty prediction and risk computation within 0.14s. Driver-in-the-loop simulations further demonstrate strong trend-level consistency between UIRAM's risk outputs and drivers' subjective risk perception.
Road traffic accident severity prediction based on large-scale data and multi-dimensional factors: an explainable ensemble learning approach.
Gao Z, Easa SM, Liu Y … +3 more , Chen Z, Ge H, Wang Y
Accid Anal Prev · 2026 Aug · PMID 42102653 · Publisher ↗
Road traffic accidents cause substantial fatalities and economic losses, yet large-scale severity analysis is often constrained by limited access to high-resolution roadway and geometric data. This study constructs a lar...
Road traffic accidents cause substantial fatalities and economic losses, yet large-scale severity analysis is often constrained by limited access to high-resolution roadway and geometric data. This study constructs a large-scale, multi-source dataset with fine-grained road-alignment characteristics to support data-driven analysis of accident severity patterns. Unlike prior work that uses coarse road descriptors, we automatically extract fine-grained horizontal and vertical geometry from extensive networks, enabling detailed geometric rarely available at a large scale. In total, 26 features were integrated, spanning environmental, roadway, and geometric alignment dimensions. A soft-voting ensemble integrating XGBoost, Random Forest (RF), CatBoost, and LightGBM (LGBM) is employed to support severity prediction, while SHAP (SHapley Additive exPlanations) is used to derive both severity-level importance rankings and sample-level explanations. Rather than focusing solely on predictive accuracy, this study uncovers mechanism-informed insights under rigorous evaluation settings designed to mitigate spatiotemporal dependence. The results show that environmental conditions are more strongly associated with lower-severity accidents, whereas roadway type and geometric alignment features become increasingly important for higher-severity accidents. Furthermore, Accumulated Local Effects (ALE) analyses reveal nonlinear patterns and threshold regions for key continuous variables, providing complementary evidence on how risk varies across operating conditions. These findings provide mechanism-informed insights that can support the development of targeted road safety strategies.
Accounting for unobserved heterogeneity, endogeneity, and temporal instability in occupational accident severity modeling.
Salam M C
Accid Anal Prev · 2026 Aug · PMID 42096812 · Publisher ↗
This study provides a simulation-based diagnostic assessment of three interrelated challenges in accident severity modeling: unobserved heterogeneity, endogeneity, and temporal instability. While these issues have been w...
This study provides a simulation-based diagnostic assessment of three interrelated challenges in accident severity modeling: unobserved heterogeneity, endogeneity, and temporal instability. While these issues have been widely examined in isolation, their combined influence on statistical inference and model reliability remains insufficiently understood. To address this gap, a transparent Monte Carlo simulation framework is developed to generate synthetic accident data with controlled structural properties, enabling systematic evaluation of model performance under known conditions. The analysis employs a mixed ordered logit specification, augmented with a control function approach for endogeneity correction and a rolling time window strategy to capture temporal variation. Results demonstrate that neglecting these factors-individually and jointly-can lead to substantial bias in parameter estimates, misidentification of statistically significant variables, and unstable inference across time periods. In particular, endogeneity is shown to induce systematic distortion in behavioral effect estimates, unobserved heterogeneity alters both magnitude and significance patterns, and temporal aggregation masks dynamic structural changes. Rather than offering direct policy prescriptions, the findings are interpreted as methodological insights that highlight the risks of model misspecification in accident severity analysis. The study contributes by providing a reproducible framework for evaluating inferential robustness and by clarifying the limitations of commonly used modeling approaches, thereby informing more rigorous empirical applications using real-world safety data.
Spatial dynamics of crash hotspots under autonomous vehicle adoption scenarios.
Oikonomou MG, Yannis G
Accid Anal Prev · 2026 Aug · PMID 42085783 · Publisher ↗
In road environments with large Autonomous Vehicle (AV) fleets and higher SAE automation levels, reliable crash data are often unavailable, making direct safety assessment infeasible. In such cases, traffic simulation of...
In road environments with large Autonomous Vehicle (AV) fleets and higher SAE automation levels, reliable crash data are often unavailable, making direct safety assessment infeasible. In such cases, traffic simulation offers a valuable alternative for evaluating safety. This study conducts a spatial modelling analysis to predict crash hotspot occurrences under different AV deployment scenarios. The study combines microsimulation-derived conflict data, a quantitative crash-risk formulation, validated using field crash data, based on Time-To-Collision (TTC) thresholds, and spatial statistical analysis using the Getis-Ord Gi* statistic to detect statistically significant hotspots of elevated crash risk. The resulting hotspots were further analysed using a binomial Generalised Additive Model (GAM) to quantify the impact of automation, roadway and spatial factors on the probability that a conflict event occurs within a hotspot area. Results show that automation significantly alters the spatial distribution of crash risk, leading to a gradual reduction and spatial diffusion of hotspots as AV penetration increases. However, a temporary rise in the probability that conflict events occur within hotspot areas occurs under moderate automation shares, highlighting the transitional instability of mixed-traffic conditions. Intersections and other high-interaction areas remained the most critical locations, while congested segments were associated with a higher probability that conflict events occur within hotspot areas. The proposed framework supports data-informed planning and policy decisions during the transition toward automated urban mobility.
From prediction to prevention: safety modeling of driver takeover time with mental workload, risk perception, and driving style in ramp scenarios.
Shao Y, Xu Y, Zhang Y … +1 more , Ye Z
Accid Anal Prev · 2026 Aug · PMID 42085782 · Publisher ↗
In conditionally automated driving, delayed or unstable takeovers can escalate into hazardous situations, making accurate prediction of driver readiness a key element of accident prevention. This study develops a predict...
In conditionally automated driving, delayed or unstable takeovers can escalate into hazardous situations, making accurate prediction of driver readiness a key element of accident prevention. This study develops a predictive framework that integrates mental workload, driving style, and real-time driving risk to anticipate takeover time and identify safety-critical conditions. Using data from 44 participants in a high-fidelity driving simulator replicating urban expressway ramps, takeover scenarios were categorized by ramp type, driver role, and driving style, with eye-tracking derived workload and risk perception metrics as inputs. The CatBoost-based model, supported by interpretability analysis, was applied to assess how individual and situational factors influence takeover performance. Results show that higher mental workload significantly prolongs takeover time, particularly in visually low-risk but cognitively demanding scenarios. Aggressive drivers respond faster but with reduced post-takeover stability, while cautious drivers show the opposite pattern. Ramp type and vehicle interaction events, such as lane cut-ins, further modulate takeover risk, with the model anticipating risk-inducing interactions up to 0.78 s before the actual interaction onset. These findings offer direct implications for adaptive takeover prompt timing, role-aware assistance, and personalized safety interventions, supporting proactive risk mitigation in automated driving.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe