Automated vehicles (AVs), by lacking conventional interaction cues derived from human driver feedback, introduce greater uncertainty in pedestrian interactions, especially when secondary task engagement heightens safety risks. External human–machine interfaces (eHMI) are widely recognized as an effective means of simulating human driver communication, but current research largely remains vehicle-oriented. With the advancement of intelligent road infrastructure, it has gradually gained the capability to convey information to road users. In the context of vehicle-infrastructure cooperation, this study considers vehicle-eHMI and road-eHMI as an integrated system to explore their potential in pedestrian-AV interaction. A video-based experiment was conducted to investigate the effects of vehicle-eHMI, road-eHMI, and secondary task complexity on pedestrian crossing behavior and subjective responses. The results indicate that both vehicle-eHMI and road-eHMI improve interaction safety and enhance trust, but their influence on behavioral decisions depends on the risk level conveyed by implicit cues such as vehicle motion. Furthermore, the combined presentation of vehicle and road eHMIs yields the best performance. Increased secondary task complexity significantly elevates cognitive workload, reduces situational awareness and trust, and leads to more hesitant crossing decisions. Notably, road-eHMI demonstrated a pronounced compensatory effect under high cognitive load, as its location within the central visual field enables more efficient detection and interpretation when attention is constrained. This study highlights the necessity and feasibility of eHMIs in pedestrian-AV interaction and suggests that future designs should consider pedestrians’ cognitive states and integrate intelligent infrastructure and cooperative communication to support multi-agent interaction, providing new design insights and theoretical support for safe and efficient interaction.

Under the development of advanced driver assistance systems (ADAS), auditory warnings have increasingly been adopted to support driver situation awareness (SA) and improve hazard response. However, the influence of warning semantic complexity and lead distance on driver behavior remains insufficiently understood. Guided by SA theory, this study designed auditory warnings at three semantic complexity levels (SA L1 to SA L3) and two lead distances: 400 m (early warning) and 200 m (late warning). A driving simulator experiment was conducted to collect behavioral, physiological, and subjective measures, with the post-warning process being analyzed across perception, preparation, and execution stages. Generalized Estimating Equation (GEE) models assessed main and interaction effects, while a fuzzy synthetic evaluation (FSE) integrated multi-dimensional indicators into a composite performance score. Results showed that, compared to the baseline no-warning condition, all three auditory warnings significantly improved drivers' hazard detection, maneuver preparation, and control stability during the lane changes. High semantic complexity (SA L3) yielded the strongest benefits under early warning conditions, while its advantage diminished under late warnings, becoming comparable to SA L2. Longer lead distances (400 m) reduced sustained visual fixation on the forward driving scene, enhanced warning information processing, and enabled earlier decisions and smoother maneuvers. These findings extend SA theory by demonstrating how warning semantics and lead distance differentially influence driver performance, and they provide practical guidance for optimizing in-vehicle auditory warning design in high-risk environments.

The automated truck platoon is one of the most promising connected autonomous vehicle technologies and is expected to become mainstream in the future. It is foreseeable that automated truck platoons and human drivers will share the roads and interact regularly. As a new traffic element, the truck platoon may influence other drivers’ behaviors and mental states, potentially compromising the safety of mixed traffic. While existing studies have extensively explored drivers’ behavioral responses to truck platoons, little is known about their psychophysiological states during such interactions, particularly how platoon organization influences drivers’ mental workload and attention. To address this gap, a comprehensive set of platoon organization factors was considered, and a high-fidelity driving simulator experiment was conducted. The study employed a 2 (platoon speed: 80 km/h vs. 100 km/h) × 2 (platoon size: three vs. five trucks) × 2 (inner gap: 5 m vs. 25 m) × 2 (traffic environment: presence vs. absence of a lead vehicle) within-subjects factorial design. Drivers’ heart rate, pupil diameter, gaze dispersion, and subjective mental workload ratings were recorded and analyzed, with data collected from 35 participants. Results showed that compared to the baseline, drivers’ horizontal gaze dispersion was more concentrated during interactions with the truck platoon. Furthermore, an inner gap of 5 m can significantly increase drivers’ mental workload compared to an inner gap of 25 m, as indicated by mean heart rate and mean pupil diameter. Regarding platoon speed, drivers’ horizontal gaze was more dispersed at a platoon speed of 100 km/h compared to 80 km/h, likely due to greater attention to maintain lateral distance from the median divider and the platoon. Moreover, drivers’ mental workload showed a significant decreasing trend with repeated interactions with the truck platoon. These findings provide insights into the operational strategies of truck platoons from a human factors perspective.

In tunnels, roadside auditory warnings provide an important safety communication channel for drivers. However, warnings transmitted at the intended volume may not remain clear when reaching drivers. This study examined the effects of broadcast voice, volume, and ambient noise on warning detection, reaction time, speech intelligibility, comprehension, listening effort, and behavioural intention using an audiovisual simulation based on in-situ tunnel recordings. Results showed that the female voice used in this study was associated with better warning detection and speech intelligibility than the male voice. Higher broadcast volume mainly improved detection, whereas high ambient noise reduced speech intelligibility. Speech intelligibility remained low overall and was strongly associated with information comprehension. Higher comprehension was also relevant to higher behavioural intention. Listening effort showed no consistent association with comprehension or behavioural intention. These findings highlight speech recognition as a major limitation and support driver-level evaluation of tunnel warnings.

ObjectiveTunnel roadside auditory warning systems are designed to deliver critical safety messages. In-cabin acoustic fluctuations, such as those caused by vehicle speed and window state, are considered important factors that may impair perceptual and cognitive processing. This study examines how these acoustic variations influence drivers' detection, comprehension, and behavioral intention regarding auditory warnings.MethodsThe study recreated ten distinct in-cabin acoustic environments by combining five vehicle speeds (0, 20, 40, 60, and 80 km/h) with two window states (closed/open). Real-world tunnel auditory warning recordings under these conditions were integrated into an audio-visual experimental platform. Thirty-six licensed drivers participated in this controlled experiment, during which multiple cognitive processing indicators were collected: auditory perception reaction time, speech intelligibility, listening effort, semantic comprehension, heart rate variability, and behavioral intention. A two-stage analytical approach was implemented: generalized estimating equations (GEE) were first used to assess how speed and window state affected warning detection, followed by structural equation modeling (SEM) to evaluate the relationships among the cognitive processing indicators.ResultsVehicle speed and window state significantly affected warning detection, with performance declining notably above 60 km/h, especially under open-window conditions. Both auditory perception reaction time and speech intelligibility were substantially impaired as speed increased. Listening effort rose with increased speech intelligibility, though this relationship demonstrated a non-linear pattern. Semantic comprehension was strongly predicted by speech intelligibility and showed significant variation across different levels of listening effort. Crucially, key information comprehension emerged as the strongest direct predictor of behavioral intention, while listening effort exhibited both direct and indirect effects on intention. The structural equation model confirmed these relationships, revealing significant mediation pathways through listening effort and comprehension in the connection between perceptual processing and behavioral outcomes.ConclusionsThis study shows that drivers' responses to tunnel roadside auditory warnings cannot be explained by detectability alone. Post-detection cognitive factors, particularly speech intelligibility and key information comprehension, play a critical role in shaping drivers' behavioral intention following successful perception of the warning. The findings underscore the relevance of incorporating drivers' perceptual and cognitive responses when assessing roadside auditory warnings in complex tunnel environments.

Understanding and monitoring driver mental workload is essential for improving road safety. This study proposes a multimodal machine learning framework to classify drivers’ mental workload using eye movement metrics, physiological signals, and driving behavior features. A driving simulator experiment was conducted with 26 participants under two workload levels induced by a secondary auditory task. Seven feature combinations and six classification algorithms were evaluated. The results showed that eye metrics were the most informative modality, and that feature selection had a greater impact on classification performance than algorithm choice. A support vector machine with optimized features was selected as the final model based on performance and stability, achieving an accuracy of 87.8% and an AUC of 0.95. To improve model transparency, SHapley Additive exPlanations (SHAP) was applied, highlighting key predictors such as blink rate and heart rate, and uncovering synergistic effects between visual and physiological variables. The model was further validated in a tunnel entrance scenario, where it identified increased workload associated with steeper longitudinal slopes. These findings emphasize the importance of multimodal data integration—particularly eye movements—for assessing mental workload. Future applications should prioritize feature diversity over algorithm complexity to enhance real-world implementation in workload monitoring systems.

Wind shear is a critical meteorological hazard in aviation, especially during takeoff and landing, where rapid changes in wind speed and direction can affect lift and airspeed, compromise aircraft performance, and threaten safety. Existing studies predominantly focus on single-horizon wind shear prediction, which limits the characterization of wind shear evolution across multiple future horizons. This limitation restricts early assessment and timely intervention; thus, multi-step forecast frameworks are necessary. This study presents a comprehensive framework for wind shear severity regime analysis and short-term multi-step time series forecasts at Hong Kong International Airport, with a focus on the central runway (RWY 07C/25C). The approach employs unsupervised methods, K-Means and Gaussian mixture models (GMM), for data-driven classification of wind shear severity, alongside a hybrid forecast framework that combines Salp Swarm Algorithm-optimized variational mode decomposition (SSA-VMD) with deep learning models, which include GRU, LSTM, and their bidirectional (BiGRU, BiLSTM) and residual (ResGRU, ResLSTM) variants. Doppler LiDAR data from July 2023 to August 2025 provide the basis for model development and evaluation. The severity classification identifies four distinct levels, with K-Means providing clearer cluster separation than GMM, with silhouette scores of 0.322 for 07C and 0.329 for 25C and reveals clear differences in event frequency and duration between the two RWY directions. The forecast results show high accuracy at short horizons, followed by a gradual decline as the forecast horizon extends. For 07C, the VMD-based models achieve R2 values of 0.995–0.997 at 1 step and 0.728–0.868 at 9 steps, while for 25C, R2 ranges from 0.995 to 0.998 at 1 step to 0.735–0.819 at 9 steps. Model performance varies across forecast horizons and RWY directions, with no single architecture providing the best performance across all cases. The ablation study confirms the contribution of VMD, with RMSE reductions of 57.88–94.03% for 07C and 61.25–96.15% for 25C across the evaluated forecast horizons. The results highlight the value of multi-scale wind shear decomposition combined with optimized deep learning models and severity-based classification for reliable short-term wind shear forecasts and proactive operational decisions in aviation.

According to the World Health Organization, road traffic accidents contribute to approximately 1.3 million deaths annually, posing a global public safety challenge. Current deployment strategies for traffic police face difficulties in balancing accident prevention and real-time response, leading to inefficiencies. This study proposes a two-phase framework for traffic police dispatch, integrating pre-deployment and dynamic emergency response. In the pre-deployment phase, a deep learning model using accident data from Yinzhou District, Ningbo City (April 2020 to October 2021) identifies high-risk zones-defined as statistically significant spatial clusters of high accident frequency-for police positioning. The dynamic dispatch phase utilizes a real-time optimization algorithm based on the golden hour (<= 10 min) to reduce response times. Experimental results show a 28.71% improvement in average response time (from 5.54-7.21 min to 4.52 min) and an 85.43% police utilization rate, outperforming baseline methods. This approach improves response efficiency and enhances officer utilization. The framework demonstrates potential scalability, offering insights for improving road safety in urban settings with similar characteristics.

Autonomous truck platooning offers a range of advantages in freight transportation, including energy efficiency and operational effectiveness, and has garnered increasing attention. But traditional truck platoons, typically aligned in a straight line, can cause channelized damage to road surfaces due to the concentrated load over short durations. To enhance pavement sustainability, the concept of misaligned truck platoons has been explored, wherein trucks within the platoon intentionally drift laterally by varying amounts. Nevertheless, the misalignment introduces concerns regarding the lateral safety performance of the trucks. To investigate the lateral safety of misaligned autonomous truck platoons under crosswinds, this paper introduces an innovative and efficient approach. Using a three-truck platoon as an example, Cross-Entropy Importance Sampling (CE-IS) is utilized to estimate the lateral risk probability of misaligned truck platoons with a randomized distribution across multiple indicators, and the results are compared with those obtained using the Monte Carlo Simulation (MCS). The results indicate that CE-IS significantly reduces computation time while maintaining high accuracy. This computational efficiency makes CE-IS particularly suitable for intelligent freight systems, where rapid and reliable risk assessment is critical for dynamic platoon coordination, real-time safety management, and optimization under uncertain conditions. Under a consistent in-lane lateral distribution control mode, misaligned truck platoons are more prone than aligned platoons to encroaching into adjacent lanes, thereby increasing the risk of accidents. However, reorganizing the middle truck in a misaligned platoon to establish a misalignment distance away from the lane marking can reduce the lateral risk probability, though it remains slightly higher than that of aligned platoons. This study contributes to the efficient and accurate quantification of lateral risk in vehicle platoon systems. Additionally, it provides valuable insights into enhancing the lateral safety of coordinated truck platoons and mitigating their impact on road surfaces. And this method combining CE-IS proposed in this study can efficiently quantify and calculate the lateral risk of vehicle platoons in intelligent freight systems, providing suggestions for real-time operational strategies.

Existing field-based lane-change risk assessment methods primarily rely on the calculation of vehicle centroid distances, which fail to accurately capture vehicle shape and dynamic interactions, leading to limitations in predicting edge collisions. To this end, this study proposes an improved potential field model to more accurately evaluate the potential collision risk of autonomous truck influenced by moving vehicles in the target lane during lane changing. The proposed improved risk potential field model represents vehicles as rectangular contours, replaces the rectangle with multiple discrete points, and considers the shortest edge-point pair distance. Based on the proposed method, simulation experiments are conducted to calculate the risk potential energy variations of connected autonomous truck lane-changing under different acceleration conditions of target lane vehicles. Compared to the traditional model, the improved model more accurately captures risk evolution, clearly distinguishing between high-risk and low-risk states. Additionally, the risk potential of the autonomous truck decreases when the target lane vehicle accelerates, indicating that a positive acceleration of the target vehicle can sometimes help reduce lane-change risk. The results of this study provide important theoretical support for safety decision-making and path planning for autonomous trucks in the future.

Vehicular crashes involving bicycles result in substantial annual fatalities, raising serious concerns for traffic safety authorities. Understanding the factors influencing crash severity is vital for designing effective countermeasures. However, the limited interpretability of many machine learning models complicates traffic safety assessments. This study introduces the Explainable Boosting Machine (EBM), a transparent glass-box model developed to predict the severity of vehicle-bicycle crashes and identify influential factors. A dataset of 5,341 crashes from the Ningbo Public Security Bureau (2020-2021) was analyzed. To address class imbalance, multiple data augmentation techniques were employed, and Bayesian optimization was used for hyperparameter tuning. EBM performance was benchmarked against black-box models, including LightGBM and XGBoost, using holdout evaluation. The EBM combined with borderline-SMOTE achieved a G-mean of 0.816 and an imbalanced accuracy of 0.651. Key predictors included weather and seasonal effects, with season-location interactions significantly influencing crash severity. This framework provides interpretable insights for data-driven traffic safety interventions and future research.

Building induced low-level wind shear may endanger the starting and landing of the aircraft. The layout of a new airport should allow for the taking off and landing of aircraft away from areas affected by buildings. This paper aims to investigate airflow patterns behind buildings of various sizes and shapes at an airfield using computational fluid dynamics (CFD). The three-dimensional dimensions of both the buildings and the study areas behind them were determined based on an investigation of existing airport structures and in accordance with design guidelines from China and the United States. The findings reveal that the buildings exerted a significant influence across all scenarios. The impact height is significantly affected by the height of buildings. Under conditions defined by moderate wind shear thresholds, the impact height ranges from 1.70 to 2.33 times the height of building. Increasing the depth of buildings at transport airports significantly reduces the impact. Increasing the depth of terminal buildings and utilizing finger piers are recommended methods for increasing terminal capacity. When multiple buildings are arranged side by side in close proximity, their spacing should be reduced. Buildings arranged with an inclination relative to the runway direction will result in a greater width of influence on the side where the building is closer to the runway. The results of this study could provide suggestions for the layout and design of airport buildings.

Intense wind shear (I-WS) near airport runways presents a critical challenge to aviation safety, necessitating accurate and timely classification to mitigate risks during takeoff and landing. This study proposes the application of advanced Residual Network (ResNet) architectures including ResNet34 and ResNet50 for classifying I-WS and NonIntense Wind Shear (NI-WS) events using Doppler Light Detection and Ranging (LiDAR) data from Hong Kong International Airport (HKIA). Unlike conventional models such as feedforward neural networks (FNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs), ResNet provides a distinct advantage in addressing key challenges such as capturing intricate WS dynamics, mitigating vanishing gradient issues in deep architectures, and effectively handling class imbalance when combined with Synthetic Minority Oversampling Technique (SMOTE). The analysis results revealed that ResNet34 outperforms other models with a Balanced Accuracy of 0.7106, Probability of Detection of 0.8271, False Alarm Rate of 0.328, F1-score of 0.7413, Matthews Correlation Coefficient of 0.433, and Geometric Mean of 0.701, demonstrating its effectiveness in classifying I-WS events. The findings of this study not only establish ResNet as a valuable tool in the domain of WS classification but also provide a reliable framework for enhancing operational safety at airports.

Accurately predicting crash injury severity in multi-class settings is vital for improving road safety, as different injury levels require tailored interventions. This study explores the effectiveness of Dynamic Ensemble Selection (DES) combined with Static Ensemble Selection (SES) classifiers for multi-class injury severity prediction. We employ diversity-driven DES methods-DES-KNN and DES-Clustering-alongside classifiers such as Extra Trees, AdaBoost, and XG-Boost. To address data imbalance, SMOTE and its variants are applied for equitable class representation. Results show that DES-KNN with XG-Boost, using SMOTE preprocessed data, achieves the best performance with a Balanced Accuracy Score of 0.56, G-Mean of 0.66, and MCC of 0.26. Additionally, LIME is used to interpret model predictions and enhance transparency by highlighting influential features. Our findings demonstrate that integrating DES with SES classifiers significantly improves predictive performance and interpretability, highlighting DES as a valuable approach for handling imbalanced multi-class crash severity data in support of sustainable transportation strategies.

Automated truck platooning is a promising technology that is expected to be mainstream within the next decade. For the foreseeable future, automated truck platoons will coexist and interact with human drivers. Resembling a train-like formation, automated truck platoons may present challenges for drivers wishing to overtake them, and it is not currently clear how these new formations affect driver behaviour. Therefore, this paper aims to examine driver behaviours in the overtaking process under various platoon organisations. A high-fidelity driving simulation experiment was conducted to investigate the influence of platoon speed (80 km/h and 100 km/h), size (three trucks and five trucks), inner gap (5 m and 25 m) and the surrounding traffic situation, e.g., the presence of a lead vehicle, on drivers' overtaking behaviour. Thirty-eight participants were recruited in the experiment. Results revealed that compared to 80 km/h conditions, the 100 km/h conditions prompted more drivers to exhibit extreme behaviours-either failing to overtake or performing a risky overtaking. Regarding platoon size, drivers tended to deviate farther from the lane center to maintain a larger lateral distance from the platoon under five-truck conditions. With respect to the inner gap, a 25 m inner gap significantly reduced the proportion of successful overtaking maneuvers. Moreover, in critical conditions, a 5 m inner gap extended drivers' response time but did not significantly impact collision probability. The presence of the lead vehicle increased drivers' mental workload and impaired longitudinal stability. These findings may offer insights for managing automated truck platoons. For instance, the platoon speed and inner gap can be regulated in different traffic conditions, to optimize efficiency, while ensuring safety for all road users.

Foggy weather is a prominent contributing factor to the malfunction of Level 3 automated driving systems, potentially impairing drivers’ perception and reaction during the takeover process. While previous studies have investigated fog’s impact on takeover safety by assessing drivers’ visual attention, the influence of fog on drivers’ recovery of situation awareness (SA) during takeovers—particularly concerning neural activities—has been overlooked. This research aims to use functional magnetic resonance imaging (fMRI) to examine how fog affects drivers’ brain activities during takeovers of Level 3 automated driving. Thirty volunteers participated in the experiment. Initially, they engaged in a non-driving-related task. On receiving a takeover request, participants pressed a button, which was followed by the presentation of a driving scenario video. The results showed that fog did not affect brain activation in non-critical takeovers, while significantly influencing brain activation in critical takeovers. Compared with the clear × critical scenario, the percentage change in signal intensity of the middle occipital gyrus and the fusiform gyrus in the foggy × critical scenario was significantly lower. Moreover, the thalamus, prefrontal cortex, and precuneus were activated only in the clear × critical scenario, suggesting that high-level cognitive functions were impeded in foggy weather. These findings indicate that foggy weather impairs drivers’ information perception (Level 1 SA) during takeovers, consequently suppressing comprehension (Level 2 SA) and projection (Level 3 SA). This highlights the potential of fMRI as an effective tool for comprehending drivers’ cognitive states throughout the takeover process.

Trapped charges are commonly observed in doped organic light-emitting diodes (OLEDs) that are especially operating at low temperatures; however, the dynamic behaviors of these charges remain poorly understood, and their potential applications are yet to be explored. Herein, using transient electroluminescence (TEL) technology, a large number of ultralong-lived trapped charges are detected in the doped OLEDs at 20 K. Through systematic studies on the device's TEL responses, we construct a clear charge-carrier dynamics model for the doped OLEDs working at low temperatures. It reveals that shallow trapped charges can spontaneously detrap under Coulomb interaction, while deep trapped charges are permanently stored at their trap states, as long as the device stays at low-temperature operation. Furthermore, we demonstrate for the first time that the spike always appearing at the TEL rising edge of the doped devices at low temperatures is generated from the radiative recombination of those deep trapped charges released by the applied external electric field. More importantly, we propose a novel application of these OLEDs with ultralong-lived trapped charges as time-temperature indicators (TTIs) for monitoring the product quality of biological agents and specialty chemicals during their low-temperature storage and transportation. Thus, this work not only elucidates the dynamics of trapped charges in low-temperature-doped systems but also expands the potential applications of OLEDs to energy and information storage technologies.

Aircraft-missed approaches pose significant safety challenges, particularly under adverse weather conditions like wind shear. This study examines the critical factors influencing wind-shear-related missed approaches at Hong Kong International Airport (HKIA) using Pilot Report (PIREP) data from 2015 to 2023. A Binary Logistic Model (BLM) with L1 (Lasso) and L2 (Ridge) regularization was applied to both balanced and imbalanced datasets, with the balanced dataset created using the Synthetic Minority Oversampling Technique (SMOTE). The performance of the BLM on the balanced data demonstrated a good model fit, with Hosmer–Lemeshow statistics of 5.91 (L1) and 5.90 (L2). The Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were slightly lower for L1 regularization, at 1528.77 and 1574.35, respectively, compared to 1528.86 and 1574.66 for L2. Cohen’s Kappa values were 0.266 for L1 and 0.253 for L2, reflecting moderate agreement between observed and predicted outcomes and improved performance compared to the imbalanced data. The analysis identified designated-approach runway, aircraft classification, wind shear source, and vertical proximity of wind shear to runway as the most influential factors. Runways 07R and 07C, gust fronts as wind shear sources, and wind shear occurring within 400 ft of the runway posed the highest risk for missed approaches. Narrow-body aircrafts also demonstrated greater susceptibility to turbulence-induced missed approaches. These findings show the importance of addressing these risk factors and enhancing safety protocols for adverse weather conditions.

To quantify the impact of lane width and truck lateral control modes on the life-cycle cost (LCC) and operational safety of exclusive lanes for autonomous trucks (ATs), this study proposes a LCC framework that considers the risk of ATs intruding into adjacent lanes under crosswind. Based on the maximum lateral deviation data obtained from XFlow-TruckSim, the gradient boosting decision trees (GBDT) model, which performed best among five machine learning methods, was utilized as the risk quantification method. The study considered three lateral control modes for ATs: central distribution, uniform distribution, and normal distribution, evaluating both their safety and the damage they cause to the exclusive lanes. Furthermore, with lane width and truck lateral control modes as decision variables, a multiobjective optimization model was established to incorporate vehicle risk into the life-cycle costs analysis of the exclusive lanes. The trend of risk probability was generally similar to the variation in total life-cycle costs, primarily due to the overwhelming importance of user safety cost. Adopting a exclusive lane width of 3.54 m for AT operation under crosswind is a favorable decision. For already constructed exclusive lanes with a width less than 3.05 m, it is preferable for ATs to adopt a uniform distribution lateral control mode. When the lane width exceeds 3.21 m, centrally traveling ATs become a better choice. If the lane width is between 3.05 and 3.21 m, central travel is generally superior, and it is least recommended for ATs to follow a uniform distribution lateral control mode. The findings provide crucial guidance for the design of exclusive lanes for ATs and the lateral control modes adopted in their operation postconstruction. The study addresses the conflicting challenge of increasing lane width, which can raise institutional costs while reducing the risk of lateral vehicle accidents.

Unsignalized intersections are high-risk areas for traffic accidents, particularly at night, making the enhancement of drivers' hazard perception a critical objective for road safety. While conventional road markings provide essential visual cues for road users, their effectiveness is limited under specific conditions. As an innovative traffic safety measure, self-luminous road markings (SLRMs) offer intelligent and controllable illumination and have been increasingly implemented in real-world settings. This study investigates the impact of SLRMs on drivers' hazard perception at night in unsignalized intersections. A driving simulator experiment was conducted under two road environments (rural and suburban) with four marking forms (continuous-illuminating text, continuous-illuminating symbol, transition-illuminating symbol, and non-luminous symbol markings). The results indicate that SLRMs significantly improve drivers' hazard perception. In terms of driving behavior, drivers exhibited improved braking responses and maintained cautious driving. In terms of eye movement, SLRMs enhanced drivers' visual search efficiency when exposed to potential hazards. Specifically, in rural environments, continuous-illuminating symbol markings were more effective in strengthening visual search efficiency, whereas in suburban environments, continuous-illuminating text markings were more effective in reducing driving speed. These findings highlight the importance of context-specific intervention strategies and provide theoretical support for the application of SLRMs in road safety, while also informing the future development of dynamic information systems in intelligent transportation.