Abstract In 2023, the World Health Organization reported approximately 1.19 million road traffic deaths annually, with over half involving vulnerable road users (VRUs) such as pedestrians, cyclists, and motorcyclists. Vehicle-to-pedestrian (V2P) systems promise to improve pedestrians’ safety by providing warnings to both pedestrians and vehicles regarding each other’s presence using connectivity. However, few V2P systems are available for widespread deployment. This paper examines the state-of-the-art in V2P systems and their components, and outlines what has been accomplished—and what still needs to be accomplished—for such systems to be able to reduce pedestrian crashes. The article reviews the following components of V2P systems: methods to detect and track pedestrians, pedestrian trajectory prediction, and the way to convey warning messages through interfaces, including the audio and visual elements found in V2P systems. Determining pedestrians’ locations and predicting their trajectory are two of the most significant challenges for effective deployment. To ensure effectiveness before deployment, pedestrian detection algorithms should be evaluated using real-world data sets, which should include a variety of scenarios. Also, a common set of performance metrics and threshold values is needed to evaluate these systems.
Abstract The highway exit ramp is an important connection area between highways and downstream urban roads. However, exit ramps are often underutilized and unable to regulate traffic flow, which can lead to congestion on the highway and downstream urban roads. To fully exploit the real-time regulatory capabilities of exit ramps, minimize average vehicle delay time, and enhance utilization efficiency of road traffic, this work proposes a collaborative control approach based on the proportional equalization of the number of vehicles in each region. Multiple relevant areas were selected, including the exit ramp area of the expressway, the upstream expressway mainline, the downstream urban connecting road, toll stations, and toll plazas. The traffic flow influencing factors in each area were analyzed. With the optimization objectives of minimizing the average vehicle travel time and maximizing the number of vehicles passing through per unit time, and taking into account constraints such as dynamic speed limits and vehicle delays, a multiarea collaborative control model was established. Taking the typical expressway exit ramps in Huangshan City as the research object, the application effect of the regulation model was verified. The results show that, compared to the SUMO software (vehicle Krauss following model), and LC2013 lane-changing model, the method proposed in this paper significantly reduces the average vehicle passage time in the exit ramp area and increases the maximum number of vehicles passing per hour at different traffic flow ratios. When the traffic flow ratio is 0.45, the optimization effect is most significant, with an average vehicle passage time reduction of 14.08% and a 13.98% increase in the maximum number of vehicles passing per hour. The research results can be effectively applied to traffic flow regulation in highway exit ramp areas and related areas, improving the quality of traffic operation.
Abstract Recent research typically formulates traffic forecasting tasks as spatiotemporal graph modeling problems. For spatial correlations, researchers typically use predefined graphs to learn short-range spatial dependencies among traffic data, neglecting the learning of long-rang spatial features that are crucial for accurately capturing the dependencies among traffic data. For temporal correlations, studies tend to focus on local continuous correlations while overlooking global temporal correlations. To this end, this paper presents an in-depth study on long-range spatial dependencies and global temporal continuity. Specifically, a novel method called adaptive diffused spatiotemporal graph convolution network (ADSTGCN) is proposed for accurate traffic flow prediction. In this approach, an adaptive adjacency matrix–based graph convolution operation is introduced to learn local spatial features. Building on this, global spatial interdependencies between traffic series are represented through graph diffusion convolution, which also relies on the adaptive adjacency matrix. Additionally, a novel module for learning continuous temporal correlations is introduced to effectively capture the sequential patterns in traffic flow data. Additionally, multihead attention is employed to model global temporal dependencies among the traffic series. The experimental results show that the proposed approach achieves superior performance compared to other state-of-the-art methods.
Abstract Growing traffic is a major concern worldwide, impacting both transportation system performance and road safety. In India, these challenges are intensified by highly heterogeneous traffic, highlighting the need to examine flow characteristics and driving behavior under mixed-class, non-lane-disciplined conditions. This study analyzes vehicular behavior at signalized intersections to quantify speed and acceleration profiles in such environments, using unmanned aerial vehicle (UAV) video data collected 100 m upstream and 40 m downstream of the stop line, segmented into 20-m intervals. The UAV-based videographic method is specifically chosen to overcome line-of-sight occlusion inherent in traditional ground-based observation techniques, thereby enabling continuous, high-quality trajectory collection in dense, mixed-traffic settings. Data were collected at four signalized intersections in the cities of Nashik and Nagpur, Maharashtra, India, covering four major vehicle classes: two-wheelers, cars, three-wheelers, and heavy vehicles (including buses). The analysis revealed that two-wheelers and cars exhibited similar speed behavior across all locations, with cars consistently showing the strongest polynomial fit and highest R 2 values, often exceeding 0.9. Two-wheelers also showed good fit, with R 2 ranging from 0.58 to 0.74, with heavy vehicles and buses displaying comparable trends. Three-wheelers showed more erratic behavior, and heavy vehicles generally exhibited the weakest correlation. Descriptive statistics and ANOVA confirmed statistically significant differences among vehicle classes. Based on these findings, a universal polynomial model framework is developed to predict mean speeds as a function of distance from the stop line, with a consistent structure across vehicle classes and locations during the green phase. The framework provides a novel basis for simulating and calibrating heterogeneous, non-lane-based traffic, with green-phase speed, acceleration, and deceleration profiles enabling accurate microsimulation calibration to support signal timing design, safety evaluation, and operational planning in complex urban settings.
Abstract Although previous studies have explored disparities in commuting burdens, the association of built environment with extreme commuting still remains unclear. This study proposes a hybrid framework combining Poisson regression and gradient boosting decision trees to capture origin-destination (OD) pair attributes and nonlinear built environment associations. An empirical analysis is conducted in Shanghai, China, using large-scale rail-transit data from September 2021 and focusing on extreme commutes lasting 60 min or more during the morning and evening peak periods. Our findings reveal a pronounced symmetry in impacts of top built environment variables on extreme commuting, particularly when analyzing origin–destination pairs and contrasting morning and evening peaks. Origin-side variables during the morning peak show nonlinear associations similar to those of destination-side variables during the evening peak; notable differences exist in their effective ranges and threshold effects. Interaction results further suggest that rail-based extreme commuting is jointly associated with living costs, workplace accessibility, and local living conditions.
Abstract As an important control method in the expressway weaving area, variable speed limit (VSL) effectively regulates the dynamic distribution of traffic flow to alleviate congestion. However, given different traffic demand conditions, the control objectives should be different. How to dynamically adjust the evaluation index weights when implementing VSL control according to traffic flow demand is particularly important. For this reason, this paper proposes a multiobjective dynamic weight allocation model and lane-level variable speed limit (LVSL) method by introducing a deep reinforcement learning (DRL) algorithm. First, LVSL control of traffic flow is modeled as a Markov decision process (MDP), and a comprehensive reward function considering traffic efficiency, safety, and environmental benefits is constructed on the scenario of weaving areas with multiple lanes. Second, a multiobjective dynamic weight allocation model and an LVSL (DW-DDPGLVSL) control method based on the deep deterministic policy gradient (DDPG) algorithm are prompted. Finally, simulation tests are conducted using real-world network data, and the results show that the proposed method can improve the safety, efficiency, and environmental friendliness of expressways.
Abstract The growing population of older drivers and the continued use of aging vehicles present increasing challenges to road safety. Older drivers, due to age-related declines in cognitive, visual, and motor abilities, are more susceptible to severe injuries in crashes. Simultaneously, older vehicles often lack modern safety features, exacerbating injury risks during severe collisions. This study investigates the combined effects of driver age and vehicle age on injury severity in single-vehicle crashes involving older drivers. Specifically, it analyzes how driver characteristics, vehicle attributes, environmental conditions, and crash dynamics influence injury outcomes when older drivers operate older versus newer vehicles. Crash data spanning 2020–2024 were obtained from Alabama’s Critical Analysis Reporting Environment system. Two datasets were developed: crashes involving older drivers in older vehicles (manufactured before 2006) and in newer vehicles (manufactured after 2017). Injury severity was categorized as major, minor, or no injury. To account for unobserved heterogeneity, a random parameters multinomial logit model with heterogeneity in means was employed. The results reveal significant differences in injury severity determinants between older and newer vehicles. Nonuse of seatbelts, vehicle rollovers, and collisions with trees were strongly associated with higher probabilities of major injuries across both groups. However, older vehicles exhibited greater vulnerability, with older drivers more likely to sustain major injuries compared to when driving newer vehicles. Environmental factors such as dark, unlit conditions and rainy weather influenced injury probabilities differently across vehicle age groups, suggesting adaptive behavior among older drivers. Crashes in residential areas and those involving pickup trucks or SUVs were associated with lower injury severity. These findings highlight the critical interaction between driver and vehicle aging in crash outcomes. The results underscore the need for targeted interventions, including promoting seatbelt use, encouraging the retirement of older vehicles, and designing safety policies tailored to older drivers.
Abstract To evaluate the dynamic risks of nonmotorized vehicle lane-changing, this study develops an analytical framework based on extreme value theory. Vehicle trajectories were extracted from unmanned aerial vehicle footage using the YOLOv8s (You Only Look Once) object detection model and the ByteTrack multiobject tracking algorithm. The minimum time-to-collision ( TTC min ) and postencroachment time ( PET min ) were modeled as response variables, with key riding behavior variables serving as covariates. Through a threshold selection method integrating mean residual life plots, threshold stability plots, and Akaike information criterion (AIC) minimization, data-driven safety thresholds were established as 0.73 s for TTC min and 0.54 s for PET min . To capture risk dynamics, a nonstationary generalized Pareto distribution (GPD) model was developed, with its scale parameter linked to the covariates. The results indicate that the nonstationary GPD models with covariates significantly outperform the stationary baseline. The optimal model identifies maximum yaw rate and minimum longitudinal distance as key covariates for TTC min , while minimum longitudinal distance is also significant for PET min . These findings provide safety thresholds and a dynamic modeling method for risk assessment, offering an analytical basis to support safety-related applications in intelligent transportation systems, such as the calibration of collision avoidance systems and real-time risk monitoring.
Abstract High-resolution trajectories generated by vehicle telematics can provide detailed, neutral, and naturalistic insights into driving behaviors across large-scale transportation systems. This research explores speeding and hard-braking behaviors extracted from the high-resolution vehicle trajectory (HRVT) data across various socioeconomic profiles, potentially complementing conventional traffic crash records and survey data. Through vehicle movement trajectory analytics, spatial characteristics of these driving events are linked to socioeconomic profiles based on HRVT departure locations. Results reveal positive correlations between speeding rates and household median income, educational attainment, drivers aged 20–59, and “Black or African American” populations, while negative correlations exist with disability rates and older drivers. Similarly, hard-braking rates positively correlate with drivers aged 20–59 and “Black or African American” populations, and negatively with disability rates and older drivers. Notably, “Non-Hispanic White” populations showed positive correlations in contrast to negative correlations with “Hispanic White” populations. Methodologically, multiscale geographically weighted regression more effectively addresses data autocorrelation in geospatial patterns compared to ordinary least squares regression. This study introduces an innovative application of emerging data sources to understand driving behavior across socioeconomic profiles, supporting transportation agencies in developing targeted safety interventions.
Abstract Connected and automated vehicles (CAVs) have the potential to significantly improve the efficiency and safety of urban intersections, yet their benefits depend strongly on penetration rates and their interaction with signal control strategies. This study proposes a joint optimization model that integrates a multiagent deep Q-network (MADQN)-based strategy for CAV trajectory control and an adaptive strategy for signal phase adjustment in mixed traffic. The proposed model incorporates key states from both vehicles and signals, including speed, acceleration, queue length, and phase information, and employs a global reward that combines average delay and energy consumption. An adaptive signal control driven by queue pressure dynamically adjusts phase durations in response to real-time traffic conditions, with its outcomes feeding back into the environment for CAVs training. Simulation experiments conducted on a real-world multidirectional intersection demonstrate that the proposed model effectively reduces delays across most movements even at low CAV penetration rates. As the CAV penetration rate increases, consistent and more pronounced reductions in both delay and energy consumption are observed for all movements.
Abstract Accurate metro ridership prediction is crucial for transit planning, particularly as urban rail networks expand and travel patterns vary across different travel scenarios. Existing models often rely on historical ridership data and may have limited adaptability to temporal and spatial heterogeneity. This study proposes a spatial proximity weighted eXtreme gradient boosting (XGBoost) ensemble to predict daily metro ridership under holidays, weekdays, and weekends. By integrating built environment variables and spatial adjacency, the proposed framework constructs local subdata sets using a nearest-neighbor approach and independently trains submodels. The model reduces dependence on continuous historical ridership sequences and is suitable for limited-data conditions. Using urban rail transit (URT) data from Nanjing, the proposed framework is compared with multiscale geographically weighted regression (MGWR) and global XGBoost. The results show that it achieves higher prediction accuracy in terms of R 2 , mean absolute error (MAE), and root mean squared error (RMSE). Feature importance analysis confirms the model’s interpretability, revealing temporal differences in the effects of variables. The proposed method effectively captures spatiotemporal heterogeneity in ridership dynamics, providing strong generalizability and practical value for urban transit planning and demand forecasting.
Abstract Roadside systems with multiple light detection and ranging (LiDAR) units offer promising capabilities for comprehensive traffic monitoring in intelligent transportation systems. However, they introduce unique challenges due to varying point cloud densities, different measurement qualities, and the need to maintain consistent object detection across overlapping sensor regions. This paper presents multi-objective optimization based 3D object detection for multiple roadside LiDARs (MulDet3D), a novel unsupervised two-stage clustering framework specifically designed for roadside multi-LiDAR object detection. Our approach uniquely combines reliability-weighted background modeling, multi-LiDAR registration, and adaptive density-based clustering with physically constrained hierarchical merging to effectively handle the complexities of multisensor point cloud data. We further introduce a multiobjective particle swarm optimization framework that automatically tunes clustering parameters to achieve optimal performance under different deployment scenarios. Extensive experiments on two real-world multi-LiDAR data sets with different sensor configurations demonstrate that our method significantly outperforms traditional clustering approaches and modern self-supervised methods in challenging scenarios. The results show that our parameter optimization framework substantially improves detection precision, with our method achieving 76.71% AP@0.1 for pedestrians, 70.23% AP@0.3 for small vehicles, and 72.62% AP@0.5 for large vehicles, significantly outperforming baseline methods. These findings highlight the effectiveness of our sensor-aware adaptive approach for roadside traffic monitoring and its potential for practical deployment in intelligent transportation systems without requiring extensive training data. The current study is scoped to dual-LiDAR roadside deployments under normal weather conditions; generalization to larger sensor arrays, adverse weather, and more complex intersection geometries represents important directions for future research.
Abstract As Slovenia’s motorway infrastructure matures and approaches the end of its design life, maintenance-related work zones are becoming increasingly frequent, making the estimation of their capacity a crucial issue for effective traffic management. In addition to maintenance activities, the planned expansion of the motorway network with additional lanes is expected to introduce numerous large-scale construction work zones in the coming years. Accurate capacity assessment is essential for minimizing congestion, ensuring safety, and supporting maintenance scheduling and construction planning on heavily used motorway sections that experience temporary lane closures and geometric restrictions. In current practice, the Highway Capacity Manual and most empirical models provide deterministic capacity estimates, whereas stochastic approaches—particularly those based on the Weibull distribution—have been successfully applied to describe capacity reliability on basic freeway segments. However, such probabilistic concepts have not yet been systematically applied to motorway work zones, where lane reductions and geometric constraints substantially alter traffic dynamics. This study applies a stochastic modeling framework that integrates survival-based capacity sampling with Weibull distribution fitting to estimate motorway work zone capacity using detailed field data from the Slovenian motorway network. The analysis was performed on several work zone configurations for which sufficient prebreakdown flow observations were available to calibrate Weibull capacity distributions. The resulting distribution parameters provide probabilistic measures of breakdown likelihood and allow comparison of reliability levels between work zone configurations under different geometric and traffic conditions. Results show significant capacity reductions in work zones involving lane narrowing and crossovers, while the Weibull-based framework effectively captures the variability and reliability of capacity across different configurations. The study demonstrates the practical applicability of stochastic capacity modeling for motorway work zones and its value for data-driven planning, reliability assessment, and management of maintenance and expansion activities in mature motorway networks.
