A Cross-Modal hierarchical enhanced fusion method for object detection in intelligent transportation systems

Efficient cell association remains a fundamental challenge in fifth-generation (5G) vehicle-to-everything (V2X) systems due to rapid topology changes, heterogeneous deployments, and stringent latency requirements. Conventional learning-based approaches often rely on shallow representations or independent optimization strategies, limiting their adaptability in dense and highly dynamic environments. To address these issues, this study introduces a multi-level graph representation framework that models interactions between vehicles and base stations across hierarchical spatial structures. The proposed approach integrates contextual node embedding with attention-driven graph learning to capture mobility patterns, signal characteristics, and network load dependencies. Additionally, a training-stage optimization mechanism is incorporated to refine attention parameters, improving convergence behavior without increasing inference complexity. The framework is evaluated using a real-world vehicular mobility dataset, demonstrating consistent improvements in association stability, handover reliability, and overall network performance compared with existing deep learning and graph-based methods. Experimental results show gains in accuracy (94.17%) and F1-score (93.93%), indicating enhanced decision robustness under dynamic conditions. Although validation is conducted on an urban dataset, the proposed architecture provides a scalable foundation for adaptive cell selection in next-generation intelligent transportation systems.

Advanced vehicle tracking systems are crucial for the development of intelligent transportation infrastructure, but existing approaches face challenges with real-time visualization, intuitive data interpretation, and effective risk assessment. This paper presents XR-VITS, an Extended Reality Vehicle Intelligent Tracking System that integrates established computer vision-based object detection (YOLO-based detector, internal variant optimized for traffic surveillance), Kalman filtering, homography mapping, and extended reality (XR) visualization techniques into a unified framework for comprehensive traffic monitoring and analysis. The primary contribution of this work lies in the systematic engineering integration of well-established algorithmic components and the comprehensive empirical validation of their combined effectiveness for operator-assisted traffic monitoring, rather than proposing novel detection or tracking algorithms. The proposed system detects and tracks multiple vehicles, maps their trajectories to real-world coordinates, predicts future paths, and assesses collision risks—all visualized through an immersive XR interface. Experimental results demonstrate that XR-VITS achieves 89.3% tracking accuracy (MOTA) while maintaining real-time performance (25 FPS) across diverse traffic conditions, including adverse weather and low-light scenarios. This work targets urban traffic monitoring scenarios where operators must rapidly interpret complex multi-vehicle interactions for safety-critical decision-making. The system’s risk assessment module shows 87.3% precision in predicting potential vehicle conflicts, with XR visualization reducing operator response time by 41.5% compared to traditional interfaces, as validated through a user study with 24 traffic management professionals. This integrated approach bridges the gap between complex traffic data and human comprehension, demonstrating practical applicability for traffic management, autonomous vehicle training, and smart city deployments.

In response to the increasingly serious traffic safety issues and regulatory challenges of electric bicycles (e-bikes), this study proposes an advanced multi-task detection model called P-YOLOv10. The model aims to achieve end-to-end unified recognition of riding safety factors and fine-grained regional attributes of license plates. To address inaccurate small-object detection and difficulty in distinguishing fine-grained features in complex real-world scenes, P-YOLOv10 introduces systematic optimizations based on the latest YOLOv10 architecture. First, it integrates the Selective Channel-Spatial Attention (SCSA) module to enhance the network's ability to capture key local features. Second, it adopts the minimum point distance intersection over union (MPDIoU) loss function to improve bounding box regression accuracy, especially for small objects such as license plates. Finally, it uses the Gaussian error linear unit (GELU) activation function to improve nonlinear representation and training stability. This study trains and evaluates the model on a self-built dataset with 2,237 images. The dataset covers diverse scenes in Guangzhou and Foshan and includes new fine-grained regional annotations. The experimental results show that P-YOLOv10 achieves excellent performance. Its overall mean average precision (mAP) reaches 96.5%, which is 1% higher than the baseline YOLOv10. It also achieves high accuracy on the newly added license plate region recognition task. The results of this study confirm the effectiveness of the integrated optimization strategy. They provide a more accurate and more comprehensive technical solution for intelligent traffic regulation systems.

Vehicular Ad Hoc Networks (VANETs) are essential for intelligent transportation systems; however, their dynamic topology and open wireless communication environment expose them to impersonation, Sybil, replay, and message modification attacks. Existing authentication schemes mainly rely on centralized authorities and conventional cryptographic mechanisms, which lack dynamic trust evaluation and fail to ensure secure authentication during Roadside Unit (RSU) handover and cluster mobility. To address these challenges, this paper proposes a blockchain-enabled trust-aware authentication framework for secure VANET communication. The framework integrates a Joint Probability and Error Unit-based Deep Learning Neural Network (JPEU-DLNN) for dynamic trust assessment and Log-based Edwards Curve Cryptography (LECC) for lightweight key generation. In addition, Gini Indexed Farthest First Clustering (GIFFC) ensures stable cluster formation, while the Directional Gannet Optimization Algorithm (DGOA) supports optimal routing. Blockchain technology provides decentralized and immutable credential storage, eliminating single-point failure and certificate forgery. Simulation results show that the proposed framework achieves 95.62% authentication accuracy and 96.94% packet delivery ratio, with reduced end-to-end delay and improved network lifetime compared to existing blockchain-based authentication schemes. Security evaluation confirms strong resistance against active and passive attacks. These results demonstrate the practical applicability of the framework for secure and reliable VANET deployments.

Short-term traffic flow prediction is crucial for enhancing the control efficiency of intelligent transportation systems. However, most existing methods are limited to single-parameter prediction at the road-segment level, failing to exploit the complex interactions among multiple traffic parameters and their spatial distribution across lanes, which constrains both prediction accuracy and potential for fine-grained management. To address this, a video-driven lane-level multi-parameter short-term prediction framework that integrates computer vision and optimization theory is proposed. The framework follows three coherent steps: first, high-precision simultaneous extraction of lane-level traffic flow, speed, and density from surveillance videos; second, a bidirectional long short-term memory network that jointly models temporal dependencies and inter-lane spatial relationships to produce short-term forecasts; third, a measurement-adjustment-theoretic lane-level joint optimization model that collaboratively corrects the initial predictions to ensure physical consistency (e.g. flow conservation). Experimental results demonstrate that the framework effectively improves prediction performance. On multiple real-world road sections, the mean absolute percentage errors for flow, speed, and density reach 2.89%, 1.01%, and 2.72%, respectively, outperforming state-of-the-art baselines significantly. This study provides a systematic technical solution for fine-grained, reliable lane-level traffic state perception and prediction, laying a solid foundation for advanced applications such as dynamic lane management and real-time congestion mitigation.

This study presents a systematic analysis framework for investigating the influencing factors of data acquisition accuracy in road geomagnetic mapping under a crowdsourcing paradigm, to solve the practical problems of uncontrollable data quality, inconsistent acquisition standards and high cost of mapping in crowdsourced geomagnetic mapping. Based on the trajectory data collected by a variety of smartphones according to the crowdsourcing mode, this study constructs a multi-dimensional crowdsourced data quality evaluation system from the perspective of positioning accuracy and geomagnetic accuracy. The impacts of key acquisition parameters-including equipment model, frequency of repeated sampling, acquisition period, and spatial environment-are comprehensively analyzed. To ensure high data quality, an acquisition boundary determination method is proposed, which provides an operational technical pathway and optimization strategy for constructing high-precision road geomagnetic maps in crowdsourced settings, thereby enhancing the reliability and usability of crowdsourced geomagnetic data. Key findings reveal that: (1) prioritizing high-performance mainstream devices (e.g., Huawei, Honor) significantly improves data quality; (2) when the frequency of repeated sampling is about 10 times, it can effectively improve the accuracy of the data, beyond which diminishing returns and saturation effects occur; (3) data acquisition during low-interference periods (e.g., nighttime or early morning) effectively reduces electromagnetic noise and improves data stability; (4) open areas exhibit superior signal conditions and measurement accuracy compared to challenging environments such as urban canyons with significant shading. These insights offer practical guidance for optimizing crowdsourced geomagnetic data acquisition and support the development of robust, low-cost, and wide-coverage data acquisition patterns. The proposed method holds promise for applications in intelligent transportation, underground navigation, and urban infrastructure monitoring, contributing to seamless indoor-outdoor positioning services.

Autonomous driving offers a promising way to tackle the rising number of fatalities from traffic accidents. An autonomous vehicle includes many features, but the ability to detect pedestrians is crucial, challenging, and relevant to various real-time situations like surveillance, tracking people, and monitoring. Accurately identifying pedestrians is difficult because they can appear in different shapes, positions, and postures. They can wear various types of clothing and sometimes be partially hidden or blend in with nearby objects. This paper focuses on the real-time detection of pedestrians for self-driving cars using a popular hardware platform: The field programmable gate array (FPGA), Ultra 96 v2. The study implements a method for pedestrian detection based on a histogram of oriented gradients (HOG) combined with a support vector machine (SVM) classifier to recognize individuals on the FPGA board, leveraging high-level synthesis (HLS) tools. The effectiveness of the system has been tested on both still images and live video. The results show that advanced FPGA boards like the Ultra 96 v2 significantly improve performance metrics. The system operates at a clock frequency of 150 MHz while using less than half of the available resources and consuming around 2.5 W of power. Also, the system reports the pedestrian detection accuracy close to 95% and other efficient metrics for detection evaluation, like precision (78.6%), recall (88.3%), and F1 Score (83.1%). In summary, the developed system can detect pedestrians in real-time and has the potential to significantly improve the development of a smart and safe transportation environment.

Accurate and rapid traffic flow prediction is crucial for traffic management and the planning of intelligent transportation systems. With the increasing complexity of traffic networks and the widespread deployment of sensors, the scale of traffic flow data has significantly increased. Prediction models designed for small-scale datasets often struggle when applied to large-scale traffic networks. They typically encounter challenges such as long training times and high computational resource consumption. This limits the scalability of existing models. To address these issues, this study proposes a traffic flow prediction model based on a Multi-Subgraph Spatio-Temporal Graph Convolutional Network (MuSTGCN). Our model divides the original graph into multiple subgraphs by identifying pivotal nodes. We employ spatio-temporal graph convolution modules for each subgraph to capture spatio-temporal features. The results of these subgraphs are integrated through a parallel output module to generate the final prediction result. Furthermore, we developed a phased training strategy to train each subgraph separately, thereby improving efficiency. Experiments conducted on real-world traffic flow datasets demonstrate that the MuSTGCN achieves high prediction accuracy on large-scale datasets containing 8600 nodes. Additionally, our model significantly reduces training time, inference time, and GPU memory usage. This study offers a novel approach to developing prediction models for large-scale traffic flow data.

Traffic accident prediction in rapidly urbanizing metropolitan regions remains a critical challenge due to the complex interplay of spatiotemporal dynamics, severe class imbalance, and the opacity of predictive models that limits actionable policy interpretation. Existing approaches tend to address these challenges in isolation—deploying graph neural networks without imbalance correction, or applying oversampling without incorporating spatial context—thereby falling short of the comprehensive decision-support capability demanded by intelligent transportation systems. This paper presents a novel integrated framework, designated SLT-SHAP, that systematically unifies spatiotemporal graph convolutional learning, Synthetic Minority Oversampling Technique (SMOTE) applied exclusively to the training partition, Long Short-Term Memory (LSTM) networks for sequential temporal dependency modeling, a Transformer encoder for long-range contextual attention across hourly traffic sequences, and SHapley Additive exPlanations (SHAP) for post-hoc model interpretability. The study employs a curated spatiotemporal dataset of 132,480 observations collected at hourly resolution across 48 administrative zones in Medan Metropolitan City, Indonesia, encompassing traffic, meteorological, infrastructural, and geospatial variables with an inherent accident class imbalance of 12.4%. Experimental results demonstrate that SLT-SHAP achieves an F1-score of 0.796, AUC-ROC of 0.963, AUPRC of 0.784, and Matthews Correlation Coefficient (MCC) of 0.783, surpassing all baseline and ablation variants. Ablation analysis confirms that each component—graph construction, SMOTE, LSTM, and Transformer—contributes independently to performance. SHAP analysis identifies congestion index, hour of day, and average speed as the three most influential predictors, with spatial heatmapping delineating persistent high-risk zones. The proposed framework offers a replicable and interpretable decision-support architecture for urban road safety analytics in the Indonesian and broader Southeast Asian metropolitan context.

As the "dual carbon" goals steadily advance， investigating the characteristics and driving factors of transportation carbon emissions in the Yellow River Basin is of great significance for promoting low-carbon transition and high-quality development. Based on transportation carbon emission data of 64 prefecture-level cities （states and leagues） in the basin from 2010 to 2022， we comprehensively applied spatial autocorrelation analysis， standard deviational ellipse， optimal parameter geographical detector （OPGD）， and geographically and temporally weighted regression （GTWR） to explore the spatio-temporal patterns and driving mechanisms of transportation carbon emissions. The findings reveal that： ① Transportation carbon emissions in the Yellow River Basin showed an increasing trend， with a spatial pattern of "high in the east， low in the west." The centroid of the standard deviational ellipse consistently remained in Changzhi City while shifting southeastward. ② OPGD identified freight volume， population size， urbanization level， and openness level as the primary driving factors， with interactions primarily exhibiting nonlinear enhancement and bifactorial enhancement. ③ GTWR results showed that freight volume， population size， and urbanization had significant positive driving effects on transportation carbon emissions in most cities， while the openness level demonstrated a negative inhibitory effect. Accordingly， we propose establishing a cross-regional carbon trading mechanism， optimizing urban spatial layouts， and advancing intelligent transportation systems， which could provide a scientific basis for reducing transport carbon emissions in the Yellow River Basin.

Congestion in urban areas continues to pose a challenge for rapidly developing cities, resulting in longer travel times, fuel consumption, and environmental degradation. Traditional shortest-path algorithms, although computationally efficient, are not very adaptable to dynamically changing congestion patterns. This study proposes a congestion-aware multi-objective Genetic Algorithm (GA) framework grounded on NSGA-II with Rolling Horizon Optimization (RHO) to improve the adaptability of routing to congestion patterns in urban transportation networks. The model was developed in the SUMO simulation framework and tested with real-world traffic data in Bethlehem City. The experimental findings show that the proposed strategy can reduce travel time by up to 16.7% in high congestion situations and intersection waiting time by up to 19.8% under high-traffic conditions. The stability and scalability of the framework were confirmed by experimental results in the presence of stochastic disturbances such as accidents, demand surges, and poor weather conditions. In contrast to most of the current GA-based methods tested on artificial data, this study focused on real-world validation, dynamic congestion integration, and multi-objective trade-off analysis through Pareto optimization. The results indicate the potential of evolutionary optimization methods for scalable data-driven intelligent transportation systems.

In recent years, the internet of vehicles (IoV) has become an important enabler of intelligent transportation systems, providing vehicle-edge computing for latency-sensitive and computation-intensive vehicular applications. In such environments, the efficient offloading of tasks is pivotal; yet existing approaches primarily focus on optimising performance, often under the assumption of benign operating conditions or by employing static, trust-based mechanisms. However, these methods fall short in practical IoV implementations due to high mobility, short-lived connectivity, and the adversarial nature of IoV, where misbehaviour and resource exhaustion can substantially compromise the system's reliability and security. In response to these problems, we design SpiralEdge-IoV, a secure and adaptive task offloading framework that tightly integrates defence and optimisation. This framework embeds a logarithmic spiral defence (LSD) mechanism that models trust as a deepening, adaptive path over time, enabling online risk evaluation and incremental offensive action against suspicious parties. We integrate these risk scores into an in-built bio-inspired Addax-optimisation-based decision model (LSD-AddaxNet) to obtain security-aware multi-objective offloading decisions that not only minimise latency, energy consumption, and execution cost, but also maximise robustness. A combination of realistic vehicular edge-offloading traces and the VeReMi misbehaviour dataset is employed to conduct an extensive simulation-based evaluation, confirming the effectiveness of the proposed framework. Relative to representative optimisation- and learning-based baselines, SpiralEdge-IoV delivers up to 18% lower average task latency, reduces energy consumption by about 15%, and increases task success rates in adversarial settings by over 20%. In addition, the analyses on convergence and scalability demonstrate that the framework enables stable optimisation with acceptable runtime overhead in dense vehicular scenarios. SpiralEdge-IoV can be helpful for attack-resilient, low-latency IoV edge computing and is thus suitable for safety-critical vehicular applications and future intelligent transportation systems, as shown in the results.

Study on a multi-factor lane-changing risk resilience assessment model based on genetic algorithm and fault tree analysis.

This article introduces a dataset that investigates the physiological responses of drivers when using advanced driver assistance systems (ADAS) in real-world traffic conditions. The study, conducted in the Federal District, Brazil, involved seven drivers in controlled driving sessions. The time of day and the days of the week were standardized to ensure comparable traffic conditions. The data collection was centered on ADAS Level 2 systems, specifically the Lane Keeping Assist System (LKAS) and the Forward Collision Warning System (FCWS). The dataset includes five physiological signals: respiration, heart rate, galvanic skin response (GSR), leg muscle activity, and brain activity. These signals were continuously acquired using a dedicated instrumentation system installed in the vehicle. Given the complexity of collecting data under real traffic conditions, the acquisition sessions generated a large volume of raw data. Considerable post-processing was conducted to identify and segment portions of the signals with sufficient integrity for subsequent analysis. The dataset is structured as time-stamped raw signal spreadsheets, each corresponding to a specific driver and direction of the pre-established route (outbound and return). Such organization enables researchers to navigate the dataset easily, explore specific segments of interest, and conduct comparative analyses across participants and varying traffic conditions. The dataset is relevant to researchers in biomedical signal processing, driver state monitoring, intelligent transportation systems, and human-machine interaction. It may be used by academic laboratories investigating physiological responses during driving tasks, as well as by engineers and developers working on advanced driver assistance systems (ADAS), including automotive manufacturers and ADAS technology suppliers. The dataset, which includes synchronized physiological and vehicle dynamics data collected under real traffic conditions may contribute to the study of human responses during semi-automated driving, supporting research and development of driver-centered mobility technologies.

Accurate detection of key traffic elements in complex road scenes is critical for autonomous driving and intelligent transportation systems. However, existing lightweight detectors often suffer from missed detections under small targets, large-scale variations, and cluttered backgrounds. To address these challenges, we propose DSE-YOLO11, a RAD-oriented lightweight adaptation of YOLO11n that integrates DynamicConv, SlimNeck, and EMA in a stage-wise collaborative manner. The main contribution lies not in introducing entirely new primitive modules, but in developing a task-specific integration strategy for improving detection robustness in complex road scenes. Specifically, a dynamic convolution-based backbone improves local feature modeling and representation of irregular and small-scale targets. A lightweight neck strengthens cross-scale feature interaction while reducing redundant fusion overhead. Additionally, an efficient attention mechanism suppresses background interference and enhances responses to key regions. Experiments on the RAD dataset show that DSE-YOLO11 improves recall from 0.744 to 0.811 and mAP50 from 0.810 to 0.856, while maintaining 2.96M parameters and 7.1 GFLOPs. These gains are practically meaningful because they indicate fewer missed detections of small, low-contrast, and safety-relevant traffic elements in complex road scenes. Additional experiments on BDD100K provide preliminary external support, although broader validation is still needed.

FNon R-CNN: A multi-scale ground object detection and recognition network

In the context of the Internet of Things (IoT) and Connected and Automated Vehicles (CAVs), eco-driving has become a key strategy for promoting sustainable and intelligent transportation by leveraging vehicle-to-everything (V2X) communication to optimize driving behavior, reduce energy consumption, and minimize environmental impact. However, achieving a dynamic balance among energy efficiency, traffic flow, safety, and driving comfort remains challenging in complex and congested urban environments such as roundabouts. This paper proposes an IoT-enabled multi-objective eco-driving framework, termed DDPG-KAN, which integrates the Deep Deterministic Policy Gradient (DDPG) algorithm with Kolmogorov–Arnold Networks (KANs) to enhance nonlinear feature representation and policy learning capability under mixed-traffic conditions. A multi-objective reward function is designed to jointly optimize safety, energy efficiency, traffic smoothness, and driving comfort, while an Action Inspector ensures collision avoidance and a Model Predictive Controller (MPC) guarantees smooth and stable control execution. Leveraging IoT-based connectivity, the framework allows cooperative perception and adaptive decision-making among CAVs in real time. Simulation results show that the DDPG-KAN approach reduces energy consumption by 31.87%, improves traffic efficiency by 10.93%, decreases carbon emissions by 31.86%, lowers collision warnings by 38.34%, and enhances driving comfort by 53.46% compared to rule-based methods. These findings demonstrate the potential of DDPG-KAN as an effective IoT-driven solution for achieving low-carbon, safe, and comfortable mobility in intelligent connected vehicle systems, contributing to the realization of sustainable and green transportation networks.

Systematic review of weaving area safety: Assessment, behavior, and countermeasures.

FedDrive-Sec: A blockchain-assisted federated deep reinforcement learning framework for secure and adaptive internet of vehicles