This study analyzes the safety risk transmission mechanism in urban logistics drone last-mile delivery within the policy-driven low-altitude economy. To address the limitations of traditional risk identification methods, which rely heavily on accident data, this research integrates the Fuzzy Decision Analysis Laboratory Method (Fuzzy-DEMATEL),Interpretive Structural Modeling (ISM), and the Matrix of Cross-Impact Multiplication (MICMAC) to construct a three-dimensional analytical framework based on causal relationships, structural hierarchy, and attribute classification.First, Fuzzy-DEMATEL is employed to quantify the strength of causal relationships among risk factors. Next, ISM is used to deconstruct the multi-level hierarchical network and identify fundamental causes within the risk system. Finally, MICMAC is applied to calculate the dependencies and driving forces of each influencing factor, helping prioritize risk governance measures. The research findings indicate that: (1) The safety risk system of urban logistics drones for last-mile delivery exhibits a “dual-core driven – multi-loop coupled” characteristic. Equipment failures act as the physical carriers of systemic failures, while the root-cause risks stem from institutional factors such as inadequate pre-service training and violations of laws and regulations. (2) The risk hierarchy follows a pyramid-shaped transmission path, with risks propagating from the root layer to the surface in successive layers. Open airspace serves as an accelerator, transforming environmental disturbances into institutional defects, which in turn lead to technical failures. (3) The dependency attributes of each factor indicate the priority order for risk prevention and control: management leverage points serve as the strategic control core, the environment-technology interaction network is central to joint prevention, standardized processes solidify basic operations, and systemic risk levels are reduced.

Vehicles comprise several critical systems, including the braking, steering, transmission, and suspension systems, which operate in concert to ensure safe and efficient movement. Research has established that vehicle malfunctions, particularly in the braking system, contribute significantly to road accidents, with technical failures accounting for approximately 15% of crashes and brake system failures responsible for 17.4% of these incidents. In light of this, an investigation was conducted to identify the factors that influence the braking coefficient and the variability of braking force in vehicle service brakes. A total of 1,018 vehicles were involved in the study, with results indicating that variables such as vehicle production year, category, place of registration, engine power and displacement, gross and curb weight, and payload significantly affect the braking coefficient. Furthermore, the analysis revealed that factors such as vehicle production year, category, registration location, gross and curb weight, and payload are prominent in determining the braking force variability. Neural network analysis was employed to further assess these influential factors, highlighting that the year of manufacture, place of registration, and vehicle payload are particularly influential in predicting both compliance with minimum braking coefficient requirements and variations in braking force. The findings underscore the importance of these factors in the development of more precise models for vehicle brake performance, with potential implications for safety standards and regulatory frameworks.

The integration of edge computing for real-time data processing in autonomous systems has been identified as a promising solution to mitigate the performance bottlenecks and latency associated with traditional cloud-based models. Autonomous systems, including vehicles, drones, and robotics, rely heavily on quick data analysis to make timely decisions. However, cloud computing, with its inherent data transmission delays, hinders the responsiveness and efficiency of these systems. To address these challenges, edge computing is proposed as a means to process data locally, at the point of use, thus enabling faster decision-making processes and reducing data transfer overhead. This approach leverages distributed machine learning for decision-making and dynamic resource allocation to balance computational tasks between edge and cloud resources. Through extensive experimentation, it has been demonstrated that the edge computing paradigm can reduce latency by up to 65%, offering substantial improvements in both energy efficiency and data processing speed when compared to traditional cloud-based methods. Furthermore, the proposed system outperforms existing alternatives in terms of computational speed, reliability, and energy consumption. The introduction of an Edge Computing Decision Model (ECDM) and a Dynamic Resource Allocation Algorithm (DRAA) is shown to optimize system performance by balancing computational demands between local edge nodes and remote cloud servers. These innovations enable autonomous systems to function more effectively and efficiently, even in resource-constrained environments. This study highlights the importance of integrating edge computing into autonomous system architectures to meet the growing demand for low-latency, high-performance applications. The potential of edge computing to significantly enhance the reliability and operational capacity of autonomous systems has been established, paving the way for more reliable and scalable solutions in real-time environments.

Real-time traffic sign recognition (TSR) plays a crucial role in intelligent transportation systems (ITS) and autonomous driving technologies. It enhances road safety, ensures efficient traffic rule enforcement, and supports the seamless operation of both autonomous and driver-assist systems. This paper proposes a hybrid TSR model that integrates mathematical morphology, edge detection, and fuzzy logic to accurately identify and classify traffic signs across diverse environmental conditions. The preprocessing stage applies contrast enhancement and Gaussian filtering to improve the visibility of key features. Next, shape- and color-based segmentation using mathematical morphology extracts regions of interest that are likely to contain traffic signs. These regions are then analyzed using a fuzzy inference system (FIS) that evaluates features such as color intensity, geometric shape ratios, and edge sharpness. The fuzzy system handles the inherent ambiguity in visual patterns, enabling robust decision-making. The entire model is developed in MATLAB R2015a, ensuring both computational efficiency and real-time performance. The integration of classical mathematical techniques with fuzzy reasoning allows the system to maintain high accuracy and reliability across a wide variety of traffic scenes. The proposed approach demonstrates significant potential for practical deployment in ITS applications, including smart vehicles and automated road safety systems.

Accurate and consistent road boundary detection remains a fundamental requirement in autonomous driving, traffic surveillance, and intelligent transportation systems, particularly under diverse lighting and environmental conditions. To address the limitations of classical edge detectors in complex outdoor scenarios, a novel multi-channel edge detection framework is proposed, termed the Multi-Channel Functional Gradient–Entropy (MC-FGE) model. This model has been specifically designed for colour road imagery and incorporates a mathematically principled architecture to enhance structural clarity and semantic relevance. The initial phase involves channel-wise normalization of RGB data, followed by the computation of a fused gradient magnitude that preserves edge information across heterogeneous spectral distributions. Two original mathematical constructs are introduced: the Spectral Curvature Function (SCF), which quantifies local geometric sharpness by leveraging first- and second-order differential operators while exhibiting resilience to noise; and the Colour Entropy Potential Function, which captures local texture complexity and intensity-driven irregularity through entropy analysis of chromatic distributions. These functions are combined into a unified Functional Edge Strength Map (FESM), designed to emphasize semantically meaningful road-related boundaries while suppressing irrelevant background textures. A central innovation is the Log-Root Adaptive Thresholding Function (LRATF), which adaptively modulates threshold sensitivity by integrating curvature and entropy cues in a logarithmic-root formulation, thereby improving robustness to illumination variability, occlusions, and shadow interference. The final binary edge map is derived through precision thresholding of the FESM and refined using morphological post-processing to ensure topological continuity and suppress artefactual edge fragments. Quantitative and qualitative evaluations conducted across varied outdoor datasets demonstrate that the MC-FGE model consistently outperforms conventional edge detectors such as Canny, Sobel, and Laplacian of Gaussian, particularly in scenarios involving texture-rich road surfaces, poor lighting, and partial occlusion. The model not only exhibits enhanced detection accuracy and edge coherence but also offers improved interpretability of road features, contributing both a rigorous theoretical foundation and a scalable computational framework for adaptive edge-based scene understanding in road environments.

A comprehensive bibliometric analysis was conducted to evaluate the evolution, thematic structure, and emerging trends in autonomous vehicle (AV) research. Scientific literature published up to 3 January 2025 was retrieved from the Web of Science (WoS), resulting in a corpus of 11,069 publications spanning 60 countries. Using VOSviewer software, a detailed examination was performed to map the intellectual structure of the field, including co-authorship patterns, citation networks, keyword co-occurrence, and institutional contributions. The findings revealed a marked increase in the volume of AV-related publications over time, indicating growing scholarly interest and investment in the domain. A total of 157 distinct scientific disciplines were identified, underscoring the inherently multidisciplinary nature of AV research, which encompasses fields such as computer science, robotics, transportation engineering, artificial intelligence, and socio-economic policy. The most prolific countries, institutions, and authors were visualised through citation and collaboration networks, revealing key contributors and international linkages. Particular emphasis was placed on the use of reinforcement learning and other machine learning methodologies in AV development, as reflected by keyword trends and thematic clustering. Additionally, attention was given to the broader socio-economic and managerial dimensions of AV adoption, including market dynamics, regulatory frameworks, and public acceptance. This analysis provides a rigorous and systematic overview of the current state of AV research and highlights potential avenues for future exploration. By synthesising large-scale bibliometric data, this study offers valuable insights for academics, policymakers, and industry stakeholders engaged in the evolving landscape of autonomous transportation systems.

Accurate detection of road surface potholes remains a persistent challenge due to environmental variability, inconsistent illumination, noise interference, and the complexity of road textures. Conventional detection methods frequently suffer from reduced performance when exposed to low-quality or noisy imagery, resulting in unreliable or delayed identification. To address these limitations, a robust and optimized image processing framework has been developed for real-time pothole detection under uncertain environmental conditions. The proposed approach employs a combination of advanced contrast enhancement techniques and adaptive convolutional processing to strengthen feature discrimination across heterogeneous road surfaces. To further improve detection reliability, a self-adaptive fuzzy refinement mechanism has been introduced, effectively delineating ambiguous or degraded regions often overlooked by deterministic methods. An energy-based functional is applied to model spatial and intensity gradients, enabling more precise localization of structural discontinuities indicative of pothole boundaries. The framework also incorporates computational optimization strategies to enhance processing speed without compromising accuracy, rendering it suitable for deployment in real-time autonomous or semi-autonomous road inspection systems. Thresholding and mask extraction operations have been systematically integrated to achieve accurate segmentation of pothole regions, even in the presence of substantial visual noise or occlusions. Experimental validations on benchmark datasets and real-world road imagery have demonstrated that the proposed method consistently outperforms existing state-of-the-art techniques with regard to detection accuracy, robustness to environmental disturbances, and computational efficiency. This approach presents a scalable and practical solution for intelligent transportation systems and automated infrastructure monitoring, contributing to improved road safety, timely maintenance, and cost-effective asset management.

Accurate detection of road surface anomalies remains a fundamental challenge in ensuring vehicular safety, particularly within the domain of intelligent transportation systems and autonomous driving technologies. Among such anomalies, crash stones—defined as irregular, protruding, and often unstructured fragments on the road—pose considerable risks due to their heterogeneous morphologies and unpredictable spatial distributions. In this study, a novel mathematical model is proposed, formulated through a functional energy minimization framework tailored specifically for the detection and segmentation of crash stones. The model incorporates three principal components: geometric edge energy to emphasize structural discontinuities, local variance descriptors to capture micro-textural heterogeneity, and fuzzy texture irregularity measures designed to quantify non-uniform surface patterns. These components are integrated into a unified total energy functional, which, when minimized, facilitates the precise localization of obstacle regions under diverse illumination, weather, and pavement conditions. Final detection is achieved through adaptive thresholding informed by fuzzy logic-based classification, enabling robust performance in scenarios with high noise or low contrast. Unlike deep learning-based methods, the proposed approach is fully interpretable, non-reliant on extensive annotated datasets, and computationally efficient, making it well-suited for real-time applications in resource-constrained environments. Experimental validations demonstrate high detection accuracy across varied real-world datasets, substantiating the model's generalizability and resilience. The framework contributes a mathematically rigorous, scalable, and explainable solution to the enduring problem of small obstacle detection, with direct implications for the enhancement of road safety in next-generation transportation systems.

Accurate identification of concrete surfaces on roadways is critical for the advancement of autonomous navigation systems and the effective monitoring of transportation infrastructure. Nevertheless, the inherently heterogeneous texture of concrete, in conjunction with environmental variables such as lighting fluctuations and surface degradation, continues to impede precise surface segmentation. To address these challenges, a novel framework has been developed that integrates Fuzzy Topological Entropy (FTE) with Multiscale Laplacian Structural Dissimilarity (MLSD) for the robust delineation of concrete regions in road imagery. Within this framework, FTE is employed to model uncertainty and spatial ambiguity through a continuous fuzzy membership function, thereby capturing the nuanced transitions between concrete and non-concrete domains. Concurrently, MLSD is utilised to quantify multiscale structural irregularities by leveraging Laplacian-based texture dissimilarity, enhancing sensitivity to surface roughness and material inconsistencies. These complementary components are embedded within a unified energy functional, the minimisation of which is conducted via an iterative optimisation strategy that avoids the need for extensive training datasets or prior scene annotations. The proposed methodology demonstrates strong resilience across a variety of environmental conditions, including shadows, glare, occlusions, and physical wear. Superior performance is observed particularly in complex or degraded urban settings, where conventional segmentation models often fail. Owing to its non-parametric nature and computational efficiency, the approach is well-suited for real-time deployment in autonomous vehicle systems, smart city infrastructure, and road condition assessment platforms. By facilitating reliable and scalable surface segmentation without reliance on deep learning architectures or exhaustive manual labelling, this work offers a significant advancement toward generalisable and interpretable road surface analysis technologies.

The efficient classification of transport vehicles is critical to the optimization of modern transportation systems, yet significant challenges persist, particularly in distinguishing Heavy Transport Vehicles (HTVs) from Light Transport Vehicles (LTVs). These challenges arise due to considerable variations in vehicle size, shape, orientation, and external factors such as camera perspective, lighting conditions, and occlusions. In this study, a novel classification framework is proposed, integrating geometric feature extraction with a soft computing approach based on fuzzy logic. Key geometric attributes, including bounding box length, width, area, and aspect ratio, are extracted through image processing techniques. Initial classification is performed via threshold-based rules to eliminate non-HTV instances using predefined feature thresholds. To address uncertainties inherent in real-world surveillance conditions, fuzzy logic inference is subsequently applied, enabling flexible and robust decision-making in the presence of imprecise or noisy data. This hybrid methodology, combining deterministic thresholding and soft computing principles, enhances classification reliability and adaptability under diverse environmental and operational conditions. Extensive real-world experiments have been conducted to validate the proposed framework, demonstrating superior performance in terms of accuracy, robustness, and computational efficiency when compared with conventional classification methods. The results underscore the potential of the framework for deployment in intelligent traffic monitoring systems where precise vehicle categorization is essential for traffic management, infrastructure planning, and safety enforcement.

Automated detection of vehicle dents remains a challenging task due to variability in lighting conditions, surface textures, and the presence of minor deformations that may mimic actual dents. This paper presents a novel hybrid framework that integrates color deviation analysis, fuzzy classification, and the Structural Similarity Index (SSI) to enhance detection robustness and accuracy. The proposed model employs an adaptive bounding box generation technique, optimized via morphological operations, for precise dent localization. A newly introduced Color Difference Metric (CDM), computed in the Hue, Saturation, and Value (HSV) color space, quantifies subtle color deviations induced by dents, improving the system’s sensitivity to minor deformations. Furthermore, a hybrid classification mechanism—merging step-function classification with fuzzy membership functions—ensures smoother transitions between dent severity levels, mitigating the risks of hard thresholding. SSI serves as a structural integrity validator, helping to differentiate true dents from surface irregularities and lighting artifacts. A Dent Confidence Score is computed as a weighted aggregation of the step-function output, fuzzy confidence levels, and SSI response, effectively balancing sensitivity and specificity. Dents are categorized into three interpretable classes: No Dent, Possible Dent, and Confirmed Dent. Evaluation on real-world datasets—encompassing diverse lighting conditions, vehicle colors, and camera angles—demonstrates the model’s superior performance. Compared to traditional approaches, our method significantly improves key metrics such as Intersection over Union (IoU), Dice Coefficient, Precision, Recall, and F1-Score, underscoring its applicability in real-world automated dent detection systems.

The mitigation of road traffic accidents remains a critical global challenge, particularly in regions where cultural norms and behavioral risk factors significantly influence driving practices. This study employs a hybrid Multi-Criteria Decision-Making (MCDM) approach, integrating Grey Theory, the Full Consistency Method (FUCOM), and the Evaluation based on Distance from Average Solution (EDAS), to systematically assess four strategic interventions: Infrastructure Improvements, Educational Programs, Policy Amendments, and Technology Integration. These strategies are evaluated based on a set of criteria that encompass attitudes toward speeding, perceptions of traffic laws, the use of safety equipment, and the prevalence of high-risk driving behaviors. The findings indicate that while Infrastructure Improvements and Technology Integration enhance the physical and technological dimensions of road safety, Educational Programs and Policy Amendments play an indispensable role in shaping driver behavior and reinforcing compliance with traffic regulations. The necessity of a comprehensive and integrated strategy that leverages both technological advancements and behavioral interventions is underscored, ensuring a holistic and sustainable reduction in traffic-related fatalities and injuries. The outcomes of this study provide valuable insights for policymakers and road safety authorities, offering a structured framework for the prioritization and implementation of road safety measures tailored to the socio-cultural and behavioral dynamics of Libya.

Accurate traffic prediction is essential for optimizing urban mobility and mitigating congestion. Traditional deep learning models, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), struggle to capture complex spatiotemporal dependencies and dynamic traffic variations across urban networks. To address these challenges, this study introduces DSTGN-ExpertNet, a novel Deep Spatio-Temporal Graph Neural Network (DSTGNN) framework that integrates Graph Neural Networks (GNNs) for spatial modeling and advanced deep learning techniques for temporal dynamics. The framework employs a Mixture of Experts (MoE) approach, where specialized expert models are dynamically assigned to distinct traffic patterns through a gating network, optimizing both prediction accuracy and interpretability. The proposed model is evaluated on large-scale real-world traffic datasets from Beijing and New York, demonstrating superior performance over conventional methods, including Spatio-Temporal Graph Convolutional Networks (ST-GCN) and attention-based models. With a mean absolute error (MAE) of 1.97 on the BikeNYC dataset and 9.70 on the TaxiBJ dataset, DSTGN-ExpertNet achieves state-of-the-art accuracy. These findings highlight the potential of GNN-based frameworks in revolutionizing traffic forecasting and intelligent transportation systems (ITS).