Mobile environmental sensing is a promising strategy for scalable and cost-effective urban monitoring. However, this approach introduces challenges due to the irregular spatial and temporal distribution of measurements, leading to sparse and asynchronous datasets. Traditional interpolation struggles with such sparsity and temporal misalignment. To address these issues, we propose an Edge-Fog-Cloud architecture that gathers data at the Edge and performs hierarchical spatio-temporal interpolation. By leveraging real-world mobility patterns, our framework adapts to varying data densities and temporal gaps, enhancing the accuracy of environmental monitoring in urban settings. Our approach combines classical spatial interpolation methods (e.g., RBF, IDW, Kriging) with temporal-aware strategies that consider the dynamic nature of mobile sensing coverage. Experimental results, conducted using instrumented vehicles and acquired environmental data in an urban setting, demonstrate that time-aware interpolation significantly improves estimation in unsampled areas, reducing the median error and variability.
Nearly five years have passed since the commercial launch of $\mathbf{5 G}$ services. Although the Sub-6 GHz band remains the primary frequency range for base station deployment, the adoption of millimeter wave (mmWave) technology, which was introduced for the first time in mobile communication, has faced challenges owing to its limited coverage and susceptibility to blockage. Consequently, the growth in the number of mmWave base stations stagnated. Addressing the coverage limitations of mmWave in 5G is an essential step toward realizing 6G, where even higher-frequency bands are expected to be utilized in the future. Among the various coverage enhancement technologies across layers 1 to 3, analog repeaters and passive reflectors in layer 1 have gained attention owing to their cost-effectiveness and signal amplification capabilities. In particular, NetworkControlled Repeater (NCR), a type of analog repeater, is standardized in 3GPP Release 18. However, existing research remains largely theoretical, and practical evaluations are limited. This paper presents the development of a prototype NCR with analog beamforming functionality and reports the experimental results for a single hop. The results demonstrate NCR's potential for coverage extension, while highlighting future challenges such as multi-hop operation and orchestrator control.
This paper proposes a Federated Edge AI framework with Adaptive Zero-Trust Security designed for resource constrained IoT networks used mainly in smart cities applications. The framework harnesses and deploys lightweight machine learning models, real-time anomaly detection capabilities, and dynamic trust scoring. Each IoT device can also deploy differential privacy and lightweight encryption as optional features. Dynamic trust scoring enables the federated learning process to identify and mitigate malicious clients in adversarial environments, while trust-weighted aggregation prioritizes reliable model updates. The framework is simulated under both IID and non-IID datasets from IoT-23 dataset, with up to $\mathbf{4 0 \%}$ malicious clients to ensure there was a significant amount of anomalous client behavior, control groups and dependent groups to make comparisons and assess precision, false positive-negative, communication and computational load. The lightweight framework is an applicable idea to real-world like smart cities applications, as it is scalable and highly privacy preserving. Use Case Examples include intelligent traffic management, smart energy grids, and security of UAV swarms - where IoT devices must operate in a secure fashion, under strict latencies, power, and resource limitations.
Understanding pedestrian visual experiences is critical for human-centered urban design, yet methods for analyzing first-person perspectives from actual movement data remain limited. This study proposes a novel methodology to reconstruct and analyze these perspectives by integrating LiDAR-collected pedestrian trajectories with a semantically-encoded 3D city model. Our approach simulates what pedestrians see along their movement paths, enabling quantitative, comparative analysis of their visual trajectories. An experiment in Tokyo captured and classified pedestrian flows. A case study analysis of representative trajectories demonstrates the framework's ability to reveal distinct visual signatures for different pedestrian routes, such as a direct departure from a building versus an ambulatory traversal through a green space. Despite using basic semantic categories, the proposed method offers an extensible foundation for evidencebased urban design by transforming subjective visual experiences into measurable data.
Anomaly detection in urban IoT networks is essential for the resilience and security of smart cities. However, the privacy of sensor data remains a critical concern. In this paper, we present a machine learning-based anomaly detection approach that operates entirely on homomorphically encrypted data using Concrete ML. To enable practical deployment under the computational constraints of Fully Homomorphic Encryption (FHE), we investigate the impact of quantization-aware training (QAT) and unstructured pruning, two established techniques for reducing model complexity. We evaluate different combinations of quantization levels and pruning degrees with respect to classification performance and resource efficiency. Our results indicate that competitive accuracy can be retained even under aggressive model compression, enabling efficient and privacypreserving inference.
Accurate forecasting of micromobility demand is critical for urban planning and the sustainable integration of shared transportation modes. This study develops a machine learning model on the demand of e-scooter and bikes using one year dataset from Paris. It also explores the transferability of micromobility demand model from Paris to a smaller urban environment with lower travel demand - Patras. Using Random Forest regression trained on hourly counts from Paris and incorporating temporal and weather-related features, the models effectively captured directional flow patterns and daily demand cycles. When applied to Patras, the models displayed low predictive accuracy, which was further enhanced through a correction stage using Generalized Additive Models (GAMs) based on local covariates. The two-step modeling approach significantly reduced prediction error, demonstrating that models developed in developed urban areas can be successfully adapted to cities with limited data. These findings support scalable approaches to micromobility demand forecasting across diverse urban contexts.
Digital Twins (DTs) are quickly becoming popular in multiple application domains, from industrial production environments to cultural heritage sites and smart cities. In this landscape, DTs for educational buildings offer a promising approach towards achieving energy efficiency, circularity and sustainability, and an important new smart city application domain, among other aspects. While there have been recent examples of DTs for university campuses, examples for primary and secondary school buildings are few and far between. In this work, we present our implementation of a DT for two secondary education buildings in Greece, involving IoT infrastructure, 3D digitization, data analytics, energy modeling, 3D annotation and segmentation, while discussing its potential application within several scenarios. We also discuss a number of the aspects related to educational building DTs. Overall, such DTs can offer several new application opportunities to the smart city community.
This paper proposes an effective energy management strategy for residential microgrid systems, specifically integrating battery energy storage systems and photovoltaic (PV) arrays. Given the increasing adoption of solar PV systems in both residential and industrial sectors, often operating in off-grid or hybrid configurations, efficient energy management is crucial. The presented strategy optimizes energy exchange among interconnected households by continuously monitoring battery state-of-charge (SOC), prioritizing critical loads, and intelligently utilizing surplus PV energy for secondary applications. This approach introduces three distinct operational modes, dynamically adapting to varying power generation and load demands. The proposed methodology ensures efficient energy distribution, prolongs battery lifespan, and significantly reduces reliance on conventional backup generators. Simulation results validate the effectiveness of the developed strategy in maintaining system stability while concurrently minimizing operational costs.
Traditional UAV authentication mechanisms rely on centralized architectures, which introduce risk of single points of failure and computationally intensive processes. To address these limitations, we propose a blockchain-based distributed authentication and on-boarding framework that decentralizes credential control, enabling user-centric management. This approach is particularly critical for UAV swarm environment, where high mobility disrupts communication stability and degrades authentication performance. Our framework integrates Hyperledger Identus as a cloud agent, KeyCloak for authorization, and NVIDIA Jetson for edge-device operations, establishing a resilient attributebased authentication system. By leveraging selective attribute disclosure without exposing raw data, the solution aligns with SelfSovereign Identity principles and privacy-by-design mandates. This eliminates dependence on central authorities while ensuring compliance with evolving security standards.
Open data plays a crucial role in fostering urban innovation and transformation. However, challenges persist due to a lack of metadata standardization, diverse data schemas, difficulties in discovering relevant datasets, and issues related to multilingualism. This study investigates the landscape of open datasets related to smart cities in Europe, based on data retrieved from data.europa.eu. We employ SPARQL queries to extract metadata and apply thematic mapping based on Giffinger's Smart City model alongside keyword-based classification. Subsequently, datasets are identified and categorized across six smart city dimensions. Our findings reveal the distribution patterns, thematic gaps, multilingual barriers, and emerging opportunities for advancing data-driven urban innovation through open data.
The Internet of Things (IoT) paradigm envisions a pervasive “network of networks,” facilitating effortless communication and data exchange across diverse domains. Nonetheless, the rise of IoT brings forth considerable challenges, especially in the realm of Secure Group Communications (SGCs), requiring strong security measures such as cryptographic techniques and encryption protocols. Since it is essential to balance security and efficiency in IoT scenarios, this paper examines encryption systems and security challenges possibly affecting IoT networks, then presenting optimization techniques including data aggregation, delta coding, compression, and selective forwarding. Then, through OMNeT++ simulations, the effects of these strategies on relevant performance metrics (such as latency, energy consumption, and data size) are described. The obtained results indicate that incorporating these optimization techniques greatly improves communication efficiency, leading to 40% latency decrease, 25% energy consumption decrease, and $\mathbf{5 0 \%}$ data size reduction. Thus, these findings highlight the effectiveness of optimization techniques in enhancing SGCs within IoT ecosystems, opening to future research activities on adaptive security mechanisms and lightweight cryptographic approaches.
Vehicular Cloud Computing (VCC) leverages the underutilised resources of vehicles, such as computing power, storage, and communication capabilities, to form a dynamic and decentralised cloud infrastructure. By bringing computation closer to data sources, VCC reduces latency and enhances context-awareness, enabling a wide range of applications, from intelligent transportation to safety-critical services. However, VCC faces challenges related to the high mobility of vehicles, frequent changes in network topology, heterogeneous entities, and limited interoperability and scalability. While previous research has addressed architectures, communication protocols, and resource management frameworks, key issues remain, particularly in integration, distribution, interoperability, and adaptability, highlighting the need for middleware systems. This paper proposes Midd4VC, a lightweight middleware designed to manage vehicular clouds by mediating communication between client applications and VCC entities, e.g., vehicles. Midd4VC abstracts communication protocols, supports flexible job assignment and resource sharing, and ensures resilience through reconnection mechanisms and parallel job execution. The current implementation employs MQTT as a transport protocol, while the architecture remains extensible to other messaging frameworks.
The increasing need for efficient emergency response in urban areas has led to the integration of Geographic Information Systems (GIS) and data-driven approaches to optimize resource allocation and reduce response times. This paper presents a new methodology for assessing emergency response in urban areas, taking historical ambulance call data as input. By utilizing geospatial analytics, we convert raw emergency data into enriched geospatial information, enabling the identification of spatial and temporal patterns in emergency incidents. For that, through the integration of tools such as PostGIS, Nominatim, and GeoPandas, it is possible to visualize hotspots and calculate Estimated Time of Arrival (ETA) across different regions, providing actionable insights for optimizing emergency services. This proposed approach was evaluated in the Federal District of Brazil, demonstrating how emergency mapping-planning can contribute to urban management and mobility improvements.
This survey presents a system-level analysis of big data analytics in smart cities, bridging data sources, analytical techniques, and deployment infrastructures. It categorizes analytics into stream, batch, predictive, semantic, and explainable methods and maps them to edge, fog, and cloud layers via distributed orchestration and standardized interfaces. Real-world applications across mobility, environment, safety, and citizen services are examined, emphasizing execution-aware design, interoperability, and compliance with ethical and regulatory frameworks of the technology. This survey highlights the integration gaps in previous studies and outlines future directions for transparent, scalable, and semantically enriched urban intelligence systems.
Modern smart cities now place top priority on keeping Critical Infrastructure Systems (CIS) like power grids, water networks, transportation systems, and telecommunications both resilient and highly efficient. Recent advances in Digital Twin (DT) technology have made it an attractive tool for managing these systems by generating dynamic, data-driven virtual models of cyber-physical assets and their interconnections. Despite widespread adoption of DTs across many industries, there is no risk-based, scenario-driven DT framework readily available to researchers, smart city operators, and CIS managers for bolstering urban resilience. To fill this gap, we present the Cyprus Digital Twin (CyDT), a one-of-a-kind digital platform designed to manage and optimize interdependent CIS. We detail CyDT's modular architecture outlining its principal components and spotlighting its novel features and capabilities and demonstrate its adaptability and effectiveness in real-life use cases that conduct risk-based resilience analyses of power and water systems.
This paper presents a robust and generalizable vision-based system for detecting, tracking, and estimating the depth of non-cooperative unmanned aerial vehicles (UAVs) in air-to-air scenarios. Advances from prior work in monocular depth estimation, robust object detection, and domain generalization to form a comprehensive real-time pipeline are integrated. The proposed system enhances a state-of-the-art detecting and tracking architecture by integrating a lightweight monocular depth estimator and incorporating domain agnostic training strategies using uncertainty-guided adversarial augmentations. Furthermore, robustness under domain shifts using a dataset enhanced with environmental and sensor corruptions such as fog, rain, blur, and noise to test data is assessed. The system was able to perform in real time in low Size Weight and Power consumption edge devices and accepting inputs from multiple camera types (i.e. RGB, low-light).
Higher education institutions face increasing challenges, necessitating advanced technological solutions. This systematic review analyzed 35 articles to map the application of Multi-Agent Systems (MAS), Agent-Based Modeling (ABM), and simulation in higher education. Findings show these AI techniques are widely used for simulating complex systems, curriculum design, academic planning, and personalized learning, with simulation also supporting virtual environments. Other AI and Machine Learning (ML) techniques, such as Learning Analytics and Reinforcement Learning, are also applied to enhance student performance, engagement, and self-regulation. Critical gaps identified include the limited adoption of emerging technologies like Generative AI and Large Language Models (LLMs), reliance on small and homogeneous samples, and a general lack of complex implementations. This review highlights these gaps, offering recommendations to guide researchers, educators, and policymakers in advancing AI integration towards smart higher education strategies.
Automotive electronic control units (ECUs) typically offer less than 32 kB of on-chip flash memory, presenting a significant challenge for deploying deep-learning models for Controller Area Network (CAN bus) intrusion detection systems (IDS). While existing deep learning methods demonstrate high efficiency in threat detection, their complex architectures often render them too resource-intensive for the stringent constraints of automotive ECUs. This paper addresses this critical gap by proposing an efficient methodology for developing an ultra-lightweight CNN-based IDS. Our approach centers on the systematic application and validation of an information-dense RGB tensor encoding for 16-frame CAN windows (adapted from prior work [1]), which effectively captures crucial temporal, identifier, and data-length characteristics. This encoding is coupled with CANET-33K, a novel and compact 32.9k-parameter CNN architecture specifically tailored for these RGB representations, and a one-shot hybrid compression strategy that combines structured pruning with 8-bit post-training quantization (PTQ). The resulting INT8 model achieves a 99.73% overall accuracy and a 0.40% false-negative rate across four common attack types (DoS, Fuzzy, Gear, RPM). Notably, the compressed model is over 420 times smaller than a 7 MB Inception-ResNet baseline and 16 times smaller than recent specialized RGB-CNNs, while maintaining comparable or superior detection performance. These results demonstrate a viable path towards deploying high-performance IDS on resource-constrained automotive hardware.
Three decades after the Gulf War created one of history's worst petroleum disasters, Kuwait's Burgan Field remains severely contaminated, offering unique insights for industrial monitoring worldwide. We developed an AI framework combining Sentinel-2A satellite imagery with machine learning to detect petroleum contamination across $195 \text{km}^{2}$ of desert landscape. Our approach analyzed over 200 multispectral images against 180 field samples containing up to $129,726 \text{mg} / \text{kg}$ Total Petroleum Hydrocarbons. The system identified 18-20 contamination sources maintaining maximum sensor values after thirty years, revealing how arid climates preserve industrial disasters indefinitely. We mapped 463 contamination hotspots and eight migration corridors threatening millions, with extraordinary spatial clustering (Moran's $I=0.9636$) showing contamination coalescence into mega-structures. Most critically, the framework provides 15-30 days advance warning of contamination spread, enabling preventive action. By targeting intervention zones representing just 4.9 % of contaminated areas, remediation costs could drop significantly. This matters because developing nations facing rapid industrialization cannot afford traditional monitoring. The framework demonstrates direct transferability to urban industrial contexts where petroleum facilities, refineries, and storage tanks pose similar contamination threats to dense populations. Our methods work with free satellite data and opensource algorithms, democratizing environmental surveillance. This transferability provides municipalities worldwide with operational capabilities for proactive contamination management.
The growing complexity of federated data ecosystems demands robust mechanisms to enforce data usage contracts and detect unauthorized data reuse. In this paper, we present the design and evaluation of the Off-Platform Contract Inspector, a core component of the PISTIS framework for trusted data sharing. The Inspector employs a hybrid similarity detection pipeline, combining path-value structural comparisons, field-aware semantic analysis, and embedding-based techniques (e.g., SBERT, E5) to detect exact matches, near duplicates, and partial dataset reuse. To support interpretability and contract governance, it generates explainable similarity reports suitable for manual review. We validate the system using a diverse set of real-world and synthetically altered datasets from the Mobility, Energy, and Automotive domains. The evaluation demonstrates high recall (92%) and precision (89%) in identifying modified datasets, along with scalable performance on modern hardware. Our findings confirm the Inspector's effectiveness for contract enforcement in decentralized data spaces, bridging technical detection methods with legal compliance requirements.
