Sky’s-Eye Perspective: A Multidimensional Review of UAV Applications in Highway Systems

Abstract

Unmanned aerial vehicles (UAVs), commonly known as drones, have emerged as promising solutions to overcome the shortcomings of traditional highway-monitoring approaches. UAVs have been used extensively for highway traffic monitoring, infrastructure inspection, safety analysis, and environmental management. This review summarizes the latest applications, contributions, and challenges of UAV technology in highway systems, highlighting their transformative impacts on traffic monitoring, infrastructure inspection, and safety assessment. Several UAV-based highway traffic datasets significantly improve research in traffic behavior analysis and automated driving system validation. The integration of UAVs with advanced technologies, such as artificial intelligence (AI), the Internet of Things (IoT), and 5G, further enhances their capabilities, enabling enhanced real-time analytics and better decision-making support. Addressing ethical, regulatory, and social implications through transparent governance and privacy-preserving technologies is essential for sustainable deployment.

1. Introduction

1.1. Background

Effective highway management and traffic monitoring are critical to modern societies. The continuous growth in traffic volume and congestion has made traffic supervision a major challenge worldwide [1]. Efficient monitoring is essential to ensure smooth traffic flow, improve road safety, and reduce the social costs of congestion and accidents. Traditionally, highways have been monitored using fixed location sensors (e.g., inductive loop detectors) and closed-circuit cameras, as well as manual observations. However, these conventional methods have significant limitations. Deploying enough loop detectors or fixed cameras to cover a whole highway network is prohibitively expensive and labor intensive, often leaving large portions of the network unmonitored [2]. Fixed sensors provide only point-based data with limited spatial coverage, and even GPS/probe vehicle data suffer from low penetration rates and accuracy problems in congested areas [3]. In practice, it is not feasible to blanket the entire highway system with static sensors, leading to ‘blind spots’, where traffic conditions are unobserved [4]. These gaps in coverage and data quality hinder the ability of traffic managers to detect incidents or analyze flow dynamics in real time.

Unmanned Aerial Vehicles (UAVs), commonly known as drones, have emerged as a promising solution to overcome the shortcomings of traditional highway monitoring [5]. UAVs offer flexibility in deployment; they can be quickly dispatched to different locations or repositioned as conditions change, unlike fixed cameras bound to one view. They also provide wide-area coverage and unique viewpoints from the sky. The bird’s-eye perspective of a drone allows a broad field of view, capturing the spatial context of traffic that ground sensors often miss [6]. This aerial vantage point makes it possible to track vehicles across a larger stretch of road and to observe traffic patterns (e.g., queue formation and dissipation) in ways that fixed sensors cannot. In fact, the overhead view allows for more accurate extraction of vehicle trajectories and positions (especially lateral lane positions), improving data obtained from traditional ground-based methods [5]. Moreover, modern small UAVs are relatively inexpensive, non-intrusive to install, and capable of collecting high-resolution video data in both space and time [7]. The result is a rich traffic dataset that can be analyzed for speeds, densities, and even individual vehicle movements in real time. For example, a large-scale field experiment in Athens demonstrated that a coordinated drone swarm could continuously monitor traffic over a 1.3 km² urban area (about 100 intersections) and record nearly half a million vehicle trajectories, an unprecedented scale of data collection [3]. This illustrates how UAVs can provide highway operators with far more comprehensive coverage and situational awareness than stationary sensors. In summary, the integration of UAV technology into highway systems brings significant advantages, flexibility of use, wide-area surveillance, and a top-down view, which address many limitations of conventional monitoring approaches [8]. These benefits have driven growing interest in the use of drones as a part of the infrastructure of intelligent transportation systems, with researchers and agencies recognizing their potential to improve traffic management and safety [9]. UAV-related research has witnessed remarkable growth, highlighting popular research themes, such as deep learning, as illustrated by word cloud analysis (Figure 1a). Furthermore, to visually illustrate the momentum of this research hotspot, we first performed a preliminary bibliometric analysis using the Web of Science (WoS) Core Collection database. This database is widely recognized as an authoritative source covering high-impact scientific literature, making it suitable for observing the field’s macro-level dynamics. Figure 1b presents the results of this preliminary analysis.

1.2. Objectives

Given the potential of UAVs for various highway applications, a systematic review of recent applications is warranted to understand the state of the art and guide future research. The first objective of this review is to catalog and synthesize the various ways in which UAV technology has been applied in highway traffic research. During the past 10 years, studies have demonstrated the use of UAVs in several key domains of highway transportation. In particular, drones have been deployed for traffic flow monitoring and analysis, where aerial videos are processed to obtain traffic counts, speeds, densities, and even detailed vehicle trajectories for model calibration [5]. They have also been used in road safety applications, such as rapid incident response and accident investigation: UAV imagery can help to reconstruct crash scenes or identify risky driving behaviors and near misses from an overhead perspective [10]. Another growing application area is highway infrastructure inspection and management, in which UAVs carry cameras or sensors to inspect the conditions of roads, bridges, and other structures (for example, detecting pavement damage or bridge defects from the air) [11]. This study systematically reviews these application areas to identify what has been accomplished with UAVs on highways and how these approaches improve or complement traditional methods. The rigorous methodology followed for the selection of the literature, including the specific inclusion criteria, is detailed in Section 2 (Review Methodology).

The second objective is to distill the core research questions and technical focus areas that have emerged around UAV-based highway monitoring. For example, how to reliably detect and track vehicles in UAV video feeds and derive traffic parameters, such as flow, density, and speed [12]. This involves significant computer vision and machine-learning challenges, as vehicles must be recognized from varying altitudes and angles, and their movements tracked across frames. Recent work has concentrated on improving the accuracy of such algorithms (e.g., using deep learning for object detection) because effective vehicle detection/tracking is fundamental to all UAV traffic applications [13]. Another important question is how to optimize UAV operations for traffic monitoring: Researchers are examining flight path planning and multi-UAV coordination strategies to maximize area coverage and persistence. Although single drones can monitor a localized area, covering an extensive highway network or an entire city requires multiple drones working in tandem. Pioneering experiments, such as pNEUMA (with 10 drones in a city center), have shown the potential of multi-UAV systems, but they also highlight open questions about coordination protocols and data fusion from many drones [3]. Integrating UAVs into existing intelligent transportation systems is another focus area. For example, drones are used as airborne sensors that feed data to traffic management centers in real time or as communication relays to vehicles in areas lacking infrastructure [14]. Through this review, we identify these central research themes and the progress made on each, providing information on the current priorities in the field.

The third objective is to highlight the gaps and challenges in the literature and suggest directions for future research. Despite advances, several obstacles still hinder the wide-scale deployment of UAVs in highway systems. Operational constraints remain a key concern. Small drones have limited battery lives and flight ranges, restricting their endurance and the area they can cover per mission [15]. Safety is paramount, particularly for operations beyond the visual line of sight, due to risks to drones and people on the ground due to malfunctions or loss of control [16]. Furthermore, regulatory frameworks have not fully caught up with drone technology; airspace regulations and privacy concerns pose nontechnical barriers to the routine use of UAVs for traffic monitoring [17]. So far, these issues have kept most UAV traffic-monitoring efforts limited to pilot studies or controlled experiments [18]. By reviewing recent studies, we pinpoint specific shortcomings in existing research. For example, there is a need for more robust autonomous flight management, better data-processing techniques for crowded traffic scenes, and comprehensive evaluations of system performance under various weather or lighting conditions. We also note that the public acceptance of drones for highway monitoring (addressing concerns such as noise or privacy) will influence their future use [19]. Building on the identified gaps, this review suggests future research directions, such as developing extended-endurance UAV platforms or automated battery swapping to allow continuous surveillance, designing multi-drone systems with collaborative coverage and collision avoidance, enhancing real-time image-processing algorithms to handle high-density traffic, and formulating clear guidelines and policies for the safe integration of UAVs into traffic management. UAV applications in highway systems are still being developed [5]. Advancements in UAV hardware, computer vision, and traffic management integration promise to unlock significant new applications and benefits, ultimately reducing congestion and improving road safety [20].

While several reviews have documented the use of UAVs in transportation, there is a critical gap for a systematic and multidimensional synthesis that bridges the technological, practical, and regulatory dimensions specific to highway systems. Many existing works either focus on a narrow application range or provide a descriptive catalog of studies without a critical comparative analysis of the underlying technologies and their tradeoffs. This review directly addresses this gap and makes the following distinct contributions to the body of knowledge:

• Provides a Rigorous and Reproducible Synthesis: Unlike narrative reviews, this study is grounded in a systematic review methodology (SLR), ensuring a transparent, comprehensive, and reproducible baseline of the state-of-the-art literature. This methodological rigor provides a solid foundation for future research and meta-analyses;
• Delivers a Critical Comparative Analysis: Instead of merely listing technologies, this review offers a critical comparison of key enabling components, including UAV platforms, sensors, and data-processing algorithms. This analysis focuses on their respective strengths, limitations, and performance tradeoffs in the context of highway monitoring, providing actionable information to practitioners and researchers;
• Establishes a Holistic Multidimensional Framework: The paper organizes the complex and fragmented literature into a cohesive, multidimensional framework. This framework interconnects technical characteristics, diverse applications (from traffic monitoring to infrastructure inspection), sociotechnical challenges (including policy and ethics), and future research trajectories, providing a clear intellectual map of the field;
• Synthesizes Key Issues and Identifies Actionable Gaps: Through a novel synthesis table, this review maps specific applications to their key corresponding issues. This structured approach allows the identification of specific and actionable research gaps and provides a nuanced overview of the complex regulatory landscape, offering practical guidance for future scientific inquiry and real-world deployment.

By providing this structured and critical synthesis, this review serves as a foundational reference and a strategic guide for advancing research and development in UAV applications for highway systems.

1.3. Outline of the Review

The review is structured as follows. Section 2 details the systematic review methodology (SLR) used for this study, including the search strategy, inclusion/exclusion criteria, and the study selection process. Section 3 discusses the technical characteristics and advantages of UAV technologies, focusing on a comparative analysis of sensor types, data acquisition features, and platform-specific considerations, which lay the foundation for future applications. Section 4 comprehensively and critically reviews the practical applications of UAVs in highway systems in multiple dimensions, including infrastructure inspection, trajectory prediction, traffic safety analysis, and traffic flow monitoring. Section 5 critically addresses the existing technical, regulatory, and ethical challenges related to the deployment of UAVs and proposes potential solutions. Section 6 explores research trends and future directions, highlighting integration strategies with emerging technologies, such as AI, 5G, and IoT, and provides recommendations for future research paths. Finally, Section 7 concludes the review by summarizing the key findings, reiterating the contributions of the article, and highlighting the potential and necessity of UAV technology while providing strategic research recommendations. The general structure of this review is organized into seven main sections, as shown in Figure 2.

2. Review Methodology

To ensure a comprehensive, transparent, and unbiased synthesis of the literature, this study uses a systematic literature review methodology (SLR). Unlike traditional narrative reviews, an SLR uses a rigorous, predefined protocol to identify, select, and critically evaluate relevant research, thus improving the reliability and reproducibility of the findings [21]. The entire process was structured according to the statement of PRISMA 2020 (Proposed Reporting Items for Systematic Reviews and Meta-Analyses), which provides an evidence-based guideline for reporting systematic reviews [22,23]. This section details the search strategy, inclusion and exclusion criteria, and the multistage process of study selection and data extraction.

2.1. Search Strategy and Identification

A systematic search was conducted in four major academic databases to ensure a comprehensive coverage of the relevant literature: Web of Science (WoS), Scopus, IEEE Xplore, and the TRID (Transport Research International Documentation) database. This selection provides a robust mix of high-impact interdisciplinary research (WoS and Scopus), specialized engineering and computer science literature (IEEE Xplore), and focused transportation research (TRID).

The search query was constructed based on three core conceptual pillars of this review: (1) technology (UAVs), (2) domain (highway systems), and (3) the scope of application (monitoring and analysis). The final Boolean search string, adapted for the syntax of each database, was as follows:

(“UAV” OR “drone” OR “unmanned aerial vehicle” OR “unmanned aircraft system”) AND (“highway” OR “freeway” OR “roadway” OR “traffic” OR “transportation infrastructure”) AND (“monitoring” OR “inspection” OR “analysis” OR “safety” OR “management”).

The search was limited to the title, abstract, and keywords of the publications to maintain a high degree of relevance. The literature search was conducted in September 2025 and covered a 10-year period from January 2015 to April 2025, a time frame that captures the most significant advances in UAV technology and its application in transportation.

2.2. Inclusion and Exclusion Criteria

To meticulously screen the identified literature, a clear set of inclusion and exclusion criteria was established.

2.2.1. Inclusion Criteria

• Publication Type: Peer-reviewed journal articles and full-length conference papers;
• Language: Publications written in English;
• Research Focus: Studies must directly investigate the application of UAVs for monitoring, inspection, or analysis within highway systems;
• Content: The articles had to present empirical results, a novel methodology, a detailed case study, or a substantive technical framework.

2.2.2. Exclusion Criteria

• Publication Type: Gray literature (including books, book chapters, editorials, dissertations, technical reports, and news articles) and other review papers were excluded to focus on primary research;
• Research Focus: Studies focusing on military applications, indoor navigation, urban (non-highway) traffic management, or theoretical communication protocols without a clear highway application were excluded;
• Content: Conceptual papers lacking technical detail and studies in which UAVs were mentioned only in passing without being a central element of the research were omitted.

2.3. Study Selection and Data Extraction

The study selection followed a multistage screening process, which is visually summarized by the PRISMA 2020 flow diagram (see Figure 3).

• Identification: The initial database search yielded a total of 1555 records;
• Screening (Deduplication and Title/Abstract Review): These records were imported into Zotero, a reference management software package, where 517 duplicates were identified and removed. The remaining 1038 articles were then screened based on their titles and abstracts against the predefined criteria, leading to the exclusion of 728 articles that were clearly out of scope;
• Eligibility (Full-Text Review): The full texts of the remaining 310 articles were thoroughly evaluated for eligibility. During this stage, 178 articles were excluded for reasons such as an inappropriate study context (e.g., urban streets instead of highways), a lack of sufficient methodological detail, or not being primary research;
• Inclusion: This rigorous process resulted in the final selection of 132 studies that were included in the qualitative synthesis of this review.

For each of the included studies, key information was systematically extracted and cataloged in a structured spreadsheet. The extracted data included (a) authors and publication year, (b) specific highway applications (e.g., traffic flow estimation and pavement crack detection), (c) UAV platform and sensor specifications, (d) data analysis techniques and algorithms used, (e) key findings and contributions, and (f) reported limitations and challenges. This structured data extraction process formed the foundation for the multidimensional analysis presented in the following sections of this review.

2.4. Bibliometric Analysis

To substantiate the qualitative synthesis and provide a data-driven overview of the intellectual structure of the research field, a bibliometric analysis was conducted. Going beyond the preliminary trend analysis, we performed a keyword co-occurrence network analysis to identify the dominant thematic clusters and research hotspots within the included literature.

This analysis was performed on the final corpus of 132 selected studies. We extracted all the keywords supplied by the authors and the index from the articles. A minimum occurrence threshold of five was set for a keyword to be included in the analysis, which helps to filter out marginal topics and focus on the most significant research themes. The network was generated based on the co-occurrence data, where the connections between keywords represent their appearance together in the same publications.

The resulting keyword co-occurrence network is illustrated in Figure 4. The network map reveals three distinct and influential research clusters, distinguished by color. The first cluster (red) is centered on methodologies, dominated by keywords such as ‘deep learning’, ‘computer vision’, ‘object detection’, and ‘trajectory extraction’. This signifies a strong focus on advanced data-processing techniques as a foundational element of the research. The second cluster (green) represents the primary application domains, connecting keywords like ‘traffic monitoring’, ‘safety analysis’, ‘trajectory prediction’, and ‘conflict analysis’. The third, and final, cluster (blue) highlights the growing research trend in asset management, linking keywords such as ‘infrastructure inspection’, ‘bridge inspection’, ‘pavement’, and ‘damage detection’. The density of this network underscores the interdisciplinary nature of the field, where methodological advances are directly fueling a diverse range of highway-monitoring applications.

3. Technical Characteristics and Advantages of UAVs in Highway Systems

3.1. Technical Features

Sensors for UAV Highway Monitoring: Modern drones can carry a variety of sensors to monitor highways, each offering unique data. Optical cameras (RGB video cameras) are the most common, providing high-resolution visual imagery for traffic surveillance [24]. They capture detailed color images and videos, enabling vehicle detection, counting, and tracking. Thermal infrared cameras detect heat signatures and have been used in conjunction with optical cameras to recognize vehicles in low light or at night [25]. This allows the detection of warm vehicles on roads, even in darkness or through smoke/fog. LiDAR (Light Detection and Ranging) sensors emit laser pulses to create 3D point clouds of the road environment. LiDAR mounted with UAVs has been tested to map road surfaces and detect pavement distress (cracks, potholes, and rutting) with high precision [26]. By capturing precise 3D geometry, LiDAR can measure the infrastructure dimensions and deformation that cameras might miss. Radar systems (e.g., millimeter-wave or ground-penetrating radar) are less common on small UAVs due to size and power demands, but research is exploring their uses. For example, the integration of ground-penetrating radar (GPR) into drones has been proposed for non-intrusive pavement inspection, although practical demonstrations are still limited [27]. In summary, optical cameras currently dominate UAV traffic monitoring, whereas thermal cameras and LiDAR provide complementary capabilities for special conditions (nighttime and 3D mapping), as summarized in

Table 1. Comparison of UAV-mounted sensors for highway monitoring.

Sensor Type	Data Characteristics	Typical Applications	Strengths	Limitations
Optical Camera	High resolution	Detection	High clarity	Light conditions
	RGB images	Counting	Low cost	Occlusion
	Videos	Tracking	Versatile
Thermal Camera	Infrared imagery	Nighttime	Effective in darkness,	Lower resolution
	heat signatures	poor-visibility monitoring	fog and smoke	limited scene details
LiDAR	3D point cloud	Road infrastructure mapping	High spatial accuracy	Costly
	Laser-based ranging	Pavement inspection	Detailed 3D data	Complex data processing

.

Data Acquisition Capabilities: UAVs offer flexible data collection with high spatial and temporal resolutions, wide coverage, and on-demand mobility. Drones can capture video at high frame rates and resolutions (e.g., 4K at 20–30 fps), providing fine-grained spatiotemporal data on traffic. Studies have shown that the use of ultrahigh-resolution video improves the accuracy of vehicle detection and tracking compared to lower-resolution footage [28]. From typical flying altitudes, drones cover a wider field of view than fixed road sensors—a single UAV can survey a large highway segment or an entire interchange by adjusting altitude and camera zoom. This wide coverage addresses the limited spatial range of ground cameras and fixed sensors [29]. UAVs can also be redeployed quickly, offering mobility to follow incidents or cover blind spots; this agility is a major advantage in rapidly evolving traffic situations [30]. Drones offer superior temporal resolution by continuously recording dynamic traffic events, such as lane changes and queue formation, unlike inductive loops that provide only aggregate counts. Drone-based traffic counts also exhibit high accuracy, with experiments showing discrepancies of only 3–5% compared to ground sensors [31]. However, certain factors affect data quality: Drone vibrations or wind gusts can introduce image blur, and higher altitudes increase the pixel footprint per vehicle (reducing detail). In addition, flight endurance constraints (the battery life is often 20–30 min for small drones) can limit continuous data collection. As Zhao et al. [32] note, battery limitations, platform instability, and wireless transmission constraints remain challenges in acquiring accurate real-time traffic data from UAVs. However, when properly installed, UAVs can efficiently collect rich traffic data in large areas in real time, far outpacing the spatial coverage of fixed sensors [33]. Thermal cameras offer a valuable complement by detecting heat signatures, making them highly effective for nighttime operations or for distinguishing active vehicles from stationary ones. However, they provide lower resolution and lack the detailed scene context of RGB cameras, making them better suited for detection rather than precise classification or tracking tasks. LiDAR sensors represent a significant step forward in terms of data robustness. By generating precise 3D point clouds, they are largely immune to lighting conditions and can provide detailed geometric data for infrastructure inspection (e.g., detecting pavement rutting) and can, to some extent, mitigate occlusion issues by capturing partial vehicle shapes. However, their high cost, complex data-processing requirements, and potential performance degradation in heavy precipitation currently limit their widespread use for real-time traffic monitoring. There is a critical tradeoff among data richness, operational robustness, and system cost. For general traffic flow analysis in clear weather, RGB cameras offer the best cost–benefit ratio. For applications requiring 24/7 operational capability or operating in environments with frequent fog or smoke, a sensor fusion approach that combines RGB and thermal imagery is superior. For high-precision infrastructure assessment or scenarios that require the highest level of robustness against occlusions, LiDAR is the most capable, albeit the most expensive, solution. Future research must focus on developing robust and cost-effective sensor fusion techniques that can dynamically take advantage of the strengths of each sensor type to provide a consistent and reliable data stream under diverse and challenging road conditions.

Computer Vision and Data Analysis Techniques: Significant advances in computer vision (CV) have enabled UAV-based traffic monitoring. Early drone traffic studies used classical vision algorithms, such as frame differencing, background subtraction, and optical flow (KLT), for vehicle detection and tracking [1]. Although effective to some extent, these traditional methods struggled with occlusions, the distortion of perspective from aerial views, and complex scenes (e.g., traffic congestion or lane changes) [34]. In the last five years, there has been a clear shift toward deep learning models for interpreting UAV imagery [35]. Convolutional neural networks (CNNs) and their variants are now routinely used to detect vehicles, often outperforming earlier techniques. For example, applying deep-CNN-based detectors (such as Faster R-CNN, YOLO, or Mask R-CNN) to drone videos has yielded higher detection rates and more robust performances under varied conditions [36]. Premaratne et al. [37] demonstrated that deep neural networks significantly outperformed traditional vision algorithms in accuracy for vehicle detection and counting, given sufficient training data. Specialized models for aerial views also exist, such as frameworks that first segment roads and then detect vehicles, focusing on road regions. Hoanh and Pham [38] presented a multitask deep learning framework that enhances vehicle detection accuracy in complex UAV imagery by simultaneously identifying road surfaces and detecting cars. In addition to detection, modern UAV analytics extract rich traffic metrics. Vehicle trajectories, speeds, densities, and even lane-changing behavior can be inferred from aerial video. The shift from classical vision algorithms to deep learning models marks a significant leap in performance, but it also introduces important tradeoffs among accuracy, computational cost, and real-world reliability, particularly in challenging highway environments. The primary deep learning models used for vehicle detection fall into two categories: two-stage detectors (e.g., Faster R-CNN) and single-stage detectors (e.g., YOLO variants). Two-Stage Detectors (e.g., Faster R-CNN and Mask R-CNN): These models are generally characterized by higher detection accuracy, especially for small or partially occluded objects, which are common in dense traffic imagery from high-altitude UAVs. Their region proposal mechanism allows for a more thorough examination of potential objects. However, this accuracy comes at a significant computational cost, making them largely unsuitable for real-time processing on lightweight, power-constrained onboard computers. They are best suited for offline high-precision analysis, such as for dataset creation or post-incident forensic studies. Single-Stage Detectors (e.g., YOLO and SSD): Models like YOLO (You Only Look Once) prioritize speed and computational efficiency, making them the de facto standard for applications that require real-time performance. Modern variants (e.g., YOLOv5/v8) have substantially closed the accuracy gap with two-stage detectors while maintaining high frame rates, even on lightweight edge devices, like NVIDIA Jetson. However, they can still struggle with detecting very small vehicles in wide-angle views or in extremely dense, overlapping traffic compared to their two-stage counterparts. Therefore, the choice of the algorithm presents a clear tradeoff: For real-time, onboard traffic monitoring and incident detection, YOLO variants offer the best balance of performance and efficiency. In contrast, for applications where the highest possible accuracy is paramount and processing can be performed offline, Faster R-CNN or Mask R-CNN is superior. The ongoing research trend is the development of lightweight and quantized versions of these models, with the aim of bringing the accuracy of two-stage detectors to the efficient framework of single-stage detectors for deployment on resource-constrained UAV platforms. For example, advanced pipelines combine object detectors with multiobject-tracking algorithms (Kalman filter or deep SORT) to follow vehicles over time and reconstruct their trajectories [39]. Such data enable lane-level analysis of traffic flow (headways, lane-specific volume, etc.) that was previously infeasible at scale. AI-powered drone video analysis has advanced anomaly detection in traffic monitoring (e.g., identifying violations or incidents) [40]. Modern computer vision, especially deep learning, significantly improves UAV traffic monitoring, providing more accurate and comprehensive aerial data compared to previous methods.

Algorithms and Processing Frameworks: To handle the torrent of UAV data, researchers are exploring efficient processing frameworks, including onboard edge computing and real-time data links. UAV traffic video is traditionally recorded and processed offline via the post-flight algorithm application [41]. Now, the trend is toward real-time processing so that traffic agencies can respond instantly to aerial observations [42]. Two complementary approaches are being pursued: improving the communication bandwidth to stream data for ground processing and performing more computations on the UAV/edge device itself. High-bandwidth wireless links (4G/5G or dedicated radio) allow drones to transmit live video to roadside units or cloud servers, where powerful computers run detection/tracking algorithms in real time. At the same time, edge computing on UAVs is gaining traction, equipping drones with lightweight AI processors (e.g., NVIDIA Jetson and RasPi-based systems) to run neural networks onboard [43]. This minimizes latency, as critical detection can occur on the drone without sending all the raw video to a distant server. Another challenge for real-time UAV monitoring is dealing with limited battery life—frequent landings to recharge disrupt continuous observation. Here, innovative edge solutions are emerging: For example, ground-based charging stations placed along highways can allow UAVs to land, recharge autonomously, and resume patrol without returning to a distant base [29]. This kind of infrastructure not only extends flight time but also provides a local hub for data offloading and processing. Li et al. [29] proposed deploying a network of such fixed automated stations, which in addition to recharging drones, maintain persistent communication links for real-time data transmission during ‘cruising’ operations on UAV highways. Using these stations, drones achieved continuous coverage of an urban road network and could transmit information in real time to traffic management centers [14]. In summary, the latest UAV data-processing frameworks emphasize real-time edge-based computation and robust connectivity. Deep learning models are being optimized to run on drone-compatible hardware, and combined with improved wireless networks and refueling strategies, modern UAV systems can now offer near-real-time traffic monitoring and incident responses from the sky [44].

3.2. UAV Platforms

Fixed-Wing vs. Rotary-Wing UAVs: UAV platforms for highway monitoring mainly include fixed and rotary wings [45]. Fixed-wing UAVs offer superior endurance, speed, and range, making them ideal for extensive corridor monitoring and large-area surveys. However, their inability to hover, the continuous forward movement requirement, and the complexity of deployment (launch and recovery) limit their use for detailed or localized inspection tasks. In contrast, rotary-wing UAVs (multirotors such as quadcopters) are more flexible due to vertical takeoff and landing (VTOL) capability, hovering ability, and ease of deployment. They are widely adopted for detailed inspections, incident monitoring, and urban scenarios, despite their shorter flight times and limited payload capacities compared to those of fixed-wing platforms. Recently, hybrid VTOL platforms have emerged that combine the benefits of fixed-wing endurance and rotary-wing maneuverability to better meet highway-monitoring needs [46]. The technical comparison of these UAV platforms and their suitable application scenarios are detailed in

Table 2. Technical characteristics of UAV platforms for highway monitoring.

Platform Type	Endurance and Range	Hover Capability	Deployment Complexity	Payload Capacity	Typical Applications	Limitations
Fixed wing	1–6 h Long range	No	High	High	Large-area surveys corridor monitoring	No hover and limited maneuverability
Rotary wing	20–40 min Short range	Yes	Low	Moderate to Low	Incident monitoring detailed inspections	Limited flight time and payload constraints
Hybrid VTOL	1–3 h Medium range	Yes	Medium	Moderate to High	Flexible monitoring and mixed tasks	Higher cost and  complexity

.

Advancements in Multi-UAV Coordination: Recent research has highlighted multi-UAV coordination to overcome the limitations of single-drone coverage. Multi-UAV systems efficiently cover larger highway networks, improve reliability, and provide persistent monitoring [47]. For example, experiments employing coordinated UAV swarms have successfully demonstrated the comprehensive collection of real-time traffic data on extensive urban and highway networks [48]. Current studies emphasize decentralized coordination strategies, including game-theoretic approaches, coverage path optimization, and dynamic task allocation, to improve collective UAV performance in traffic surveillance [49].

UAV Hardware and Technical Requirements: Successful highway deployment of UAVs requires specific technical capabilities, including adequate flight endurance, payload capacity, flight stability, and robust data transmission links [50]. Limited battery life remains a significant operational challenge, prompting the use of ground-based recharge stations or battery swap systems to maintain continuous monitoring. Stable high-bandwidth communication is vital for real-time data relay, necessitating advanced onboard systems and dedicated communication infrastructure. Safety regulations and collision avoidance systems further shape hardware requirements, highlighting the need for robust and fail-safe UAV designs and skilled operation [51].

3.3. Summary and Research Gaps of Technical Characteristics

This section provided a comprehensive analysis of the technical characteristics of UAVs and their specific advantages for highway-monitoring systems. Performing a systematic review of the current literature, it was found that UAV technologies significantly surpass traditional fixed-ground sensors in terms of mobility, spatial and temporal resolutions, flexibility in deployment, and data richness. In particular, advances in sensors (e.g., optical, thermal, and LiDAR), computational frameworks (edge computing and deep-learning-based video analytics), and multi-UAV coordination strategies have contributed extensively to real-time and large-scale highway-monitoring capabilities. Despite these technological advances, several critical gaps and challenges remain, which present valuable opportunities for future research.

Sensor Integration and Fusion: Existing research heavily relies on optical cameras, whereas the integration of multiple sensor types (e.g., LiDAR, thermal imaging, and radar) for improved robustness under adverse conditions (e.g., nighttime and poor weather) remains underexplored. Standardized methodologies for sensor fusion and calibration in UAV-based highway applications are insufficient, limiting the reproducibility and scalability of research outcomes.

UAV Platform Selection and Coordination: Comparative analyses among fixed-wing, rotary-wing, and hybrid UAV platforms that focus specifically on their operational effectiveness and cost–benefit analyses in diverse highway scenarios are limited. Research on robust multi-UAV collaborative systems, including standardized protocols for decentralized coordination, collision avoidance, and autonomous flight management, remains inadequate, especially for extensive and complex highway networks.

Computational Efficiency and Real-Time Analytics: Although deep learning techniques significantly improved detection and tracking accuracies, lightweight deep learning models optimized for advanced UAV-computing platforms are still immature, particularly in handling densely trafficked and occluded highway environments. Real-time processing frameworks that address bandwidth constraints, latency issues, and computational bottlenecks for seamless integration with existing intelligent transportation systems are still lacking.

4. Applications of UAVs in Highway Systems

4.1. Infrastructure Inspection and Asset Management

Infrastructure inspection and asset management using UAV technology have increasingly focused on structural integrity [52], road condition assessment [53], and disaster response [54], demonstrating significant improvements over traditional methods in efficiency, precision, and safety.

Bridge inspection has benefited significantly from UAV-based technologies. UAVs equipped with optical and infrared sensors have demonstrated capabilities for detailed ${360}^{\circ }$ inspections of bridges, effectively overcoming inspection blind spots traditionally encountered beneath bridge structures and in complex terrains [11]. For example, UAVs equipped with cameras and infrared sensors provided comprehensive data capture, 3D modeling, and defect localization for bridges in Alaska, significantly reducing data storage requirements and inspection times compared to those of traditional methods [55]. Similarly, the integration of UAVs with automated systems demonstrated their efficacy in detecting fatigue cracks in the fracture-critical inspection of steel bridges, highlighting their potential despite challenges related to flight stability and camera resolution under adverse conditions [56]. Automated robotic systems using UAVs further streamline bridge inspection pipelines by automating the entire process from data acquisition, 3D reconstruction, and defect detection to reporting, demonstrating high efficiency and accuracy despite limitations such as sensor resolution and weather impacts [55].

In pavement and road infrastructure management, UAV-based inspection techniques significantly improved the accuracies of damage detection and analysis. Enhanced deep learning methods, such as the UM-YOLO algorithm [57], effectively identified various maintenance objects on the road in real time with high precision and recall rates, especially under complex background conditions. UAV-based crack detection leveraging semantic segmentation models, such as U-Net, PSPNet, and DeepLabv3+, addressed challenges in dynamic object occlusion and crack quantification, achieving high accuracy and robustness in complex road scenarios [58]. Furthermore, employing UAV infrared thermography (IRT) techniques effectively detected subsurface defects caused by damaged culverts, demonstrating the potential for early identification and preventive maintenance, although limited by depth detection capabilities [59].

The inspection of landslide and geotechnical hazards also benefited significantly from UAV technology. UAV-assisted photogrammetry, combined with 3D modeling and finite element analysis, accurately mapped and analyzed slope failures, providing detailed forensic analysis and effective monitoring of embankment failures along highways. Li et al. [60], Nobahar et al. [61] and Zhu et al. [62] underscored the potential of the UAV to provide critical information during landslide emergencies, although challenges, such as limitations in real-time monitoring and dependency on environmental factors, persisted. In-depth forensic analyses using UAV and electrical resistivity imaging (ERI) highlighted the effectiveness of non-destructive combined methods for the accurate detection and evaluation of slope stability failures, enhancing disaster mitigation strategies [63].

Despite these advances, challenges persist, including the stability of the UAV system under adverse conditions, sensor limitations, and the complexity of integrating multimodal data. Future research should emphasize improving UAV flight stability and precision under adverse conditions, enhancing sensor resolutions for detailed inspections, integrating multimodal sensor data, and developing robust real-time monitoring systems [64].

4.2. Trajectory Prediction and Behavior Analysis

Trajectory prediction and behavior analysis have become integral to improve highway traffic safety and traffic management efficiency and the advancement of autonomous driving technology. Recent advances have extensively utilized UAV-based high-resolution trajectory datasets, significantly improving both prediction accuracy and behavior analysis.

Classical statistical and heuristic methods have shown utility in less complex vehicle interactions or specific traffic maneuvers. For example, density-based clustering combined with Support Vector Machines (SVMs) effectively predicted lane-changing events, demonstrating good accuracy with minimal computational overhead [65]. Similarly, Poisson regression models successfully analyzed influential variables affecting passing rates in short passing zones, considering geometric characteristics and traffic flow [66].

Machine-learning approaches integrating traffic context and driver heterogeneity have significantly advanced predictive accuracy. Models utilizing XGBoost and Long Short-Term Memory (LSTM) networks substantially improved lane-change predictions by incorporating traffic contexts, vehicle dynamics, and driver characteristics [67]. Multilevel generalized linear regression methods effectively identified influential factors on lane-change rates in weaving segments, highlighting the impacts of road configuration and flow conditions [68]. In addition, unsupervised clustering was applied combined with logistic regression to classify driving styles, improving the accuracy of predicting driving intentions [69].

Deep learning methodologies have emerged as dominant tools for complex trajectory prediction tasks. Transformer-based time series models demonstrated excellent short-term vehicle speed prediction performance at highway interchanges [70]. The integration of graph neural networks (GNNs) and neural ordinary differential equations (neural ODEs) successfully captured multiagent interactions, enabling reliable multimodal trajectory predictions in dynamic environments [71]. CNN-based models that use bird’s-eye-view representations [72], variational autoencoders with interpretable latent spaces [73], and attention-based recurrent neural networks [74] have also contributed significantly to improve the accuracy of trajectory predictions.

Innovative approaches combining physical models with deep learning techniques represent another promising research direction [13]. Physics-informed transformer models integrating intelligent driver models (IDMs) have improved trajectory prediction accuracy by embedding physical constraints [75]. The incorporation of kinematic wave theory into deep convolutional neural networks further enhanced the estimation of traffic speed [76]. Trajectory predictions based on velocity vector fields provided substantial improvements in short-term predictions for automated vehicles [77].

Multidimensional methods addressing complex interactions and environmental variability have recently gained attention [78]. Methods employing spatiotemporal graph transformations [79,80], personalized predictions using temporal graph neural networks [81], and hierarchical attention mechanisms [82] have significantly improved predictive robustness. Furthermore, multimodal trajectory prediction methods [83] and federated learning models based on Siamese networks effectively [84] addressed data privacy and individualized driving behaviors.

Behavior analysis focusing on patterns in lane changes has also been extensively explored. Methods combining heuristic attention models [85], interaction-aware predictions during merging [86], and detailed behavior analyses under weak lane discipline conditions [87] demonstrated substantial advancements in understanding and predicting complex driving behaviors and decision-making processes.

Despite considerable advancements, challenges remain, such as model adaptability to varying conditions, real-time processing limitations, and dataset restrictions [88]. Future research should focus on model generalization, real-time applicability, and developing standardized frameworks for evaluation and comparison in diverse scenarios [13]. Beyond predictive accuracy, a critical evaluation of trajectory prediction models must consider their scalability and the cost/benefit tradeoffs for real-world deployment. Physics-based models, such as the intelligent driver model (IDM), are computationally lightweight and offer high interpretability, making them scalable for simulating large networks. However, they often fail to capture the complex nonlinear interactions and heterogeneous driving behaviors present in real traffic, limiting their predictive accuracy in complex scenarios, like merges or dense congestion. In contrast, deep learning models, particularly those based on LSTMs, Graph Neural Networks (GNNs), and transformers, have demonstrated superior accuracy by learning these complex interactions directly from data. However, this high performance comes with significant costs. These models require large-scale, high-quality, and meticulously labeled trajectory datasets for training, which are expensive to collect and process. Furthermore, their high computational complexity poses a major challenge for scalability and real-time onboard inferences. Running a complex transformer model to predict trajectories for hundreds of vehicles simultaneously on a UAV’s edge computer is currently infeasible. This presents a clear cost/benefit tradeoff: Deep learning models offer the highest accuracy but are costly to train and challenging to scale for real-time, system-wide applications. Physics-informed neural networks, which embed physical constraints (e.g., kinematic equations) into a deep learning architecture, represent a promising compromise. They aim to achieve high accuracy with smaller training datasets and better generalization, potentially offering a more scalable and cost-effective solution for future intelligent transportation systems.

4.3. Traffic Safety and Conflict Analysis

UAVs have proven to be valuable in the analysis of traffic safety and conflict events by allowing the detailed observation and modeling of vehicle behaviors and interactions on highways. Recent studies have used UAV trajectory data to explore relationships among traffic flow characteristics, driver behavior, and collision risks [89].

Data collected from drones have significantly advanced the assessment of highway design safety and conflict prediction. For example, UAV-captured lateral acceleration data were used to assess the safety of highway loop designs, demonstrating that geometric configurations considerably impact driving behaviors, such as lateral acceleration. A proactive risk-based assessment model (LSL) was developed that incorporates lateral acceleration indices, effectively diagnosing high-risk geometric configurations of highway loops [90].

In predicting the risk of collisions, UAV data facilitated the real-time evaluation of traffic conflicts, improving the accuracy of predicting imminent crashes. By applying extreme value theory (EVT) to vehicle pre-crash trajectories, Chen et al. [10] found that the time to collision (TTC) was superior in predicting collisions occurring within short prediction windows, demonstrating the benefit of UAV trajectory data in capturing critical conflict indicators. Moreover, surrogate modeling frameworks that employ UAV data (such as the SM-OPMS framework) have enabled the adaptive selection of optimal conflict prediction models, substantially improving computational efficiency and maintaining high prediction accuracy Wu et al. [91].

In addition, UAV datasets, such as CitySim and HighD, were used to develop individualized conflict prediction models. Studies employing deep learning models, including LSTM, CNN, and AT-LSTM, improved the precision of conflict predictions in weaving segments and merging areas by capturing complex driver interactions and individualized driving behaviors [92]. Liu et al. [93] and Yu et al. [92] effectively integrated driving risk maps and multimodal trajectory predictions, enhancing risk evaluation and trajectory prediction accuracy by addressing the complexities of interactions and data imbalance.

The analysis of traffic-conflict mechanisms has also benefitted from UAV data. Bayesian network models and spatiotemporal transformer models provided information on the causal relationships between vehicle interactions and crash risks [94], particularly in merging zones of highways [95]. This enabled a clearer understanding of conflict mechanisms and identified key factors that influence crash occurrences [96].

Despite these advances, several limitations remain, such as dataset homogeneity, limited geographic coverage, and challenges related to data imbalance. Future research directions include expanding UAV data collection under diverse geographic and traffic conditions, integrating multimodal trajectory predictions, addressing data imbalance issues, and improving the robustness of models by considering additional contextual factors, such as driver attributes, weather conditions, and roadway geometries [91].

4.4. Traffic Flow and State Analysis

UAV technology has become crucial for understanding traffic state transitions, bottleneck formation, and heterogeneous traffic flow dynamics, significantly enhancing advanced traffic management and control [97]. UAV-based methods have validated fundamental traffic flow theories and provided insight into vehicle behavior near critical bottlenecks. For example, UAV-based analysis was utilized to study traffic state transitions near merge highway bottlenecks, confirming that lane changes under low headways significantly affect transitions, aligning well with three-phase traffic theory [98].

In heterogeneous traffic flow contexts, UAV-derived data have contributed to refining fundamental diagrams and driving behavior models. Recent studies have shown that incorporating UAV data into macroscopic traffic models improves accuracy by capturing diverse driving behaviors and interactions, crucial for better management of complex traffic scenarios [99].

Advanced UAV-based communication techniques have also improved traffic state estimation efficiency. Specifically, studies exploring orthogonal frequency division multiplexing (OFDM) for UAV-based small cell communications highlighted that compressed sensing-based channel estimation significantly improved the reliability and precision of high-mobility traffic surveillance [100].

For real-time traffic state estimation, integrating UAV-acquired video data and ensemble classifiers with optical flow techniques has achieved accurate estimation of traffic parameters, such as flow, density, and speed, facilitating the effective monitoring and management of multiregional traffic networks [101]. Similarly, a study introduced a real-time UAV-based estimation approach for multiregional traffic networks, demonstrating a greater ability to dynamically assess complex traffic network conditions [102].

Furthermore, closed-loop control architectures based on UAVs, which integrate wireless sensor networks (WSNs), have significantly improved highway surveillance efficiency by autonomously detecting traffic incidents, such as non-recurrent congestion, thereby reducing response time and enhancing overall monitoring capabilities [103].

UAV datasets also facilitated the selection and validation of suitable car-following models for automated vehicles. Using UAV data, the improved intelligent driver model (IIDM) showed superior performance in simulating the behaviors of automated vehicles on highways, demonstrating the potential of UAV data to refine automated driving models [104].

Despite these advances, limitations persist, such as constraints on UAV flight durations, data continuity, and geographic coverage limitations. Future research should focus on expanding UAV data collection efforts in various geographic contexts, optimizing UAV and sensor integration frameworks, improving real-time communication systems, and incorporating more complex traffic dynamics and driver behavior factors into traffic state estimation models [104].

4.5. Dataset Creation and Benchmarking

Several UAV-based highway traffic datasets have emerged, significantly improving research capabilities in traffic behavior analysis, safety assessment, and automated driving system validation. Prominent datasets include A43 [105], CitySim [106], CQSkyEyeX [107,108], exiD [109], HighD [110], Mirror-Traffic (URL (accessed on 30 August 2025) http://www.scenarios.cn/html/index.html), Ubiquitous Traffic Eyes (UTE) (URL (accessed on 30 August 2025) http://www.seutraffic.com/#/home), and AD4CHE [111]. As shown in

Table 3. Comparison of representative UAV-based highway traffic datasets.

Dataset	Region	Scenarios	Data Acquisition Methods	Trajectory Precision	Application Highlights
A43 [105]	Germany	ramp bottleneck, mixed congestion, free-flow traffic	YOLOv5, 3D camera calibration	10 cm	bottleneck analysis, ramp-induced congestion, safety analysis, behavior modeling
CitySim [106]	Multi-region	highway straight merging/diverging safety events	Mask R-CNN, manual correction	high accuracy, detailed vehicle motion	safety analysis, digital twin simulations, behavior modeling
CQSkyEyeX [107,108]	China	Highway weaving merging/diverging, free-flow, congested	YOLOx, DeepSort	<10 cm, 30 Hz	behavior modeling, safety analysis, weaving area dynamics
exiD [109]	Germany	entrance/exit, merging/diverging	YOLO, DeepSort, high-definition maps	Decimeter-level, detailed dynamics	ramp interaction, ADAS validation, micro-simulation
HighD [110]	Germany	Highway straight, lane changes, congestion	deep learning	<10 cm, 25 Hz	behavior modeling, safety assessment, automated driving validation
Mirror-Traffic	China	ramps, straight segments	tracking, manual validation	Centimeter-level	trajectory prediction, heterogeneous analysis, automated driving validation
UTE	China	ramps, merging/diverging	YOLOv4, OpenCV tracking	Decimeter-level, 25 Hz	Ramp flow dynamics, behavior modeling, automated driving validation
AD4CHE [111]	China	Congested, frequent stops, lane changes	deep learning	High precision, georeferenced, 30 Hz	Traffic jam, congestion strategies, behavior modeling

, representative UAV-based highway traffic datasets are compared regarding scenarios, data acquisition methods, trajectory precision, and key application highlights.

HighD provides extensive naturalistic vehicle behavior data from German highways, using UAV-captured high-resolution videos processed through deep learning models. The dataset includes diverse driving scenarios, especially valuable for validating automated driving systems [110]. Similarly, the exiD dataset focuses on complex interactions on highway ramps, providing critical detailed trajectories to model interactive traffic behavior and validate safety-critical scenarios [109].

CitySim and CQSkyEyeX datasets emphasize safety-oriented research, particularly in Chinese expressway environments. CitySim captures numerous safety-critical events (e.g., merging, diverging, and crossing maneuvers), processed via advanced computer vision algorithms, such as Mask R-CNN, and object-tracking techniques [106]. CQSkyEyeX employs YOLOx and DeepSort algorithms to extract high-precision vehicle trajectories, demonstrating high accuracy suitable for safety assessments [108].

The AD4CHE dataset specifically targets typical congestion scenarios prevalent on Chinese highways, providing rich behavior insights into driver responses under congested conditions. Using UAV-based video collection, the AD4CHE dataset supports a detailed analysis of driver behavior models and automated congestion management systems [111].

The A43 dataset includes off-ramp and congestion traffic scenarios on German highways, combining YOLO-based detection with advanced tracking methods to deliver reliable trajectory data for traffic flow and accident risk analysis [105]. Additionally, datasets such as Mirror-Traffic and UTE provide scalable and real-time data extraction frameworks, emphasizing usability under a variety of traffic conditions.

Compared to others, datasets such as HighD, exiD, and CitySim offer comprehensive coverage with high accuracy, supporting precise benchmarking in complex scenarios. In contrast, datasets such as AD4CHE and A43 focus on specialized traffic scenarios, providing valuable context-specific data crucial for targeted applications.

These datasets collectively establish a robust benchmarking foundation, significantly advancing research in traffic safety analysis, state estimation, and automated driving validation.

4.6. Other Applications

4.6.1. Network Coverage, Resource Scheduling, and Patrol

UAVs significantly improve network coverage and resource scheduling in highway management scenarios [112]. UAV-enabled Mobile Edge Computing (MEC) systems optimize the scheduling of computationally intensive tasks within the Internet of Vehicles (IoVs), minimizing response times through Markov network-based coevolutionary algorithms and particle swarm optimization strategies [113]. Furthermore, vehicular ad hoc networks (VANETs) assisted by UAVs, using clustering algorithms and nonorthogonal multiple access (NOMA), effectively handle vehicle-to-vehicle communication disruptions, improving bandwidth utilization and energy efficiency [114]. In addition, mathematical programming and heuristic optimization methods have been proposed to efficiently schedule UAV patrols on highways, ensuring effective monitoring despite physical constraints, such as battery life and flight range [115]. Deep-reinforcement-learning approaches have also shown significant improvements in UAV trajectory control, achieving adaptive network coverage, and reducing the number of UAVs required for comprehensive surveillance [116].

4.6.2. Environmental Monitoring

UAV technology also significantly advances environmental monitoring on highways. By vertically measuring particulate matter (PM) concentrations near roads, UAVs help to assess the environmental benefits of roadside vegetation in reducing PM2.5 and PM10 levels, providing practical data for sustainable highway management [117]. Similarly, vertical monitoring enabled by UAVs reveals the effectiveness of highway-cover structures (e.g., highway caps) in mitigating particulate matter and noise pollution in adjacent areas [118]. Studies integrating vertical monitoring based on UAVs with ground sensors show that vehicle types, emission standards, and wind speeds and directions significantly influence particulate dispersion, highlighting the importance of precision environmental assessment in transportation management [119]. Furthermore, UAV-assisted wireless-energy redistribution mechanisms within multiscale Internet of Things (IoT) frameworks have been proposed to optimize renewable energy (solar and wind) distributions for highway-lighting systems, contributing to sustainable and efficient highway infrastructure [120].

4.6.3. Policy, Regulations, and Human Factors

The integration of UAVs into highway systems introduces critical considerations that go beyond technology into the realms of policy, regulation, and public acceptance. Navigating this complex sociotechnical landscape is paramount for the safe, ethical, and scalable deployment of UAVs in transportation.

Regulatory frameworks for low-altitude airspace, particularly in relation to operations near critical infrastructure, such as highways, are still evolving and vary significantly among jurisdictions. This fragmentation presents a major challenge for the standardization and deployment of technology. A comparative analysis between the United States and the European Union, two leading regions in the regulation of UAVs, highlights these differences (see

Table 4. Comparative analysis of UAV regulatory frameworks for highway operations.

Feature	United States (FAA)	European Union (EASA)
Regulatory Body	Federal Aviation Administration (FAA)	European Union Aviation Safety Agency (EASA)
Core Principle	Performance- and risk-based rules. BVLOS and advanced operations typically require waivers.	Unified, risk-based categories (‘Open’, ‘Specific’, ‘Certified’) applicable across all member states.
Airspace Management	Developing UTM (UAS Traffic Management) in partnership with industry.	Mandating the development of U-space, a set of services for managing UAV traffic automatically and safely.
Operations Near Highways	Governed by “Operations Over Moving Vehicles” rules. Restrictions are based on the UAV’s kinetic energy and safety features.	Falls under the ‘Specific’ category, requiring a formal risk assessment (SORA) and operational authorization.
Privacy and Data Protection	No single federal privacy law. Governed by a mix of state laws and sector-specific regulations.	Strictly regulated under the General Data Protection Regulation (GDPR), requiring data minimization and a clear legal basis for data processing.

).

The US, under the Federal Aviation Administration (FAA), has adopted a performance-based approach, particularly with its rules on “Operations Over People and Moving Vehicles”, which categorize UAVs based on risk [121]. Operations in more complex scenarios, such as beyond the visual line of sight (BVLOS), typically require specific waivers. In contrast, the European Union, through the European Union Aviation Safety Agency (EASA), has established a more structured, risk-based framework divided into ‘Open’, ‘Specific’, and ‘Certified’ categories, coupled with the development of a U-space system for automated traffic management [122].

These different approaches create a complex compliance environment. For example, while the EU GDPR provides a stringent and unified framework for data privacy, privacy regulations in the US are a patchwork of federal and state laws, adding complexity to data handling for UAV operations that span multiple states [123].

The high-resolution sensors aboard UAVs inherently raise significant privacy concerns. To build public trust and ensure ethical operation, concrete privacy-preserving workflows must be integrated into every stage of UAV deployment. This goes beyond simple policy statements to include technical and procedural safeguards. A multistage workflow can be adopted.

• Privacy Impact Assessment (PIA): Before any mission, a formal PIA should be performed to identify potential privacy risks and establish mitigation strategies.
• Geofencing and Mission Planning: Flight paths should be meticulously planned to avoid sensitive areas (e.g., residential properties adjacent to highways). Dynamic geofencing can create virtual barriers to prevent unintentional data collection outside the target corridor.
• Onboard Anonymization: To minimize the collection of personally identifiable information (PII), such as faces and license plates, UAVs can be equipped with lightweight edge-computing processors running real-time anonymization algorithms. These algorithms can automatically apply blurring or masking to sensitive areas in the video feed before the data are even stored or transmitted.
• Data Minimization: Operators should adhere to the principle of data minimization, using the lowest possible sensor resolution that still meets the technical requirements of the mission (e.g., using lower resolution for traffic counting vs. higher for crack detection).
• Secure Data Handling: All the data collected must be encrypted both in transit and at rest. Access should be strictly controlled and logged based on the principle of least privilege.
• Differential Privacy: When sharing datasets for research or public use, techniques such as differential privacy can be applied. This involves adding a calibrated amount of statistical “noise” to the data, which protects individual privacy while preserving the overall statistical properties and utility of the dataset.

Finally, the human factor of driver distraction remains a key safety concern [124]. This reinforces the need for regulations that specify the minimum operational altitudes and lateral distances from active traffic lanes, ensuring that UAVs do not become a hazard themselves. Addressing these policy, ethical, and human-centric challenges is as crucial as solving technical ones for the successful integration of UAVs into highway systems.

4.7. Summary and Research Gaps of Applications

Before delving into a detailed discussion of research gaps, it is useful to synthesize the main UAV applications in highway systems and map them to their corresponding key technical and operational issues.

Table 5. Synthesis of UAV applications in highway systems and key corresponding issues.

Application Domain	Key Objectives	State-of-the-Art Approaches	Key Issues and Challenges
Traffic Flow and State Analysis	Estimate macroscopic parameters (flow, density, speed); Detect congestion, queues, and shock waves.	Deep learning (YOLO, Faster R-CNN) for vehicle detection. Tracking algorithms (DeepSORT, Kalman Filter) for trajectory extraction. Macroscopic traffic flow models.	Data Continuity: Limited UAV flight endurance restricts long-term, continuous monitoring. Real-Time Processing: High computational cost of algorithms challenges onboard, real-time analytics. Occlusion: High-density traffic leads to vehicle occlusions, reducing detection accuracy.
Infrastructure Inspection and Asset Management	Detect structural defects (cracks, corrosion) on bridges and pavements; Monitor slope stability; Asset inventory.	High-resolution RGB imagery with CNNs for crack detection. LiDAR for 3D modeling and deformation analysis. Thermal imaging for subsurface defect detection (e.g., delamination).	Data Quality: Environmental factors (wind, rain, poor lighting) degrade sensor data and flight stability. Scalability: Manual flight and data analysis are labor-intensive; automated inspection is still maturing. Multimodal Fusion: Lack of standardized methods for fusing data from multiple sensors (e.g., LiDAR + RGB).
Traffic Safety and Conflict Analysis	Extract vehicle trajectories to calculate surrogate safety measures (TTC, PET); Analyze driver behavior in near-miss events.	High-precision trajectory extraction from aerial video. Deep learning models (LSTM and GNNs) to predict conflicts. Risk assessment models based on kinematic data.	Dataset Limitations: Lack of diverse, large-scale, and geographically varied datasets with labeled conflict events. Contextual Factors: Insufficient integration of driver behavior, weather, and road geometry into models. Validation: Difficulty in validating surrogate measures against actual crash data.
Network Coverage and Environmental Monitoring	Provide communication relays for V2X; Monitor air quality (PM2.5) and noise levels along corridors.	UAVs as mobile edge computing (MEC) nodes. Reinforcement learning for patrol path optimization. Onboard environmental sensors for vertical profiling of pollutants.	System Integration: Seamless integration with ground-based ITS and communication networks is complex. Standardization: Lack of standardized protocols for environmental data collection and measurement accuracy. Energy Efficiency: Balancing communication/sensing tasks with limited battery life is a major constraint.

provides such a summary, providing a clear overview of the state of each domain and the persistent challenges that future research must address. This synthesis helps to contextualize the specific research gaps identified in this section. This section comprehensively reviewed the diverse applications of UAVs in highway systems, covering infrastructure inspection, trajectory prediction, safety assessment, traffic flow analysis, dataset creation, and emerging application scenarios. UAV-based applications demonstrate substantial advances, outperforming traditional methods by providing high resolution, extensive coverage, and flexible data collection in challenging highway environments. Despite considerable achievements, several research gaps and practical limitations remain. As illustrated in Figure 5, UAVs support a broad spectrum of highway tasks.

• Infrastructure Inspection: There is limited research on automated and real-time damage detection methods that integrate multimodal sensor data (e.g., LiDAR, infrared, and radar) to accurately assess the condition of the infrastructure. Current UAV inspection frameworks face environmental challenges, such as strong winds, rain, and fog, affecting data quality and inspection reliability.
• Trajectory Prediction and Behavior Analysis: Models lack sufficient generalization across varied traffic contexts due to limited dataset diversity, especially under adverse conditions (e.g., nighttime or inclement weather). Real-time trajectory prediction algorithms are rarely validated in actual UAV deployments due to computational constraints and latency issues.
• Traffic Safety and Conflict Analysis: The heterogeneity and geographic coverage of the conflict analysis datasets available using UAVs remain limited, restricting the development of robust and widely applicable predictive models. There is insufficient integration of additional contextual factors, such as driver behavior characteristics, road geometry, and weather, into current conflict prediction models.
• Traffic Flow and State Analysis: UAV-based studies often rely on short-duration data due to UAV flight time constraints, limiting the continuous monitoring and comprehensive evaluation of the traffic state. Integration frameworks combining UAV data with ground-based sensors and communication technologies to enable real-time, comprehensive traffic state estimation are currently underdeveloped.
• Dataset Creation and Benchmarking: Although UAV-based datasets, such as HighD and CitySim, have advanced benchmarking capabilities, there is still a lack of standardized data collection protocols, annotations, and performance evaluation metrics across different datasets, limiting comparative analysis and universal applicability.
• Emerging Applications (Environmental Monitoring and Network Coverage): UAV applications in environmental monitoring (e.g., air quality and noise) lack standardization, particularly with regard to vertical spatial coverage and measurement accuracy. UAV-assisted vehicular networks for enhanced coverage and resource scheduling remain at the early stages of research, especially in terms of scalability and robustness under real-world operational constraints.

5. Challenges and Solutions

5.1. Technical Challenges in Data Acquisition

Adverse weather conditions significantly affect the performance of UAV-based highway-monitoring systems [7]. Strong winds can destabilize UAV flight paths, leading to erratic movements and compromised data accuracy. Rain and fog reduce visibility and sensor effectiveness, resulting in degraded image quality and unreliable data collection. These environmental factors pose substantial challenges in maintaining consistent and accurate UAV operations for traffic monitoring.

Visual occlusions caused by infrastructure elements, vegetation, or overlapping vehicles hinder the ability of the UAV to capture complete traffic data [125]. Such obstructions lead to incomplete or inaccurate vehicle detection and tracking. Furthermore, data noise arising from UAV vibrations and image compression artifacts further deteriorates data quality, complicating subsequent analysis and interpretation.

Ensuring the accuracy and reliability of UAV-collected traffic data is challenging due to factors such as GPS drift, limited sensor resolution, and battery constraints [126]. The positional inaccuracies introduced by standard GPS receivers, coupled with the limited battery capacity of drones, reduce both the spatial accuracy and the continuity of data collection, ultimately affecting the completeness and reliability of the dataset for real-world applications.

5.2. Data Processing and Analytical Challenges

Integrating multisource heterogeneous data from UAV videos, ground sensors, and vehicle communication systems poses substantial challenges due to differences in data formats, spatial resolutions, and synchronization accuracy [9]. Data acquired from drones often require sophisticated fusion techniques to ensure a coherent representation of traffic dynamics, but discrepancies in time synchronization and coordinate alignment can hinder effective data integration, limiting the reliability of subsequent analyses.

Despite advances in deep learning techniques, current UAV-based algorithms for object detection, tracking, and trajectory prediction face significant challenges [5], such as reduced detection accuracy for small or partially occluded objects, limited robustness under varying illumination conditions, and sensitivity to UAV camera resolution and flight altitude. Algorithms such as YOLO variants and SORT-tracking approaches, while effective under clear conditions, often experience substantial accuracy drops in complex, densely trafficked environments or under variable lighting conditions, reducing their effectiveness in real-world applications.

Real-time processing and response capabilities remain critical yet challenging for UAV-based traffic-monitoring systems [127], primarily due to computationally intensive image analysis methods and constraints on onboard processing power. Techniques like optical flow computation, deep neural networks for object detection, and trajectory prediction frequently exceed the computational capacity available on UAV-embedded platforms, causing delays in processing. This latency restricts UAVs’ potential for timely intervention under rapidly changing traffic conditions, such as incident detection or immediate safety warnings.

5.3. Safety and Regulatory Issues

Integrating UAVs into transportation systems requires clear regulatory frameworks, especially in relation to low-altitude airspace management [17]. Current guidelines vary considerably between regions, leading to uncertainty and potential operational conflicts, highlighting the need for standardized international airspace management practices.

Privacy concerns arising from high-resolution UAV imagery present significant ethical challenges [5,128]. Although existing methods, such as data anonymization, mitigate privacy risks, these techniques can negatively impact the analytical utility of datasets.

In addition, ethical issues related to data sharing and open access remain unresolved. Balancing data transparency for research and commercial purposes with privacy protection requires comprehensive ethical guidelines and transparent governance [5,128].

5.4. Proposed Solutions

Integrating data from multiple sensors, such as cameras, LiDAR, and radar, enhances the accuracy and reliability of UAV-based highway monitoring. By combining visual and nonvisual data, these techniques mitigate issues such as visual occlusion and adverse weather conditions, providing a comprehensive understanding of traffic dynamics. This fusion enables more robust vehicle detection and tracking, even in complex environments [129].

Developing lightweight deep learning models tailored for UAV platforms addresses the challenges of limited computational resources and real-time processing requirements. These algorithms maintain high performance in object detection and tracking while being efficient enough to run on UAVs, facilitating prompt data analysis and decision making during flight operations [57].

Implementing technical solutions, such as data anonymization and differential privacy, helps to address privacy concerns associated with UAV data collection. These measures ensure that sensitive information is protected while allowing effective traffic monitoring. In addition, establishing clear airspace management protocols and incorporating secure communication channels are essential for safe and compliant UAV operations [130,131,132].

6. Research Trends and Future Directions

6.1. Research Trends

The deployment of UAV technology in highway transportation is no longer characterized by standalone systems but by a deep convergence with a trinity of cutting-edge technologies: artificial intelligence (AI), 5G communications, and the Internet of Things (IoT). This technological synergy is the primary driver of current research trends, transforming UAVs from simple data collection platforms to intelligent, interconnected nodes within a larger transportation ecosystem.

AI fundamentally shifts UAV operations from reactive monitoring to proactive and predictive management. Beyond improving the accuracy of object detection algorithms, AI is enabling full autonomy and advanced data analytics. Enhanced Autonomy and Swarm Intelligence: AI-driven algorithms are the core of autonomous navigation, enabling dynamic route planning, real-time collision avoidance, and intelligent patrol strategies that adapt to changing traffic conditions without human intervention [133]. Furthermore, AI is crucial for the coordination of large-scale UAV swarms, allowing them to collaboratively monitor extensive highway corridors, optimize network coverage, and execute complex tasks, such as the reconstruction of post-accident scenes in a decentralized manner [134]. Predictive Analytics and Edge AI: Onboard AI processors (Edge AI) allow for real-time data processing directly on the UAV. This minimizes latency by enabling critical analytics, such as identifying precursor behaviors to traffic incidents or detecting structural fatigue on bridges, without the need to transmit raw data to a central server. This onboard intelligence is essential for immediate safety alerts and rapid decision making [135].

The rollout of 5G networks provides the high-throughput and low-latency communication backbone required for advanced UAV applications. It overcomes the limitations of previous wireless technologies and unlocks new capabilities. Ultra-Reliable Low-Latency Communication (URLLC): 5G’s low latency is critical for real-time command and control (C2) links, enabling precise teleoperation in complex environments and supporting vehicle-to-everything (V2X) communication, where the UAV acts as a communication relay between vehicles and infrastructure [136]. Enhanced Mobile Broadband (eMBB): The high bandwidth of 5G allows for the real-time streaming of multiple high-resolution data feeds (e.g., 4K videos, LiDAR point clouds, and thermal imagery simultaneously) to traffic management centers, providing a rich, multimodal, and real-time view of the highway environment [137].

Within the IoT paradigm, UAVs function as dynamic mobile nodes in a vast network of interconnected devices, creating a holistic and intelligent transportation system (ITS). UAVs as Mobile IoT Nodes: UAVs act as “flying sensors”, collecting and transmitting data that complement the information gathered by static IoT devices, like roadside units (RSUs) and in-pavement sensors. This creates a more complete and resilient data ecosystem. Synergy with Smart Infrastructure: The integration of UAVs with ground-based IoT infrastructure enables the creation of a “digital twin” of the highway network. In this model, real-time data from all the sources are fused to create a virtual replica of the physical environment, allowing for complex simulations, predictive maintenance scheduling, and optimized traffic flow control [138].

Together, these technologies create a powerful feedback loop: IoT sensors provide ubiquitous data, 5G ensures their instantaneous transmission, and AI delivers the intelligence to analyze them and command UAVs to act. This technological convergence is also the primary enabler of the emerging “low-altitude economy”, which refers to a variety of economic activities and industries centered around civil manned and unmanned aerial vehicles operating in low-altitude airspace (typically below 1000 m), encompassing applications from logistics and transportation to infrastructure management [139].

6.2. Future Research Recommendations

Future research should systematically address critical areas to further mature and expand the applications of UAV technology within highway transportation. First, establishing standardized data acquisition and sharing frameworks is critical. Unified, accessible, and interoperable data platforms should be developed to minimize redundancy, improve data integrity, and encourage collaborative research between transportation researchers and stakeholders. Second, standardized methodologies for UAV-based highway monitoring must be prioritized. Establishing robust protocols for sensor calibration, data preprocessing, and validation procedures will ensure replicability, enhance accuracy, and support greater adoption in diverse transportation scenarios. Finally, the future of UAV technology lies in the creation of integrated, safety- and efficiency-focused intelligent transportation ecosystems. Researchers should explore innovative mechanisms to improve real-time interactions among UAV platforms, smart roadside infrastructure, connected and autonomous vehicles (CAVs), and centralized traffic management systems. Developing proactive coordination algorithms and advanced digital twin technologies could substantially increase traffic safety, operational efficiency, and sustainability. Given the rapid growth of low-altitude airspace economies and evolving regulatory landscapes, future research must also rigorously address privacy, regulatory compliance, ethical considerations, and human factors to facilitate the responsible and beneficial integration of UAVs into society.

7. Conclusions

This review has systematically explored the applications, contributions, and challenges of state-of-the-art unmanned aerial vehicles (UAVs) in highway systems, highlighting their transformative impacts on traffic monitoring, infrastructure inspection, safety analysis, and environmental management. UAVs offer significant advantages, including dynamic mobility, extensive coverage, high-resolution data acquisition, and cost effectiveness, making them highly suitable for addressing the limitations of traditional roadside equipment and sensor networks. The integration of UAVs with advanced technologies, such as artificial intelligence (AI), the Internet of Things (IoT), and 5G, has further enhanced their ability to handle complex highway scenarios, providing enhanced real-time analytics and improved decision-making support.

However, despite these promising developments, several technical and practical challenges remain prominent. UAV data acquisition still faces critical issues resulting from adverse weather conditions, visual occlusions, and data accuracy and reliability concerns. Additionally, bottlenecks in real-time data processing and analytical complexity, especially in multisource data fusion and adaptive model development, continue to limit the full exploitation of UAV-collected information. Safety, privacy, regulatory compliance, and ethical considerations related to the deployment of UAVs also present substantial challenges, highlighting the need for robust and standardized policy frameworks to guide operations and data-sharing practices.

Looking ahead, future research should emphasize the standardization of data collection protocols and dataset creation to improve comparability and reproducibility among studies. Establishing unified platforms and standardized methodologies will foster data sharing, support collaborative research, and accelerate methodological innovations. In addition, the integration of UAVs with intelligent transportation infrastructures and emerging technologies, such as 5G, IoT, and edge computing, is a promising direction that warrants further investigation to develop fully connected, intelligent, and adaptive highway management systems. Finally, addressing ethical, regulatory, and social implications through transparent data governance mechanisms and privacy-preserving technologies will be crucial to the sustainable and responsible deployment of UAVs in transportation applications.