Urban Traffic Monitoring and Analysis Using Unmanned Aerial Vehicles (UAVs): A Systematic Literature Review

: Unmanned aerial vehicles (UAVs) are gaining considerable interest in transportation engineering in order to monitor and analyze traffic. This systematic review surveys the scientific contributions in the application of UAVs for civil engineering, especially those related to traffic monitoring. Following the PRISMA framework, 34 papers were identified in five scientific databases. First, this paper introduces previous works in this field. In addition, the selected papers were analyzed, and some conclusions were drawn to complement the findings. It can be stated that this is still a field in its infancy and that progress in advanced image processing techniques and technologies used in the construction of UAVs will lead to an explosion in the number of applications, which will result in increased benefits for society, reducing unpleasant situations, such as congestion and collisions in major urban centers of the world.


Introduction
Unmanned aerial vehicles (UAVs), commonly known as drones, are gaining considerable interest in applications such as surveillance, mapping, and remote sensing [1].The growing interest in the use of UAVs is based on many factors, including the cost of acquiring these systems, the availability of trained operators, low risk to human life, and ease of use.Due to these advantages, as well as offering a good resolution and tracking capabilities, they are starting to be used increasingly in more fields.
After they were first used in geomatics applications, providing alternatives to classical photogrammetry [2] and 3D mapping [3] to present data in a suitable format for architects and engineers [4], UAVs became commonly used tools for data acquisition.Through photogrammetry techniques and remote sensing, structure from motion (SfM) applications allow for the creation of 3D models of different objects, buildings, or areas [5].
In the last few years, UAVs have found their applicability in the field of civil engineering, especially in transportation engineering, in order to supervise and monitor traffic [6].The main benefit of traffic monitoring with UAVs is that they can be deployed to many different places where, for example, a local council may want to gather information on the use of infrastructure, such as roads, bridges, train tracks, and so on.The same goes for monitoring people and animals for conservation purposes [7,8], or, more recently, they have been used to combat the coronavirus disease (COVID-19) pandemic [9].Because of their mobility, traffic monitoring UAVs are able to collect high-resolution data, which can then be analyzed in real time.The results can then be displayed or printed, and in some cases sent to a central server or cloud for further analysis.
The growth in traffic volume and the growth of global travel makes traffic monitoring a problem of interest and a major challenge in many countries around the world.In this context, it is expected that UAVs will be an emerging solution to this challenge [10].The bird's eye-view of the camera provided by UAVs improves the traditional methodologies used in traffic monitoring [11], but the recognition and tracking of moving vehicles still remains a challenging problem, depending on the accuracy of image registration methods [12].
The use of UAVs for monitoring purpose is a relative new emerging field that requires development and validation of new solutions [13].UAVs represent a potential solution to support many aspects of the existing traffic monitoring systems such as surveillance and collision avoidance [14].In a similar way, UAVs have been applied for the monitoring of environmental parameters, e.g., air pollution, land surface temperature, flood risk, forest fire, road surface distress, land terrain monitoring etc. [15][16][17][18][19][20], but also for pedestrian traffic monitoring or disaster evacuation [21][22][23].
Currently, UAVs are very vulnerable to adverse weather conditions, such as wind, fog, and rain [24].However, there have been significant efforts to improve the robustness of the systems using different types of sensors.For instance, GPS-enabled UAVs have been shown to provide a reasonable degree of robustness and accuracy in challenging environments [25].Moreover, the use of inertial measurement units (IMUs) for UAV stabilization has received significant attention and has been applied to different types of UAVs including quadcopters [26].
In traffic monitoring, precision is needed to collect and send real-time vehicle data to traffic processing centers for efficient traffic management.This is of special significance to cities where traffic and road conditions are monitored every day.In most cases, wireless sensors are deployed on the road and connected to each other via wireless communication networks to obtain real-time traffic data within the intelligent transportation system (ITS) [27].In addition, in the case of vehicle monitoring, it is necessary to identify the speed, distance, and current location of the vehicle.In this context, UAVs have significant advantages in traffic information collection because they provide a global perspective of the road and they can obtain traffic parameters that cannot be extracted by conventional monitoring methods [28].
Traffic monitoring represents a challenge not only for police and traffic authority departments, but also for individual drivers.There is great potential in UAVs for assisting drivers in a variety of traffic-related applications, including safety, incident detection, and vehicle tracking [11].Some problems that can be solved with UAVs in the future are: traffic congestion [29,30], collision avoidance [31], safety analysis [32], and roundabout flow analysis [33].The driver assistance can be provided via UAV-to-car communication [34].
The aim of this paper is to analyze the main applications regarding the use of UAVs in traffic monitoring.A systematic literature review was conducted for this purpose and the results provide a base for future research and development in this field.The study also highlights the current surveys related to the use of UAVs in the civil engineering field.

Related Work
Research on the on the use of UAV in civil engineering related to transportation is relative limited, including several literature reviews that summarize a wide range of applications (Table 1).These studies address the following topics:


In [35], a review optimization approaches for drone operations and drone-truck combined operations in civil applications is provided.Drone operation and applications, some previous works, and issues like mathematical models, solution methods, and synchronization between a drone and a truck are presented in the study, also suggesting some possible research directions.


The recent advances of UAVs and their roles in current and future transportation systems are presented in [10].The paper summarizes the emerging technologies of UAV in transportation, highlighting performance measures, network and communications, software architecture, privacy, and security concerns.The challenges and opportunities of integrating UAVs in ITS are discussed and some potential research directions are identified in the paper.


In [36], a literature review of 111 publications related to the use of civil drones for transportation is provided.The focus is on passenger transportation drones, but applications from the urban and transportation planning fields are also reviewed.Potential problems are identified, and proposed solutions are given for different areas of application.


Emerging issues in civilian UAV usage and case studies for various fields are presented in [37], a review article that tries to analyze the potential implementations of drones in the economic system and how these implementations can be managed.


The state of the art of UAV for geomatics applications is reported in [3].The survey gives an overview of different UAV platforms, also presenting various applications, approaches, and perspectives for UAV image processing. [38] provide an extensive review of optimization approaches for the civil application of UAVs.The study addresses different aspects related to UAV operation, such as area coverage, search operations, routing, data gathering and recharging, communication links, and computing power.


In [11], the applications of UAVs in three domains of transportation (road safety, traffic monitoring, and highway infrastructure management) are reviewed.The paper discusses topics related to vision algorithms and image processing systems used in accident investigation, traffic flow analysis, and road monitoring.


An overview of advances in the vision-based condition assessment of civil infrastructure, civil infrastructure inspection, and monitoring applications is presented in [39].
The study reviews relevant findings in computer vision, machine learning, and structural engineering, highlighting some key challenges and concluding with ongoing work.


Another study [40] presents the research on using UAVs for vehicle detection by means of deep learning techniques.The work is focused on accuracy improvements and computation overhead reduction, showing similarities and differences of various techniques.


A comprehensive study focused on UAV civil applications and their challenges is presented in [12].Research trends, key challenges related to charging, collision avoidance, networking and security, and future insights are featured in the paper.


In [41], a critical review of UAVs remote sensing data processing and their application is performed, focusing on land-cover classification and change detection and discussing potential improvements and algorithmic aspects.Despite the diversity of UAV analyses in transportation-related areas, less attention has been paid to advances in traffic monitoring techniques using UAV data, which are briefly addressed in the presented studies.Thus, to our knowledge, there is no overview of the acquisition and processing of data received from UAVs in traffic monitoring applications in urban areas.There is only one slightly older conference paper that deals exclusively with this topic, presenting the advantages and disadvantages of various researches in universities and research centers [42].This paper is therefore a first attempt to review a study that strictly addresses this topic and opens the door to further research in drone monitoring applications that use various detection algorithms.Although there have been many articles investigating the use of UAVs in various fields (mining [43], architecture and urbanism [44], glacial and periglacial geomorphology [45], agriculture [46], geology [5], forest regeneration [47], water monitoring [48] etc.), there is no study yet to exclusively summarize the applications of UAVs in urban traffic monitoring and analysis.
The high percentage of drones in various applications can be seen in the large number of review studies that systematize the work in various fields corresponding to the latest technologies, such as: path planning techniques [49], computer vision algorithms [50], application of blockchain [51], swarm communication and routing protocols [52], configurations, flight mechanisms [53], optical remote sensing applications [54], communication and networking [55], regulation policies and technologies [56], mobile edge-computing for Internet of Things (IoT) applications [57], photogrammetry and remote sensing [58], deep learning approaches for road extraction [59], and advances toward future transportation.As shown in paper [49], it is expected that the percentage of UAVs in the transportation system to be 81% by the year 2022.

Materials and Methods
A systematic review of the literature covering relevant research over the last 10 years was performed.The papers were selected according to the recommendations of Systematic Review and Meta-Analysis (PRISMA) (Salameh, 2020).

Protocol and Registration
The methods and the hypothesis of the review were prepared a priori, but they were not registered on PROSPERO.

Eligibility Criteria
The papers were selected according to the following inclusion criteria: articles addressing UAV with focus on traffic monitoring or traffic analysis; articles published in English; articles published in peer-review journals; articles published from 2010 onwards; research articles.
The exclusion criteria were the following: duplicate articles; articles addressing the use of UAV in contexts other than car traffic; articles published in languages other than English; articles published before 2010; conference papers, book sections, editorial letters; reviews, conceptual papers.Articles that focused on simulations instead of using realworld data were also excluded.
Although they may provide interesting and valuable works, publications that did not meet these criteria were not included in the study in order to ensure a high-quality standard of investigation.

Information Sources
The research was carried out on five electronic databases: Scopus, Web of Science, Science Direct, IEEE Xplore, and Springer.The filtering facilities provided by the electronic databases were used to identify the items according to the eligibility criteria presented above.The search has been performed on 25 August 2021.

Search
Search terms included were: UAV, unmanned aerial vehicle, uncrewed aerial vehicle, drone, unmanned aerial system, traffic, transport, flow, road, analysis, monitoring, surveillance, management, observation, vehicle detection, congestion, urban, city, intersection.The keywords were combined with Boolean operators according to the search possibilities provided by each database.
The results obtained were exported to EndNote (Clarivate™, Philadelphia, PA, USA).Using this software, the duplicates were removed and an initial screening of titles and abstracts followed to extract relevant studies.
Figure 1 shows the author keywords visualization related to UAV use for traffic monitoring.It was obtained using VOS viewer software and it can be observed that several clusters are formed according to the author keywords: aerial vehicle, drone, traffic, image, communication.

Study Selection
The authors of this article (R.G.B. and E.V.B.) performed the search and selection of papers to be included in the study.When disagreements arose between the two reviewers, they were resolved by consensus.Papers were included in this review if they were relevant to a UAV system used for traffic surveillance purposes.

Data Extraction
Data extraction was performed independently by the two reviewers (R.G.B. and E.V.B) and disagreements were also resolved by consensus.The following data from each study were extracted: author, publication year, country, paper objective, UAV type, camera resolution, flying height, software technique, urban area, outcomes, vehicle type, main findings, future work.This information was added to Microsoft Excel (Microsoft, Redmond, WA, USA) for further analysis.

Study Selection
A flow-chart diagram showing the selection process according to PRISMA guidelines is presented in Figure 2. A total of 2557 articles were found, but 191 were duplicated papers.After their removal, 2366 studies were screened by title and abstract and 2278 were excluded because they were not relevant to our study.The remaining 88 papers were selected for full-text screening.Of these articles, 54 were excluded (no full-text available: two, magazine article: one, review article: three, no relevant to the study: 48).Finally, 34 papers were considered eligible to be included in the review.

Study Characteristics
The review process identified a total of 34 studies:  [92].
The quantitative analysis of the publications is presented in Figure 3, where the distribution of papers in terms of journal, publication year, and country is presented.This was realized online [93].As can be seen, the analyzed articles were published in top ranking journals like Automation in Construction (AC), Transportation Research: Part A (TR_A), IEEE Internet of Things (IoT) Journal and so on.Most of the articles identified have been published in Remote Sensing (six studies), then Accident Analysis & Prevention (AAP) (four studies), IEEE Access (two studies), IEEE Transactions on Intelligent Transportation Systems (two studies), Transportation Research: Part C (two studies).
Regarding the year of publication, there is a growing trend from 2014 to 2021.This is to be expected given the continuous growth of the UAVs market [94].Further, it was considered the country where the experiment was conducted or where the research center is located.As can be seen, the country that dominates overwhelmingly in terms of the number of publications on the proposed topic is China, with 16 studies, followed by Greece with three studies and Belgium, Germany, and Italy with two studies each.For one of the studies, even if the authors belong to an institution in Switzerland, the experiment described was performed in Athens, so Greece was considered as the host country.

Synthesis of Results
The synthesis of the results is provided in Table 2, where some significant data are extracted, and Table A1, where objective findings and future work for each study are summarized.

Main Purpose of the Study
The selected works were classified according to their main purpose into two main categories, as can be seen in Table 3: traffic analysis and traffic monitoring.For each of these categories, the basic objective of the study was extracted and several subcategories could be identified.As can be observed, most studies in the first category address issues related to vehicle trajectory extraction, traffic parameter estimation, congestion analysis, or conflict evaluation.In the category of articles referred to traffic monitoring, works that mainly address vehicle detection, vehicle tracking, or vehicle collision detection were included.The articles were classified in the two categories, taking into account the following criteria: in the traffic monitoring category, the studies focused only on the identification and/or tracking of vehicles in traffic, often in real time were included.In the category of traffic analysis, studies that present a more detailed analysis of certain traffic parameters, such as traffic density estimation, recognizing vehicle behavior, or assessing the risk of collisions were included.In most cases, traffic analysis studies include elements of traffic monitoring: vehicles are first detected, tracked, then further analyses are performed.The analysis, visualization, and interpretation of data obtained from UAV cameras requires intelligent processing systems [77].Thus, for the trajectory extraction of multiple vehicles, several steps are required: preprocessing, stabilization, georegistration, vehicle detection and tracking, and trajectory management [74].The first four steps are usually common for both traffic monitoring and traffic analysis applications.
Regarding the variables that were taken into account for the evaluation of the proposed system, some authors used parameters such as speed [60,64,78,85], traffic density [66,73], vehicle counting [77,92], vehicle trajectory [80,91], and parameters related to the performance of developed method: precision [79], accuracy [81], F1 score [87], correctness, completeness, and quality [84,90].In the vast majority of studies, there is no difference between the types of vehicles identified, but in some of the them, vehicles are classified in various categories, like cars, buses, trucks, motorbikes, and even pedestrians are detected in several studies.
The drones used for data acquisition are of various types, the most used being produced by DJI Technology Co., Ltd., Shenze, China, especially Phantom model (in seven studies, Phantom 2, 3, and 4 were used), Inspire 1 (three studies), Mavic Pro (two studies), Matrice 100 (two studies).Thirteen studies did not mention the UAV model.The flying height varies from 50 to 281 m, but this parameter is also reported in a few studies.The resolution of images varies from 960 × 540, 24 frames per second (fps) to 5184 × 3456.
The objective of the studies, the main findings, and the future work for each selected study are presented in Table A1 from Appendix A.

Discussion
The applications related to traffic monitoring and analysis identified in the literature review include different techniques for vehicle detection and tracking, and estimation or extraction of different traffic parameters.The variety of approaches can be divided into two categories: conventional machine vision techniques and deep learning machine vision techniques [95].The conventional motion-based methods use traditional machine learning and computer vision techniques to detect and track vehicles, e.g., background subtraction, optical flow, blob analysis [74], histogram of oriented gradient (HOG), Haar-like features, speeded-up robust features (SURF), and so on.The most recent techniques are based on deep learning and it has been shown that they outperform the traditional ones, providing better feature representation and processing time [70].There are many object detection methods based on deep learning, but they are often divided into one-stage or two-stage detectors [87].While one-stage detectors use region of interest (ROI) directly from the image, two-stage detectors use first some techniques such as region proposal network (RPN) to predict the location of potential objects [66].YOLO and SSD are CNN one-stage detectors [79], while R-CNN, Fast R-CNN, Faster R-CNN, and Mask R-CNN are two-stage detectors [70].Even if two-stage detectors are more advanced, they require high hardware performance [62].Moreover, it was shown that YOLO v3 outperforms R-CNN and runs significantly faster [78].
In the following, some aspects related to the type of UAVs used for traffic monitoring purposes will be discussed.Depending on the construction of the flying mechanism, UAVs can be classified in: fixed-wing, rotary-wing and hybrid UAV [57].Fixed-wing were prevalent for traffic monitoring applications a few years ago [75], but today small rotarywing are preferred [11] due to the fact that they are low-cost and require less experience and training [64].The first type of UAV have some advantages, e.g., increased flight endurance [53], faster travel and ability to carry heavier payloads [51], and ability to fly along linear distance [62], but they are larger in size and depend on airfield for take-off [96].Rotary-wing UAVs are lighter in weight, capable for vertical take-off and landing (VTOL), provide significant advantages in enclosed or constrained environments [43], and are capable to hover and to get very close to objectives [62], providing very high spatial resolution [64].On the other hand, they have less mobility and consume more power [51].A compromise of this types is the hybrid fixed/rotary-wing UAV, that can provide operation modes for both high speed flight and low speed flight, including hovering [80].However, all types presented are limited to climate factors, the presence of physical obstructions, as well as instrumental or legal factors [97].More details, classifications and characteristics of UAVs can be seen in [50].In the analyzed studies, the rotary-wing type predominates, especially quadcopter DJI drones (Phantom 2, Phantom 3, Phantom 4, Mavic Pro, Inspire 1 Pro, Matrice 100), Argus-One quadcopter, and hexacopters (see Table 3).This is consistent with the findings of a study showing that rotary-wing UAVs are more efficient for use in urban environments [98].
Another important issue when it comes to using UAVs for traffic monitoring in urban environments is the safety of UAV against ground vehicles.The certification of UAV operation is regulated by authorities for each state in order to avoid different situations like crashing into pedestrians or buildings, collision with other aircrafts, or disturbance [87].With the spread of drones and their types, the risk of accidents also increases [35] and that is why strict rules are needed to control UAV operations and avoid their unsafe and unnecessary use.The cooperation between all authorities is of great importance to ensure the uniformity of regulations [3].In some countries, the operation of UAVs can be performed only if the operator holds a certificate recognized by the Federal Aviation Administration (FAA) [99].In order to ensure the safety of UAV and to reduce the risk of collisions, in most cases adequate separation of people, buildings, and traffic is sufficient.Thus, some countries have imposed on UAV operators well-defined limits (i.e., 30 or 50 m) for the flight of drones to any person or structure [100].However, this separation is difficult to achieve in urban environments and different solutions were proposed for this problem: an air tunnel designed as an air tunnel for movement of UAV in areas of transport infrastructure facilities [101], risk maps to define the risk associated to accidents [102], safe landing systems able to identify obstacles [103] or to identify landing zones [104].The various techniques of safe landing zone detection are reviewed in [96].
In most of the cases, a single UAV was used in the analyzed literature for urban traffic monitoring, capturing portions of roads [60,66,72,85], intersections (one intersection [64,74,75], two [67], five [92], or ten intersections [86]), roundabouts [61,69], a toll plaza area [88,89], and so on.One single paper presented a large-scale field experiment, using observation taken by a swarm of 10 UAVs from a large congested area covering 1.3 km 2 with around 100 busy intersections [65].It is obvious that a collaborative formation of UAVs can provide faster, more effective and more flexible monitoring [55].A performance comparison of single and multi-UAV systems is provided in [52].Moreover, different solutions for traffic monitoring and management using multiple cooperative UAVs were proposed [105,106].However, there are also limitations of multi-cooperative UAV systems like 'blind' gaps, as can be seen in [76].
The low-altitude traffic management should also be taken into account when developing systems for monitoring traffic in urban areas since the flight environment in these areas is increasingly complex with the development of UAVs.Progress has been made in this regard as well through the development of public air route networks for UAVs [107] based on aerial corridor systems [108] or airspace geofencing volumization algorithms to support unmanned aircraft management of low-altitude airspace [109].Another solution is represented by a multilayer network of nodes and airways [110].Nevertheless, these aspects are not discussed in the papers selected for this analysis because they are strictly focused on describing the algorithms for traffic monitoring.As stated in [82], the requirements for real-time traffic management and control generated broad attention in the field of traffic monitoring and new frameworks for low-altitude UAV systems were developed in many countries.
In this field of urban traffic monitoring, appropriate spatial and temporal resolutions are required to capture details related to three-dimensional traffic.The spatial resolution determines the quality of an image, the smallest pixel area that can be identified [111].It represents a key aspect for determination of traffic flow parameters [66].Since the UAVs fly at lower altitudes, they can achieve high spatial resolutions [64].For instance, in [61] the spatial resolution is mentioned as having the value of 10.5 cm, in [66] it has the value of 13 cm and in [81] it is 2 cm.Compared to satellite remote sensing, that can achieve highresolution of up to 0.3 m [112], UAVs provide ultra-high spatial resolutions at cm-level.A research on the evaluation of the impact of spatial resolution on the classification of vegetation types in provided in [113].Another research on post-fire mapping and vegetation recovery highlights the advantages of UAV-based systems compared to satellite-collected imagery in terms of spatial and temporal resolutions [114].Temporal resolution is also of great importance for remote-sensing applications in urban environments because of its dynamic nature.Traffic monitoring and analysis must be performed promptly and consistently.UAV platforms are efficient in this regard, providing appropriate high temporal resolutions [112,115].However, in the analyzed studies, these parameters were reported to a very small extent.
Finally, some issues related to different laws and regulations of airspace management will be addressed.In light of the growing number of UAVs, countries around the world are trying to develop policies and means to control the operations of low-altitude aircrafts to ensure their safety and the environment in which they fly.The main regulations are related to the controlled use of airspace, operational limitations, and administrative procedures [10].There are three categories of the management of low-altitude UAVs: the registration of flight activity, the limitation of maximum flight height for different types of UAVs, setting the area for different flight activities [107].As an example, the Federal Aviation Administration of USA mention that the maximum allowable altitude is 400 ft (about 122 m) above the ground [116].The same maximum height is stated by the European Aviation Safety Agency (EASA) [117].A review of maximum flying height for different countries is provided in [56].Regarding the different zones where the UAVs are allowed to fly, there are different legal requirements defined by each country.For instance, in Belgium and the Netherlands, the use of UAVs for flying above crowds of people or urban areas, and in Canada, UAVs must not approach more than 120 m to people, animals, buildings or vehicles [118].In Romania, the Romanian Civil Aviation Authority (AACR) provides that a safety distance of 500m from buildings, people, vehicle and animals is required [119].Moreover, in order to take pictures or videos in this country, pre-approval is required from the Ministry of Defense.Other EU regulations and requirements regarding policies and authorizations can be found in [120].
Given these regulations, authorities around the world are trying to find solutions to implement a secure UAV operating environment.At the EU level, a report states that there is a need to develop and validate UAV capabilities in certain key areas, such as: urban air mobility, air traffic management, advanced services and technologies [121].Since in the future the airspace will be very crowded, as multi-purpose UAV applications are developed, it is necessary to implement policies and regulations for a safe, reliable, and efficient use of these flying vehicles, and this can be obtained by digitally sharing flight details in traffic management systems, with minimal human intervention [10].It has become obvious that there is a need for a common system to control the flight of UAVs and large aircraft, referred to by specialists as 'low altitude airspace management systems (LAAM)' [37].Moreover, future smart cities must provide the necessary infrastructure for UAV to vehicle (UAV-2-V) communication [122], which ultimately has to be adopted in the vehicle-to-everything (V2X) space, involving serious data security issues [85] and issues of processing large volumes of data.A new approach to addressing these issues is proposed in [123], a blockchain-based solution for unmanned traffic management.Beyond the current limitations and barriers of UAVs (such as reduced flight time, legal issues, lack of acceptance, economic barriers, and so on [36]), solutions must be found for optimal planning routes, the development of computer vision systems, infrastructure for processing large data sets [124], and UAV positioning algorithms [125].Some future research directions are provided in [11,12,35,56,57].

Conclusions
In this work, we provided an overview of the UAVs application for traffic monitoring and analysis.The main conclusions that can be drawn are the following:


There is a growing trend in the use of drones to monitor traffic in recent years, with a significant increase in the last three years. China has supremacy in terms of the number of applications in this field, as well as the source of data acquisition equipment (i.e., UAV models).


In terms of the construction of flying mechanisms, rotary-wing UAVs were preferred for data collection, especially quadcopters.


Various image processing methods were proposed for vehicle detection and tracking, but approaches based on deep learning have been preferred in recent years.


Most of the identified studies are based on vehicle detection and tracking techniques, but also the extraction of the trajectory of the vehicles and the evaluation or prediction of a collision.


There is a vast literature on the use of drones in various fields, but there is still much to add to traffic monitoring.This article is part of a series of those aiming to provide help to researchers and practitioners who contribute to this field.
For future work, we plan to expand the investigation and to include more studies, to add the current ones and to analyze in more detail every aspect related to the use of drones in transportation field.Obviously, this article has limitations that will be covered in a future paper.
The utilization of a UAVbased geospatial analysis technique for accurate extraction of longitudinal and lateral distances between vehicles to determine the relationship between macroscopic and microscopic parameters of traffic flow.
 Lateral gaps between vehicles are inversely related to traffic density;  Higher heterogeneity and aggressive driver behaviour, which also increases the risk of accidents;  A policy framework is needed to reduce the heterogeneity of the traffic stream and induce some discipline in the traffic stream.
 Studying the relation of traffic mix on the behaviour of fundamental diagrams and drivers' behaviour.

Apeltauer et al., 2015 [61]
A new approach for simultaneous detection and tracking of vehicles moving  The approach showed sufficient performance for automatic  [63] Address PTW (Powered Two-Wheeler) overtaking phenomena using a twostep modelling approach based on optimized and meta-optimized decision trees.
 UAV can contribute substantially towards creating detailed naturalistic trajectory datasets;  A detailed dataset is the most important step when it comes to data mining techniques and understanding a phenomenon.
 The combination of the advanced data gathering tools and ML models that can advance the design of Advanced Driver Assistance Systems (ADAS).

Barmpounakis et al., 2019 [64]
Examination of the potential of using UAVs as part of the ITS infrastructure as a way of extracting naturalistic trajectory data from aerial video footage from a low volume four-way intersection and a pedestrian passage.
 Accuracy is highly dependent on the stabilization of the video and the geo-reference procedure.
 High accuracy and fast communication protocols are required to send the information back to the ground (for example researchers, traffic centres and managers).
Barmpounakis and Geroliminis 2020 [65] Recording traffic streams in a multi-modal congested environment over an urban setting using UAS that can allow the deep investigation of critical traffic phenomena.
 Tremendous possibilities of the specific dataset to be share with the rest of the community;  It can be a benchmark dataset for both existing and future modelling approaches for several disciplines.
 This dataset can be utilized by the whole research community of transportation science and other disciplines, such as Machine Learning or Artificial Intelligence, to study, model and improve traffic congestion.

Brkić et al., 2020 [66]
Proposing a new, low-cost framework for the determination of highly accurate traffic flow parameters.
 The proposed framework provides a simple and accurate method for plotting vehicle trajectories and continuous headway measurements at road sections;  Vehicle detection achieved a recall of 0.994 and vehicle speed estimation with a MAPE was 0.92%.Proposing a novel framework for real-time traffic flow parameter estimation from aerial videos.The proposed system identifies the directions of traffic streams and extracts traffic flow parameters of each traffic stream separately.
 The experimental results show that the proposed method achieves about 96% and 87% accuracy in estimating average traffic stream speed and vehicle count, respectively;  The method also achieves a fastprocessing speed that enables real-time traffic information estimation.
 Testing the method on heavily congested traffic conditions and curved road segments, and adjusting it to improve performance would also be insightful;  Improving accuracy for estimation of large vehicles and improving the overall performances.
Khan et al., 2017 [74] Processing and analysis of UAV-acquired traffic footage.A detailed methodological framework for automated UAV video processing is proposed to extract the trajectories of multiple vehicles at a particular road segment.
 The results generated depict the overall applicability of the system.Such a systematic framework may prove to be helpful for future traffic-related UAV studies as well by streamlining the processes involved.It may also serve as a comprehensive guide for the automated and quick extraction of multivehicle trajectories from UAV-acquired data.
 The extension of the applications of the proposed framework within the context of UAV-based traffic monitoring and analysis;  The proposed framework will be extended to implement real-time processing and analysis of UAV-acquired data.vehicle trajectories' analysis, including the 'critical point' approach, will be explored in more detail;  The prospects of real-time processing and analysis of traffic data obtained via UAVs will be inspected.
Khan et al., 2020 [76] Proposing a smart traffic surveillance system based on Unmanned Aerial Vehicle (UAV) using 5G technology.
 The results show that those violations when to overcome, the number of accidents per year falls to 299,317 leading to 4868 deaths and 33,199 injuries for 1st year, and in the next five years the number of deaths and will be decreased to 3745 and injuries to 16,600 based on the current data available.
 Autodetection of lane switching, other traffic violations, and warning to drivers about these violations will promote better lane discipline among drivers in Saudi Arabia.

Kujawski and
Dudek 2021 [77] Presenting methods of data acquisition from cameras mounted on unmanned aerial vehicles (UAV) and their further analysis, which may be used to improve urban transportation systems and its sustainability.The analysed data concerning the situation of urban transport in points of intersection of national and local roads.
 Results can be used in the future together with the data from existing intelligent transportation systems (a fusion of such data will be needed);  The application of used methods allow to extend and support the existing approaches to manage public and freight transport in cities.
 Improving the power supply of flying vehicles so that it is possible to power them continuously, because currently there is only possibility to fly up to 30 min on one battery. The detailed influence needs to be investigated in order to further enhance prediction accuracy;  The interpretability and convenience of non-parametric models could be improved, which may impede the practicability compared with statistical models;  The unobserved heterogeneity should be analysis by employing the advanced modelling techniques.

Xu et al., 2016 [90]
Proposing a new hybrid vehicle detection scheme which integrates the Viola-Jones (V-J) and linear SVM classifier with HOG feature (HOG + SVM) methods for vehicle detection from lowaltitude unmanned aerial vehicle (UAV) images.
 A comprehensive evaluation shows that the switching strategy, combined with the road orientation adjustment method, can significantly improve the efficiency and effectiveness of the vehicle detection from UAV images;  The proposed vehicle detection method is competitive compared with other existing vehicle detection methods.
 Future research will be focusing on expanding the current method for detecting other transportation modes such as buses, trucks, motors, bicycles, and pedestrians.

Zhu et al., 2018a [91]
Presenting an all-in-one behaviour recognition framework for moving vehicles based on the latest deep learning techniques.
 The approach outperformed all other methods in terms of both single class performance and overall performance;  T-BiLSTM achieved an accuracy of 0.965 on the "go straight" types.
 Vehicle trajectory analysis also has more applications which we will consider in future works: for example, illegal lane changes, violations of traffic lines, overtaking in prohibited places, and illegal retrograde;  An artificial-intelligencebased transportation analytical platform can be implemented and it can be integrated into the existing intelligent transportation system in order to improve the driving experience and safety of drivers.Zhu et al., 2018b [92] Presenting an advanced urban traffic density estimation solution using the latest deep learning techniques to intelligently process ultrahigh-resolution traffic videos taken from an unmanned aerial vehicle (UAV).

Figure 1 .
Figure 1.Network visualization of author keywords co-occurrence.

Table 1 .
Related review papers on UAV application in civil engineering.

Table 2 .
Summary of information related to data acquisition and analysis.

Table 3 .
Synthesis of the results related to the main purpose of the studies.
The enhanced single shot multibox detector (Enhanced-SSD) outperforms other DNN-based techniques and the deep learning techniques are more effective than traditional computer vision techniques in traffic video analysis;  ultrahigh-resolution video provides more information that enables more accurate vehicle detection and recognition than lower resolution contents.