Involvement of surveillance drones in smart cities: A systematic review

Drones, or Unmanned Aerial Vehicles (UAVs), are among the most beneficial and emerging technologies, with a wide range of applications that can support the sustainability concerns of smart cities and ultimately improve citizens’ quality of life. The goals of this systematic review were to explore the involvement of surveillance drones in smart cities in terms of application status, application areas, proposed models, and characteristics of drones. We conducted this systematic review based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We systematically searched the Web of Science and Scopus for journal articles and conference papers written in English and published up to August 2021. Of the 323 records identified, 43 met the inclusion criteria. Findings showed that surveillance drones were used in seven distinct research fields (transportation, environment, infrastructure, object or people detection, disaster management, data collection, and other applications). Air pollution and traffic monitoring were the dominant application areas. The majority of reviewed models were based on the application of rotary-wing single-drones with the camera as the aerial sensor. Reviewed models showed that the adoption of a single or multiple UAVs, either as a stand-alone technology or integrated with other technologies (e.g., internet of things, wireless sensor networks, convolutional neural networks, artificial intelligence, machine learning, computer vision, cloud computing, web applications), can offer efficient and sustainable solutions compared to conventional surveillance methods. This review can benefit academic researchers and practitioners.


I. INTRODUCTION
Unmanned Aerial Vehicles (UAVs), or drones, are high-end cyber-physical systems (CPSs) for numerous data collection and monitoring tasks because of their capability to perform complex computations via wireless communication channels, high mobility, and automated operation [1]. UAV can act as an Internet of Things (IoT) device for data sharing, make real-time data available to be input in 'big data' applications, and act as an enabler for efficient decision-making [2]. UAVs are one of the advanced technologies that, along with the other eleven technologies, are used to manage smart cities [3]. Bouassida et al. [4] classified UAV applications into data covering, e.g., surveillance and event covering, data relaying, e.g., delivery and emergency services, and data dissemination, e.g., cartography and precise agriculture. The surveillance task is about monitoring a target, which can be a person, a group of people, behaviors, activities, air pollutants, infrastructure, or buildings, and its typical applications are border patrol, construction management, power grid inspection, traffic monitoring, environmental monitoring, etc. [5]. When compared to traditional surveillance methods, using UAVs to perform complex surveillance tasks is a more beneficial and sustainable option because they can cover large and difficult-to-access areas in a short amount of time, reducing human intervention and manpower requirements, operating during and after natural disasters, positioning in precise locations, and so on.
Population growth is occurring in several cities around the world. On this note, statistics provided by the United Nations show that the world's urban population reached 4.2 billion (55% of the world's population) in 2018 from 751 million in 1950. Furthermore, the United Nations projects that the world's urban population is expected to increase by 68% by mid-century and that there will be 43 megacities with at least 10 million people by 2030. This population's rapid surge in megacities urges city planners and municipalities to develop strategically sustainable solutions to meet the demands of increasing citizens (e.g., including infrastructure improvement and expansion, providing adequate services, generation of new jobs, etc.). In this regard, the relatively new concept of the smart city can assist city management authorities in dealing with urbanization growth issues more efficiently and sustainably. A smart city is a large and complex system (or framework) that enables the improvement of quality of life and overall safety of city residents by leveraging multiple heterogeneous technologies mainly based on Information and Communication Technologies (ICT) and IoT. Indicators of a smart city are smart government, economy, environment, people, living, and mobility. It is worth mentioning that, however, a successful smart city strategy would be more dependent on the identification and utilization of the wise methodology that could address the values, needs, and expectations of all actors and residents of the city than on the technological solutions.
A few review studies on the applications and challenges of drones in the smart city context exist in the literature. Mohammed et al. [6] and Mohammed et al. [7] reviewed UAV opportunities and related issues such as privacy, safety, and ethical use, respectively, for Dubai smart city and its general form. Vattapparamban et al. [8] presented cybersecurity, privacy, and public safety challenges of drones in future smart cities. Guvenc et al. [9] reviewed UAV's associated threats (cyber and physical) and different techniques for detection, tracking, and interdiction of malicious drones. Haouari et al. [10] presented a comparative overview of fog and cloud computing plus applications of fog computing in smart cities. Alsamhi et al. [11] comprehensively reviewed various methods and implications of collaborative UAVs and IoT for the smartness improvement of smart cities. Al-Turjman et al. [12] reviewed drone applications in Software-Defined Networking (SDN)enabled Drone Base Stations (DBS), surveillance monitoring, and emergency networks, as well as performance assessment approaches and related cybersecurity issues. The SDN is an infrastructure option for robust and secure management of multiple UAV communications. The applications, implications, challenges, and regulations of drones in smart cities were discussed by Mohamed et al. [13]. Dilshad et al. [14] surveyed the video surveillance operation of drones equipped with vision sensors for various applications for smart cities. In another study, Outay et al. [15] systematically reviewed developments in Computer Vision (CV) algorithms along with applications of drones in Intelligent Transport Systems (ITSs) and future smart cities. Alsamhi et al. [16] presented the involvement of drones in greening IoT for sustainable smart cities plus its associated potential opportunities and barriers. Finally, Pakrooh and Bohlooli [17] classified different UAV-assisted services in IoT for smart cities and discussed the design of UAV-assisted IoT systems. These studies, however, have not systematically reviewed the literature on the involvement of surveillance drones in smart cities. Conducting a systematic review enables a thorough examination of current literature and the drawing of conclusions based on previous research. This review thus aimed at exploring the applications of surveillance drones in smart cities and seeks to answer the following questions:  RQ1: What is the application status of surveillance drones in the context of smart cities?  RQ2: What application areas of surveillance drones have been addressed in the literature associated with smart cities?
 RQ3: What solution models are proposed, and what UAV characteristics are used for each application area of surveillance drones in the literature associated with smart cities?
The remainder of this review article is organized and presented as follows. Section II explains the different stages of the implemented method to select the eligible articles for this research. In Section III, we revealed the results of the bibliometric analysis to investigate the status of applications of surveillance drones in smart cities (RQ1). This Section also contains the classification of different application areas into distinct categories (RQ2). Section IV provides a comprehensive review of proposed methods and developed systems (RQ3). Section V discusses the characteristics of UAVs in terms of number, types, and aerial sensors being used (RQ3). The conclusion of the paper is described in the last section.

II. METHODOLOGY
The methodology of this review article was designed based on Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) to answer the research questions. The flow diagram of this strategy includes identification, screening, eligibility, and inclusion stages [18].
The PRISMA-based flowchart process of our systematic review is shown in Figure 1. To collect relevant academic papers in the identification stage, the Scopus and Web of Science (WoS)-clarivate analytics databases were searched on August 27, 2021. For this purpose, we used an advanced search tool and the search string in their title, abstract, and keywords. The complete used search string was ("drone*" OR "unmanned aerial vehicle*" OR "uav*" OR "unmanned aircraft system*" OR "uas*" OR "remotely piloted aircraft*" ) AND ("surveillance" OR "monitoring" OR "inspection") AND ("smart cities*" OR "smart city*"). The initial search results revealed that the respective populations of academic publications covered by the Scopus and WoS databases are 197 and 125. The initial search results of both databases showed a similar source (Mohammed et al., [6] published in 2014) as the oldest published article. Therefore, 323 records appeared as the cumulative search results of both databases from 2014 to August 27, 2021. Then, we limited the Scopus results (197 records) based on document type (article and conference paper), source type (journal and conference proceeding), and language (English). The respective results of these limitations indicated 159, 146, and 145 records. Regarding WoS search results (125 records), they were limited based on document type (article and meeting), database (WoS core collection), and language (English), which resulted in 118, 114, and 113 records, respectively. Thus, a total of 258 records (Scopus (145) and WoS (113)) remained. Checking these records showed that 92 out of 258 records were duplicated, which were then removed. Thus, 166 records were identified in the identification stage. In the screening stage, the authors checked the titles and abstracts of the 166 records and found that 72 records (survey and review (19) and not the actual application (53)) were irrelevant to this study. Thus, a total of 94 records remained. In the eligibility stage, we carefully screened the full text of all 94 articles. Among them, 51 articles addressed different technical and non-technical issues of drone operations, which are out of the scope of this review. Consequently, a total of 43 articles were included in our systematic review.

III. STATUS OF APPLICATIONS AND APPLICATION AREAS OF SURVEILLANCE DRONES IN SMART CITIES
The bibliometric analysis of all included articles was performed to demonstrate the current applications of surveillance drones in smart cities in terms of the annual number of publications, conference and journal papers, publishers, and countries. The annual number of publications  (6), followed by Spain (5) and Philippines (2), as well as Italy, the United Kingdom, Saudi Arabia, Jordan, Tunisia, Poland, Bangladesh, Thailand, Brazil, Pakistan, Taiwan, and Mexico (1 each). Studies on the use of drones for surveillance purposes in smart cities have focused on several disciplines in recent years. The authors of this review classified 43 included papers into seven distinct categories. Transportation, environment, infrastructure, object or people detection, disaster management, data collection, and others are among the categories. Notice that some articles could be considered in multiple categories. We picked the most related category among alternatives to assign a single category to each article. For instance, research into condition monitoring of road pavement is more suitable for the infrastructure category than transportation. Table 1 shows the classification and distribution of included articles based on application areas. The results showed that most of the included articles belong to the transportation and environment categories (20.9% each), followed by infrastructure (16.2%). Furthermore, included articles addressed thirty different application areas of surveillance drones in smart cities. Air pollution monitoring gained the highest population (7) among other application areas, followed by traffic monitoring (6), power line inspection and traffic safety (2 each), and each of the remaining application areas addressed in one article.

IV. APPLICATION AREA DEVELOPMENTS OF SURVEILLANCE DRONES IN SMART CITIES
This section contains our review of all 43 proposed models and consists of 7 sub-sections. The detailed information (application area, aim, research gap/problem, results, future recommendation studies, and evaluation type) of the proposed models for transportation, environment, infrastructure, object or people detection, disaster management, data collection, and other categories are respectively presented in Tables 2-8. Note that these tables are sorted based on the publication year of the articles.

A. TRANSPORTATION
Nine of the 43 (20.9%) included articles were about surveillance drone applications in transportation. Of those, 6 articles (66.6%) reported the usage of surveillance drones for traffic monitoring in smart cities. In this direction, one article addressed the concept of Cooperative Intelligent Transportation Systems (C-ITS). De Frias et al. [19] proposed a cooperative system for traffic monitoring in urban areas by implementing drones and a lightweight semantic neural network. This system enables the generation of a segmented image of the vehicles from the RGB images, the 2D real-world position of the detected vehicles, and the dissemination of information about surrounding vehicles. The evaluation results showed the applicability of the proposed model for traffic monitoring purposes since it includes a low mean error in vehicle position estimation. Two articles addressed the utilization of multiple drones. In the SwarmCity project, Roldan et al. [20] explored the effectiveness of using aerial swarm instead of placing sensors in fixed locations or in public transportation systems, which have the respective problems of blind zone production and non-controllable sensor movement, to monitor the traffic in a simulated city (SwarmCity). The results showed that aerial swarms are suitable for data collection and traffic modeling in smart cities. In a similar project, Garcia-Aunon et al. [21] proposed an approach to monitor the traffic by developing and utilizing a behavior-based surveillance algorithm that allows the controlling of multiple aerial agents (drones). They argued that the configured algorithm has acceptable performance with low standard deviation, is suitable for different traffic volumes, and is scalable. Two articles presented the integrated drone applications with web applications and CV methods. Hossain et al. [22] proposed a model to reduce the congestion on roads in urban areas by offering a path with a minimal number of vehicles to the users. In their model, drones are utilized in every junction in the area to record the real-time video from the covered path and transmit the collected information (the number of vehicles) to a local cloud server. The collected data is then transmitted to the main server, which processes the collected data from all drones to select the path with the fewest vehicles among all potential paths upon receiving a request from a user. The authors evaluated the model's performance. However, the data collection and analysis were performed using software (sky software and PTV Vissim 2020) instead of actual drones. Shirazi et al. [23] developed a deep visual tracking system for intersection monitoring (vehicle detection and tracking) using drones and CV techniques. The tracking system is also improved using the optical flow method. The authors evaluated the performance of the proposed system using Unmanned Aerial Vehicle Benchmark Object Detection and Tracking (UAVDT) video data. The outcomes showed the capability of the proposed system regarding vehicle tracking and providing traffic measurements. Furthermore, two articles have addressed traffic safety issues. Yang et al. [24] proposed a method to accurately and automatically detect traffic safety anomalies. A UAV is used to provide traffic video data, and the Functional Data Analysis (FDA) technique is implemented for safety indicator time series modeling. The authors used a rotarywing drone with an onboard camera to perform the field test of the proposed method at the signalized intersection of a roadway. They found a good separation between safetyrelated anomalies and non-anomalies. However, more research is required to validate the proposed functional method. Bouassida et al. [4] investigated the effectiveness of using a drone for improving road safety and energy efficiency in a case (car-pedestrian accident) that may happen in C-ITS and smart cities. A UAV was implemented at the scene (where there was a crosswalk, vehicle, and pedestrian) for information collection and communication with the vehicle (U2V) and infrastructure (U2I) to avoid potential accident occurrences. Their findings show that using a UAV in this situation reduces accident rates, braking distance, and occasional visibility while saving energy. In another study, Beg et al. [25] presented a system based on existing traffic policing infrastructure and autonomous UAV-enabled solutions to address the inadequacies of traditional traffic monitoring and policing tactics. Their simulated results indicate that the proposed system could effectively and intelligently handle different tasks such as congestion and accident detection, traffic rerouting, traffic light violations, autonomous emergency response, etc. Besides, one study focused on smart parking management applications. Gogoi et al. [26] proposed a system to monitor and manage parking spaces smartly and accurately. The system includes wirelessly interconnected sensors (magnetic and ultrasonic distance) assisted by a UAV equipped with a camera. The proposed system could accomplish the task with low power consumption and high precision.

B. ENVIRONMENT
Drone utilization for environmental applications in smart cities was described in 9 (20.9%) of the 43 eligible articles. Five articles addressed the measurement of airborne Particulate Matter (PM), which is the mixture of solid particles and droplets in the air. Chen et al. [27] designed a LoRa-based air quality monitor mounted on a UAV to measure PM2.5 in the air, which is an irregular and uncertain air pollutant, and a web user interface that enables the user to configure the UAV's route and view the sensed data with no delay. Their field test showed that the UAV with a LoRa-based PM2.5 sensor could successfully perform data collection and transmit the data to the server in real-time with minimum human intervention. In another study, Mayuga et al. [28] tested their aerial sensor system (which consists of a UAV with PM, temperature, and humidity sensors) to measure the PM in different elevations in high and light traffic conditions. Hu et al. [29] implemented an IoT-based 3D system to measure and monitor air quality using a combination of aerial and ground sensing. The system has different layers (sensing, transmission, processing, and presentation), and it is capable to provides information in a real-time, fine-grained, and power-efficient manner. The sensing layer of the system consists of several tiny sensors on the ground and a UAV equipped with sensors to address the issue of changing air quality from meter to meter in city areas. Yadav et al. [30] presented a system consisting of a solar-powered UAV prototype capable of performing perpetual flight and a data fusion module. They conducted a practical test by integrating the module into a UAV to obtain the PM2.5 real-time data. They also tested the module when mounted in a car to demonstrate how this low-cost module may be widely used on public transportation systems in urban areas. Finally, Gao et al. [31] presented a novel system to measure PM in urban areas. The framework of the system includes three layers (data collection, processing, and presentation), and it uses a visual approach that uses a UAV with the onboard camera to provide 360-degree aerial panoramic haze images. Evaluation of the system performance showed that high-level recognition accuracy can be achieved using a small-scale pre-training dataset. Also, deployment of the system in a large-range area can reduce its energy consumption since it benefits from the provided genetic-based location selection algorithm. Two articles addressed the measurement of multiple air pollutants. Hernández-Vega et al. [32] proposed and evaluated a prototype model to monitor different air pollutants (sulfur dioxide, nitrogen dioxide, particulate material, lead, carbon monoxide, and ozone). They used a quadcopter drone equipped with a Data Acquisition System (DAQ) that communicates with the Ground Control Station (GCS). The DAQ includes a board card that converts, filters, and amplifies the signal emitted by the sensors. The data is then transmitted from the DAQ to the GCS using Radio Frequency (RF) technology for further processing. Gu and Jia [33] proposed a consumer UAV-based air monitoring prototype that enables us to provide real-time information on multiple air pollutants. The components of the system are a UAV, a data fusion module, and two sensors mounted on the UAV platform. They performed a field test to monitor PM and nitrogen oxide pollutants. One article addressed waste incineration management. Lu et al. (2019) [34] proposed a CPS for effective dealing with waste incineration pollution. It benefits from UAVs equipped with micro-sensors for verification and post-evaluation purposes. The system enables pollution avoidance by VOLUME XX, 2017 3 controlling the whole life cycle of waste incineration from pre-treatment to pollutant emission, and it is not limited to on-site pollution monitoring. The authors argued that the adoption of this method overcomes conventional Continuous Emission Monitoring Systems (CEMS) shortcomings. The application of surveillance drones in weather monitoring is presented in Chodorek et al. [35]. They proposed the open universal system for continuous monitoring of urban areas and industrial regions based on UAV integrated with IoT measurements and Web Real-Time Communications (WebRTC), which is for transmission of reliable data and real-time video. The system architecture consists of a network that connects an air station to a ground station. The field-test results of the system showed its capability in performing the task with high temporal and spatial resolution and in poor visibility conditions.

C. INFRASTRUCTURE
The usage of surveillance drones for infrastructure fields in smart cities has been reported in 7 (16.2%) articles. Two studies addressed bridge inspections. Seo et al. [36] analyzed the effectiveness of drones for bridge inspection in terms of image quality and damage identification on a specific bridge type (three-span glued-laminated timber girder with a composite concrete deck). The method includes bridge information review, site risk assessment of surrounding areas, pre-flight setup, drone-enabled bridge inspection, and damage identification stages. The results suggested the utilization of a drone (equipped with a digital camera) and photogrammetry software for the highly detailed bridge inspection tasks. Shang et al. [37] presented a data fusion platform capable of bridge deck condition monitoring and defects detection. It is composed of the data collection unit (UAV and ground inspection), image processing unit (perspective projection and image stitching), and web-based user-interface unit (map and defects layer). The platform can fuse the multi-scale aerial and ground images to identify, describe, and locate the surface cracks and sub-surface delamination. One study addressed both bridge and pavement inspections. Wu et al. [38] proposed a framework based on the combined use of UAV and deep learning for infrastructural cracks and damages detection and classification. High-resolution pictures and infrared thermal images taken by a UAV are used to train deep Convolutional Neural Networks (CNNs). The authors have successfully applied the method on the concrete slab of a parking garage, road, and parking lots pavement. One study addressed pavement inspections. Roberts et al. [39] proposed a method for pavement distress detection by providing 3D models from images taken from the pavement's surface using UAVs and using segmentation algorithms to assess the 3D models of pavement sections. Their results from a real-case study yield that the method is a reliable, flexible, and low-cost technique for pavement condition monitoring. Two studies addressed power line inspections. Barreto et al. [40] introduced the SIAD-AERO Project for autonomous and automatic inspection and anomalies detection in different city utilities. The project is an Unmanned Aerial System (UAS) consisting of different systems (decision-making system and an advanced Artificial Intelligence (AI) system), command and control stations (mobile and portable), and aerial platforms (fixed-wing UAS and Quadcopter UAS). The impact business analysis results showed the company's capacity growth and reduction in operational time, human intervention, and labor risk. Pan et al. [41] proposed a system for automatic power lines recognition and broken strands detection based on UAV images. It includes three main processes: oversampling (to provide more training data with balanced normal and broken strand samples), image transformation (to filter numerous redundant image backgrounds), and Machine Learning (ML) algorithms (for automatic power lines recognition and broken strands detection based on UAV aerial images). Finally, building inspection is presented in Latha et al. [42]. They conducted a pilot study regarding automatic city-buildings information extraction using a UAV and cloud image processing platform. They tested the method for two types of buildings, and comparison results showed that the method has potential for urban development measure applications in smart cities.

D. OBJECT OR PEOPLE DETECTION
The deployment of surveillance drones for the object or people detection was described in 6 (13.9%) of the 43 included articles. Surinta and Khruahong [43] presented a model that can be implemented in autonomous UAVs for human tracking (face recognition) and object detection (color recognition) purposes. The components of the model are face recognition (using Haar-cascade classifier and Max-Margin Object Detection with Convolutional Neural Network (MMOD-CNN) methods), color recognition (using the HSV color space to detect the Region Of Interest (ROI)), and tracking algorithm (using image processing and ML techniques). The experimental results showed that the model can perform object detection faster than people detection and that the model is capable of small objects detection. Yadav et al. [44] proposed a system based on the application of UAV (real-time image and video capturing) and ML (object classification) for efficient management of urban green resources considering the various types of trees. The system architecture consists of training (image resizing, feature extraction, labeling, and classification), classification (classify image objects using logistic regression model), and segmentation (color assignment, post-processing, and result's improvement using boundary extraction method) phases. The outcomes of the field test confirmed the capability of the system. Also, multiclass logistic regression is the most accurate classifier. Kim et al. [45] proposed a method for the detection of obstructions (e.g., trees and berm) around runways based on UAS and visual assets (still images, point cloud models, and infrared photography). Three stages of the method are data collection of 2D images using UAS, providing a 3D point cloud model using UAS photogrammetry, and running analysis using Point Cloud Library (PCL). The field experiment showed that detailed information of obstructions around runways can be obtained using UAS-based point cloud data. Wang et al. [46] developed an efficient and robust method to improve the quality of images taken by UAVs at night-time for pedestrian detection. Image enhancement was performed using a combination of algorithms (Hyperbolic Tangent Curve (HTC) and Block-Matching and 3D filtering (BM3D)). Their comparative and experimental test showed that the performance of the proposed method is higher than other methods in terms of image quality enhancement and detection accuracy. Munyer et al. [47] developed a dataset of Foreign Object Debris (FOD) using CV, UAS, and machine learning technology to assess the potential of a lightweight automated detection system. The field-test results showed that the method detects the majority of FODs correctly. Wan et al. [48] suggested a smart system for threat detection, criminal tracking, and emergency response in urban areas.
The system consists of different working layers including a central agent, UAVs, multiple robots (several independent and intelligent agents on the ground), and a sensor network (several static distributed-connected sensors). These layers can transmit the data between and within themselves. The central agent is responsible for synthesizing the data from the layers and controlling the entire system. UAVs detect threats and transport robots to the scene. Robots collect detailed data on the scene and handle emergency and security issues. Sensors are the assistants of UAVs and robots, and they are responsible for constant area monitoring.

E. DISASTER MANAGEMENT
The surveillance drone applications for disaster management appeared in five of 43 (11.6%) articles, 2 of which described fire-related disasters. Sharma et al. [49] presented a system to detect an early-stage fire around smart cities and forests using UAVs, cloud computing, Wireless Sensor Networks (WSN), and image processing techniques. UAVs provide the region's real-time images, map, and localize the target area. The experiment conducted by the authors verified the costeffectiveness of the method for real-time data collection and monitoring and showed its ability for accurate fire occurrence detection. Zadeh et al. [50] developed a UAV design with a shooting and dropping mechanism for firefighting services in urban areas. The designed UAV can shoot and drop the fire distinguisher balls to decrease fire spread. The UAV communicates with a controller using a radio frequency transmitter, and it is equipped with a night vision camera, GPS, servo motor, and gyroscope. The field test showed a high-performance rate of the designed UAV. However, the shooting accuracy may be affected due to UAV's sensitivity to the weather condition.
Sahil and Sood [2] proposed a framework that provides an evacuation strategy and timely medical support for panicked stranded individuals after a disaster occurrence. It includes physical (IoT sensor layer, stranded individuals, and evacuation personnel) and cyber (fog and cloud layers) subsystems. The outcomes indicated the functional role of a UAV in more efficient evacuation map building. Tariq et al. [51] presented DronAID, a real-time autonomous system based on UAV technology for detecting human bodies buried under the debris as a result of natural or man-made disasters. DronAID is composed of five modules: control (microcontroller), monitor (passive infrared sensor), capture (camera), data storage (SD card), and communication (Wi-Fi). The sensor of the system detects infrared radiation generated by the human body.
Alsamhi et al. [52] presented an embedded system based on cooperation between UAVs and Internet of Public Safety Things (IoPST) devices for public safety purposes. It consists of multiple drones and Search and Rescue (SAR) responders connected via mobile ad hoc network (MANET) and Ad-hoc On-demand Distance Vector (AODV) protocol for service routing. Evaluation of the system showed its capability for providing quick connectivity and fast data communication services in a large area while keeping high QoS and lessening economic losses.

F. DATA COLLECTION
Data collection implications discussed in three articles (6.9%). UAVs can act as data collectors from different devices for various applications in smart cities. Khalifeh et al. [53] presented a framework for data collection by integrating UAVs and WSNs. In this method, WSNs are ground-static sensors responsible for data collection from a designated area. A UAV acts as a data mule by enabling covered area extension, sensing above-the-ground, collecting the collected data via WSNs, and transmitting it to a remote center. They performed three field tests with three different channel conditions between sensor and destination nodes based on two wireless sensor nodes. The results stated that highenergy, high-range wireless sensor nodes perform significantly better than cost-effective, low-range ones. Garge and Balakrishna [54] investigated how utilization of UAVs can be beneficial for high-performance needed surveillance tasks such as multimedia streaming. They found that UAVs with onboard Mobile Edge Computing (MEC) that hosts a Transmission Control Protocol (TCP) proxy are viable solutions for multimedia streaming applications in smart cities. Drones can act as flying infrastructure extenders, provide on-demand Quality of Service (QoS), and cause enduser experience improvement. Roldán-Gómez et al. [55] proposed a system of the aerial robotic swarm for city monitoring (collection of traffic, pedestrians, climate, and pollution data). The system comprised of different modules: city simulator that includes different models, drone swarm which is controlled by a behavior-based algorithm, base station that is used for data analysis and maps production using a data fusion algorithm, and operator interface for city monitoring. They have used a virtual city called SwarmCity to validate the data collection, data processing, and data presentation.

G. OTHERS
The remaining four articles presented the application of surveillance drones in farming, city and extensive ocean monitoring, 3d city modeling, and mapping. Marín et al. [56] proposed a system for urban lawn monitoring that facilitates decision-making regarding proper irrigation and planting. The system's architecture consists of a drone with a preplanned flight, controlling algorithm, and image processing. Their experimental results demonstrated the usage of the algorithm in the classification of plots based on the coverage (high, low, and very low), and the proposed system is a timeefficient method to cover a large surface of urban lawns. Kim et al. [57] presented differential frameworks using multiple heterogeneous smart UAVs capable of simultaneous detection of multiple events at a smart city (tight plane-based framework) and monitoring and seamless surveillance over an extensive ocean (loose hierarchical-based framework). In the city framework, UAV Ground Stations (UGSs) are installed in possible public areas, and current public transport systems are utilized to charge UAVs. This framework includes three steps: periodic report and event detection, scheduling workable UAVs, and scheduling returning UAVs, and it supports multiple pair communications (UGS-UGS, UGS-UAV, and UAV-UAV). Pannozzi et al. [58] presented a 3D simulated real-world environment of a city (Turin, Italy) for city monitoring simulations and modeling. The model can pilot and navigate drones with onboard sensors either through a GCS or through manual direct piloting. It is beneficial for the development and execution of case studies before undertaking field experiments. Wu et al. [59] proposed a UAV-based framework for 3D spectrum mapping. It includes pre-sampling, spectrum situation estimation, iterative Region Of Interest (ROI)-driven UAV deployment, and spectrum map recovery. The authors tested different schemes of UAV implementations (ROI-driven, random, and ROI-only) in a simulation environment, and found that the ROI-only scheme has the best performance. However, due to the high energy usage of this scheme, it can't be a good option when energy saving is the main criterion.   Application of this method on real-world experiment showed the capability of the method with more than 99% precision and accuracy *Improve this process for autoremoval of FOD after inspection *Automated UAS navigation and item retrieval *Perform FOD detection by adopting enhanced CV-UAS based system Experiment

V. CHARACTERISTICS OF IMPLEMENTED DRONES
The characteristics of the drones used in earlier studies in terms of the number of UAVs, type of UAV, and aerial sensors are presented in Table 9. These results indicate that the use of multiple UAVs appeared in about 20% of the articles. The remaining articles dealt with single-UAV applications. Fixed-wing and rotary-wing types of UAVs appeared in 2 and 33 articles, respectively, which indicates the dominance of the rotary-wing type. Cameras were the most-used aerial sensors in all categories except the environment category. Several aerial sensors for the measurement of various air pollutants were implemented. MQ7, MQ8, and MQ135 are semiconductor gas sensors for the respective detection of Carbon Monoxide (CO), Hydrogen (H2), and Carbon Dioxide (CO2) based on the temperature cycling (high and low) method. Their sensing material is Stannic Oxide (SnO2). It has low conductivity in clean air. Its conductivity rate increases when the gas concentration rises. MQ131 is a sensor for Ozone (O3) gas detection, and its sensing material is Tungsten Oxide (WO₃ ). The main advantages of these MQ sensors are high sensitivity, low cost, long lifespan, and simple drive circuits. The Adafruit HTU21D-F is a breakout board comprises of temperature and humidity sensors. It enables to measure the temperature and humidity with high accuracy of respective ± 1° C and ± 2% within the range of -30° C to 90° C and 5% to 95% humidity. However, its accuracy drops in outside of these ranges. The 4-Electrode NO2 sensor is used for urban air quality monitoring, and it measures the low-concentration nitrogen dioxide. Its benefits include accuracy, ultra-high resolution, long-term detection performance, sensitivity, and short response time. The Particulate Matter (PM) pollutant is measured by different aerial sensors: LoRa-based PM2.5, Plantower PMS3003 Laser Dust, A3-IG, OPC-N2, and OPC-R1. LoRa is a spread spectrum modulation technique derived from Chirp Spread Spectrum (CSS) technology which supports the data rate between 300 bps to 50 kbps based on the spreading factor and the channel bandwidth [60]. The LoRa PM2.5 is a sensor node for precise measurement of PM2.5 up to 600 μg/m 3 (microgram per cubic meter) with a transmission range of 2 to 5 km in urban areas. Its key advantages are low-power consumption, long-range transmission, and application in indoor and outdoor environments. Plantower's laser PM2.5 PMS3003 is a digital particle concentration sensor that operates on the basis of laser light scattering. It detects PM2.5 in the air with great precision and sends serial data (in the form of a digital interface) to the host at a time interval between 200 to 800 ms (milliseconds). Real-time reaction, in-time data correction, zero false alarm rate, and detection of particles with diameters ranging from 0.3 to 10 m (micrometers) are among its key benefits. A3-IG is the digital sensor for PM2.5 measurement. This low-cost sensor has low-power consumption, high-temperature resistance between -10° C and 85° C, a range of 6000 μg/m 3 , and measuring particles with a diameter range of 0.3 to 10 μm. OPC-N2 is the Analog PM2.5 sensor that provides PM readings and real-time particle size histograms. It detects particles with a diameter range between 0.38 to 17 μm in an environment with a temperature range of -10° C to 50° C. OPC-R1 is the digital sensor based on the laser scattering theory. It detects particles with a diameter range between 0.4 to 12.5 μm in a temperature range similar to the OPC-N2.
FLIR A651 is a thermal imaging temperature sensor based on infrared technology to detect hot spots with the underlying anomaly. The high-quality, low-noise thermal images enable easy tracking of even very small temperature changes. The sensor is suitable for process control/quality assurance, fire prevention, and condition monitoring. Ofil Swift is the ultraviolet camera to detect corona, which creates corrosive materials such as Ozone and Nitrogen oxides, on an asset. The utilization of ultraviolet sensors allows pinpointing corona and arcing, identifying the location and severity of the failure, demonstrating the emitting objects and the emitted radiation, and predicting the inevitable crisis. PointGrey Grasshopper is an RGB camera that uses Charge Coupled Device (CCD) and Complementary Metal Oxide Semiconductor (CMOS) technology to produce high-quality images. It can be used as a visual sensor to detect and locate different anomaly types.

VI. Conclusion
Drone technology is one of the most recent and advanced technologies that offers more efficient and sustainable solutions for solving numerous surveillance problems compared to conventional methods. In this direction, different drone-based methodologies for various surveillance tasks in the context of smart cities have been proposed by researchers recently. Conducting a systematic review is essential to highlight the current situation of the topic in the academic literature and analyse the relevant studies. The contribution of the current review lies in the classification of application areas of surveillance drones in the smart city research domain, investigation of the drone's integration with other technologies to perform different tasks, and identification of aerial sensors used in surveillance operations. In this work, eligible articles were classified into seven distinct categories, and complementary information, including application area, aim, research gap, outcome, future work recommendation, and validation method for each article, was provided. In addition, the characteristics of utilized UAVs in terms of number, type, and the aerial sensor, were explored. This review found that the given topic is in an early-stage status in academic research. IEEE Explore is the main outlet that publishes most of the related articles. Surveillance drones have been deployed in different application areas, including transportation, environment, infrastructure, object or people detection, disaster management, data collection, and other applications in smart cities. Transportation, environment, and infrastructure are the most attractive categories, and traffic and air pollution monitoring are the dominant application areas. Adoption of a drone (or multiple drones) either as a stand-alone technology or combined with other recent technologies (e.g., IoT, WSNs, CNNs, AI, ML, C-ITS, CV, WebRTC, cloud computing, web applications, etc.) enables us to provide real-time data and efficiently perform the complex surveillance tasks in smart cities. Also, it allows to overcome the shortcomings of conventional methods and offers cost-effective, time-saving, laborintensive, and highly accurate solutions. A minority of the proposed models were validated based on experimental and real-world scenarios, while the majority used simulation as the validation method. The real-case executions of surveillance drones in smart cities are still limited. Their large-scale use necessitates overcoming various challenges and barriers, such as operational issues, policies and legislation, and the safety and security of citizens. The rotarywing UAV is the predominant type for surveillance operations compared to the fixed-wing type. Cameras were used as aerial sensors in most of the reviewed models for visual data collection, and a few of the models were based on laser scanners. In comparison to multiple-drone applications, the majority of the examined models concentrated on singledrone applications. The collected images from the onboard cameras on UAVs can be converted into point clouds using image-based photogrammetric methods. The point clouds generation is useful in different application areas in smart cities since it enables us to provide 3D models of the area of interest. However, the precision of the 3D models depends on the processing, analyzing, and visualizing of raw point cloud data. Thus, further studies are required to investigate the impacts of these tasks on providing accurate 3D city models. Besides, this study also suggests more research on employing vision-based approaches to assess multiple air pollutants. Note that the outcomes derived from this systematic review were based on multiple limitations. Firstly, the relevant documents in any scholarly databases, except Scopus and Web of Science, were not taken into account in this review. The second is that this review is limited to those studies that propose methods for different applications of surveillance drones in the context of smart cities. Research works that address the challenges of drone operations were excluded (e.g., authentication security, drone communications modeling, path planning algorithms, cyber threats, data transfer, etc.). Thus, further review research to explore the technological challenges of drones in smart cities is required. Finally, we only considered articles and conference papers in English.