RASID: a secure UAV-based platform for intelligent traffic accident assessment with cryptographic verification and AI-driven analysis

Traffic accident management typically deals with delays from the time an accident is reported to the time of the actual submission of the final report, and this ultimately causes traffic congestion. The process can be done in a significantly shorter time compared to the traditional way by utilizing unmanned aerial vehicles (UAVs) in accident management, especially drones. This project aims to provide a simulation of a secure drone platform to assess vehicle traffic accidents. This approach eliminates the demand for an investigator's presence on the scene, which speeds up the process of submitting accident reports and cuts down on response time. Furthermore, the research proposes security measures to ensure the integrity and confidentiality of all gathered data by a drone in both aspects of in-transmission and storage. The common risks of gathering data by drone include unauthorized interception, access, and possible alteration of data in transmission between the drone and the ground station. The current traffic accident management mostly experiences delays between the incident reporting and final documentation, which creates a jam on the streets and ineffective response by authorities. This study introduces RASID, a secure drone-based system that aims to automate the incident assessment process, assure the integrity and confidentiality of data, and speed up reporting. The project simulates realistic drones through the employment of the AirSim tool, the authentication and encryption methods were professionally verified using ProVerif, and utilized YOLOv8-based AI models for incident investigation and automated liability assessments. High-resolution photographs of the incident scene are automatically taken by the drones, and TLS encryption is implemented to transfer the data to a secure cloud. After that, the data is encrypted with AES-256 and verified using OpenID Connect. The ProVerif results showed that messages could not be accessed or altered without authorization, proving that the exchanges among the nodes were private and authentic. The AI module achieved a precision of 0.6919, a recall of 0.6244, F1 score of 0.6564, and mAP@50 of 0.6717. It was most precise in two scenarios: rear-end and front-end collisions. The findings demonstrate that the RASID system is capable of securely collecting, transmitting, and analyzing accident data, enabling nearly real-time crash assessments. This study provides the improvements of efficiency, accuracy, and cybersecurity of traffic accident management via the integration of secure drone operations, well-known and proven encryption mechanisms, along with AI-powered analytics, when compared to the traditional crash assessment methods.

1 Introduction

The Internet of Things (IoT) is a network of various devices that use sensors to gather and share data over the Internet. Such technology develops efficiency and decision-making across multiple fields by enabling real-time data exchange from items found at work, home, or any type of vehicle (Bouzidi et al., 2017). The Unmanned Aerial Vehicles (UAVs), usually known as drones, are one of the different applications of IoT and the most trending. Drones are remote-controlled aircraft that do not require a human pilot. They are equipped with sensors and cameras that collect data from their environment and send it back to operators directly (Bouzidi et al., 2017). UAVs are highly useful tools due to their real-time communication capabilities through IoT connectivity (Boursianis et al., 2022). This makes UAVs potentially useful in a lot of sectors, including traffic accident management. Moreover, they can get to the accident location faster than humans and capture images of the accident scene (Boursianis et al., 2022). However, UAVs are becoming a target for malicious users due to the enhanced dependence on technology in emergencies and disasters. Cyber-attack risks would include the UAV losing control or disconnecting. The most common security risks that UAVs encounter include Denial-of-Service attacks (DoS), which could prevent the UAV accessible, and signal hijacking, in which the malicious intruders control the drone. Additionally, the UAV's navigation system can be tricked by GPS spoofing attacks, which might cause it to function incorrectly or affect its mission. The UAVs make an ideal target for eavesdropping and data theft because of the sensitive data they gather, such as photos and accident details (Mohsan et al., 2023). This study aims to simulate the implementation of UAVs in the accident management process by managing and documenting traffic accidents. By resolving these issues, the project seeks enhancement in both cybersecurity and operations areas, making UAVs a practical tool in the traffic accident management field.

1.1 Motivation

The existing accident management systems depend on human data collection and processing, which are vulnerable to risks such as traffic congestion, secondary incidents, and delayed response to emergencies. At the same time, the area of the UAV-integrated systems enhances new cybersecurity challenges that could not be ignored, especially considering the sensitivity of the transmitted data, such as fault percentages, accident images, and location details. Traffic congestion has raised significantly, and one of the reasons is traffic accidents. The time it takes to manage and deal with accidents is around 1–2 h. This will increase the traffic, and people will be late for their work or appointments. UAVs such as drones have different applications as they can collect data, deliver packages and surveillance, and provide liveness detection proof in case of any obligation. Our solution is to use drones in road accident management and ensure that all transmitted and stored data is protected. The main reason for implementing our solution is to ensure the privacy and secrecy of the data, to provide an effective way that can replace the traditional way of handling traffic accidents. The traditional way takes a lot of time, which increases the traffic, and it may lead to another accident. We will simulate a drone capturing photos and videos of the accident scene and securely transmitting this data to the authority using encrypted communication protocols. The simulation will incorporate authentication methods such as multi-factor authentication (MFA) and security measures such as hashing and encryption to ensure data integrity and confidentiality. Also, with the use of artificial intelligence, such as deep learning or machine learning, that can let the drone can decide the fault percentage. This makes it possible to reduce traffic, accidents, and response time. Although most studies on traffic accidents focused on the causes and mortality rate, examined the travel delays cost associated with car accidents. The research estimated a total delay of more than 700 thousand vehicle-hours in Hong Kong traffic crashes in 2021. According to the financial impact exceeded 11 million US dollars (Mohsan et al., 2023). The necessity to address the risks and ensure secure UAV operations against various cyber threats is what motivated this project. Eventually, it seeks to enhance operations automation, turning UAVs into a safe and reliable tool for traffic accident management.

1.2 Contributions

This research suggests a UAV-integrated accident management system that automates the collection, analysis, and transmission of accident data while simultaneously processing data confidentiality, integrity, and availability. Moreover, it will develop road safety by speeding up the accident response time. As a result, it will lessen traffic congestion and the financial loss brought on by delays. The main findings of this study can be listed as follows:

• Design RASID, a secure data transmission framework

By developing an effective and reliable communication protocol that guarantees the secure transmission of incident data from the drone to the authorities. Employment of privacy-preserving techniques to protect the data collected by drones, like vehicle details and an individual's personal information, from being intercepted and analyzed by intruders.

• Simulating a safe UAV-based system

Real-time implementation helps to get the data captured by drones in real-time for analysis. AirSim is utilized to design and implement RASID, a complete UAV-assisted accident assessment framework that identifies the mechanism of collecting and sending evidence securely and in real-time environment. The simulation results prove that the system is able to capture, analyse, and securely send accident data in various situations. In addition, ProVerif is used to check that the system's authentication and secure communication methods are verified. These methods include AES-256 encryption, TLS channel protection, and OpenID Connect (OIDC) authentication. The results confirm that the framework met seven verified security standards, such as protection against replay and impersonation threats, privacy, and integrity.

• AI-based accident liability assessment

Rasid is an intelligent accident analysis framework that is implemented by using YOLOv8 and Convolutional Neural Networks (CNNs) to automatically identify cars, examine collision patterns, and identify faults. The model performance is precise with an accuracy of 0.69, a recall of 0.62, F1 score of 0.65, and mAP@50 at 0.67. This shows that Rasid is effective at identifying rear-end and front-collision scenarios under different conditions.

• Cloud dashboard with authentication

Providing a scalable feature that enables the system to be deployed in various regions using cloud computing. A web dashboard is developed using Flask that allows coordinators and administrators to log in with role-based control and OIDC token validation. The interface enables real-time monitoring of UAV operations, access to AI-generated accident reports, and safe storage of encrypted evidence in the cloud. This will guarantee the data integrity and end-to-end accountability.

This study will discuss first the literature review of the current and existing studies to define the gap. Followed by the system design section and implementation of the idea. Moreover, the results of the simulation and verification of the method will be presented.

2 Literature review

2.1 Background

Unmanned Aerial Vehicles (UAVs), commonly referred to as drones, are present and used in various trending applications across multiple domains. The number of potential cybersecurity threats targeting UAVs is growing in relevance to the continuously increasing rate of their deployment. Due to drones' dependency on wireless communications and other complex technology, they inherit vulnerabilities to relative attacks such as spoofing, malware, jamming, and denial of service attacks (DoS) (Khan et al., 2020). Addressing these challenges is essential to ensure secure and reliable operations. This requires professionals to address these challenges to ensure secure and reliable operations of the drones. Through this chapter, we are exploring UAV studies, mainly the ones with a focus on security and traffic management applications. This chapter is divided into seven sub-sections, each focuses on a significant area, including cybersecurity challenges in UAVs, enhancement of security in UAV communication, the utilization of blockchain and machine learning in UAV operations, the application of UAVs in traffic accident management, system analysis in the UAV system, and lastly, gap analysis.

2.2 Cybersecurity challenges in UAVs

The UAVs' reliance on wireless connections and complex technology made them a great target for cyberattacks. Some vulnerabilities, such as the threats of jamming, spoofing, malware, and physical attacks, which interrupt the operations and expose sensitive data. Environmental factors, such as changes in weather, can further complicate UAV security and performance (Malik, 2024). To determine and reduce risks, advancements in penetration testing techniques are used. While automated tools use AI for fast vulnerability detection, methods such as Red Teaming copy real attacks. UAV systems are tested against data and communication protocol vulnerabilities using fuzzing and wireless penetration testing. In-depth assessments of security are assisted by tools such as DepthK and DRAT. Using strong encryption, like the Advanced Encryption Standard (AES), is important to protect UAV communications. Moreover, these mechanisms and other frameworks develop UAV security by mitigating risks, including denial-of-service and data hijacking (Malik, 2024). However, there are several serious challenges related to security vulnerabilities and performance that affect the communication protocols used in UAVs. Despite being widely adopted for their lightweight and scalable features, many current protocols, including MAVLink, lack security, leaving them as potential targets to numerous attacks that could have serious consequences, like unauthorized data access or guarantee control over UAV operations to intruders. On the other hand, the implementation of security measures and encryption techniques often adds overhead to the communication exchange, which results in decreasing performance. The affected applications, like military operations or search and rescue missions, are crucial, where it is essential to guarantee timely and reliable communication. To create balance between efficiency and security in UAV operations, it is necessary that smart communication protocols be developed that allow adjustment to their security levels based on mission criticality (Khan et al., 2020).

Michailidis and Vouyioukas (2022) address Internet of Drones (IoD) in regard to both software-based and hardware-based authentication mechanisms. In the context of IoD, where drone communications are extremely vulnerable to intrusion, it describes the need for secure authentication mechanisms. The authentication schemes in this study are divided into multiple categories: mutual authentication, which involves the secure data exchange between two nodes; drone authentication, through unique features like physically unclonable functions (PUF) chips or acoustic signals; user authentication through password or biometrics; and operator authentication using behavioral biometrics. To guarantee access only by authorized entities, the multiphase authentication procedure in IoD includes key generation, registration with Ground Control Stations (GCS), and dynamic management of drones within the network. A number of software-based IoD authentication solutions are covered in Michailidis and Vouyioukas (2022). Hash-based Authentication is one of the significant examples; the underlying cryptography of hash functions ensures low energy consumption and computational overhead while offering resilience against many kinds of attacks. Additionally, another significant solution is Public Key Infrastructure-based methods, which includes a sentinel lightweight mechanism that uses binary certificates offering faster authentication in comparison with traditional Transport Layer Security (TLS) protocols. Although these methods are considered to be secure, potentially malicious drones with valid certificates pose a threat. By utilizing a variety of techniques, such as radio frequency analysis and flight behavior, ML-based methods enable real-time identification of UAVs. However, in Michailidis and Vouyioukas (2022) which discusses hardware-based authentication mechanisms in IoD, several contributions have been made using various solution approaches, including Trusted Platform Module-Based (TPM), Hardware Security Module-Based (HSM), and PUF technologies. While TPM- and HSM-based techniques attempt to provide a secure platform that stores cryptographic keys and protects them from unauthorized access, PUF accomplishes the same goal by offering distinctive identities without the use of internally stored keys for cryptography.

To enable security in IoD, Yang et al. (2022) discuss some important security concerns and related solutions. This study provides an overview of the growing use of drones in various domains, such as military, agriculture, and healthcare, as well as developments that increase their vulnerability to attacks like jamming, hijacking, and tampering. It focuses on confidentiality, integrity, availability, authenticity, and privacy as IoD essential security criteria, with a particular focus on the challenges brought on by drones' restricted resources. Furthermore, many solutions will be discussed, especially the ones relevant to the potential contribution of blockchain technology to enhance IoD security (Adel and Jan, 2024). Blockchain-based authentication and data management frameworks are promising approaches for IoD networks, it structure their communication through the implementation of cryptography and distributed ledgers to be more secure, with the ultimate goal of preventing unauthorized access and attacks. Finally, the paper highlights the trade-off between security and efficiency and urges more investigation into enhancing IoD system resilience through optimized solutions, such as lightweight cryptographic algorithms and decentralized architectures. Simultaneously, a review of the current drone security solutions and their challenges, an in-depth analysis of drone anatomy, artifacts, and potential vulnerabilities and attacks, as well as an examination of proposed security solutions, is provided by a systematic literature study in Semenov et al. (2025). The analysis of the drone components allowed the authors to classify the drone's data into artifacts based on their sources, such as flight logs, GPS data, photo files, and configuration files. UAVs, communication systems, control hubs, command centers, and associated equipment are the five divisions into which the authors categorized potential drone attacks according to their attack surfaces. The study discussed some potential attacks like signal jamming, network intrusion, data interception, and eavesdropping. The authors also examined various suggested security solutions and their defense mechanisms. One of the approaches was to utilize advanced encrypting algorithms to protect against eavesdropping by encrypting signals in transit between drones and ground control stations. Moreover, for secure data storage and signal transmission, they examined other solutions based on blockchain, which maintains the security of communication protocols to prevent any unauthorized alteration. Additionally, to detect threats, a machine learning algorithm for anomaly and pattern detection through a random forest classifier was discussed.

2.3 Enhancing UAV communication security

UAVs' integration into smart city infrastructure presents enormous security challenges, in particular data transmission protection against cyber-attacks. Automatic Dependent Surveillance-Broadcast (ADS-B) system vulnerabilities were evaluated by researchers at Semenov et al. (2025). UAVs rely on ADS-B systems for exchanging location and functional information, making them vulnerable to attack. In order to hide UAV identification within ADS-B signals, steganographic algorithms are designed with Fourier transforms to enhance security and keep signal distortion at a minimum, which was confirmed by testing. Therefore, steganographic algorithms have become a viable UAV-based systems security improvement.

Two secure drone communication cryptographic methods were presented in Zhang et al. (2023), Ephemeral Diffie-Hellman over COSE (EDHOC) and Lightweight Authentication Protocol based on Elliptic Curve Cryptography (LAPEC). EDHOC lightweight cryptography made it suitable for drones and limited-resource environments. On the other hand, LAPEC addresses identity fraud prevention and backwards security and therefore ensures session key security and protection from future comprise. AES encryption with the encrypt-then-authenticate-then-translate (EAX) mode can be used to protect UAV data integrity and confidentiality. AES ensures data is properly encrypted, while EAX provides an authentication tag for its integrity. EAX will first encrypt, then authenticate, and finally translate the data. EAX capacity to handle different message lengths made it perfect for UAVs transferring data of various lengths in real-time operations (Cecchinato et al., 2023).

In another work, Khan N. A. et al. (2022) proposed a solution to enhance the MAVLink protocol security during real-time UAV and ground station communication. The proposed approach encrypts MAVLink communication through a cryptographic mechanism that utilizes a unique mapping algorithm and a reverse Caesar cipher. The suggested approach ensures transferred data safety by randomly selecting and updating serial numbers, that are used to identify the correct encryption keys, for every connection request. In addition, to further prevent attacks, each packet is encrypted uniquely. This approach yields packet interception useless for attackers without the proper decryption key. Performance test reports indicate that this technique secures MAVLink communications without affecting the protocol's efficiency.

In a similar study, a secure key-pairing authentication system was proposed by Rajkumar et al. (2023). This system tackles UAVs communication protocols challenges. Authors proposed schema aims to enhance security by incorporating elliptic curve cryptography (ECC), which provides higher security with low computational cost in comparison to other traditional methods.

Additionally, the approach is designed to enable scalability features, which is essential in IoT-based networks as the number of connected devices continues to grow. The proposed scheme consists of four main phases. First, authority control selects an elliptic curve and a collision-free hash function as well as a random integer as a privet key, which is used to compute a public key and then the authority control distributes the necessary parameters. Second, the registration phase, where participants, in this case drones must register with a trusted authority to gain access, then each participant chooses random integers as ephemeral secret elements and computes a message digest and a point on the elliptic curve, which are then sent to the AC for registration. Third, the authentication and key agreement phase. This phase ensures that participants can verify each other and create a secure key for communication. Lastly, the communication phase, where participants can securely communicate through the uses of securely established session keys (Rajkumar et al., 2023).

In addition, paper Ramos et al. (2021) analyzed the extent to which strong and user-friendly communication frameworks assist the growth of unmanned vehicle (UV) services. The manner of the users with the UV by means of communication should be natural, straightforward, and easy to configure tasks. One way is to use the TCP/IP protocols, which allow using the same network, for example, a Wi-Fi router, to communicate between UVs and their users. This enables the connection over the internet to facilitate interaction between users and vehicles, which requires connecting things like IP address support. Despite of this, to make remote communication less complex, Ramos et al. (2021) approach aims to simplify and streamline all communications between users and unmanned vehicle systems, ensuring that services can be easily integrated, regardless of the vehicle's type or geographical location. The proposed architecture tackles issues of remote interaction's critical requirements, scalability, and geographic spread. The architecture consists of three components: the platform, which is the cloud, the vehicle stations, and the users. The platform serves as a bridge to link the users to the services and manages the operational side of the vehicle and user management. The user interacts with the platform, whereas vehicle stations contain UVs for supporting the platform to allocate tasks to the UVs.

In another study Chang et al. (2023), they discuss flying base stations and mobile networking security. The Flying base station connects the traditional ground base stations of UAVs with the user by using a UAV drone that functions as a mobile station. This approach contributes to improving area coverage, especially where the infrastructure is limited. In their research, the authors investigated previous research on improving the security of UAVs, including the Medium Access Control (MAC) protocol to protect and manage resources for communication between the base station and the users, which can support preventing Denial-of-Service (DoS) attacks. Moreover, a study in, Chang et al. (2023) states that to protect communication, cryptographic algorithms such as Diffie–Hellman Key Exchange and physical-layer and radio-based authentication can be used. Furthermore, authors emphasized securing distance measurements between nodes, such as advancing distance and time-of-arrival measurements, to protect against injection attacks and guarantee the integrity of the location coordination.

Software-defined drone network (SDDN) is proposed by Kumar et al. (2022) where functionality and security are enhanced by combining several features. SDDN combines: (i) multi-layered drone movement strategies for collision avoidance and operation management in traffic monitoring situations. (ii) lightweight security protocol: to generate and encrypt communication keys. (iii) sensors and image processing techniques: for data processing to allow comprehensive traffic data collection and analysis. (iv) ad-hoc networks construction: to provide connectivity and interoperability with vehicular communication systems and provide efficient drone-to-drone data exchange. The proposed SDDN system design supports the evaluation and maintenance of critical parameters for quality of service (QoS) in real-time traffic monitoring applications such as delay and jitter.

2.4 Blockchain and machine learning in UAV operations

Blockchain technology is similar to a revolutionary system built to enhance the security and transparency of data by enabling the decentralized recording of a network of computer transactions. Every transaction is divided into blocks that are cryptographically linked, making the data immutable and resistant to any alteration. Key features of blockchain include decentralization, integrity, and transparency, enabling users to share and verify data without intermediaries (Chang et al., 2023). The use of blockchain technology and federated learning is a promising method to maintain security and efficiency in drone tasks. Recent research has approved the use of decentralized architectures that allow IoT devices to authenticate UAVs securely, mitigating the risks of illegitimate access (Alsumayt et al., 2023). Using Multi-Access Edge Computing, which brings cloud-like processing and storage closer to UAVs by using nearby edge nodes, encourages gathering data in real-time from UAVs, enhancing operational responsiveness. It has been shown that approaches such as DenseNet, a type of neural network designed to solve vanishing gradients, which is a problem where updates to earlier layers of a deep network become tiny during backpropagation, making learning hard. By connecting every layer to the next layers, DenseNet can successfully solve this issue and develop feature extraction from environmental data. Backpropagation is the behavior of changing the weights of a neural network by going backwards through it to decrease errors during training. Moreover, off-chain solutions such as Inter Planetary File System (IPFS) enable effective management of huge datasets while preserving the integrity of the blockchain. Sensitive data is kept safe due to the integration of all these technologies, which also facilitates decentralized machine learning and develops confidentiality and integrity (Alsumayt et al., 2023).

UTM-Chain is a Blockchain-Based Secure Unmanned Traffic Management for the Internet of Drones architecture. It is composed of five main elements: the number of UAVs, ground control stations (GCS), users, cloud servers, and the blockchain network. UAVs in this framework use sensors to capture various types of data, for instance, power level, speed, altitude, RGB images and thermal images. GCS has an essential role in the system; it sends commands to control the drone and receives data from it accordingly. Users in this architecture are third parties that need to access the parameters of the drones and the GCS to get useful information. The blockchain network is concerned about providing an immutable, decentralized, distributed database of all actions, commands, and collected data from the UAVs and GCS. To address UAVs' limited computing resources, a cloud server offloads intensive computations, optimizing drone operations and extending flight time. OrbitDB with IPFS is used as an off-chain solution to store large UAV data, with IPFS generating hashes to store immutable transactions on the blockchain (Allouch et al., 2021).

In this paper Hafeez et al. (2023), the stage is set for the development of effective privacy and security mechanisms within UAV networks, with special consideration for the usage of blockchain technology. The study examines the protocols that facilitate the movement of the drones from one location to another, especially D2G (Drone to Ground station) and D2S (Drone to Satellite) communications. They observed that a reliable communication means between UAVs and ground control systems (GCS) is crucial in D2G, more so for flight route surveillance and automated take-off and landing. In addition, satellite communication (D2S) is stated as critical for the out-of-sight camera systems, where ground-based systems are not in use. This aspect is particularly important during military operations and humanitarian missions in drones where network coverage is unavailable. Several layers of vulnerabilities are identified within the survey referring to UAV networks while the attention is redirected to communication level attacks. For example, physical and MAC layer threats are very prevalent in wireless D2G networks comprising UAVs, where commercial off-the-shelf Wi-Fi-based UAVs are exposed to zero-day attacks. One of the important findings in this survey is that widely used communication protocols (e.g., MAVLink) are vulnerable not only to denial of service against the transport layer but also to data injection over the network. The survey recommends using a robust transport layer protocol to defend UAV communication against these threats. Blockchain has been suggested as a potential solution for the security issues that affect UAV communication systems because of its decentralized, distributed ledger system. Given that blockchain components like cryptographic hashing, digital signatures, and consensus methods maintain the security of data integrity or privacy and resist third-party interference. The study illustrates that coupling blockchain with UAV communication eliminates multiple threats, such as jamming assaults and routing protocol weaknesses in Flying Ad-hoc Network (FANETs). Another example is the ability of blockchain to preserve public keys, which drones could use for decentralized peer-based public-key management, making them ideally equipped to operate temporarily but in high-risk dynamic environments without single points of failure. The survey provides practical measures to secure the (UAV) communication system assisted by blockchain. It is composed of designing blockchain architecture, developing smart contracts, setting up network infrastructure, implementing cryptography techniques, designing user interfaces, and performing proper system testing and evaluation. Protecting transactions of UAV data with blockchain networks requires cryptographic protocols such as AES and Rivest-Shamir-Adleman (RSA). It also provides that the blockchain authentication system for UAV networks is based on BETA-UAV, implementing secure and Proof-of-freshness or authentication protocols that enable mutual authentication in UAV environment.

The paper Kurunathan et al. (2024) presents an overview of ML techniques specially designed for UAVs. UAVs can perform tasks including picture classification, segmentation, trajectory planning, caching, scheduling, and monitoring of UAV-assisted communication when machine learning is integrated with the UAVs. This integration builds huge opportunities for real-time controlling, data collection, and processing in various fields. Many ML mechanisms used in UAV operations include both supervised and unsupervised learning. Supervised learning is essential to achieve accurate classification and segmentation by applying labeled training sets to train models, while unsupervised learning reveals patterns in unlabeled data, helping in clustering and association activities when inherent relationships within the dataset are unclear. Thus, supervised learning gives a variety of beneficial methods for UAV applications. Supervised techniques, such as Convolutional Neural Networks (CNNs), could be used to detect objects in the site and classify them accordingly. In addition, Recurrent Neural Networks (RNNs) are used to analyze videos. Furthermore, many supervised techniques can be used in capturing images and videos, extracting features from images and videos, such as speed, and making decisions based on this information. Moreover, Kurunathan et al. (2024) proposed various concerns about the security and privacy of the gathered data. The data transmissions are vulnerable to modifications and eavesdropping, and the intruders might change the trajectory of the UAV.

Khan M. A. A. et al. (2022) suggest a new method to detect road damage using deep learning and UAVs. Their design uses a deep learning model, specifically CNN, to identify and categorize road damage types. Additionally, a drone will capture an image of the road and forward it to the model. The drone will subsequently send the image with the associated data, such as time and geographic coordinates, to the web application, which serves as the third component. The web application architecture consists of three layers: the presentation layer for end-user interfaces, the application layer for handling API requests, and the data layer for information storage. Furthermore, it will regulate the user's connection to perform services like rejecting or accepting the reports, managing the drone, and monitoring it. Eventually, the model emphasizes the importance of deploying separation and control of the user interfaces; hence, the interfaces become concealed from unauthorized users with various roles.

Jha et al. (2024) propose a design that operates UAVs, deep learning image processing, and the Internet of Things (IoT); real-time transmission to cloud servers to capture potholes and speed bumps in Indian roads and alert authorities as illustrated in Figure 3. The proposed system implemented the You Only Look Once (YOLO) algorithm for image analysis and object identification. YOLO uses CNN to detect any obstacles in input pictures, only in a single shot, eliminating the need for multiple analyses of an image and demonstrating a balance between accuracy and speed. Moreover, the YOLO algorithm version applied by the authors has a lower computational cost as it is a lightweight algorithm, making it very suitable for resource-constrained hardware devices. The proposed mechanism represented an accurate detection of potholes, achieving an 85% true positive rate, effectively underscoring the effectiveness of their solution for road safety management (Jha et al., 2024).

2.5 UAVs in traffic accident management

UAVs, widely known as drones, have emerged in many applications. Their role in traffic management has enabled the authorities to gather real-time data and enhance response strategies. In highway infrastructure management, they play a crucial role in bridge inspections and distress recognition, allowing for efficient monitoring and maintenance of physical infrastructure. Additionally, UAVs are increasingly being integrated into intelligent transportation systems (ITS) to support communication networks, optimize traffic flow, and assist in urban planning. Their ability to capture high-resolution aerial imagery and process it through advanced computer vision algorithms further enhances their utility in engineering surveys and road design (Outay et al., 2020). Current studies on Urban Air Mobility (UAM) and drone safety have highlighted the importance of safe and efficient transportation systems in cities. A study (Baek and Kim, 2025) analyzes the safety features of UAM aircraft and drones, predicting crash rates from known safety indicators such as Boeing's Statistical Summary of Commercial Jet Airplane Accidents (SSCJA), ICAO's Target Level of Safety (TLS), and the UK's Risk of Travel (ROT). The findings indicate crash rates remain low, particularly with automated flights, and this supports the need for strong infrastructure and advanced collision avoidance systems.

Dronova et al. (2022) studied using UAVs in traffic monitoring to get detailed information about road accident reasons, circumstances, and conditions. To assess road conditions UAVs with GPS can be used over the route. Moreover, several sensors deployed on UAVs are used to track meteorological variables like humidity and air temperature in real-time. In addition to giving information regarding the type of incidents, when they happen, and the level of vehicle damage, specialized UAVs can identify areas that are vulnerable to traffic accidents. While advanced UAVs that capture vehicle details like make, model and license plates help law enforcement with their analytical and operational capabilities, there are worries that some of these drones' functions, like photography and video recording, may violate traffic regulations.

Meng et al. (2024) proposes a drone technology-based approach for improvement in the assessment and investigation of road accidents. Drones can be installed in advance, so that as soon as an accident is reported, traffic police may quickly deploy them to save time in the evaluation. This study distinguishes the causes of the recurrent and non-recurrent congestion of traffic. It addresses that non-recurrent congestion is usually due to some unexpected situation like an accident. It is possible to reduce the assessment time of traffic by just 1 min, thereby reducing up to about 5 min of vehicle delays. Drones can accomplish tasks often performed by traffic police in minor accidents. In more significant situations, however, they help gather vital information, such as measurements and photos, which are then sent to the appropriate authorities, such as insurance companies and traffic police. Drones must be positioned as efficiently as possible to be deployed to accident scenes quickly. The study shows increased emergency services response and efficiency by using the robust stochastic optimization (RSO) model, which focuses on drone pre-positioning, improving the traffic accident management procedure.

Saveliev et al. (2022) aim at making a road accident map using a fully autonomous UAV. It detects and keeps in view objects around the scene with full accuracy. The improved recognition techniques will allow the drone to capture the entire site of the accident for the accurate measurement of distances between important objects. Data collection can follow two kinds of trajectories: either a type that the UAV decides on a circular path based on a given central point and radius, or it can allow the user to flexibly choose areas of interest, thus taking different geometric shapes. The first type of trajectory is considered more efficient. Furthermore, the use of panoptic segmentation improves the UAV's ability to recognize objects in the scene, making the road accident map accurate.

The paper Kristiansson et al. (2024) studied how drones can save time in responding and taking live-view photos during critical incidents that need immediate medical services. Drones have shown great promise in emergency medical situations, especially in delivering crucial medical supplies during critical events such as out-of-hospital cardiac arrests (OHCA). Studies suggest that drones can arrive, on average, 3 min before ambulances, thus improving response times. Equipped with cameras, drones can capture the first images of incident scenes and transmit them to the dispatch centers to enhance situational awareness for better patient care. Indeed, one elaborate four-month-long study needed the deployment of drones in 59 of the total 440 emergency cases that resulted in a final success rate of 98% reaching the scene of identified coordinates, out of which the drones succeeded in sending preliminary scene assessments as photos back to dispatch in 20 traffic incidents. However, there were limiting factors, such as reliance on black-and-white images instead of live videos, weather conditions, and regulatory constraints on flying drones.

In a comprehensive systematic review of enhancing Road Accident Management (RAM) by employing UAVs, Gohari et al. (2023) highlight the advantages of UAV integration in RAM, which encompass gathering real-time data from accident-prone areas, traffic bypassing, and increasing the chance of saving human lives. In addition, the authors point out that many research papers utilize UAV photogrammetry to reconstruct the accident scenes in 3D models, thus effectively locating the people and debris involved in the accident. However, the review outlines the key challenges that UAV-RAM faces, such as safety threats in adverse weather conditions and inaccessible zones. Also, the authors emphasize the necessity of advanced security algorithms and techniques to ensure confidentiality and availability. While several examined papers by the authors proposed frameworks to capture high-resolution images with eliminated distortion, the vast majority are still in simulation stages, with limited real-life case studies.

2.6 Security analysis in the UAV system

A research in Mohammed et al. (2025) proposes the Enhanced UAV Forensic Framework (EUAVFF) where a modular, forensic-by-design model that uses blockchain-based audit trails, secure logging, telemetry offloading, and lightweight encryption appropriate for UAVs systems. Usage of a comprehensive literature review and a stakeholder survey (n = 100), the findings confirm notable gaps in awareness and preparedness. Up to 70% of respondents were unaware of UAV-related cyber threats and risks, and existing drones were rated poorly in critical forensic capabilities such as handling of digital evidence and tamper-proof logging and proper. Just around 28% of participants showed familiarity with drone-specific threats, underscoring a concerning lack of readiness across the field. These results emphasize the pressing need to embed forensic readiness directly into UAV system design. The proposed EUAVFF gives a structured pathway toward more secure, accountable, and resilient UAV operations, assisting organizations in navigating the growing challenges of cybersecurity and digital evidence in modern airspace environments. Another study in Omolara et al. (2023) poses a comprehensive overview of drone security and privacy challenges, testing both cyber and non-cyber threats. It classifies cybersecurity issues into nine categories—including communication, hardware, software, and physical attacks—and highlights broader concerns such as terrorism, illegal surveillance, and smuggling. Using survey data from aviation stakeholders and the public. The results illustrate that 70% of respondents were unaware of drone-related cyber risks, underscoring the urgent need for better awareness and protection. Analyses attack patterns like DDoS, spoofing, and hijacking, the paper proposes practical solutions and future research directions to enhance drone resilience and support the development of secure and cloud-based defense systems. Another study in Abiodun et al. (2022) this study identifies the challenges of tracking the origins and movement of sensitive data within cloud environments, where data often comes from various sources and changes rapidly during processing. It illustrates the role of data provenance—the process of recording data origins, usage, and updates—as a crucial tool for tracing malicious activities and identifying system vulnerabilities. Despite the fact of its importance, data provenance still a significant challenge in cloud storage, particularly in areas such as wireless sensor networks, IoT, blockchain, and digital forensics. An essential issue lies in decreasing the complexity of digital evidence and capturing volatile data before it is lost.

2.7 Comparative and gap analysis

Table 1 below presents gap analysis for each reviewed paper, highlighting its strengths and limitations. This helps find gaps in our research that our project aims to address.

Table 1

Table 1. Gap analysis.

2.7.1 Comparative analysis

Most studies, such as Malik (2024), Cecchinato et al. (2023), and Alsumayt et al. (2023), focused on risks related to UAV communications were managed by adopting various encryption and authentication methods. Different unique encryption approaches, lightweight authentication schemes, and various strong protocols have been analyzed in these research works to improve security during data transmission. Other studies, such as Khan M. A. A. et al. (2022) and Allouch et al. (2021) rely on machine learning-based methods, together with decentralized architectures. These studies did not focus on any security methods but showed distributed systems, integration of machine learning for increased communication and operational efficiency. Further, some studies used blockchain for improvement in the security and integrity of data, such as Semenov et al. (2025) and Kumar et al. (2022), while some studies, such as Semenov et al. (2025), Hafeez et al. (2023), and Kristiansson et al. (2024), focused on useful applications in accident management and ensuring road safety without solving the security issues. It has been indicated that various UAV research objectives are missing, while underlining the need to establish more comprehensive ways by integrating application and security-focused tactics.

2.7.2 Gap analysis

The reviewed papers on UAVs for accident management show gaps in comprehensive data security for transmission and storage. While some studies target specific security issues, such as communication protocols and encryption, they frequently lack comprehensive solutions for protecting sensitive data throughout the process. Our proposed framework enhances the above aspects by providing an integrated secure accident reporting and response system using drone technology. When an accident occurs, one participant in the accident will report it, and the location details will be transmitted automatically. A coordinator will receive the accident report and dispatch a drone to the provided location. At the scene of the accident, the drone captures photos and videos. The data will be sent to the cloud for secure storage using security protocols. Additionally, the drone can determine the fault percentage and send it securely with other data. After the drone finishes assessing the scene, the coordinator will be able to access the cloud and review the case.

3 System design and implementation

3.1 System name overview

Our developed system name is derived from the Arabic term RASID, which means observer or watcher. This term reflects one of our primary objectives in the project, a system that continuously monitors and analyzes traffic accidents while ensuring correct response and data security. RASID is an observer who autonomously gathers and transmits accident data using UAVs.

3.2 System architecture

The proposed system, namely RASID, is designed to automatically investigate road accidents by using Unmanned Aerial Vehicles (UAVs) with the help of a secure cloud infrastructure and a graphical dashboard interface. The system has a number of key components:

UAV: Simulated drones are used at accident scenes to acquire high-resolution images as well as relevant situational data. These drones are equipped with onboard sensors and camera modules, and their operations are simulated through AirSim, which was chosen for its ability to mimic real-time drone operations accurately. AirSim provides a realistic and flexible environment that supports complex simulations, including environmental factors such as weather and lighting, making it an ideal tool for testing and validating drone-based systems in accident assessment scenarios.

Cloud infrastructure: Information gathered is securely forwarded into a cloud-based system for storage as well as processing. The cloud system provides scalability, secure data storage, as well as effortless integration with the dashboard interface and with the modules for artificial intelligence processing.

Dashboard: A web-based interface, built using Flask, provides administrators and coordinators with real-time access to information related to accidents, the status of drones, incident reports, and outcomes of AI-based liability assessments. The interface also supports report generation, drone-related task assignment.

Authentication system: The system utilizes role-based access control, where permissions are assigned based on the role of a user, for example, admin or coordinator. OpenID Connect (OIDC) is also employed to authenticate users and drones. OIDC provides ID tokens with identity, roles, and permissions, as well as access tokens to authorize operations such as drone control, data access, and report generation. This standardized mechanism strengthens both authentication and secure communication, which guarantees system integrity and end-to-end security.

Figure 1 illustrates the relationships and workflow of the system components. The process starts when an accident is reported, the user (coordinator) receives an accident notification (step 1). Followed by activation and authentication of the drone through RASID interface (step 2). The drone then takes off to the accident scene (step 3), where it collects accident data, conduct liability assessment, and securely transmits the data to the cloud (Step 4). Eventually, the data is securely stored on the cloud to generate a final report (step 5). Later, the final report can be viewed by the coordinator and approved.

Figure 1

Figure 1. Overview of the RASID system workflow for accident response and reporting.

Thus, the steps can be summarized as follow:

1. - Intake of incidents: An accident alert appears on the dashboard for a coordinator.

2. - UAV identification and dispatch: Before mission launch, OIDC is used to activate and authenticate the UAV remotely

3. - Capture of a scene: In addition to recording imagery and scene context and (optionally) conducting an initial, on-board assessment, the UAV navigates to the supplied coordinates.

4. - Safe transfer: Evidence is sent to the cloud service via a TLS-secured channel.

5. - Analysis of AI: After processing the submitted photos, the AI module creates organized accident narratives and estimates culpability.

6. - Creation of reports: The cloud stores encrypted data and AI evaluation results, which are then made available to the dashboard for coordinator clearance and review.

The following is the sequence diagram (Figure 2) for the RASID system that describes the interaction between User, RASID Website, Drone, and Cloud Storage.

Figure 2

Figure 2. Sequence diagram of RASID system.

3.3 Security mechanisms

RASID employs a multi-layer security paradigm to ensure that data is real, correct, and private throughout the handling process:

Reciprocal verification and access control. The identities of UAV operators and the drones themselves are confirmed via OIDC. Control commands and evidence viewing are only available to confirmed and authorized individuals or objects.

Safe communication. All communications between the UAV, the cloud backend, and the dashboard are protected by Transport Layer Security (TLS), which makes sure that a man-in-the-middle cannot intercept, replay, or attack them.

Encrypted storage. To keep them safe, accident photos, operator data, and reports are kept in the cloud using AES-256 encryption.

Verifying the integrity. The hashes are compared when the reports and pictures are sent. They are hashed (for example, using SHA-based hashing) to do this. Any modifications performed during transportation can be located.

Safe exposure of the interface. Because the dashboard is hosted over HTTPS, session tokens and report information are protected while they are being examined, and credentials are prevented from leaking.

These safeguards ensure that sensitive information is not disclosed to unauthorized parties and that accident evidence remains admissible in court (i.e., can be used to refute allegations of tampering).

The privacy and access-control model used in RASID is depicted in Figure 3. Before transmitting, the UAV encrypts all of its collected imagery and uses artificial intelligence (AI)-based anonymization filters to mask personally identifiable information. Data is stored in the cloud using AES-256 encryption and transmitted via a TLS-secured channel. Only role-based users who have been authenticated by OpenID Connect (OIDC) are allowed access, and all access events are recorded and digitally signed for auditability. These safeguards guarantee that privacy is maintained at every stage of the system's lifespan, from gathering data to producing reports.

Figure 3

Figure 3. Data security workflow.

3.4 AI-based accident assessment

To automatically evaluate accident scenes and provide a preliminary responsibility estimate, RASID incorporates an AI element.

1. Data collection. Above and surrounding the collision site, the UAV takes high-resolution aerial photos.

2. Preprocessing and ingestion. Images are preprocessed (e.g., resized, normalized) and made ready for inference before being sent to the cloud.

3. Identifying objects and comprehending scenes.

- To quickly identify and locate pertinent items (vehicles, road markers, static obstructions, and pedestrians), a YOLOv8-based object detection model is employed.

- Structured feature extraction, including relative impact zones, lane placement, and vehicle orientation, is accomplished by Convolutional Neural Networks (CNNs).

4. Estimating liability. The method generates an initial estimate of fault distribution among concerned parties using rule-based mappings generated from traffic regulations and the retrieved scene elements. This estimate is not immediately enforced as final legal attribution; instead, it is shown on the dashboard for coordinator evaluation.

Dataset. A annotated dataset of accident-scene photos depicting various crash types—including rear-end, side-impact, front collision, and collisions with stationary objects—was used to train the AI model. Bounding boxes for cars and pertinent scene components (such as road limits, signage, and debris fields) were added to each image. Data augmentation techniques were used to increase diversity in lighting, weather, and viewing angle in order to improve resilience. These approaches included motion blur, weather-like noise, and brightness/contrast variations to imitate day/night situations. This enhances generalization to actual driving situations in cities.

Performance. Precision = 0.6919, recall = 0.6244, F1-score = 0.6564, and mAP@50 = 0.6717 were all attained by the integrated model. While collisions with static objects were more difficult (precision 0.5484), rear-end and front-collision scenarios had the highest per-class precision (0.7227 and 0.8331). These findings show that the model is sufficiently accurate to prioritize cases for prompt response and to support first-pass culpability assessment in typical accident types.

3.5 Simulation & verification

To ensure the security and reliability of the RASID system, both the simulation and verification approaches are combined. AirSim is a simulation platform that provides realistic flight dynamics, testing for sensor operations, image acquisition scenarios, and accident scene assessments in a controlled environment. AirSim can be directly integrated with Unreal Engine, which makes it ideal for drone simulation. ProVerif is the tool that is used for the purpose of protocol verification in order to determine the security components related to the system's authentication and the communication models. It includes the analysis of critical components like message confidentiality, integrity, and the authenticity robustness, ensuring that the OIDC-based authentication protocol and the secure communication model are well-protected from threats. It has been widely used in research for protocol verification.

4 Results

4.1 ProVerif findings

Verifying the security mechanisms in the RASID system using ProVerif was done in two aspects: secure communication and ODIC-based authentication. First, the data exchange between the drone and the server, which is encrypted using AES-256, was verified to ensure data integrity throughout the communication. In detail, the server will compare the SHA hash received with a hash calculated from the decrypted message. As a result, the server decides to start processing the message or discard an altered one. ProVerif approved the security of this encryption and integrity validation mechanism, confirming the data authenticity. Moreover, communication between the drone and the RASID system will utilize Transport Layer Security (TLS). The ProVerif results for this algorithm demonstrated that the TLS handshake establishes a secure channel, ensuring confidentiality and mutual authentication. Algorithms 1 and 2 show the tested pseudocode for communication in ProVerif.

Algorithm 1

Algorithm 1. Secure Drone Communication.

Algorithm 2

Algorithm 2. UAV Secure Handshake and Data Transfer.

Second, the authentication mechanism by submitting the Hash-based Message Authentication Code (HMAC) and OpenID Connect (OIDC) token generation process into ProVerif. Algorithm 3 shows the tested pseudocode for the authentication. The result proved that the algorithm effectively ensures valid user identity, preventing unauthorized access.

Algorithm 3

Algorithm 3. User Authentication with HMAC and OIDC.

Figure 4 shows the result from ProVerif. Where true means secures against attacks.

Figure 4

Figure 4. (a) ProVerif results of Algorithm 1; (b) ProVerif results of Algorithm 2; (c) ProVerif results of Algorithm 3.

The verification questions and results are compiled in Table 2. ProVerif demonstrated that the property is true using the conventional Dolev–Yao attacker model, as indicated by the statement “Satisfied = True.”

Table 2

Table 2. RASID secure communication and authentication verification summary by ProVerif.

The assertion that RASID is secure-by-design is reinforced by these quantitative verification results, which show that it is not only built with “secure” components but is also officially tested against active adversaries that have the ability to intercept, inject, or replay messages.

4.2 AirSim simulation result

AirSim is one of the sophisticated simulation tools for unmanned aerial vehicles developed by Microsoft that mimics accident assessment scenarios in an accurate manner. The drone is controlled using the RASID dashboard, which initiates automated operation that instructs the drone to take off and move directly toward the coordinates where the accident has taken place. When the drone reaches the accident location, the drone has the responsibility for taking high-definition images and video footage around the area and therefore collecting critical visual information from various angles. During this time, real images around the accident are sent to the cloud, where the photos would be evaluated using a machine learning model to analyze and interpret the information based on the accident determination criteria that have been prescribed by the authorities. From there, the drone flies back to the launch point. Eventually, the model analyzes the accident scene and assigns fault percentages to all the parties involved and directly reflects the percentage on the RASID dashboard. Following this approach would create a timely and accurate accident assessment method that strengthens the decision-making and response processes. We can regularly test UAV dispatch, pathing, evidence collection, and return-to-base behavior using our simulation framework without being constrained by privacy, safety, or regulatory issues that would prevent us from doing crash-site testing in the real world.

4.3 AI-based accident assessment

The YOLOv8 accident detection model and the CNNs employed to classify and extract features are implemented by the AI module of the system to analyze accidents automatically. CNNs are used to process the location of accident images and improve object detection c, whereas YOLOv8 is known for its fast speed and accuracy in real-world scene analysis. To enable liability assessment based on accident patterns, the model analyzes real-time images from the drone in simulations to identify cars, people, road boundaries, and collision areas. The training data includes thousands of labeled accidents with different lighting, weather, and road conditions, which makes the model more robust. The drone takes photos of the accident scene in simulation and sends them to the cloud to be processed. Relevant objects, like the location of cars, their proximity to traffic signs, and obvious damage, are identified by the YOLOv8 model in combination with CNNs. Based on defined rules from local traffic authorities, it then determines the percentage of fault for each party involved. The model performed well overall, achieving precision of 0.6919, recall of 0.6244, F1-score of 0.6564, and mAP@50 of 0.6717 (Figure 5). The greatest precision scores, 0.7227 and 0.8331, respectively, were obtained in rear-end and front collisions. On the other hand, static object collisions performed worse, with a precision of 0.5484. These results indicate that the model can cope with different automobile crash scenarios. For further model performance improvement and accuracy, further research will explore optimizing model performance using an even larger and diverse dataset.

Figure 5

Figure 5. (a) Overall model performance; (b) Per class performance comparison.

A multi-class accident dataset with tagged photos of four different collision types—rear-end, side-impact, front-end, and collision with static objects—was used to train the model. Bounding boxes for infrastructure, cars, and areas of obvious damage were added to each image (Figure 6). Both the collision type and contextual cues (such as which car seems to have violated right-of-way) are included in the labels. We used motion blur, viewpoint variation, rain/fog overlays, illumination shifts (day/night), and other data augmentation techniques to increase resilience. Because accident scenarios can happen at night, in bad weather, or with partial occlusion, this augmentation is required (Figure 7).

Figure 6

Figure 6. Fault analysis results.

Figure 7

Figure 7. Image analyzed by the model. Courtesy of Nikita Nikitin at Pexels https://www.pexels.com/.

Quantitative results. The model accomplished the following on the validation set:

Accuracy = 0.6919

0.6244 is the recall value.

F1-score is 0.6564.

mAP = 0.6717 at 50

Rear-end and front crashes had the highest per-class precision (0.7227 and 0.8331), suggesting good performance in typical, structured accident types. Because of less obvious vehicle orientation patterns and visual ambiguity (such as guardrails, poles, and barriers), collisions with static objects were more difficult (precision 0.5484).

Analysis of confusion. We created a confusion matrix spanning the four collision types in order to better comprehend model behavior. The majority of mistakes were made when attempting to distinguish between (i) side crashes and rear-end crashes (especially when the camera position is not exactly perpendicular to the road) and (ii) front collisions and static-object collisions (when the damaged automobile is facing a barrier). This implies that spatial context—such as lane markings, stop lines, and skid patterns—is just as crucial as object detection, which encourages future research to incorporate more scene semantics, including road topology layers from HD maps.

5 Discussion

5.1 Enhancing security & accuracy with RASID

Traditional accident investigation methods rely mainly on humans. The system currently in use is as follows: first, you contact the accident investigation organization, then you wait for the accident investigator to arrive and assess the accident, and finally, you will receive the liability report for the accident. The currently implemented system is good; however, it lacks several improvements which have been solved by RASID. First, secure image transmission UAVs will encrypt images and accident-related data using AES-256, ensuring data confidentiality. Moreover, RASID provides verification of integrity. The images are hashed using SHA, preventing tampering and ensuring authenticity. Additionally, the use of an AI model will enhance security in two ways: automated, real-time fault assessment and reducing human errors and potential biases. Lastly, ensuring that the evidence is tamper-proof by utilizing cryptographic techniques, we can ensure that accident data remains unaltered and trustworthy, which is crucial for legal and insurance purposes.

5.1.1 Performance advantage over existing approaches

RASID exhibits three main advantages over traditional or partially autonomous UAV systems:

End-to-end security verification: RASID formally verifies each cryptographic and authentication procedure using ProVerif, ensuring mathematically proven confidentiality and integrity, in contrast to previous frameworks that use encryption heuristically.

Automated fault assessment: Manual imagery interpretation is typically used in legacy UAV-based systems. By integrating AI analysis, RASID shortens the turnaround time for reports from hours to minutes.

Operational resilience: While unverified systems had integrity failures (about 6%), RASID's secure communication approach stopped 100% of simulated unwanted access attempts and eradicated all data-corruption occurrences.

All things considered, RASID enhances response efficiency and credibility by integrating formal verification, secure communication, and AI automation—elements that are rarely combined in current solutions (Table 3).

Table 3

Table 3. Trade-off analysis.

With an acceptable latency overhead of about 23%, RASID improves successful report completion from 87.5% to 99.3% and decreases integrity failures from 6.3% to 0%, according to empirical simulation results (Table 3). When compared to current UAV techniques that do not have validated security layers, these results validate the system's efficacy.

5.1.2 Scalability considerations

There are organizational and technical difficulties when RASID is scaled to the local or regional level:

• Computational Load: Several UAVs broadcasting HD images at once have the potential to overload cloud processing nodes; adaptive task scheduling and edge-AI inference are designed to reduce latency.

• Communication Bandwidth: Many drones have higher bandwidth requirements for secure TLS sessions; load balancing is being assessed using a message-queue broker and compression protocol (MQTT + gRPC).

• Multi-Drone Coordination: Prioritizing tasks and avoiding collisions are essential for central control of hundreds of UAVs. It is suggested that dynamic frequency allocation and swarm-coordination algorithms be combined to prevent channel congestion.

• Administrative Integration: RASID needs to work with current traffic authority databases and incident-response platforms via secure APIs in order to be implemented on a big scale.

5.2 Limitations & challenges

RASID faces several challenges that may affect its performance. At present, it can identify some accident types, such as side and rear-end crashes, but it may have trouble correctly identifying more complicated or unusual accidents can be harder to identify accurately. Rain, poor lighting, or blocked views are examples of environmental factors that can degrade drone image quality, which could affect AI research. Strong network coverage is also necessary for securely and swiftly transferring high-quality photos or videos to the dashboard, which isn't always possible in crowded or rural locations. The technology is susceptible to threats like GPS spoofing and signal jamming even if it employs encryption and multi-factor authentication. Drone flight routes and data collection methods are also restricted by legal and privacy requirements. Last but not least, the AI model used to evaluate culpability might not always provide a clear explanation for its conclusions, which could be problematic in circumstances involving law or insurance. These challenges highlight areas where RASID could be improved to better handle real-world conditions.

In order to evaluate the overall performance of RASID, we compared two configurations:

Baseline UAV configuration (without ProVerif/AES validation): The drone uses a regular, unconfirmed communication route to provide pictures and telemetry without using AES encryption or formal security verification.

Setting up a secure RASID (using ProVerif-verified AES/TLS + OIDC): includes OIDC authentication, TLS-based secure communication, AES-256 encryption, and ProVerif protocol verification. Table 4 illustrates a complete comparison between the two situations.

Table 4

Table 4. Comparative evaluation.

Handshake and encryption authentication add only a tiny amount of latency (approximately 23%), but they dramatically and promisingly increase system security and dependability. In test circumstances, the secure configuration successfully stopped all unauthorized access attempts while maintaining total data integrity during all transmissions.

5.2.1 Simulation-only validation limitations

Several real-world aspects are still outside the current simulation scope, even though AirSim and ProVerif offer a strong and controlled validation framework:

Environmental complexity: The fluctuating lighting, turbulence, and signal interference found in actual accident sites cannot be accurately replicated in simulations.

Network dynamics: Unlike real-world 4G/5G or ad hoc UAV networks, which suffer from jitter and delays, AirSim's communication model assumes steady latency and little packet loss.

Legal and ethical restrictions: Privacy issues, restrictions on public airspace, and bystander interactions during data collection cannot all be properly captured by simulation.

Hardware-specific failures: AirSim does not accurately mimic sensor calibration issues, GPS drift, or battery degradation.

Therefore, real-world performance measures, including network reliability, latency under congestion, and AI model behavior on real imagery, require empirical evaluation even while simulation validation validates architectural feasibility and protocol soundness.

5.3 Potential future enhancements

The current AI model handles four categories of car crash: rear-end, collision with static object, side crash, and front crash. There are however several areas where areas for future development remains. One of them is to generalize the model's ability to handle more complex situations involving accidents, including multi-car crashes. In addition, the model's robustness to different conditions of the environment, such as lighting and rough weather conditions, can be improved so that it attains higher accuracy in actual deployment. Another direction for future work is integrating the AI model with advanced traffic simulation systems, which would provide more dynamic and realistic accident situations for model training. Finally, training the model with a large and heterogeneous dataset will increase its capacity to generalize across different types of accidents as well as situations.

5.3.1 Future research: toward pilot trials in the real world

Pilot-scale field deployments will be the main focus of future research in order to validate RASID under real-world operating circumstances. Among the planned extensions are:

• Combining onboard edge-AI modules with actual UAV prototypes to carry out local inference before uploading to the cloud.

• Collaborating with local authorities to verify end-to-end latency, coverage, and evidence integrity through controlled accident scene reconstructions.

• Extending multi-drone cooperation to evaluate scalability for handling accidents in multiple urban zones at the same time.

• Use red-team simulations to assess cybersecurity resistance against actual network threats (such as jamming, man-in-the-middle attacks, and GPS spoofing).

• By bridging the gap between operational deployment and simulation-based verification, these pilot trials will allow RASID's secure communication stack and AI-driven liability assessment pipeline to be improved for practical implementation.

6 Conclusion

This study looked at a secure and intelligent drone-based system that included formal security checks, real-time data collection, and AI-powered liability analysis. Instead of using standard procedures, the RASID technology can help identify what happened at an automobile accident scene. The system's primary features include its secure design and data security. It makes use of OpenID Connect (OIDC) authentication, TLS-based secure communication, and AES-256 encryption. We used ProVerif to verify these steps and guarantee that seven critical security aspects, including mutual authentication, secrecy, and integrity, are operational and effective. The RASID system had around 23% longer access times than standard drone-based systems, but it was more secure and reliable for data. The proposed system effectively resolved all integrity concerns and stopped unauthorized access attempts. The AI model using YOLOv8 and CNNs has an accuracy of 0.69, recall of 0.62, F1 score of 0.65, and mAP@50 of 0.67. The RASID technology performed well in identifying and assessing automobile collisions from all perspectives. The ProVerif protocol verification proved the effectiveness of the implemented security measures. Moreover, AirSim-based operational simulations proved that the RASID architecture is fully functional. These findings show that it can protect secret data, secure conversations, and validate individual identities even against simulated threats.

The RASID architecture has numerous benefits in real-world application. The involved authorities, such as the police, can gather evidence from accident scenes more quickly and accurately by reducing the possibility of human errors and employing cryptographic encryption to secure digital evidence. On the other hand, insurance companies can handle claims faster and more accurately with the AI-based automated liability calculations. And this results in avoiding arguments and saving an individual's money and time. RASID integrates drone-assisted surveillance, coordinated response systems, and secure data analytics to support smart transportation programs as part of the broader smart city picture. This enhances urban transit systems' effectiveness and safety.

Future studies will focus on developing new techniques to further improve RASID's intelligence and scalability. Also, the accuracy and robustness of the AI model will be enhanced through training and testing it on numerous information, such as weather, lighting, and vehicle positions. To evaluate system performance under realistic situations, including network instability, latency, and GPS interference, local authorities will assist in conducting real-world pilot-scale drone tests. Additionally, we will investigate how blockchain technology might be used to ensure secure documentation of evidence and authentication events. This will facilitate the retrieval of data and prevent disagreements over actions taken. Furthermore, to allow individuals to make faster on-site decisions and eliminate dependency on central authority for approvals, we will investigate the integration of edge AI processing, and we will seek further developments, including the establishment of a coordination system between multiple drones for widespread deployment. And future research will focus on developing explainable AI systems for automated liability assessments in the judicial system.

In conclusion, RASID identified the feasibility of using safe AI-powered drones to identify traffic accidents in real-time and reduce traffic congestion. The integration of rigorous security inspections, automated analysis, and reliable communication represents significant progress toward creating reliable, self-sufficient, and scalable incident management systems. Such technologies will assist both traffic authorities and police, guide decisions made by insurance companies, and contribute to smart city infrastructure by making transportation systems more efficient, safe, and intelligent.

Data availability statement

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

Author contributions

AAls: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Writing – original draft, Writing – review & editing. AAlm: Formal analysis, Investigation, Supervision, Visualization, Conceptualization, Writing – original draft, Writing – review & editing. FA: Investigation, Methodology, Project administration, Resources, Supervision, Writing – original draft. HA: Conceptualization, Data curation, Formal analysis, Methodology, Supervision, Writing – original draft. LA: Conceptualization, Data curation, Methodology, Project administration, Writing – original draft. SAlm: Conceptualization, Investigation, Methodology, Project administration, Resources, Writing – original draft. ZA: Conceptualization, Investigation, Methodology, Project administration, Writing – original draft. SAlg: Conceptualization, Formal analysis, Investigation, Methodology, Resources, Writing – original draft.

Funding

The author(s) declare that financial support was received for the research and/or publication of this article. We would like to thank the SAUDI ARAMCO Cybersecurity Chair for funding this project.

Acknowledgments

We would like to thank the SAUDI ARAMCO Cybersecurity Chair for funding this project. In addition, we would like to thank the Eastern Province General Department of Traffic for their cooperation in completing the research.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declare that no Gen AI was used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Abiodun, O. I., Alawida, M., Omolara, A. E., and Alabdulatif, A. (2022). Data provenance for cloud forensic investigations, security, challenges, solutions and future perspectives: a survey. J. King Saud. Univ. Comput. Inf. Sci. 34, 10217–45. doi: 10.1016/j.jksuci.2022.10.018

Crossref Full Text | Google Scholar

Adel, A., and Jan, T. (2024). Watch the skies: a study on drone attack vectors, forensic approaches, and persisting security challenges. Future Internet 16:250. doi: 10.3390/fi16070250

Crossref Full Text | Google Scholar

Allouch, A., Cheikhrouhou, O., Koubâa, A., Toumi, K., Khalgui, M., Nguyen Gia, T., et al. (2021). UTM-Chain: blockchain-based secure unmanned traffic management for Internet of Drones. Sensors 21:3049. doi: 10.3390/s21093049

PubMed Abstract | Crossref Full Text | Google Scholar

Alsumayt, A., El-Haggar, N., Amouri, L., Alfawaer, Z. M., and Aljameel, S. S. (2023). Smart flood detection with AI and blockchain integration in Saudi Arabia using drones. Sensors 23:5148. doi: 10.3390/s23115148

PubMed Abstract | Crossref Full Text | Google Scholar

Baek, H. Y., and Kim, J. H. (2025). Prediction of urban air mobility and drone accident rates and the role of urban management systems. Urban Sci. 9:24. doi: 10.3390/urbansci9020024

Crossref Full Text | Google Scholar

Boursianis, A. D., Papadopoulou, M. S., Gotsis, A., Wan, S., Sarigiannidis, P. G., Nikolaidis, S., et al. (2022). Internet of Things (IoT) and agricultural unmanned aerial vehicles (UAVs) in smart farming: a comprehensive review. Internet Things 18:100187. doi: 10.1016/j.iot.2020.100187

Crossref Full Text | Google Scholar

Bouzidi, M., Gupta, N., Cheikh, A., Shalaginov, A., and Derawi, M. A. (2017). Novel Architectural Framework on IoT Ecosystem, Security Aspects and Mechanisms: A Comprehensive Survey. New York: IEEE Access.

Google Scholar

Cecchinato, N., Toma, A., Drioli, C., Oliva, G., Sechi, G., Foresti, G. L., et al. (2023). Secure real-time multimedia data transmission from low-cost UAVs with a lightweight AES encryption. IEEE Commun. Mag. 61, 160–165. doi: 10.1109/MCOM.001.2200611

Crossref Full Text | Google Scholar

Chang, S-. Y., Park, K., Kim, J., and Kim, J. (2023). Securing UAV flying base station for mobile networking: a review. Future Internet 15:176. doi: 10.3390/fi15050176

Crossref Full Text | Google Scholar

Dronova, O., Makhovikova, N., Kononov, S., Sedykh, D., Dobrynin, N., Polyakov, M., et al. (2022). Unmanned aerial vehicles as element of road traffic safety monitoring. Transp. Res. Procedia. 63, 2308–2314. doi: 10.1016/j.trpro.2022.06.263

Crossref Full Text | Google Scholar

Gohari, A., Hosseinzadeh, M., Noor, R. M., Razzaque, M. A., Alsaeedi, A., Gohari, F., et al. (2023). Drones for road accident management: a systematic review. IEEE Access 11, 109247–109256. doi: 10.1109/ACCESS.2023.3321142

Crossref Full Text | Google Scholar

Hafeez, S., Ullah, F., Abbas, N., Saleem, Y., Hasan, Q., Shahzad, F., et al. (2023). Blockchain-assisted UAV communication systems: a comprehensive survey. IEEE Open J. Veh. Technol. 4, 558–580. doi: 10.1109/OJVT.2023.3295208

Crossref Full Text | Google Scholar

Jha, S. K., Srivastava, R., Behuria, C., Mishra, A., Panda, S., Yadav, D. K., et al. (2024). “Road safety monitoring using deep learning and unmanned aerial vehicle,” in 2024 1st International Conference on Cognitive, Green and Ubiquitous Computing (IC-CGU) (Bhubaneswar: IEEE), 1–6.

Google Scholar

Khan, M. A. A., Akhtar, S., Rahman, M., Ahmad, M., Raza, M., Rahman, F., et al. (2022). “Road damages detection and classification using deep learning and UAVs,” in 2022 2nd Asian Conference on Innovation in Technology (ASIANCON) (Ravet: IEEE), 1–6.

Google Scholar

Khan, N. A., Jhanjhi, N. Z., Brohi, S. N., and Almazroi, A. A. (2022). A secure communication protocol for unmanned aerial vehicles. Comput. Mater. Continua. 70, 601–618. doi: 10.32604/cmc.2022.019419

Crossref Full Text | Google Scholar

Khan, N. A., Jhanjhi, N. Z., Brohi, S. N., and Nayyar, A. (2020). “Emerging use of UAV's: secure communication protocol issues and challenges,” in Drones in Smart-Citie (Amsterdam: Elsevier), 37–55.

Google Scholar

Kristiansson, M., Andersson, P., Larsson, L., Svensson, J., Almqvist, R., Sundström, A., et al. (2024). Drones can be used to provide dispatch centres with on-site photos before arrival of EMS in time critical incidents. Resuscitation 202:110312. doi: 10.1016/j.resuscitation.2024.110312

PubMed Abstract | Crossref Full Text | Google Scholar

Kumar, A., Yadav, A. S., Gill, S. S., Pervaiz, H., Ni, Q., Buyya, R. A., et al. (2022). Secure drone-to-drone communication and software defined drone network-enabled traffic monitoring system. Simul. Model Pract. Theory 120:102621. doi: 10.1016/j.simpat.2022.102621

Crossref Full Text | Google Scholar

Kurunathan, H., Huang, H., Li, K., Ni, W., and Hossain, E. (2024). Machine learning-aided operations and communications of unmanned aerial vehicles: a contemporary survey. IEEE Commun. Surv. Tutor. 26, 496–533. doi: 10.1109/COMST.2023.3312221

Crossref Full Text | Google Scholar

Malik, S. (2024). Security of unmanned aerial vehicle systems through advanced penetration testing. TechRxiv. doi: 10.36227/techrxiv.172296783.30458380/v1

Crossref Full Text | Google Scholar

Meng, Z., Zhu, N., Zhang, G., Yang, Y., Liu, Z., Ke, G. Y., et al. (2024). Data-driven drone pre-positioning for traffic accident rapid assessment. Transp. Res. E Logist. Transp. Rev. 183:103452. doi: 10.1016/j.tre.2024.103452

Crossref Full Text | Google Scholar

Michailidis, E. T., and Vouyioukas, D. A. (2022). Review on software-based and hardware-based authentication mechanisms for the Internet of Drones. Drones 6:41. doi: 10.3390/drones6020041

Crossref Full Text | Google Scholar

Mohammed, U. M., Omolara, E. O., Abiodun, O. I., Rasheed, J., Osman, O., Oyadiran, P. A., et al. (2025). Cyber threat in drone systems: bridging real-time security, legal admissibility, and digital forensic solution readiness. Front. Commun. Netw. 6:1661928. doi: 10.3389/frcmn.2025.1661928

Crossref Full Text | Google Scholar

Mohsan, S. A. H., Othman, N. Q. H., Li, Y., Alsharif, M. H., and Khan, M. A. (2023). Unmanned aerial vehicles (UAVs): practical aspects, applications, open challenges, security issues, and future trends. Intell. Serv. Robot. 16, 109–137. doi: 10.1007/s11370-022-00452-4

PubMed Abstract | Crossref Full Text | Google Scholar

Omolara, A. E., Alawida, M., and Abiodun, O. I. (2023). Drone cybersecurity issues, solutions, trend insights and future perspectives: a survey. Neural Comput. Appl. 35, 23063–23101. doi: 10.1007/s00521-023-08857-7

Crossref Full Text | Google Scholar

Outay, F., Mengash, H. A., and Adnan, M. (2020). Applications of unmanned aerial vehicle (UAV) in road safety, traffic and highway infrastructure management: recent advances and challenges. Transp. Res. Part A Policy Pract. 141, 116–129. doi: 10.1016/j.tra.2020.09.018

PubMed Abstract | Crossref Full Text | Google Scholar

Rajkumar, S. C., Jegatha, D. L., Vijayakumar, P., and Karthick, K. R. (2023). Secure session key pairing and a lightweight key authentication scheme for liable drone services. Cyber Secur. Appl. 1:100012. doi: 10.1016/j.csa.2022.100012

Crossref Full Text | Google Scholar

Ramos, J., Ribeiro, R., Safadinho, D., Barroso, J., Rabadão, C., Pereira, A., et al. (2021). Distributed architecture for unmanned vehicle services. Sensors 21:1477. doi: 10.3390/s21041477

PubMed Abstract | Crossref Full Text | Google Scholar

Saveliev, A., Lebedeva, V., Lebedev, I., and Uzdiaev, M. (2022). An approach to the automatic construction of a road accident scheme using UAV and deep learning methods. Sensors 22:4728. doi: 10.3390/s22134728

PubMed Abstract | Crossref Full Text | Google Scholar

Semenov, S., Krupska-Klimczak, M., Mazurek, P., Zhang, M., and Chernikh, O. (2025). Improving unmanned aerial vehicle security as a factor in sustainable development of smart city infrastructure: automatic dependent surveillance–broadcast (ADS-B) data protection. Sustainability 17:1553. doi: 10.3390/su17041553

Crossref Full Text | Google Scholar

Yang, W., Wang, S., Yin, X., Wang, X., and Hu, J. A. (2022). review on security issues and solutions of the Internet of Drones. IEEE Open J. Comput. Soc. 3, 96–110. doi: 10.1109/OJCS.2022.3183003

Crossref Full Text | Google Scholar

Zhang, S., Liu, Y., Han, Z., and Yang, Z. A. (2023). Lightweight authentication protocol for UAVs based on ECC scheme. Drones 7:315. doi: 10.3390/drones7050315

Crossref Full Text | Google Scholar