Context-Aware Identity Validation for UAV-Assisted Urban Mobility and Traffic Monitoring Environments

1  Introduction

The transport systems of cities are becoming more complex, leading to a high demand for smart, real-time, and large-scale traffic monitoring [1]. The relevant unmanned aerial vehicles (UAVs) have demonstrated the feasibility of intelligent urban mobility, as it provides deployability, flexible coverage, and an aerial view that is complementary to the road shed unit, as well as ground-based cameras and other coverage schemes [2,3]. They are highly flexible and can be applied in modern traffic monitoring applications because they can track traffic levels, detect congestion, and support rapid incident analysis.

Although these advantages apply, the entry of UAVs into urban mobility environments brings up serious security issues. The traffic monitoring procedure employs UAVs to sustain the continuous communication between UAVs, vehicles, road-view framework, and edge or cloud frameworks to solve the threats of identity spoofing, weaponized data assaults, replay assaults, Sybil assaults, and path manipulation assaults [4]. In these safety-sensitive environments, using a false identity may lead to inaccurate traffic patterns and misjudged work decisions [5]. Despite studies on authentication, privacy, and trust management, most solutions continue to conceptualize identity primarily as a cryptographic notion [6]. Consequently, they are not able to set standards for determining the consistency of any claimed identity with actual physical and environmental behavior.

To overcome this limitation, this study proposes a CIV-UAV scheme for the UAV-assisted urban mobility and traffic surveillance setting. The framework generated integrates cyber credentials with mobility behavior, visual evidence, road network limitations, temporal continuity, and joint multi-UAV observations. This makes identity validation transcend message authentication and is resolved by traffic in reality. This will enable the system to not only reject invalid credentials but also to reject valid-looking, contextually erroneous identity claims. The key value additions of this paper are:

•   This study proposes a context-aware identity validation framework, named CIV-UAV, for UAV-assisted urban mobility and traffic monitoring environments. The framework validates vehicle identities by combining cryptographic credentials with physical and environmental evidence.

•   This work develops a cross-modal trust scoring model that integrates cryptographic validity, mobility consistency, visual observation, road-network constraints, temporal continuity, anomaly behavior, and multi-UAV consensus into a unified identity validation decision.

•   The proposed framework introduces a risk-adaptive validation strategy to dynamically adjust the strictness of identity verification according to anomaly severity, contextual uncertainty, and mobility inconsistency, thereby balancing security, reliability, and computational efficiency.

•   To assess the proposed framework under realistic adversarial conditions, this study designs a traffic-specific evaluation model covering identity spoofing, Sybil attacks, trajectory forgery, fake event injection, and vision-communication mismatch attacks.

The rest of this paper is structured as follows. Section 2 provides a review of the relevant literature and identifies the research gap this study will address. Section 3 provides the context for urban mobility environments by explaining the problem setting for CIV-UAV using UAVs. Section 4 develops a proposed identity validation model and determines the key context, trust components, and decision components. Section 5 gives the performance analysis, simulation, and analysis under various attack and traffic conditions. Lastly, the paper concludes with Section 6, which outlines potential future research directions.

2  Related Work

In recent years, intelligent transportation systems have been enhanced through machine learning, visual sensing, and UAV-assisted monitoring for mobility management in urban transportation. Ref. [7] demonstrated that machine learning can be applied to aid the analysis of transport infrastructure connectivity and conflict resolution in a complex mobility environment. Similarly, traffic prediction research using visual attributes, a traffic space-time graph learning model, and heterodox traffic flow modeling has enhanced understanding of dynamic traffic states [8–10]. Flexible coverage, wide-area traffic monitoring, and real-time monitoring are also the reasons why UAV-based traffic surveillance has become important. Swin Transformer with recurrent neural networks was used by [11] to implement UAV-based intelligent traffic surveillance. The majority of traffic monitoring studies, however, have investigated the accuracy of detection, prediction, and surveillance, and identity claims of vehicles are yet to be validated in context.

The second class of studies concerns modeling, predicting, and modeling risk-aware driving behavior. However, Ref. [12] introduced a twisted Gaussian risk model based on longitudinal and lateral motion, and Ref. [13] analyzed the short-term lateral reasoning behavior based on driver preview characteristics. Ref. [14] created Trajectory Prediction by using Lane Crossing Behavior and Good driving styles. These studies are significant as identity validation for UAV-assisted traffic systems will be necessary to confirm that the claimed identity is compatible with realistic motion, lane handling, and road network restrictions. They primarily focus on prediction and planning, however, and do not specifically consider malicious attempts to manipulate identities, spoof, Sybil attacks, or mobility lies. The perception of images and the combination of different sources of information and images are also in focus for intelligent mobility systems today. Causal intervention for target-driven visual navigation has been used by Refs. [15,16] proposed a re-projection multi-task multi-sensor fusion system for autonomous driving 3D object detection and occupancy perception. Localization robustness in challenging city environments has also been recently studied in the field of GNSS/IMU and urban navigation [17–19]. These works employ various spatial, visual, and temporal consistency checks for traffic monitoring using UAVs. However, they do not involve visual observations, identity authentication in multi-UAV consensus, or anomaly scoring within a unified validation framework.

In recent research related to UAVs and aerial networks, deployment, beamforming, resource control, and task offloading have been studied in the context of advanced communications. The deployment and edge association of aerial vehicles for energy-efficient FL were presented by [20], and predictive beamforming based on radar was studied by [21] for the UAV-assisted network. Other works addressed UAV-assisted ISAC systems, full-duplex ISAC for covert communication, ISAC for trajectory control of multiple UAVs, and 6G resource allocation [22,23]. These studies demonstrate the need for UAV-assisted communication and sensing; they focus mainly on network performance, resource efficiency, and/or trajectory optimization, while context-dependent identity trustworthiness of UAVs is rather neglected. There are other studies that offer background context to context-aware traffic intelligence, namely, crash hotspot and road-risk studies. To find out road clusters and intersections with high crash rates, Refs. [24,25] applied spatial data mining and spatial weight matrices. The results of these works indicate that the structure of the road network and spatial context are important for mobility risk analysis. In the case of identity validation, it could involve determining whether it is feasible for a vehicle, given its location and trajectory, to be using a specific road segment, whether this is in line with the road at that time, and/or whether it is spatially consistent. However, these studies have not been geared toward UAV identity verification or aggressive traffic-monitoring environments.

In terms of related literature, traffic monitoring, traffic prediction, trajectory modeling, visual perception, urban navigation, and UAV-assisted communication have advanced, but three gaps remain. First, many studies treat traffic monitoring as a perception or prediction task and do not consider attacks from an identity perspective. Second, authentication-based methods only accept credentials and take no account of physical movement, appearance, travel space limitations, and time continuity. Finally, none of the current UAV-assisted systems are readily integrated to incorporate multiple adaptive trust factors such as cryptographic validation, mobility consistency, visual matching, map constraints, anomaly scoring, and multi-UAV consensus. To fill these gaps, CIV-UAV’s design is to validate a vehicle’s identity using Cyber credentials and the physical and environmental context, helping detect spoofing, Sybil attacks, trajectory forgery, fake event injection, and mismatches in vision-communication.

3  System Model

This section presents the system architecture, key elements, and mathematical model of the proposed CIV-UAV framework for UAV-enabled urban mobility and traffic monitoring scenarios. Fig. 1 illustrates the integration of UAV and vehicle layers with context-aware trust evaluation, optimization constraints, and risk-adaptive identity validation for secure decision-making. The system model of the UAV-assisted urban traffic monitoring environment is also given in this section. It specifies the key components of the system, such as vehicles, UAVs, RSUs or edge nodes, and the road network in the city, as well as the information exchanged between these components. This section aims to outline the functioning at the monitoring environment and the available observations for identity validation. The mathematical trust computation, validation, decision and optimization formulation are independently performed in Section 4.

Figure 1: Overall system context of UAV-assisted urban traffic monitoring and interaction among the vehicle layer, UAV layer, and decision-support components.

We consider an urban traffic monitoring system consisting of UAVs, vehicles, roadside units (RSUs) or edge nodes, and an urban road network. Let 𝒰={1,2,…,U} denote the set of UAVs serving as aerial sensing and validation agents, 𝒱={1,2,…,N} the set of vehicles generating mobility data and identity claims, and ℛ={1,2,…,R} the set of RSUs or edge nodes supporting validation and coordination. The urban road network is modeled as a graph:

𝒢=(ℒ,ℰ),(1)

in Eq. (1), ℒ represents road segments or lanes and ℰ denotes their connectivity. The system operates over discrete time slots t∈𝒯={1,2,…,T}. At each time slot, every vehicle i∈𝒱 periodically transmits an identity message containing its claimed identity, motion state, application data, and cryptographic proof:

Mi(t)={IDi(t),xi(t),vi(t),di(t),σi(t)},(2)

in Eq. (2), IDi(t) denotes the claimed identity or pseudonym, xi(t)=(xi(t),yi(t)) is the spatial position, vi(t) is the velocity vector, di(t) is the reported traffic-related data, and σi(t) is the attached cryptographic signature. Each UAV u∈𝒰 observes only a subset of vehicles that fall within its communication or sensing range, denoted by 𝒱u(t)⊆𝒱. Along with received identity messages, each UAV also gathers visual observations from onboard sensing devices such as cameras, represented as 𝒪u(t)={ou,1(t),ou,2(t),…}, where every observation represents an identified object with position and motion-related features. This architecture can concurrently leverage communicated identity assertions and real-time sensing data to perform reliable identity validation in UAV-assisted urban traffic monitoring conditions.

According to the system model, the identity-related communication information of every vehicle is broadcast, such as its purported identity, position, velocity, and traffic data. The UAVs receive external visual data of the area covered. The inputs from these communications and sensors are compared with spatial and motion cues to identify prominent contextual features, including cryptographic validity, mobility consistency, road-network feasibility, visual matching, temporal continuity, and cooperative UAV evidence. The features that are extracted are then put together to produce a trust score, which is used to determine whether a given identity IDi(t) will be accepted or rejected. Unlike traditional methods that rely on credential verification, the framework determines whether the reported identity matches the physical and environmental conditions there. To represent this decision, a binary validation variable is defined as follows:

δi(t)={1,if identity is accepted,0,otherwise,(3)

in Eq. (3), a value of 1 indicates acceptance, while 0 denotes rejection. The final validation yields a context-aware trust score Ti(t), calculated in the following subsection and used to validate identities. In analyzing the authenticity of stated identity, the framework uses several dimensions of context, such as cryptographic, mobility, spatial, visual, temporal, and cooperative evidence. Rather than using a single measure, identity legitimacy is evaluated using a set of complementary context scores that test the consistency between the reported identity and the measured traffic environment. Cryptographic validity verifies whether the received message has a valid digital signature:

Cicrypto(t)={1,if signature σi(t) is valid,0,otherwise.(4)

In Eq. (4), Cicrypto(t) represents the cryptographic validity score of vehicle i at time t, where a value of 1 indicates that the signature σi(t) is valid and a value of 0 indicates an invalid signature. Mobility consistency evaluates whether the reported motion follows physically feasible dynamics. It is modeled as an exponential function of the deviation between reported and predicted velocity:

Cimob(t)=exp⁡(−α1‖vi(t)−v^i(t)‖),(5)

in Eq. (5), v^i(t) is the predicted velocity and α1 controls sensitivity. Map consistency ensures that the reported position lies within valid road segments, i.e., Cimap(t)=1 if xi(t)∈ℒ and 0 otherwise, with possible extensions to lane and direction constraints. Vision-based consistency compares the reported position with UAV observations:

Civision(t)=maxo∈𝒪u(t)exp⁡(−α2‖xi(t)−xo(t)‖),(6)

in Eq. (6), α2 controls spatial matching sensitivity. Temporal consistency ensures stability of identity over time by penalizing abrupt position changes:

Citemp(t)=exp⁡(−α3‖xi(t)−xi(t−1)‖).(7)

In Eq. (7), α3 controls sensitivity to abrupt position changes between consecutive time slots. To improve robustness, multi-UAV consensus aggregates validation evidence from multiple UAVs. Finally, an anomaly score captures suspicious behavior patterns, combining trajectory and behavioral anomalies:

Ai(t)=β1Aitraj(t)+β2Aibehavior(t).(8)

In Eq. (8), Aitraj(t) and Aibehavior(t) denote trajectory-level and behavior-level anomaly scores, respectively, while β1 and β2 control their relative contributions. All these context dimensions represent identity legitimacy in totality by an amalgamation of the communication validity and mobility behavior, visual confirmation, temporal continuity, and cooperative sensing, the foundation of trust computation and decision making. Once the individual contextual components have been derived, the next step is to merge them into at single trust measure to validate identity. In the proposed framework, a total trust score is determined by incorporating aspects of cryptographic validity, mobility consistency, confirmation by vision, map consistency, multi-UAV consensus, and temporal consistency, which penalizes deviant behavior. This provides an integrated assessment of how well a purported identity can be countered by both digital authentication demands and the physical context. The overall trust score is expressed as:

Ti(t)=w1Cicrypto(t)+w2Cimob(t)+w3Civision(t)+w4Cimap(t)+w5Cicons(t)+w6Citemp(t)−w7Ai(t),(9)

in Eq. (9), w1,w2,…,w7 are weighting parameters that determine the relative importance of each context factor. The anomaly score is subtracted, and an increase in the anomaly score indicates a higher probability of suspicious or malicious behavior. To have equal contribution, the weights meet the condition of normalization ∑kwk=1. This construction of trust enables the system to move beyond stand-alone checks and to consider identity legitimacy as a product of heterogeneous evidence in its entirety. Consequently, identities with valid credentials may receive low trust scores when they fail to align with mobility, map, visual, or temporal contexts.

This is the level at which validation is so strict as to balance the false acceptance rate and the false rejection rate. An increased threshold enhances security by reducing the likelihood of accepting malicious identities, but also increases the acceptance of legitimate entities and heightens security risk. The trust score Ti(t) offers the primary view of determining whether a claimed identity is regular to the seen communication, mobility, visual, spatial, temporal, and cooperative surroundings. Nevertheless, the accept/reject rule is not determined at this point, since the level of validation strictness can vary depending on the severity of the anomalies, sensing uncertainty, and the system’s risk level. As such, a formal validation decision is developed in Section 4 as a formulation that takes an adaptive threshold. The formulation of the proposed identity validation can be cast as an optimization problem, with the objective of maximizing correct identity validation and minimizing false acceptance and false rejection, subject to trust, resource, and coverage constraints. Based on this framework, the system is trying to find an optimal set of binary validation decisions for all vehicles and time slots:

maxδi(t)∑i∈𝒱∑t∈𝒯δi(t),(10)

Eq. (10) represents the optimization objective for selecting validation decisions across all monitored vehicles and time slots. This optimization is subject to several constraints. First, the validation decision must satisfy the trust condition, meaning that an identity can only be accepted if its trust score exceeds the decision threshold:

δi(t)≤I[Ti(t)≥τ],(11)

in Eq. (11), I[⋅] denotes the indicator function. Each UAV also operates under limited computational or validation capacity. Thus, the total validation cost for all vehicles observed by UAV u at time t must remain within the available resource budget, as expressed by ∑i∈𝒱u(t)ci(t)≤Cu, where ci(t) is the validation cost of vehicle i and Cu is the resource capacity of UAV u. In addition, identity validation can only be performed for vehicles located within the observation or coverage region of a UAV, that is, i∈𝒱u(t).

This formulation integrates communication-layer authentication with physical and contextual validation into a single system for identity verification in the context of UAV-assisted traffic monitoring. The framework can also identify context-incongruent identities by introducing mobility consistency, map constraints, visual confirmation, temporal continuity, and collaborative UAV observations and identities, even when credentials appear legitimate.

4  Problem Formulation and Proposed CIV-UAV Framework

This section develops the CIV-UAV problem and gives the proposed framework. Using the system model in Section 3, the identity claims i∈𝒱 produced by each vehicle Mi(t) at time t∈𝒯 are determined by the decision variable δi(t), and the system decides whether to accept or reject the identity claim. As the method of urban traffic monitoring by UAVs involves sensing noise, occlusion, limited UAV resources, dynamic movement, and malicious attacks such as spoofing, forging trajectories, and producing false events, UAV-based identity validation cannot be performed solely using binary authentication. Thus, the problem is modelled as a multi-objective optimization problem that transforms communication and sensing evidence into a set of context scores, compands them into a trust score, and risks adaptive validation to sum up the remembrance of true acceptance to false rejection, false acceptance to the falsely rejected, latency, and resource cost. These are reflected in the objective function as follows:

maxδi(t)∑t∈𝒯∑i∈𝒱[λ1⋅TPi(t)−λ2⋅FAi(t)−λ3⋅FRi(t)],(12)

in Eq. (12), TPi(t) is the proper acceptance of a legitimate identity, FAi(t) is the improper acceptance of a malicious identity, and FRi(t) is an improper rejection of a legitimate identity. The relative weight of these components of performance is determined by the coefficients λ1, λ2, and λ3. All these weights enable this framework to trade-off security and reliability depending on the requirements of the application, which offers a principled approach to optimizing CIV-UAV platforms into new UAV-assisted schemes of urban traffic monitoring.

The offered optimization system can be influenced by a range of factors, and identities can be verified in a UAV-based city traffic control environment. Firstly, there should be a threshold condition of trust in the acceptance or rejection of an asserted identity. The validation decision depends on whether the computed trust score exceeds a predefined threshold that may vary according to system state or risk level:

δi(t)={1,Ti(t)≥τ(t),0,otherwise,(13)

in Eq. (13), δi(t) denotes the final validation decision, Ti(t) is the context-aware trust score, and τ(t) is the adaptive trust threshold adjusted according to contextual uncertainty and anomaly severity. Each UAV is also constrained by limited computational resources. Since identity validation incurs a processing cost for each monitored vehicle, the total validation effort assigned to UAV u at time t must remain within its available resource budget, that is, ∑i∈𝒱u(t)κi(t)≤Cu, where κi(t) is the validation cost of vehicle i and Cu is the computational resource budget of UAV u.

Moreover, for vision-based validation, the comparison should be between the vehicle under consideration and the object within the UAV’s field of view, which imposes a latent restraint on sensing. Communication bandwidth also limits the system, as UAVs have a limited capacity to process and exchange identity and sensing data. This communication constraint is formulated as:

∑i∈𝒱bi(t)≤Bu,(14)

in Eq. (14), bi(t) denotes the communication load associated with vehicle i at time t, and Bu represents the communication bandwidth available to UAV u. All of these limitations combine to maintain the proposed framework within reasonable bounds in the trust, sensing, computation, and communication domains, enabling it to be utilized in practical applications for dynamic UAV-based traffic monitoring.

The framework presupposes a risk adaptive strategy, and instead of the standard strategy of every identity claim being validated, dynamically adjusts the strength of validation in accordance with the perceived risk level of each entity. This enables effective resource use by applying lightweight validation to low-risk cases and rigorous validation to high-risk situations. The risk score is defined in Eq. (15) by combining anomaly information with context inconsistencies:

Ri(t)=γ1Ai(t)+γ2(1−Cimob(t))+γ3(1−Cimap(t))(15)

where Ai(t) indicates the score in anomaly, and mobility and consistency of the map represent the differences between the desired behaviour and the observed behaviour. The coefficients γ1,γ2,γ3 control the contribution of each component. Based on this score, validation is categorized into three levels: low risk (Ri(t)<θ1), medium risk (θ1≤Ri(t)<θ2), and high risk (Ri(t)≥θ2), with increasing validation strictness across levels. To support this adaptive process, the validation threshold is adjusted according to risk level, as defined in Eq. (16):

τ(t)={τL,low risk,τM,medium risk,τH,high risk.(16)

This algorithmic, risk-adaptive method is more efficient and robust, distributing validation resources based on contextual risk, and better balances security and computational cost in dynamically deployed UAV-assisted traffic monitoring systems. The complete procedural flow of the proposed context-aware identity validation process is summarized in Algorithm 1.

The proposed CIV-UAV framework has several significant benefits, such as the ability to identify context-inconsistent identities even in cases where valid credentials are provided, flexibility in the adjustment of the validation efforts to risk, and adaptability due to scalability to multi-UAV cooperation, and the fact that it is consistent with the existing and practical constraints of urban traffic. The proposed formulation is based on a real-life context of mobility and space, unlike traditional authentication-only systems, making it more trustworthy and adaptable to the UAV-assisted smart city use case.

5  Simulation Setup and Performance Evaluation

In this section, the suggested CIV-UAV framework is tested in simulated urban traffic monitoring conditions with discrete-time and agent-based Python simulation. The environment is built with NumPy, NetworkX, and matplotlib, in which vehicles propagate on a graph-based road network, and UAVs are aerial agents for sensing and communication with fixed altitude and range, subject to sensing noise and communication constraints. At the time slot, vehicles send identity messages; the UAVs gather communication and visual data; attacks are introduced; and the CIV-UAV algorithm is used to compute trust scores and decisions for validation. The performance is evaluated based on accuracy, strength, false acceptance/rejection, attack detection, latency, and computational overhead.

We model a dense urban traffic environment as a road graph 𝒢=(ℒ,ℰ), where N vehicles move under lane and road-network constraints within a 1 km×1 km area, while U UAVs monitor vehicles within their coverage regions at fixed altitudes. The vehicles in every simulation run will be set up on the valid portion of the road with their respective velocities, and mobility and sensing noise will be introduced to mimic realistic movement and aerial perception uncertainty. These attacks are injected by modifying identity claims, positions, trajectories, or event messages per the attack model. The CIV-UAV algorithm is run at every time slot, and its accept/reject decision is compared with the ground-truth label to evaluate accuracy, FAR, FRR, detection rate, latency, and computational load. The main simulation parameters are summarized in Table 1. The vehicle motion model is defined in Eq. (17) as:

xi(t+1)=xi(t)+vi(t)Δt+εi(t)(17)

where εi(t) represents small random disturbances in motion. Similarly, UAV observations are assumed to be affected by sensing noise, and the observed position is modeled in Eq. (18) as:

xo(t)=xi(t)+η(t)(18)

where η(t) denotes the observation noise. The overall interaction among UAVs and vehicles is illustrated in Fig. 2,

Figure 2: Number of vehicles covered per UAV.

To assess the strength of the suggested framework, several adversarial scenarios are considered, summarized in Table 2. These are identity spoofing, in which malicious entity produces valid credentials and falsifies their location; Sybil attacks, and where a single attacker produces multiple fake identities; trajectory forgery, where the reported motion is violating of physical or road-network constraints; fake event injection, where false events involving the traffic are reported; and mismatch between vision and communication, where no corresponding UAV observation can be deployed corresponding to a claimed identity. The attack scenarios will model realistic threats in an urban mobility setting and demonstrate how the identified framework can effectively uncover anomalies between digital identity assertions and the physical world.

The proposed CIV-UAV framework is compared with four baseline approaches, as summarized in Table 3. These include Cryptographic Authentication (CA), where identity is only accepted as it is being tested, Location-Based Validation (LBV), where identity is tested as long as it is being observed by the UAV without communication-layer validation, and Trust-Based Model (TBM), where identity is tested according to historic ratings of reputation. These benchmark schemes serve as reference points for comparing the relative effectiveness of the proposed CIV-UAV framework.

To fully evaluate the effectiveness of the proposed framework, several performance metrics are considered, including identity validation accuracy, false acceptance rate (FAR), false rejection rate (FRR), attack detection rate, validation latency, and computational overhead. These evaluation metrics and their definitions are summarized in Table 4.

The suggested CIV-UAV architecture is critically discussed in various contexts to evaluate its potential to improve identity validation performance, withstand attacks, and conserve computational resources, as presented below. The results are contrasted with baseline schemes and assessed using different performance measures to highlight the benefits of the proposed strategy.

Identity Validation Accuracy: The overall identity validation accuracy of different schemes is presented in Fig. 3. The study findings demonstrate that the proposed CIV-UAV framework is the most accurate at a variety of vehicle and UAV configurations. An accurate improvement of 15%–25% as compared to CA and LBV schemes is realized by CIV-UAV. This may be attributed to the fact that it has incorporated the notion of multi-dimensional context, including consistency in mobility, confirmation of vision, and constraint of map, which minimizes ambiguity and further elaborates on the performance of identity validation.

Figure 3: Accuracy comparison under varying traffic density.

False Acceptance Rate (FAR): The false acceptance rate (FAR) performance under different attack scenarios is illustrated in Fig. 4. The CA scheme is characterized by a high FAR, particularly compared to spoofing and Sybil attacks, as it relies solely on cryptographic integrity. On the other hand, the proposed CIV-UAV system will help reduce FAR by 40%–60% by highlighting discrepancies between reported identities and measured context. This has been boosted mainly by cross-modal validation, which synergistically uses communication, vision, and mobility information to more effectively reject malicious identities.

Figure 4: FAR comparison under different attack scenarios.

False Rejection Rate (FRR): Comparison of false rejection rate (FRR) is in Fig. 5. Strict validation schemes may reject valid identities when they are not supposed to, and though the CIV-UAV scheme proposed should have a relatively low FRR, it is risk-adaptive because the validation scheme minimizes risk. Lightweight validation is advantageous in low-risk environments, and stricter validation is offered in high-risk environments to enhance security without compromising honest users. This equal action allows the CIV-UAV to achieve better performance than the TBM and VOV schemes in terms of validation reliability.

Figure 5: FRR comparison under different schemes.

Detection of Complex Attacks: The detection rates of the various types of attacks can be found in Table 5. The results reveal that CIV-UAV achieves high detection rates across all attack types, with the highest success rates observed for trajectory forgery, mismatched vision communication, and injection of fake events. Unlike the baseline schemes, the integrated scheme is recommended to include multiple dimensions of validation, thereby allowing it to detect sophisticated attacks that would not be detected otherwise.

Impact of Multi-UAV Cooperation: The cooperative validation effect is depicted in Fig. 6. The resulting trust score during UAV validation is more valid because information is fused through consensus among the UAVs. The results prove that multi-UAV cooperation improves validation accuracy on average by 10%–15% and reduces uncertainty significantly in situations when only a portion of the object is seen because of occlusion. This observation highlights the role of distributed, collaborative validation in assessing dense urban landscapes.

Figure 6: Validation accuracy across various numbers of UAVs.

Validation Latency and Computational Overhead: The results of validation latency and computational overheads are discussed in Fig. 7 and Table 6, respectively. Despite the fact that CIV-UAV requires more computation than the CA and LBV schemes, the proposed risk-adaptive mechanism ensures efficient resource utilization. Specifically, low-risk identities are handled with low overhead, whilst typical high-risk cases are more intensively vetted. As a result, the mean increase in latency remains within a reasonable range of 10%–20%, while the framework achieves high security performance.

Figure 7: Average validation latency comparison.

The results clearly demonstrate that the proposed CIV-UAV framework provides a significant improvement over traditional validation approaches. Context-aware analysis helps the system recognize malicious identities that would otherwise go undetected due to cryptographic checks. Moreover, cross-modal verification, adaptive decision-making, and multi-UAV cooperation can be integrated to provide a more robust, scalable, and use-case-specific solution for monitoring real-world traffic within the city. Overall, the analysis will demonstrate that CIV-UAV is a vital element for building credibility in UAV-assisted mobility systems, and the novel CIV-UAV framework envisioned in the analysis has the potential to meet this need.

6  Conclusion

This study proposes a CIV-UAV framework for UAV-based urban mobility and traffic surveillance applications. Compared with classic methods, where cryptographic authentication is usually the main tool, the framework introduced accounts for a variety of contextual variables, including mobility behavior, visual perception, technological features of the road network, systematic consistency, and cooperation among multiple UAVs. By implementing these points into a single, trust-based, risk-adaptive solution, more reliable identity verification could be achieved in dynamic urban traffic conditions.

The simulation results showed that the CIV-UAV framework is more effective in validation accuracy, reducing false acceptance and false rejection, and improving the detection of more complex attacks, including identity spoofing, Sybil attacks, trajectory forgery, and vision-communication discrepancies. Despite the high effectiveness indicated in the framework, we should examine its performance under harsh sensing uncertainty, in large-scale deployments, and with realistic traffic data in the future. In addition, advanced learning models and decentralized identity management could also be included to further enhance it.

Acknowledgement: Not applicable.

Funding Statement: This research is supported by the Korea Institute of Marine Science and Technology Promotion (KIMST), in 2022 through the Project is Development and Demonstration of Data Platform for AI Based Safe Fishing Vessel Design (Grant Number: RS-2022-KS221571).

Author Contributions: The authors confirm contribution to the paper as follows: Conceptualization, Kuldashbay Avazov, Kudratjon Zohirov, and Alpamis Kutlimuratov; methodology, Charos Khidirova, and Jasur Sevinov; software, Urishev Omadjon; validation, Adilbek Dauletov, and Akmalbek Abdusalomov; formal analysis, Kuldashbay Avazov; investigation, Kudratjon Zohirov; resources, Alpamis Kutlimuratov; data curation, Urishev Omadjon, and Young Im Cho; writing—original draft preparation, Kuldashbay Avazov, and Kudratjon Zohirov; writing—review and editing, Charos Khidirova, Jasur Sevinov, Urishev Omadjon, and Adilbek Dauletov; visualization, Akmalbek Abdusalomov; supervision, Young Im Cho; project administration, Young Im Cho; funding acquisition, Young Im Cho. All authors reviewed and approved the final version of the manuscript.

Availability of Data and Materials: Data will be available on reasonable request from the authors.

Ethics Approval: Not applicable.

Conflicts of Interest: The authors declare no conflicts of interest.

References

1. Zhang J, Zhang D, Chen X, Zhang T, Wang S. UAV-assisted geo-edge routing protocol for internet of vehicles based on reinforcement learning strategy. IEEE Trans Veh Technol. 2026:1–14. doi:10.1109/tvt.2026.3653679. [Google Scholar] [CrossRef]

2. Bakirci M. Internet of Things-enabled unmanned aerial vehicles for real-time traffic mobility analysis in smart cities. Comput Electr Eng. 2025;123(2):110313. doi:10.1016/j.compeleceng.2025.110313. [Google Scholar] [CrossRef]

3. Ali R, Ali A, Naeem HMY, Asad M, Alsarhan T, Heyat BB. A comprehensive survey of deep learning-based traffic flow prediction models for intelligent transportation systems. ICCK Trans Adv Comput Syst. 2024;1(3):117–37. doi:10.62762/tacs.2025.795448. [Google Scholar] [CrossRef]

4. Miao J, Wang Z, Ning X, Shankar A, Maple C, Rodrigues JJC. A UAV-assisted authentication protocol for internet of vehicles. IEEE Trans Intell Transp Syst. 2024;25(8):10286–97. doi:10.1109/tits.2024.3360251. [Google Scholar] [CrossRef]

5. Ouyang W, Mu J, Jing X, Wang Y. Efficient vehicle recognition and tracking for UAV-enabled intelligent transport systems: a multi-agent reinforcement learning method. IEEE Trans Intell Transp Syst. 2025;26(11):20930–40. doi:10.1109/tits.2025.3601740. [Google Scholar] [CrossRef]

6. Bashir N, Boudjit S, Zeadally S. A closed-loop control architecture of UAV and WSN for traffic surveillance on highways. Comput Commun. 2022;190(7547):78–86. doi:10.1016/j.comcom.2022.04.008. [Google Scholar] [CrossRef]

7. Luo J, Wang G, Li G, Pesce G. Transport infrastructure connectivity and conflict resolution: a machine learning analysis. Neural Comput Appl. 2022;34(9):6585–601. doi:10.1007/s00521-021-06015-5. [Google Scholar] [CrossRef]

8. Wang Q, Chen J, Song Y, Li X, Xu W. Fusing visual quantified features for heterogeneous traffic flow prediction. Promet Traffic Transp. 2024;36(6):1068–77. doi:10.7307/ptt.v36i6.667. [Google Scholar] [CrossRef]

9. Chen J, Zhang S, Xu W. Scalable prediction of heterogeneous traffic flow with enhanced non-periodic feature modeling. Expert Syst Appl. 2025;294(11):128847. doi:10.1016/j.eswa.2025.128847. [Google Scholar] [CrossRef]

10. Fan Y, Chen J, Xu W, Peng W. DADiffNet: delay-aware diffusion networks with adaptive subgraphs for large scale traffic forecasting. Neural Netw. 2026;200:108756. [Google Scholar] [PubMed]

11. Alshehri M, Wu T, Almujally NA, AlQahtani Y, Hanzla M, Jalal A, et al. UAV-based intelligent traffic surveillance using recurrent neural networks and Swin transformer for dynamic environments. Front Neurorobot. 2025;19:1681341. doi:10.3389/fnbot.2025.1681341. [Google Scholar] [PubMed] [CrossRef]

12. Zhou Z, Wang Y, Zhou G, Nam K, Ji Z, Yin C. A twisted gaussian risk model considering target vehicle longitudinal-lateral motion states for host vehicle trajectory planning. IEEE Trans Intell Transp Syst. 2023;24(12):13685–97. doi:10.1109/tits.2023.3298110. [Google Scholar] [CrossRef]

13. Zhou Z, Wang Y, Liu R, Wei C, Du H, Yin C. Short-term lateral behavior reasoning for target vehicles considering driver preview characteristic. IEEE Trans Intell Transp Syst. 2021;23(8):11801–10. doi:10.1109/tits.2021.3107310. [Google Scholar] [CrossRef]

14. Liu X, Wang Y, Zhou Z, Nam K, Wei C, Yin C. Trajectory prediction of preceding target vehicles based on lane crossing and final points generation model considering driving styles. IEEE Trans Veh Technol. 2021;70(9):8720–30. doi:10.1109/tvt.2021.3098429. [Google Scholar] [CrossRef]

15. Zhao X, Wang T, Li Y, Zhang B, Liu K, Liu D, et al. Target-driven visual navigation by using causal intervention. IEEE Trans Intell Veh. 2024;9(1):1294–304. doi:10.1109/ccdc58219.2023.10327097. [Google Scholar] [CrossRef]

16. Ren Y, Wang L, Li M, Jiang H, Cui Z, Yang M, et al. RM2Occ: re-projection multi-task multi-sensor fusion for autonomous driving 3D object detection and occupancy perception. IEEE Trans Intell Transp Syst. 2025;26(11):20864–81. [Google Scholar]

17. Li J, Sun R, Wang Y, Ochieng WY. A robust time synchronization algorithm for GNSS/IMU integrated navigation in urban environments. Meas Sci Technol. 2025;36(3):036302. doi:10.1088/1361-6501/ada976. [Google Scholar] [CrossRef]

18. Cheng Q, Chen W, Wang J, Mi X, Yang Y, Sun R. Multi-constellation instantaneous single difference triple-carrier ambiguity resolution in urban environments. GPS Solut. 2025;29(4):186. doi:10.1007/s10291-025-01946-1. [Google Scholar] [CrossRef]

19. Sun R, Sheng Q, Cheng Q, Shang X, Ochieng WY. 3D grid-based resilient pseudorange error prediction for adaptive GNSS/IMU integrated navigation in urban areas. IEEE Internet Things J. 2025;12(12):19264–79. doi:10.1109/jiot.2025.3533077. [Google Scholar] [CrossRef]

20. Wu T, Li M, Qu Y, Wang H, Wei Z, Cao J. Joint UAV deployment and edge association for energy-efficient federated learning. IEEE Trans Cogn Commun Netw. 2025;11(6):4126–40. doi:10.1109/tccn.2025.3543365. [Google Scholar] [CrossRef]

21. Yin R, Peng J, Cai Y, Wu C, Champagne B, Al-Dhahir N. Radar-assisted predictive beamforming for UAV-aided networks: a deep-learning solution. IEEE Trans Veh Technol. 2025;74(10):16079–93. doi:10.1109/icc51166.2024.10622815. [Google Scholar] [CrossRef]

22. Yao Y, Xiao W, Miao P, Chen G, Yang H, Chae CB, et al. UAV-RHS-enabled full-duplex ISAC covert system: robust beamforming and trajectory optimization. IEEE Trans Commun. 2026;74:5637–53. doi:10.1109/tcomm.2026.3668166. [Google Scholar] [CrossRef]

23. Yin R, Peng J, Cai Y, Wu C, Champagne B, Al-Dhahir N. Intelligent 3D trajectory and resource control for multi-UAV 6G networks via GNN and deep unfolding. IEEE Trans Commun. 2026. doi:10.1109/tcomm.2026.3655771. [Google Scholar] [CrossRef]

24. Zhang Z, Ming Y, Song G. A new approach to identifying crash hotspot intersections (CHIs) using spatial weights matrices. Appl Sci. 2020;10(5):1625. doi:10.3390/app10051625. [Google Scholar] [CrossRef]

25. Zhang Z, Ming Y, Song G. Identify road clusters with high-frequency crashes using spatial data mining approach. Appl Sci. 2019;9(24):5282. doi:10.3390/app9245282. [Google Scholar] [CrossRef]