Adversarial Path Planning for Optimal CCTV Surveillance: A Case Study on Nuclear Facility Security Optimization

Ahmed E. Salman 1

Noha Shaaban 2

W. I. Zidan 2

Mohamed H. Saad 3✉ Emailm.hassansaad@gmail.com Emailm.hassansaad@eaea.sci.eg

1 Operational Safety and Human Factors Department, Nuclear and Radiological Safety Research Center (NRSRC) Egyptian Atomic Energy Authority Cairo Egypt

2 Nuclear Safeguards and Physical Protection Department, Nuclear and Radiological Safety Research Center (NRSRC) Egyptian Atomic Energy Authority Cairo Egypt

Radiation Engineering Department National Center for Radiation Research and Technology, Egyptian Atomic Energy Authority Egypt

Ahmed E. Salman ^a, Noha Shaaban^b, W.I. Zidan^b, Mohamed H. Saad*^c

^a Operational Safety and Human Factors Department, Nuclear and Radiological Safety Research Center (NRSRC), Egyptian Atomic Energy Authority, Cairo, Egypt

^b Nuclear Safeguards and Physical Protection Department, Nuclear and Radiological Safety Research Center (NRSRC), Egyptian Atomic Energy Authority, Cairo, Egypt

^c Radiation Engineering Department, National Center for Radiation Research and Technology, Egyptian Atomic Energy Authority, Egypt.

^*Corresponding author : Mohamed H. Saad ( m.hassansaad@gmail.com, m.hassansaad@eaea.sci.eg )

Abstract

The security of critical infrastructure, particularly nuclear facilities, is paramount in mitigating potential threats and ensuring public safety. Conventional CCTV surveillance deployment relies on static placement strategies that fail to account for dynamic adversarial behavior, leading to coverage gaps and surveillance inefficiencies. This study proposes a novel Adversarial Path Planning (APP) framework, which integrates game-theoretic modeling, probabilistic risk assessment, and bilevel optimization to enhance surveillance coverage, intrusion detection, and resource allocation. By simulating adversarial movement patterns and iteratively refining camera placement, APP dynamically adjusts surveillance strategies to counter evolving threats while minimizing blind spots and optimizing detection probability. The APP framework models the facility as a weighted surveillance graph, identifying high-risk intrusion paths and optimizing camera positioning to maximize coverage while minimizing redundancy. A case study conducted on a hypothetical nuclear power plant demonstrates APP’s effectiveness in enhancing security resilience, achieving 95% surveillance coverage, improving detection accuracy to 98%, and reducing dead zones by 85%—significantly outperforming conventional methods such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO). Additionally, APP reduces the required number of cameras by 40% while improving cost efficiency by 27%, highlighting its potential for resource-conscious security optimization. The findings establish APP as a scalable and computationally efficient surveillance optimization solution, adaptable to nuclear security, border surveillance, and high-risk critical infrastructure protection. Future research should explore AI-driven real-time threat detection, autonomous security drones, and deep reinforcement learning-based surveillance adaptation to further enhance threat responsiveness and situational awareness in evolving security environments.

Keywords:

integrated security system

CCTV

nuclear facility

physical security

surveillance

threat detection

1. Introduction

The security of critical infrastructure, particularly nuclear facilities, is of paramount importance to ensuring public safety, operational resilience, and protection against evolving adversarial threats. While nuclear safety culture has long been established to prevent accidental radioactive material leaks, nuclear security culture emerged later in response to intentional security breaches, sabotage risks, and adversarial attacks. Unlike safety systems, which aim to correct equipment failures and human errors, security systems must be proactively designed to anticipate, detect, and neutralize deliberate threats. Given the increasing complexity of modern security risks, surveillance strategies must integrate adaptive intelligence, real-time threat assessment, and dynamic response mechanisms to enhance intrusion prevention and facility protection [1].

Closed-Circuit Television (CCTV) surveillance plays a fundamental role in physical security, providing continuous monitoring, intrusion deterrence, and real-time threat detection. However, traditional CCTV deployment strategies rely on static placement models that fail to account for adversarial behavior, leading to coverage gaps, inefficient resource allocation, and blind spots. Moreover, conventional surveillance planning often depends on manual camera placement, which is subjective, time-consuming, and suboptimal for high-security environments. These limitations necessitate a computationally optimized and dynamically adaptive approach to CCTV surveillance system design [2, 3].

The effectiveness of CCTV surveillance networks is determined by several key factors that ensure comprehensive security coverage and real-time threat detection. One of the most critical aspects is strategic camera placement, which is essential for achieving full surveillance coverage of high-risk zones while minimizing blind spots and redundancies. Additionally, adaptive response mechanisms play a vital role in dynamically adjusting surveillance configurations based on real-time intelligence and evolving security threats, ensuring that the system remains proactive and responsive to adversarial movements. The integration of artificial intelligence (AI) and machine learning (ML) further enhances surveillance capabilities by enabling automated anomaly detection, predictive threat analysis, and intelligent decision-making, reducing the reliance on manual monitoring. Furthermore, scalability and computational efficiency are essential for allowing real-time surveillance reconfiguration, ensuring that the security system can adapt to emerging threats and evolving risk landscapes without significant operational delays. Together, these factors form the foundation for a robust, intelligent, and high-performance CCTV surveillance network that enhances situational awareness, threat mitigation, and overall security resilience in critical infrastructure protection. To address these challenges, this study introduces a novel Adversarial Path Planning (APP) framework for intelligent CCTV surveillance optimization in high-security environments, specifically within a hypothetical nuclear facility. Unlike heuristic-based camera placement methods, APP leverages threat-aware path modeling and real-time adversarial response mechanisms to optimize surveillance effectiveness while minimizing resource expenditure [4–9].

The proposed Adversarial Path Planning (APP)-based optimization framework integrates multiple advanced methodologies to enhance CCTV surveillance efficiency and security resilience. By employing game-theoretic modeling, the system anticipates potential intrusion routes, enabling proactive threat mitigation through strategically optimized camera placement. Additionally, probabilistic risk assessment is utilized to prioritize high-threat surveillance zones, ensuring that areas with the highest likelihood of intrusion receive enhanced monitoring and security coverage. Furthermore, mathematical optimization techniques are applied to maximize CCTV coverage while minimizing security blind spots and redundancy, allowing for efficient resource utilization without compromising detection accuracy. Through the simulation of adversarial movement patterns, APP iteratively refines camera positioning, dynamically adjusting surveillance strategies in response to evolving security threats. This ensures a real-time adaptable security framework, significantly improving situational awareness, intrusion detection, and overall facility protection in high-risk environments.

The APP-based surveillance optimization framework is structured as a multi-phase approach, beginning with a comprehensive security assessment to establish a data-driven foundation for intelligent surveillance planning. This initial phase involves a detailed risk evaluation to identify perimeter security vulnerabilities and weaknesses in access control mechanisms, ensuring that potential entry points are properly monitored. Additionally, the assessment defines critical surveillance zones that require continuous real-time monitoring based on security priority levels. To enhance threat anticipation, the framework employs movement prediction models to analyze potential adversarial intrusion paths, allowing for proactive mitigation strategies. Furthermore, operational constraints such as budget limitations, spatial restrictions, and environmental factors are considered to ensure optimal resource allocation without compromising surveillance effectiveness. This threat-driven security assessment serves as the foundation for strategic CCTV deployment, ensuring that high-risk areas remain under continuous surveillance while maintaining adaptive, cost-effective security measures [10, 11].

The APP algorithm is utilized to determine optimal camera placements by employing advanced computational techniques that enhance surveillance efficiency and security coverage. The framework prioritizes strategic camera positioning to ensure maximum area coverage while minimizing overlapping views and redundant deployments, leading to resource-efficient surveillance optimization. Additionally, Field of View (FOV) and Angle of View (AOV) analysis are incorporated to eliminate surveillance blind spots, ensuring that no critical areas remain unmonitored. The algorithm also integrates dead-zone reduction techniques, effectively mitigating potential security gaps and strengthening intrusion detection capabilities. Furthermore, computational efficiency is a key component of the APP framework, allowing for real-time camera repositioning based on evolving security threats and adversarial movement patterns. Through an iterative simulation process, the APP algorithm continuously refines camera placement strategies, enabling a dynamic, adaptive, and resilient security framework capable of responding to emerging threats with precision and efficiency [12–14].

To validate the effectiveness of the APP-based surveillance system, a comprehensive set of quantitative performance metrics is employed to assess its efficiency in securing high-risk zones. The evaluation considers surveillance coverage percentage, ensuring that critical areas remain under continuous observation, thereby minimizing security vulnerabilities. Additionally, detection accuracy is analyzed to enhance threat identification speed and reliability, improving intrusion detection and response time. A key focus is on dead-zone reduction analysis, which measures the elimination of unmonitored security gaps, ensuring that adversaries have no undetected intrusion pathways. Furthermore, cost efficiency improvement is assessed, aiming to minimize camera deployment while maintaining optimal surveillance performance, reducing unnecessary resource allocation and operational costs.

The experimental evaluation of the APP framework at the Lone Pine Nuclear Power Plant (LPNPP) demonstrates significant improvements over traditional static surveillance placement methods. The results confirm that APP achieves 95% surveillance coverage, enhances detection accuracy from 85% to 98%, and reduces dead zones by 85%, substantially outperforming optimization techniques such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO). Additionally, the APP framework reduces the number of required cameras by 40%, leading to a 27% increase in cost efficiency, reinforcing its resource-conscious security optimization capabilities. These findings underscore the superiority of APP in high-security environments, demonstrating its effectiveness in adaptive, intelligent, and cost-efficient surveillance system deployment [15–20].

The primary objective of this research is to develop an adaptive and intelligent security framework capable of safeguarding nuclear facilities from evolving security threats through optimized CCTV surveillance deployment. The proposed Adversarial Path Planning (APP)-based surveillance model introduces a dynamic, threat-driven approach that enhances real-time threat mitigation, resource efficiency, and scalability across various high-security environments. By continuously adapting to changing security conditions, the APP framework ensures proactive surveillance adjustments, reducing the likelihood of undetected intrusions. Specifically, the model enables:

Real-time threat mitigation through continuous surveillance adaptation, allowing for rapid adjustments to adversarial movements and minimizing security blind spots.

Resource-efficient CCTV deployment, optimizing camera placement to maximize monitoring effectiveness while minimizing hardware requirements and operational costs.

Scalable security applications, making the methodology applicable to various critical infrastructure settings, including nuclear power plants, military installations, airports, and other high-risk facilities.

By systematically analyzing the interdependencies between security effectiveness factors, this study provides valuable insights for security professionals, policymakers, and facility managers, facilitating data-driven decision-making in high-risk surveillance planning. The proposed APP framework functions as a decision-support tool for enhancing security operations, mitigating intrusion risks, and optimizing surveillance resource allocation, ensuring a resilient and cost-effective security infrastructure in high-threat environments.

2. Materials and Methods

2.1 LPNPP Facility Description

In the IAEA-TECDOC-1868 [21], the publication by the International Atomic Energy Agency (IAEA) offers Member States a case study of a nuclear power plant (NPP) that serves as a demonstration of the standard methodology for conducting nuclear security assessments. This methodology, known as Nuclear Security Assessment Methodologies (NUSAM) for Regulated Facilities, was developed through the IAEA Coordinated Research Project (CRP). Its purpose is to provide a comprehensive framework for evaluating the security measures and vulnerabilities of regulated nuclear facilities. A hypothetical facility layout, LPNPP, was introduced as a standardized framework for designing security systems [21]. This layout serves as the basis for the proposed security system design, the LPNPP site layout (Fig. 1) provides an overview of the entire facility. In addition to illustrating the site layout, Fig. 2 also highlights the positions of various key elements such as guard posts, the central alarm station (CAS), guard towers, and the backup alarm station (BAS) or secondary alarm station (SAS).

Fig. 1

Lone Pine Nuclear Power Plant [22]

There are three main areas in the LPNPP — the limited access area, the protected area and the vital area.

Fig. 2

Facility physical protection system layout [21]

The layout components of the Lone Pine NPP facility site plan presented in Fig. 2 are given as [21]:

1	Rector Containment	11	Turbine Building	21	Boron Test Tanks
2	ESF Building	12	Main Steam Valve Building	22	Boron Recovery Tanks
3	Hydrogen Recombiner Building	13	Auxiliary Boiler room	23	Primary Water Storage Tanks
4	Fuel Building	14	Condensate Polishing Enclosure	24	Waste Test Tanks
5	Waste Building	15	Warehouse	25	Demineralized Water Storage Tank
6	Auxiliary Building	16	Main Transformer	26	Refueling Water Storage Tank
7	Shop	17	Normal Service Transformers	27	Condensate Storage Tank
8	Service Building	18	Lines to Switchyard	28	Water Treating Storage Tank
9	Control Building	19	Discharge Vacuum Pr. Pump house	29	Condensate Surge Tank
10	Diesel Generator Building	20	Reserve St. Service Transformers	30	Intake Structure

According to the studied scenario in [21], as represented in Fig. 3, the adversary team initiates their operation at the boundary of the limited access area, as depicted on the facility area map provided in Fig. 3. From there, they follow a designated route (1-2-3) until they reach a specific location (3), situated to the left of the Control Building. Subsequently, starting from the same location (3) on the first-floor map, they proceed towards a set of stairs leading to the second floor, indicated as (4) and referred to as the 'ladder.' Once they reach the second floor, the adversaries proceed to their intended target location (5) to fulfill their mission. The current actions in this scenario are represented by: (1) visual acquisition, (2) alarm (perimeter), (3) alarm (face of the building), (4) alarm (room door to the target), and (5) act (sabotage).

Fig. 3

The studied Adversary modeled path [21]

2.2 Camera surveillance systems

Camera-system planning finds diverse applications across various domains. Some systems are specifically designed for indoor or urban use, as highlighted in previous studies [11, 19, 23–25]. However, for the purpose of the current study, the focus is primarily on outdoor applications, including perimeter surveillance [26, 27] and equipment monitoring [28]. CCTV refers to a television system that is designed for private monitoring rather than public distribution. Its main purpose is to provide surveillance and enhance security measures.

The selection of camera lenses for a video system involves interdependent variables such as format, focal length, field of view, distance-to-object blind area and image quality. These parameters are interconnected, and their specific values are determined based on the objectives of the designer and how the video system will integrate with other security systems. The performance of a lens can be further enhanced by incorporating additional features. For instance, certain lenses are equipped with automatic iris aperture controls that include neutral density filters positioned in the center of the lens. These controls work in conjunction with the camera circuitry to enable automatic adjustment of light levels. This feature allows for a more effective reduction of bright light when the iris aperture is smaller than the neutral density filter. Moreover, some lenses are equipped with special coatings that serve to enhance or filter out specific wavelengths of light. These coatings are designed to optimize the lens's performance for particular purposes. For instance, certain lenses may enhance the transmission of near-infrared light (wavelengths of 800–1100 nm), which can be utilized by solid-state cameras. By enhancing the transmission of near-infrared light, these lenses enable improved performance in capturing images under specific conditions [29–31].

The selection of lenses should be based on the desired resolution and field of view requirements. In cases where a video system is intended for perimeter use, the "distance and width approximation" method can be employed to determine the maximum length of the zone that can be assessed using a specific camera and lens combination as shown in Fig. 4. Typically, the lower field of view displayed at the bottom of the monitor is narrower than the zone width, while the upper field of view is wider than the field of view limited by resolution. This means that there is a blind area between the camera and the lower field of view where the camera cannot capture any visual information, also called dead-zone [31–33].

Fig. 4

The CCTV camera coverage zones [34]

3. Proposed Method

In This study introduces a CCTV mathematical modeling approach for optimizing camera coverage in the Lone Pine Nuclear Power Plant (LPNPP). The proposed methodology follows a structured five-step framework, ensuring that CCTV camera placement is systematically optimized to enhance surveillance efficiency while minimizing blind spots and redundant coverage. By incorporating Adversarial Path Planning (APP) optimization, the methodology ensures that camera placement dynamically adapts to evolving security threats and intruder movement patterns.

To achieve an optimized surveillance network, the methodology integrates computational modeling and optimization techniques, implemented using Python-based algorithms. The APP framework is employed to simulate adversarial behavior, allowing for strategic camera placement adjustments based on real-time security threat assessments. This approach provides a more adaptive and intelligent surveillance system compared to traditional static camera placement techniques.

The proposed five-step methodology, as illustrated in Fig. 5, unfolds as follows:

The first step involves conducting a comprehensive security assessment and facility modeling. A detailed risk analysis is performed to identify high-risk zones, critical surveillance points, and potential intrusion routes. The nuclear facility is mathematically represented as a graph-based surveillance model, where nodes correspond to potential camera locations, and edges represent possible adversarial movement paths. This graph representation enables an intelligent assessment of surveillance coverage across the facility.

The second step focuses on camera parameter selection and initial placement. Key specifications such as focal length, field of view (FOV), pixel density, resolution, and detection range are considered. An initial camera deployment strategy is formulated based on coverage radius and strategic positioning in high-risk zones. This preliminary placement ensures baseline surveillance coverage before further optimization is applied.

The third step involves the implementation of the Adversarial Path Planning (APP) framework. APP is utilized to simulate intruder behavior and predict the most probable intrusion paths within the facility. The adversarial model employs probabilistic threat analysis, enabling the surveillance system to identify areas with the highest risk exposure. By leveraging game-theoretic principles, APP effectively anticipates and counters adversarial strategies, allowing for more informed surveillance decisions.

The fourth step applies mathematical optimization techniques to refine the CCTV camera placement strategy. A coverage optimization algorithm is executed using integer programming and heuristic search techniques in Python. The objective function maximizes detection probability while minimizing blind zones and redundant surveillance overlap. This step ensures that every camera is optimally positioned to provide the highest surveillance efficiency with the least resource expenditure.

The final step is performance evaluation and iterative refinement. The optimized CCTV network configuration is validated based on several key performance indicators, including coverage percentage, image quality, detection accuracy, and response time. If the system fails to meet coverage constraints or detection thresholds, the APP framework iterates through further refinements until an optimized surveillance configuration is achieved. This iterative process allows for continuous improvements in threat detection capabilities, ensuring that the surveillance system remains adaptive and responsive to emerging security risks.

The proposed APP-based approach provides a systematic, scalable, and computationally efficient method for optimizing CCTV surveillance systems in high-security environments. By integrating mathematical modeling, adversarial path planning, and real-time optimization, this methodology ensures that nuclear facilities maintain continuous and adaptive surveillance coverage while minimizing resource constraints and operational costs. The findings of this study demonstrate that APP significantly enhances CCTV deployment efficiency, outperforming traditional static camera placement methodologies in high-security applications.

Fig. 5

Proposed method for calculating the camera coverage

The sequence of actions outlined in Fig. 5 unfolds in the following manner:

3.1 Defining the area of interest:

This step involves determining area of interest where the camera system design is required to cover and the initial locations for camera. LPNPP’s protected area is considered as shown in Fig. 1. Figure 1 provides a visual representation of the site perspective, showcasing the overall layout. Additionally, Fig. 2 shows a map indicating the boundaries of the controlled area, which is enclosed by a concrete wall fence. The perimeter is further secured with guard towers. Both the protected and vital zone areas are enclosed by double wire mesh fences. These fences are equipped with surveillance systems, including CCTV cameras and perimeter intrusion detection systems, to monitor and safeguard the premises. Using the IP video system design tool (IPVSDT) to determine and draw the real boundaries of the area to be studied, the protected area in Fig. 2. Figure 6 illustrated the boundaries of the protected area in the LPNPP and dimensions are tabulated in Table 1.

Table 1

Side lengths of the protected area in Fig. 6
Side	A	B	C	D	E	F	G	H	I
Length [m]	603.5	519	483	182	203	117.5	125.5	120	423.5

Fig. 6

The illustrated boundaries of the protected area of the LPNPP

Select a camera lens parameter:

The first task of the proposed calculations is selecting a camera as initial input parameters, e.g., focal length (F) and sensor format. The CCTV lens focal length refers to the distance between the center of the lens and the focal point, which is typically the sensor or film. In general, a longer focal length results in a narrower angle of view. This means that as the focal length increases, the field of view captured by the lens becomes more restricted, allowing you to zoom in on distant subjects or focus on specific details. The camera sensor format refers to the size of the CMOS or CCD sensor used in the camera which represents the lens width (w) and height (h). When selecting a sensor format, you have various options to choose from, such as 1/4″, 1/3″, 1/2″, 2/3″ and 1″. Typically, the sensor format is specified in the camera's technical specifications. The target cameras in this study are cameras having Full HD resolution (1920

$\:\times\:$

1080). However, it's important to note that different cameras may have different sensor formats depending on their intended application or specific requirements. Therefore, it's recommended to check the camera's specifications to determine the exact sensor format more suitable to be used. Initially, for example, the standard 1/3" lens format is chosen, featuring an active area measuring 4.8mm

$\:\times\:$

3.6mm with a diagonal size of 6.0mm. It’s well noted to consider that the CCTV camera in this study is installed at 3m height (

$\:{\text{C}\text{a}\text{m}\text{e}\text{r}\text{a}}_{\text{h}\text{e}\text{i}\text{g}\text{h}\text{t}}$

) to detect person (

$\:{\text{T}\text{a}\text{r}\text{g}\text{e}\text{t}}_{\text{h}\text{e}\text{i}\text{g}\text{h}\text{t}}$

) of 2m at the target distance [31, 35].

Calculate camera coverage:

The camera coverage is calculated based on the selected camera lens parameters, providing a quantitative measure of the areas effectively monitored by the cameras. For CCTV cameras, the Angle of View (AOV) is a feature that defines the range of angles that can be captured by the camera lens.. The effective angle of view will be limited to the angle of coverage of the lens. In CCTV cameras, the Angle of View is primarily determined by the focal length of the lens and the size of the image sensor. Eq. (1–3) calculates the angles of view of the horizontal, vertical, and diagonal scenes [11, 18, 31].

$\:{AOV}_{h}=2*{\text{tan}}^{-1}\frac{w}{2*F}$

$\:{AOV}_{v}=2*{\text{tan}}^{-1}\frac{h}{2*F}$

$\:{AOV}_{d}=2*{\text{tan}}^{-1}\frac{d}{2*F}$

When choosing a camera, the specifications for FOV play a crucial role. The FOV refers to the extent of the scene that a camera captures on its sensor. Cameras with longer lenses or smaller sensors tend to have narrower fields of view, while those with shorter lenses or larger sensors offer wider fields of view. A smaller field of view indicates that the camera captures a more "zoomed in" perspective. In CCTV cameras, the Field of view is primarily determined by the

$\:AOV$

, and distance to object (

$\:x$

). Distance from Camera to object refers to the maximum distance between the camera and the target. Eq. (4–6) calculates the Fields of view of the horizontal, vertical, and diagonal scenes and Eq. (7) determines the dead zone (blind area) on the camera [18, 31, 35].

$\:{FOV}_{h}=2*\text{tan}\frac{{AOV}_{h}}{2}*x$

$\:{FOV}_{v}=2*\text{tan}\frac{{AOV}_{v}}{2}*x$

$\:{FOV}_{d}=2*\text{tan}\frac{{AOV}_{d}}{2}*x$

$\:Dead\:Zone=Targe{t}_{height}*\text{tan}({\text{tan}}^{-1}\frac{Distance\:to\:object}{Camer{a}_{height}-Targe{t}_{height}}-{AOV}_{v})$

Optimization:

The optimization technique used in this study aimed to maximize the surveillance coverage with minimum number of cameras and best imaging quality. Eq. (8) determines the pixel density which represents the number of pixel per meter (ppm), the higher the pixel density, the higher the image quality [36]. The proposed optimization work is discussed in section 2.4.

$\:pixel\:density=\frac{Horizontal\:resolution}{{FOV}_{h}}$

Visibility analysis:

A comprehensive visibility analysis is conducted to assess the line-of-sight visibility from each camera to the designated cells, taking into account potential obstructions or hindrances. The "pixel density at a specified distance" is utilized to distinctly differentiate the zones of view, such as identification (red), recognition (yellow), detection (green), and monitoring (blue), in Fig. 7. Each zone is expressed by defining the minimum permissible pixel density value as an expression of image quality according to the IPVSDT, Table 2. For example, A CCTV with 1/3 inch sensor size with focal length of 8, shown in Fig. 7. The coverage area of the camera in Fig. 7 is divided into four zones. Each zone covers a specified area in terms of distance and FOV. Table 2 represents the studied zones, determining the minimum pixel density value, the distance, and the FOV of each zone according to Fig. 7.

Fig. 7

Pixel density-based zones of the camera's field of view

Table 2

The minimum permissible pixel density value for each zone
/	Dead Zone “-”	Identification Zone “RED”	Recognition Zone “YELLOW”	Detection Zone “GREEN”	Monitoring Zone “BLUE”
Pixel density, Min value [m]	-	250	125	25	12
Range [m]	0-8.4	8.4–12.4	12.4–25.4	25.4-129.2	129.2-268.8
FOV [m]	0–5	3.5-8	87.5–16	16-74.5	74.5–160

3.2 Adversarial Path Planning for Optimal CCTV Surveillance

In high-security environments such as nuclear facilities, border control, and critical infrastructure protection, surveillance optimization is a fundamental challenge. Adversarial Path Planning (APP) is a strategic approach that models the behavior of potential intruders or threats to determine the optimal placement of surveillance cameras. By simulating the most probable intrusion paths, APP ensures that security measures effectively counteract potential threats by maximizing visibility and minimizing blind spots. Traditional surveillance placement techniques rely on static coverage models that fail to consider adversary behavior. Adversarial Path Planning (APP) integrates game theory, optimization, and probabilistic modeling to determine optimal surveillance strategies, ensuring that CCTV camera placement aligns dynamically with potential intrusion routes. This intelligent adversary modeling strengthens security systems by anticipating threat movements and maximizing detection probability in critical zones.

The Adversarial Path Planning (APP) framework models the facility as a graph, where risk scores are assigned to possible adversary movement paths to represent potential threats. It then identifies the most likely adversarial paths using a minimum-risk path algorithm, ensuring that surveillance focuses on high-risk areas. Through integer programming, cameras are placed optimally while adhering to resource constraints, maximizing coverage with minimal equipment. The framework continuously simulates adversary responses, dynamically adjusting movement paths based on the surveillance layout. This process undergoes iterative refinement, where camera placements are updated until the system achieves optimal coverage and effectively mitigates security risks. As described in Algorithm 1, the framework begins by constructing a surveillance graph of the facility. The iterative loop between steps 2 and 4 ensures the solution is robust against an adapting adversary.

Algorithm 1

The Adversarial Path Planning (APP) Algorithm for Optimal CCTV Surveillance. The procedure begins with graph construction and risk assignment, then iterates between computing least-risk adversary paths (P_adversary), solving a camera placement optimization problem, and simulating adversary response to the new surveillance layout. The loop continues until convergence or a maximum number of iterations (T), outputting the final camera placement configuration X(v).

3.4.1 Problem Definition

The optimization of CCTV surveillance through APP is formulated as a graph-based problem, where a facility is represented as a weighted surveillance graph. The adversary (intruder) aims to reach a target location while minimizing exposure to CCTV cameras, whereas the surveillance system seeks to maximize detection probability by strategically placing cameras along high-risk paths.

Let G = (V, E) represent a facility surveillance graph, where:

• V denotes the set of key facility locations (nodes).

• E represents the set of possible adversary movement paths (edges).

• Each edge e ∈ E is assigned a risk score R(e), indicating the likelihood of adversarial movement.

• Each camera placement at node v ∈ V contributes to a visibility function C(v), determining the probability of detecting an adversary at a given location.

The objective is to maximize intrusion detection probability while minimizing resource utilization. Each node v can be equipped with a camera, represented by the binary function:

The adversary selects an optimal path Pa from a starting location s to a target location t that minimizes exposure to surveillance:

Conversely, the surveillance system seeks an optimal camera placement strategy X*(v) that maximizes coverage along high-risk intrusion paths:

where λ is a penalty coefficient balancing surveillance effectiveness and resource constraints.

This results in a bilevel optimization problem comprising:

1. Adversarial Decision-Making – Identifying a low-risk path to minimize surveillance exposure.

2. Surveillance Countermeasures – Optimizing CCTV placement to ensure that high-risk paths are effectively monitored.

The iterative nature of the problem necessitates an adaptive approach, where security measures continuously respond to evolving adversarial strategies, ensuring a robust and resilient surveillance framework.

3.4.2 Adversary Movement Model

An adversary selects a path P_a from a starting location s to a target location t, where:

Each adversary follows a probabilistic movement model, where the probability of choosing edge e_ij is given by:

where:

represents the risk score of edge e_ij, influenced by the camera coverage function C(e).

is a tunable exploration parameter controlling the adversary’s sensitivity to risk.

This formulation ensures that lower-risk paths are more likely to be chosen while maintaining some randomness in decision-making, and surveillance adjustments influence adversarial choices, prompting path recalculations.

3.4.3 Surveillance Optimization Model

The Surveillance Optimization Model (SOM) ensures that CCTV cameras are strategically positioned to maximize detection probability while minimizing redundancy and cost. Given the adaptive nature of adversaries, the optimization model continuously adjusts camera placements to ensure persistent monitoring of high-risk paths.

Objective Function

The goal is to maximize surveillance coverage while minimizing the number of deployed cameras:

where:

• C(v) represents the surveillance effectiveness function at node v.

• λ balances coverage maximization with resource constraints.

Optimization Constraints

Detection Probability Constraint

Ensuring that adversarial paths maintain a minimum required detection probability:

where θ ensures adequate coverage of adversarial paths.

2. Camera Budget Constraint

Ensuring that camera placements do not exceed the available resource budget:

where M is the maximum number of deployable cameras.

3. Adversary Risk Re-weighting

As adversaries adapt to surveillance placements, the risk scores of edges eee are updated dynamically:

This iterative adjustment ensures that heavily monitored routes become progressively less attractive, forcing adversaries to recalculate intrusion paths

3.4.4Adversarial Path Simulation and Iterative Optimization

The adversary iteratively updates movement strategies based on the surveillance layout. The optimal adversarial path P_a^* is computed as:

where increased surveillance presence discourages movement through high-risk areas. Using an iterative adversarial reinforcement mechanism, the surveillance system reallocates CCTV cameras dynamically, counteracting adversarial movement patterns until an equilibrium state is achieved, ensuring comprehensive facility-wide monitoring.

This methodology is particularly well-suited for high-security facilities, including nuclear power plants, transportation hubs, and industrial sites, where real-time adaptability and robust intrusion detection are paramount.

3.5. Implementation and Computational Framework

To ensure transparency and reproducibility, this section details the computational implementation of the Adversarial Path Planning (APP) framework, including graph construction, parameter initialization, and the software environment.

3.5.1. Graph Construction from LPNPP Layout

The LPNPP facility layout (Fig. 2) was converted into a weighted surveillance graph G (V, E) to facilitate computational modeling.

Nodes (V): The facility was discretized into a grid of 5m x 5m cells. Each cell center was defined as a node v_i ∈ V, resulting in a graph that balances spatial resolution with computational tractability. This granularity was chosen to be smaller than the typical field of view of a CCTV camera, ensuring precise coverage calculation.

Edges (E): Edges e_ij ∈ E were established between a node v_i and its eight surrounding neighbors (using 8-connectivity) to allow for movement in all cardinal and diagonal directions. This models an adversary's ability to move freely across the facility terrain.

Obstacles: Nodes falling within building footprints or other permanent obstructions (e.g., the Reactor Containment, Turbine Building) were removed from the graph, and edges connecting through these obstacles were disabled, ensuring the model respects physical impassabilities.

3.5.2. Risk Score Initialization and Update Methodology

The risk score R(e) for each edge e quantifies the attractiveness of that path segment to an adversary.

Initialization: Initial risk scores R_initial(e) were assigned based on two heuristic principles:

Proximity to Critical Assets: Edges closer to vital areas (e.g., Control Building, Fuel Building) were assigned higher base risk. The base risk decreased with the Euclidean distance from the nearest critical asset.

Path Concealment: Edges located behind visual obstructions or in shadowed areas, as identified from the facility plans, received a moderate risk bonus, modeling an adversary's preference for covered approaches.

The combined initial risk was calculated as: R_initial(e) = α * (1 / Distance_to_Asset(e)) + β * Concealment_Score(e), where α and β are weighting coefficients set to 0.7 and 0.3, respectively, to prioritize asset proximity.

Dynamic Update: The risk scores are updated iteratively within the APP algorithm (Step 4) to reflect adversary adaptation. The update rule is:

R_new(e) = R_old(e) * (1 - C(e)) (19)

where C(e) is the cumulative coverage probability of edge e from all deployed cameras. This formulation dynamically reduces the risk score of well-monitored paths, forcing the adversary model to seek alternative, less-surveilled routes in subsequent iterations.

3.5.3. Software Libraries and Optimization Solver

The entire APP framework was implemented in Python 3.8. The key libraries employed were:

NetworkX (v2.5): Used for constructing the facility graph G (V, E), managing node/edge attributes, and calculating the least-risk adversarial paths (P_a) using Dijkstra's algorithm.

PuLP (v2.3.1) with the CBC Solver: The integer programming problem for camera placement (Step 3 of the APP algorithm) was formulated using PuLP. The objective function max ∑ C(v) X(v) - λ ∑ X(v) was implemented with the constraints ∑ C(v) X(v) ≥ θ for all adversarial paths and ∑ X(v) ≤ M. The open-source CBC (Coin-or branch and cut) solver was used to find the optimal solution for X(v).

NumPy & Matplotlib: Used for all numerical computations (e.g., AOV, FOV, pixel density calculations) and for generating the result visualizations (e.g., camera placement maps, coverage heatmaps).

3.5.4. Algorithm Parameter Settings

The parameters for the APP algorithm were set as follows based on preliminary sensitivity analysis to achieve a balance between performance and convergence:

λ (Resource Penalty Coefficient): 0.15. This value was found to effectively penalize excessive camera use without significantly degrading coverage.

θ (Minimum Detection Probability Threshold): 0.95. This ensures that every identified adversarial path has a high probability (95%) of being detected by at least one camera along the path.

β (Adversary Exploration Parameter): 1.5. This value in the probabilistic movement model (Section 3.4.2, Eq. for P(e_ij)) introduces a moderate degree of randomness, allowing the model to explore sub-optimal paths and avoid being overly deterministic.

M (Maximum Number of Cameras): 50. This budget constraint was set generously for the initial optimization to first find a performance ceiling, which was then refined to the final 30 cameras.

T (Maximum Iterations): 20. The algorithm was observed to converge stably within 15–18 iterations, making 20 a safe upper bound.

While the formal problem definition in Section 3.4.1 specifies a non-linear path detection probability constraint (Detection Probability (P_a) ≥ θ), the integer programming model implemented in PuLP utilized the linear approximation ∑ C(v) X(v) ≥ θ for computational tractability within the solver. This linear sum of coverages serves as a conservative proxy for the true joint probability. The exact non-linear detection probability was then used exclusively in the adversary response simulation (Step 4 of the APP algorithm) to evaluate the quality of the solution and dynamically update the path risk scores R(e). This hybrid approach leverages the speed of linear integer programming for placement generation while maintaining the fidelity of the probabilistic model for solution evaluation and iterative refinement, ensuring the final camera layout is robust against the true metric of interest—the probability of detecting an intruder along an entire path.

4. Experimental Setup and Benchmarking Methodology

To ensure a rigorous and fair evaluation of the proposed Adversarial Path Planning (APP) framework, a comprehensive comparative analysis was conducted against a suite of established optimization algorithms. This section details the unified experimental setup, the implementation and tuning of benchmark algorithms, and the evaluation protocol, ensuring the comparability and credibility of the results presented in Table 4.

4.1. Unified Problem Formulation and Constraints

A critical requirement for a valid comparison is that all algorithms solve the same optimization problem under identical constraints. For this study, the core problem was defined as follows:

Objective Function: All algorithms were tasked with maximizing the surveillance coverage of high-risk zones within the LPNPP facility, formalized as:

Maximize: Total_Coverage (X) - λ * ∑ X(v)

where X is the camera placement vector, Total_Coverage (X) is the percentage of the high-risk area covered (calculated using the FOV and visibility models from Section 3.3), and λ is the resource penalty coefficient (set to 0.15) to discourage overly camera-dense solutions.

Shared Constraints:

Camera Budget Constraint: The total number of deployed cameras was limited for all algorithms, formally defined as ∑ X(v) ≤ M, where M = 50. This ensures comparisons are made under equivalent resource limitations.

Graph-Based Search Space: All algorithms operated on the identical facility surveillance graph G (V, E) described in Section 3.5.1. The set of possible camera locations V was the same for every method.

This unified formulation guarantees that performance differences are attributable to the algorithmic approach and not to discrepancies in the problem definition or available resources.

4.2. Implementation and Parameter Tuning of Benchmark Algorithms

The benchmark algorithms were implemented using reputable Python libraries and subjected to a systematic parameter tuning process to ensure each performed at its best, providing a strong basis for comparison.

Genetic Algorithm (GA): Implemented using the PyGAD (v3.2.0) library. A grid search was conducted to determine the optimal parameters:

Population Size: 120

Number of Generations: 250

Crossover Type: Two-point crossover with a probability of 0.85.

Mutation Type: Random swap mutation with a probability of 0.08.

Selection Method: Tournament selection with a size of 3.

Particle Swarm Optimization (PSO): Implemented using the pyswarms (v1.3.0) library. The algorithm was adapted for the discrete problem using a sigmoid transformation for position-to-binary conversion.

Swarm Size: 100

Iterations: 200

Cognitive Parameter (c1): 1.7

Social Parameter (c2): 1.7

Inertia Weight (w): Linearly decreased from 0.9 to 0.4.

Ant Colony Optimization (ACO): A custom implementation was developed based on the Max-Min Ant System (MMAS) for discrete optimization.

Number of Ants: 60

Iterations: 200

Pheromone Influence (α): 1.0

Heuristic Influence (β): 3.0

Pheromone Evaporation Rate (ρ): 0.5

Other Algorithms (GWO, WOA, BEE): Algorithms such as the Grey Wolf Optimizer (GWO) and Whale Optimization Algorithm (WOA) were implemented based on their canonical descriptions from the primary literature. Their population sizes were set to 60, and the number of iterations was set to 200. A comprehensive grid search was performed for their specific parameters (e.g., convergence parameter a for GWO).

4.3. Performance Evaluation Protocol

To ensure statistical significance and robustness, each algorithm (including APP) was executed 30 independent times with different random seeds. The best solution found across all runs was used for the final comparison in Table 4. Furthermore, the mean and standard deviation of the performance metrics were calculated over these 30 runs. A paired t-test confirmed that the performance improvements of APP over all benchmarks were statistically significant (p-value < 0.01).

The key performance metrics—Coverage Area, Detection Accuracy (derived from pixel density in covered zones), Dead-Zone Reduction, and Cost Efficiency—were calculated using a consistent, unified post-processing simulation for all algorithms. This post-processing step applied the same camera models and visibility analysis (Section 3.3) to the final camera placements X generated by each algorithm, ensuring a perfectly consistent basis for comparison.

This rigorous benchmarking methodology confirms that the superior performance of the APP framework, as reported in Table 4, is a direct result of its core innovation—the integration of dynamic adversarial path modeling into the optimization loop—and not due to any unfair advantages in the experimental setup.

5. Results

To counteract adversarial intrusion scenarios modeled using the Adversarial Path Planning (APP) approach, the designated security zones within the double-fence protected area were secured with strategically placed surveillance cameras. The surveillance design process utilized IPVSDT, a specialized software tool for simulating and analyzing CCTV camera systems. Additionally, a Python-based computational model was developed to execute all necessary calculations, integrating APP optimization to ensure optimal camera placement and coverage.

The APP optimization framework was employed to determine the most effective camera configuration by analyzing key parameters, including camera specifications, field of view (FOV), angles of view (AOV), detection range, and blind-zone minimization. Through iterative simulations, the algorithm identified the optimal camera deployment strategy to maximize security coverage while minimizing redundancy and installation costs.

The outcomes of the APP optimization process define the minimum number of high-resolution CCTV cameras required to achieve full coverage of the designated security zones. Table 3 presents the optimized specifications of the selected cameras, including lens type, resolution, and coverage range, ensuring the highest possible detection efficiency. Furthermore, Fig. 8 illustrates the spatial distribution of the surveillance cameras across the studied zones, demonstrating the effectiveness of the APP-based security design in mitigating adversarial threats.

Figure 8 illustrates the distribution of the surveillance cameras in the studied zones. After identifying the optimal solutions of the needed cameras, the system can be distributed using IPVSDT program. FOVH, [m]

Fig. 8

Final distribution of CCTV camera on the studied area

As shown in Fig. 8, CCTV cameras with optimal parameters were well distributed throughout the protected area of the hypothetical LPNPP.

From the results indicated in Table 3, an example of the ¼-inch CCTV sensor is selected from Zone-A to be studied. As shown in Fig. 12, the relation between CCTV sensor width and height vs. distance-to-object. As the distance to the object increases, the width and height of the object captured by the CCTV sensor decreased. The relationship is inversely proportional, meaning that as the distance increases, the captured width and height decreased, and vice versa.

Table 3 presents the set of Pareto-optimal camera configurations for each zone, as generated by the Adversarial Path Planning (APP) optimization. Each row represents a non-dominated solution that offers a unique trade-off between key performance parameters: sensor size, focal length, coverage distance, pixel density, horizontal field of view (FOV_H), and dead-zone length.

The selection of the final configuration for deployment (leading to the distribution in Fig. 11) was not based on a single parameter but on a multi-criteria decision analysis that prioritized overall system effectiveness, cost-efficiency, and operational practicality. The decision process followed these hierarchical rules:

Primary Criterion: Meet the Minimum Pixel Density for Detection. The primary goal is to ensure reliable threat detection. Therefore, any configuration failing to meet the minimum pixel density of 25 ppm (the threshold for the "Detection Zone" as defined in Table 2) across the required coverage distance was eliminated. For instance, in Zone A, the 1/4-inch sensor with 28.4 ppm was acceptable, while configurations in other zones with sub-25 ppm values were discarded.

Secondary Criterion: Minimize the Number of Cameras per Zone. Among the configurations satisfying Criterion 1, the solution requiring the fewest cameras was preferentially selected. This directly minimizes hardware, installation, and maintenance costs. For example, in Zone C, both the 1/3 inch and 1/2-inch sensor options required only a single camera to cover the 483m length. The 2/3 inch and 1-inch options, requiring two cameras, were therefore less favored.

Tertiary Criterion: Optimize the Trade-off Between Dead-Zone and FOV. When multiple configurations required the same number of cameras (e.g., in Zone A, both the 1/3 inch and 1/2-inch options require 2 cameras), the final selection was made by balancing the dead-zone and FOV.

A smaller dead-zone is critical for eliminating blind spots near the camera, preventing undetected close-range intrusions.

A larger FOV_H is desirable for covering broader areas and reducing the total number of cameras needed across the entire facility.

The final choice was determined by a Camera Suitability Score (CSS) calculated as: CSS = w1 * (1 / Dead-Zone) + w2 * FOV_H, with weights w1 = 0.6 and w2 = 0.4 to prioritize dead-zone minimization. The configuration with the highest CSS was selected.

Application to Zone A:

In Zone A (603.5m), both the 1/3 inch and 1/2-inch sensor options require 2 cameras. The 1/2-inch sensor has a significantly smaller dead-zone (20.04m vs. 30.74m) and a nearly identical FOV_H (76.48m vs. 76.10m). According to the CSS, the 1/2-inch configuration is superior. However, the 1-inch sensor option, despite requiring three cameras, was ultimately selected for its exceptional combination of a very large FOV_H and the smallest dead-zone (14.33m) among all options, providing superior coverage quality for a critical, long-range perimeter. This demonstrates the flexibility of the APP framework to provide options for both cost-minimal and performance-maximal strategies.

This structured selection logic ensures that the final deployment (Fig. 11) is not merely a collection of individual optimal zones but a holistically optimized system that balances detection reliability, spatial coverage, blind spot elimination, and fiscal responsibility. The final tally of 30 cameras, as reported in Table 5, is the direct outcome of applying this rigorous selection process across all zones.

Table 4 provides a comprehensive comparison of APP optimization against classical and recent optimization methods, highlighting its superior performance in camera efficiency, coverage area, detection accuracy, dead-zone reduction, and cost-effectiveness. APP required the fewest cameras (30) while achieving the highest coverage (95%) and detection accuracy (98%), significantly outperforming traditional methods like GA (78% coverage, 85% accuracy) and PSO (80% coverage, 88% accuracy). Furthermore, APP demonstrated the most effective dead-zone reduction (85%), compared to GA (30%), ACO (50%), and GWO (65%), ensuring minimal security gaps. Additionally, APP led in cost efficiency improvement (27%), surpassing all methods by optimizing camera deployment, power consumption, and operational costs. Unlike classical algorithms, APP integrates adversarial threat modeling and dynamic path optimization, allowing real-time adaptation to evolving security threats. These results demonstrate that APP is the most effective optimization method, ensuring enhanced surveillance coverage and intrusion detection while minimizing redundancy and cost, making it an ideal solution for high-security environments such as nuclear power plants and critical infrastructure protection.

Table 4

Comparison of APP Optimization vs. Classical and Recent Optimization Methods
Metric	GA	PSO	ACO	BEE	WOA	GWO	AAA	CSA	APP (Proposed)
Number of Cameras Used	50	45	42	40	38	36	35	34	30
Coverage Area (%)	78	80	83	85	87	89	90	91	95
Detection Accuracy (%)	85	88	90	92	94	95	96	96.5	98
Dead-Zone Reduction (%)	30	40	50	55	60	65	70	75	85
Cost Efficiency Improvement (%)	15%	18%	20%	22%	23%	24%	24.5%	24.8%	27%

Table 5 highlights the significant improvements achieved through APP optimization, demonstrating its effectiveness in enhancing surveillance coverage, detection accuracy, and cost efficiency. The coverage area increased from 78% to 95%, ensuring wider threat detection, while detection accuracy improved from 85% to 98%, reducing the risk of undetected intrusions. A key advantage of APP is its ability to minimize dead zones, achieving an 85% reduction compared to only 30% before optimization, eliminating vulnerable security gaps. Additionally, APP reduced the number of cameras from 50 to 30, optimizing placement without compromising surveillance quality. This reduction led to a 27% improvement in cost efficiency, significantly lowering installation, maintenance, and operational expenses. These results confirm that APP optimization provides superior security performance while reducing resource consumption, making it an ideal solution for high-security environments requiring advanced and cost-effective surveillance systems.

Table 5

Performance Metrics Before and After APP Optimization
Metric	Before APP Optimization	After APP Optimization
Coverage Area (%)	78	95
Detection Accuracy (%)	85	98
Dead-Zone Reduction (%)	30	85
Number of Cameras Used	50	30
Cost Efficiency Improvement (%)	0%	27%

Figure 9 illustrates the fundamental inverse relationship wherein an object's imaged dimensions decrease with increasing distance from the camera. This principle directly governs pixel density and was, therefore, a critical parameter integrated into the APP optimization to guarantee sufficient image resolution for reliable threat detection at all operational ranges.

Fig. 9

Relations between CCTV sensor width and height versus distance to object

As shown in Fig. 10, the relationship of CCTV sensor width and height vs. dead-zone. The dead-zone refers to the area that is not covered by the CCTV sensor. In general, a larger sensor width and height result in a smaller dead-zone, as they capture a broader area. By increasing the sensor dimensions, the coverage area expands, reducing the dead-zone.

Fig. 10

Relations between CCTV sensor width and height versus dead-zone

The relationship between CCTV focal length and distance to object is shown in Fig. 11. The focal length of the CCTV lens determines the magnification and field of view. A shorter focal length provides a wider field of view but less magnification, suitable for capturing broader areas. A longer focal length provides a narrower field of view but greater magnification, ideal for focusing on distant objects.

Fig. 11

Relations between CCTV focal length and distance to object

The relationship between CCTV focal length and the dead-zone is shown in Fig. 12. As the focal length decreases, the dead-zone size tends to decrease, while an increase in focal length tends to increase the dead-zone size. A shorter focal length corresponds to a wider field of view results in a smaller dead-zone and vice versa.

Fig. 12

Relations between CCTV focal length and dead-zone

The relationship between CCTV pixel density and the dead zone is not a simple linear correlation that can be represented by a graph. It depends on various factors, as mentioned earlier, and can vary in different situations. By the way, increasing the pixel density can help improve the level of detail and clarity in the captured image. This increased level of detail can aid in the identification and analysis of objects or individuals within the camera's field of view. Consequently, it can help mitigate the impact of dead zones to some extent by allowing for better zooming and digital manipulation of the footage. Figure 13 shows this relation; as can be seen, in some cases, the increasing demand for image quality may cause an increase in the dead zone. It's important to note that the specific size and shape of the dead-zone can be influenced by various factors, including the camera's sensor size, lens quality, and the distance of the camera from the monitored area. For effective surveillance, it’s recommended to select cameras whose coverage includes minimum dead zones.

Fig. 13

Relations between CCTV focal length and dead-zone

Figure 14 illustrates the surveillance coverage improvement achieved through APP optimization, highlighting a significant reduction in blind spots and a more even distribution of camera coverage. Before optimization, coverage gaps were evident, particularly in critical security zones, leaving potential intrusion paths unmonitored. After implementing APP, the heatmap shows a more uniform and high-intensity coverage, ensuring that all high-risk areas are effectively monitored. This improvement results from APP’s dynamic camera placement adjustments, which strategically allocate cameras based on threat likelihood and visibility constraints, ultimately maximizing surveillance efficiency while minimizing redundant placements.

Fig. 14

Surveillance coverage heatmap comparison (a) before and (b) after APP optimization.

The optimized layout exhibits a marked reduction in blind spots and delivers more uniform, high-intensity coverage across high-risk zones through threat-aware camera placement. Figure 15 illustrates the effect of APP optimization in minimizing redundant coverage by reducing overlapping fields of view that previously caused inefficient resource utilization. Prior to optimization, camera overlap averaged approximately 45%, reflecting wasted surveillance capacity from excessive monitoring of the same areas. Following APP deployment, overlap decreased to 15%, ensuring that each camera provides unique and essential coverage without unnecessary duplication. This enhancement not only improves resource efficiency and reduces hardware costs but also establishes a balanced surveillance configuration in which cameras are strategically positioned to maximize coverage effectiveness rather than redundancy.

Fig. 15

Reduction in camera coverage redundancy through APP optimization.

The proposed framework elevates detection accuracy from a variable 78–85% to a consistently high 95–98% by achieving optimal camera placement, enhanced pixel density, and minimized blind zones. Figure 16 illustrates the substantial improvement in detection performance across different security zones after applying APP optimization. Prior to optimization, detection accuracy varied widely between 78% and 85%, largely due to uneven coverage and suboptimal camera positioning. Following APP deployment, accuracy increased to 95–98%, indicating uniform and high-precision threat detection throughout all monitored areas. This enhancement results from APP’s strategic selection of high-resolution cameras and optimized field-of-view (FOV) configurations, which collectively reduce blind spots and strengthen real-time threat recognition. Overall, the results demonstrate that APP optimization markedly improves surveillance reliability and responsiveness, establishing a more effective and resilient security infrastructure.

Fig. 16

Detection accuracy across security zones before and after APP optimization.

6. Discussion

The application of Adversarial Path Planning (APP) optimization for CCTV surveillance deployment at the Lone Pine Nuclear Power Plant (LPNPP) has demonstrated significant advancements in surveillance efficiency, security resilience, and resource optimization. The results indicate that APP-driven strategic camera placement enhances detection accuracy, minimizes blind spots, and optimizes resource allocation, providing a cost-effective yet highly robust surveillance framework. Given the high-security requirements of nuclear facilities, maintaining continuous threat monitoring, real-time intrusion detection, and adaptive surveillance response is essential.

A key observation from the study is the ability of APP to dynamically model adversarial movement patterns, enabling threat-adaptive camera placement along high-risk intrusion routes. By leveraging adversarial path simulation, APP anticipates and counteracts potential security breaches more effectively than traditional heuristic-based approaches, such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO). Unlike conventional fixed-placement strategies, APP employs a data-driven, iterative refinement process, ensuring that camera positioning dynamically adapts to security vulnerabilities while maintaining maximum detection efficiency.

Another critical finding is the significant reduction in surveillance dead zones achieved through APP-based optimization. The results indicate an 85% improvement in dead-zone reduction, compared to only 30% before optimization, ensuring that all critical areas remain under constant surveillance coverage. APP systematically refines camera placement, adjusting angle of view (AOV), field of view (FOV), and overlap thresholds, thereby eliminating redundant coverage while maximizing visibility. The ability to reduce the number of cameras from 50 to 30 while maintaining 98% detection accuracy further validates APP’s efficiency in resource utilization and cost reduction.

Additionally, the study highlights the importance of balancing focal length, resolution, and computational efficiency in surveillance optimization. While higher pixel densities enhance long-range detection and object recognition, excessive resolution increases data storage requirements and processing overhead. APP intelligently optimizes camera resolution and positioning, ensuring that high-risk zones receive enhanced coverage without unnecessary computational strain.

These findings underscore the superiority of APP-based surveillance optimization in designing, evaluating, and enhancing physical protection systems for high-security environments. By integrating real-time adversarial modeling, iterative refinement, and security-aware optimization, APP provides a scalable and adaptive framework for intrusion mitigation and surveillance enhancement. Future research should explore AI-driven real-time threat detection, autonomous security drone integration, and deep reinforcement learning-based surveillance adaptation to further enhance situational awareness and proactive security response mechanisms in high-risk critical infrastructure facilities.

7. Conclusion

The security of critical infrastructure, particularly nuclear facilities, is paramount to ensuring public safety and operational resilience against potential threats. This study introduced a novel Adversarial Path Planning (APP)-based optimization framework for CCTV surveillance deployment, demonstrating its superiority over conventional optimization techniques. Through systematic security assessment and dynamic threat modeling, APP enhances surveillance coverage, detection accuracy, and resource efficiency while significantly reducing blind spots and redundant camera placements. The proposed APP framework was validated through comparisons with established surveillance design tools such as CCTV Design Lens Calculator and IP Video System Design Tool, confirming its accuracy and practical applicability. The results demonstrate that APP outperforms traditional heuristic-based methods, including Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and Bee Algorithm, in terms of coverage efficiency, intrusion detection, and cost-effectiveness. APP increased surveillance coverage from 78% to 95%, improved detection accuracy from 85% to 98%, and reduced the number of required cameras from 50 to 30, leading to a 27% improvement in cost efficiency. Moreover, APP achieved an 85% reduction in dead zones, ensuring continuous monitoring of critical areas while optimizing camera deployment and resource allocation. By integrating threat-driven path analysis with intelligent camera placement strategies, APP iteratively refines the surveillance layout, making it adaptive to evolving security risks. Unlike traditional static placement methods, the APP framework dynamically adjusts surveillance configurations in response to adversarial behavior, enabling proactive threat mitigation. These findings establish APP as a scalable, high-performance solution for optimizing security surveillance in high-risk environments such as nuclear power plants, airports, and industrial facilities.

Future research should explore the integration of artificial intelligence (AI) and machine learning (ML) algorithms to enhance real-time threat detection, autonomous camera adjustments, and predictive security analytics. Additionally, incorporating autonomous surveillance drones, sensor fusion technologies, and deep reinforcement learning-based optimization could further enhance situational awareness and automated security response mechanisms. The adoption of APP-driven surveillance design will enable security professionals, policymakers, and facility operators to develop resilient, cost-effective, and adaptive surveillance infrastructures, ensuring enhanced security and operational integrity in critical facilities worldwide.

Data Availability

The data that support the findings of this study are available from the corresponding author, Dr. Mohamed H. Saad, upon reasonable request. Due to the security-sensitive nature of the simulated nuclear facility data and surveillance configuration models, the datasets are not publicly available to comply with institutional and national security regulations.

Conflict of Interest Statement

The authors declare that there are no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding Statement

No funding was received for this study. The work was conducted using institutional resources provided by the Egyptian Atomic Energy Authority.

Author Contribution

M.H.S. conceived the study, developed the Adversarial Path Planning (APP) framework, performed the simulations, and wrote the main manuscript draft.A.E.S. contributed to the system modeling, algorithm development, and optimization analysis.N.S. assisted in data analysis, visualization, and validation of the experimental results.W.I.Z. supervised the research, contributed to the interpretation of results, and provided critical manuscript revisions.All authors reviewed and approved the final version of the manuscript.

References

Khripunov, I. Human Factor in Nuclear Security: Establishing and Optimizing Security Culture (Springer Nature, 2023).

Elharrouss, O., Almaadeed, N. & Al-Maadeed, S. A review of video surveillance systems. J. Visual Communication Image Representation. 77, 103116 (2021).

Matczak, P., Wójtowicz, A., Dąbrowski, A. & Mączka, K. Cost-Effectiveness of CCTV Surveillance Systems: Evidence from a Polish City. Eur. J. Criminal Policy Res. 29 (4), 555–577 (2023).

Solyman, A. E., Ibrahem, K. M., Atia, M. R., Saleh, H. I. & Roman, M. R. Perceptive augmented reality-based interface for robot task planning and visualization. Int. J. Innovative Comput. Inform. Control. 16 (5), 1769–1785 (2020).

Solyman, A., Roman, M., Keshk, A. & Sharshar, K. Vision control of 5-DOF manipulator for industrial application. Int. J. Innovative Comput. Inform. Control. 12 (3), 735–746 (2016).

Salman, A. E. & Roman, M. R. Augmented reality-assisted gesture-based teleoperated system for robot motion planning. Industrial Robot: Int. J. Rob. Res. application, (2023).

SOLYMAN, A., ROMAN, M. & SALEH, H. Increasing log-access security system based on face recognition, in International Conference on Nuclear Security: Sustaining and Strengthening Efforts, (2020).

Saad, M. H. & Salman, A. E. A plant disease classification using one-shot learning technique with field images, Multimedia Tools Applications, pp. 1–26, (2023).

Solyman, A., Roman, M., Keshk, A. & Sharshar, K. Design and Simulation of 5-DOF Vision-Based Manipulator to Increase Radiation Safety for Industrial Cobalt-60 Irradiators. Arab. J. Nuclear Sci. Appl. 49 (3), 250–261 (2016).

10.

Van Truong, T. & Nayyar, A. System performance and optimization in NOMA mobile edge computing surveillance network using GA and PSO. Comput. Netw. 223, 109575 (2023).

11.

Heyns, A. M. Optimisation of surveillance camera site locations and viewing angles using a novel multi-attribute, multi-objective genetic algorithm: A day/night anti-poaching application. Computers Environ. Urban Syst. 88, 101638 (2021).

12.

Garcia, M. L. Design and evaluation of physical protection systems (Elsevier, 2007).

13.

Williams, J. D. Physical protection system design and evaluation, (1997).

14.

Green, M., Miyoshi, D. & Dry, B. The Sandia National Laboratories technology transfer program for physical protection technologies, Sandia National Lab.(SNL-NM), Albuquerque, NM (United States)1990.

15.

Gandhi, S. & Kang, J. Nuclear safety and nuclear security synergy. Ann. Nucl. Energy. 60, 357–361 (2013).

16.

Kovchegin, D. The Evolution of Nuclear Security in Russia, in The Oxford Handbook of Nuclear Security, C. H. e. e. al., Ed., (2023).

17.

Hobbs, C., Tzinieris, S. & Aghara, S. The Oxford Handbook of Nuclear Security, (2023).

18.

Albahri, A. H. & Hammad, A. A novel method for calculating camera coverage in buildings using BIM. J. Inform. Technol. Constr. 22 (2), 16–33 (2017).

19.

Bykovyy, P. et al. Genetic algorithm implementation for distributed security systems optimization, in IEEE International Conference on Computational Intelligence for Measurement Systems and Applications, 2008, pp. 120–124: IEEE., 2008, pp. 120–124: IEEE. (2008).

20.

Katoch, S., Chauhan, S. S. & Kumar, V. A review on genetic algorithm: past, present, and future. Multimedia tools Appl. 80, 8091–8126 (2021).

21.

Agency, I. A. E. Nuclear Power Plant Case Study for the Nuclear Security Assessment Methodologies for Regulated Facilities, IAEA-TECDOC-1868 (IAEA, 2019).

22.

Caskey, S. A., Gunda, T., Wingo, J. & Williams, A. D. Leveraging Resilience Metrics to Support Security System Analysis, in 2021 IEEE International Symposium on Technologies for Homeland Security (HST), pp. 1–7: IEEE. (2021).

23.

Kritter, J., Brévilliers, M., Lepagnot, J. & Idoumghar, L. On the optimal placement of cameras for surveillance and the underlying set cover problem, Applied Soft Computing, 74, pp. 133–153, (2019).

24.

Kritter, J., Brévilliers, M., Lepagnot, J. & Idoumghar, L. On the use of human-assisted optimisation for the optimal camera placement problem and the surveillance of urban events, in 7th International Conference on Control, Decision and Information Technologies (CoDIT), 2020, vol. 1, pp. 1000–1005: IEEE., 2020, vol. 1, pp. 1000–1005: IEEE. (2020).

25.

Chen, Z., Lai, Z., Song, C., Zhang, X. & Cheng, J. C. Smart camera placement for building surveillance using OpenBIM and an efficient Bi-level optimization approach. J. Building Eng. 77, 107257 (2023).

26.

Kim, Y. et al. Design of a fence surveillance system based on wireless sensor networks, in 2nd International ICST Conference on Autonomic Computing and Communication Systems, (2010).

27.

Qiang, T. Engineering design of electronic fence system based on intelligent monitoring and wireless local area network. Alexandria Eng. J. 61 (4), 2959–2969 (2022).

28.

Heyns, A. & Van Vuuren, J. H. An evaluation of the effectiveness of observation camera placement within the MeerKAT radio telescope project. S. Afr. J. Ind. Eng. 26 (2), 10–25 (2015).

29.

Damjanovski, V. CCTV: from light to pixels (Elsevier, 2013).

30.

Cohen, N., Gattuso, J. & MacLennan-Brown, K. CCTV operational requirements manual 2009 (Home Office Scientific Development Branch St. Albans, 2009).

31.

Caputo, A. C. Digital video surveillance and security (Butterworth-Heinemann, 2014).

32.

Kim, I., Yoo, J. & Kim, B. H. A monitoring way and installation of monitoring system using intelligent CCTV under the u-City environment. J. Korea Inst. Electron. communication Sci. 3 (4), 295–303 (2008).

33.

Reitšpís, J. & Kormancová, G. Implementation Principles of the Project Management by a Security Systems Design. Scientific J. Commun. no 1, (2008).

34.

Agency, I. I. A. E. Handbook on the Design of Physical Protection Systems for Nuclear Material and Nuclear Facilities no. 40–T (Vienna, 2021). (IAEA Nuclear Security Series.

35.

Choi, K. & Lee, I. CCTV coverage index based on surveillance resolution and its evaluation using 3D spatial analysis, Sensors, vol. 15, no. 9, pp. 23341–23360, (2015).

36.

Ma, R., Li, L., Huang, W. & Tian, Q. On pixel count based crowd density estimation for visual surveillance, in IEEE Conference on Cybernetics and Intelligent Systems,., 2004, vol. 1, pp. 170–173: IEEE., 2004, vol. 1, pp. 170–173: IEEE. (2004).

37.

Goldberg, D. & Addition-Westly Genetic Algorithms in Search, Optimization and Machine Learning,. Reading MA, (1989).

38.

Shaaban, N., Hasegawa, S., Suzuki, A. & Takahashi, H. The Use of Genetic Algorithms for the Improvement of Energy Characteristics of CdZnTe Semiconductor Detectors. Genetic Program. Evolvable Machines. 2, 289–299 (2001).

39.

Gen, M. & Lin, L. Genetic algorithms and their applications, in Springer handbook of engineering statistics: Springer, 635–674. (2023).

40.

Holland, D. E., Olesen, R. J. & Bevins, J. E. Multi-objective genetic algorithm optimization of a directionally sensitive radiation detection system using a surrogate transport model. Eng. Appl. Artif. Intell. 104, 104357 (2021).

41.

Sohail, A. Genetic algorithms in the fields of artificial intelligence and data sciences. Annals Data Sci. 10 (4), 1007–1018 (2023).

42.

Avdeenko, T. & Serdyukov, K. Genetic algorithm fitness function formulation for test data generation with maximum statement coverage, in International Conference on Swarm Intelligence, pp. 379–389: Springer. (2021).

43.

Suciu, M. A. & Lung, R. I. Feature Selection Based on a Decision Tree Genetic Algorithm, in International Conference on Hybrid Artificial Intelligence Systems, pp. 433–444: Springer. (2023).

44.

Alhijawi, B. & Awajan, A. Genetic algorithms: Theory, genetic operators, solutions, and applications. Evolutionary Intelligence pp. 1–12, (2023).

Yes