UAV-Based Infrared Thermography for Qualitative and Quantitative Building Energy Assessment: A Review

Abstract

The growing demand for energy-efficient buildings and the urgent need to retrofit aging infrastructure have driven increased interest in advanced diagnostic technologies. Among these, unmanned aerial vehicle (UAV)-based infrared thermography (IRT) has emerged as a promising non-destructive technique for assessing the thermal performance of building envelopes. This review examines recent developments and applications of dynamic infrared thermography (IRT) in the building sector for both qualitative and quantitative thermal assessment, based on previously conducted studies. It highlights the increasing adoption of integrated UAV-based IRT for building inspection and diagnostics, and critically reviews the operational, technical, and methodological advancements in dynamic thermography achieved over the past decade. Furthermore, the review presents a comprehensive framework for operational planning, encompassing environmental conditions, infrared camera configuration, and optimal UAV flight parameters. The key findings identify major challenges associated with dynamic IRT applications, particularly those related to measurement accuracy that currently limit its use for quantitative assessments and synthesize proposed methodologies to address these limitations. The review also highlights the absence of standardized procedures for determining emissivity and reflected apparent temperature in dynamic measurement setups and discusses potential approaches to overcome these gaps. Finally, it outlines priority directions for future research to support the reliable and consistent application of dynamic IRT in quantitative analysis and provides a reference for energy auditors and thermography practitioners to inform the selection of appropriate procedures for accurately quantifying heat loss in building envelopes.

1. Introduction

In the context of the ongoing focus on energy conservation, the construction industry is under increasing focus not just to create designs for energy-efficient buildings but also to develop novel retrofitting approaches and improve existing structures [1]. As per the International Energy Agency (IEA) [2], the collective energy consumption stemming from both building operations and construction activities contributes to 30% of the global energy usage. Additionally, these sectors are accountable for 26% of both direct and indirect greenhouse gas (GHG) emissions [2]. In Canada, buildings play a significant role, with residential buildings accounting for 13% and commercial buildings for 12% of total energy demand [3]. Consequently, there is a strong emphasis on enhancing the energy efficiency of buildings. For instance, the Government of Canada has implemented an updated building code that places greater emphasis on energy efficiency in new structures [4]. While these policies primarily focus on new constructions, existing buildings, which comprise 66% of the building sector, still face several energy efficiency challenges including poor insulation, air leakages and moisture damages [5]. More than 60% of existing buildings are over 30 years old and about 40% are over 50 years old [6] and these types of building usually face energy efficiency issues mainly due to the aging and material degradation [7]. Aging and material degradation lead to increased energy consumption. In addition, issues such as mold, moisture intrusion, thermal bridges, and missing or deteriorated insulation further elevate the energy demand of buildings. Consequently, these buildings require proper assessment and the adoption of effective retrofit strategies to reduce their energy consumption. Therefore, reducing building energy demand is essential, and any strategy that contributes to this goal is highly valuable [8].

Several methods are available for assessing the thermal performance of building envelopes, including heat flux measurement, blower door testing, moisture content measurement, and infrared thermography. Infrared thermography (IRT) is one of the most widely used methods for assessing the thermal performance of buildings. It is a non-destructive, non-contact technique that utilizes infrared cameras to measure surface temperatures and generate 2d thermal maps of objects for visualizing their thermal patterns [9,10]. In the building sector, IRT has long been employed to identify thermal anomalies such as thermal bridges, cracks, air leakage, and missing insulation, as well as to estimate parameters like the thermal transmittance (U-value). With advancements in technology, the integration of infrared cameras with unmanned aerial vehicles (UAVs) has gained increasing popularity. UAV-based IRT allows for faster data collection compared to stationary systems and enables access to difficult-to-reach areas, such as tall façades and rooftops [11]. Traditionally, such inspections relied on scaffolding, ladders, or rope-access technicians, which are time-consuming, costly, and pose safety risks. UAVs allow remote data collection, reducing these hazards while maintaining high-resolution imaging. Despite regulatory and operational challenges, drones offer a safer, more efficient alternative for building envelope diagnostics, especially on tall or complex structures. As a result, the use of dynamic IRT in building diagnostics continues to grow rapidly.

Despite the growing interest in UAV-based infrared thermography (IRT) for building diagnostics, most existing review studies have primarily focused on specific aspects of the technology, such as environmental inspection conditions, UAV flight planning strategies, or the qualitative detection of thermal anomalies [12]. As a result, a comprehensive synthesis that integrates both qualitative and quantitative applications of UAV-based IRT remains limited. In particular, the literature lacks a systematic discussion of the operational requirements, environmental constraints, camera configuration parameters, and UAV flight settings that influence the reliability of thermographic measurements under dynamic conditions. The purpose of this review is therefore to provide a comprehensive and structured overview of UAV-based IRT applications for building inspection and diagnostics, with particular emphasis on bridging the gap between qualitative anomaly detection and emerging quantitative thermographic assessments. Unlike previous reviews, this work synthesizes the literature from both methodological and operational perspectives by examining environmental inspection conditions, camera stabilization requirements, and UAV flight parameters, and by proposing application-oriented guideline tables and inspection workflows derived from the literature. Through this synthesis, the review identifies key methodological challenges affecting quantitative accuracy in dynamic UAV inspections and highlights areas where standardized inspection protocols and improved measurement methodologies are still needed. By consolidating these findings, this study aims to provide a reference framework for researchers and practitioners, supporting the development of more reliable and standardized UAV-based thermographic diagnostics in the building sector. As a result, a comprehensive and systematic overview of UAV-based IRT applications encompassing both qualitative and quantitative assessment remains lacking. The purpose of this review is to provide a comprehensive overview of current applications of unmanned aerial vehicles (UAVs) in building inspection and diagnosis, with a focus on the most commonly adopted qualitative and quantitative IRT techniques. The review analyzes prevailing research trends and key research domains within the existing literature, and critically examines analytical methods applied during inspections, with particular emphasis on factors influencing quantitative accuracy under dynamic conditions. Furthermore, it highlights technical and methodological challenges that require continued investigation and aims to serve as a reference framework to support future research and the advancement of quantitative UAV-based thermographic diagnostics in the building sector.

The structure of the paper is as follows: (1) Methodology of the Review, which describes the literature search strategy, including the databases consulted, keywords used, and inclusion and exclusion criteria applied to select relevant studies; (2) Background, which introduces the fundamental principles and key components of infrared cameras, discusses relevant standards for selecting appropriate devices, and reviews different types of UAV platforms; (3) Applications of UAV-Based IRT in the Building Sector, which reviews existing qualitative and quantitative techniques and summarizes key findings related to building energy assessment using UAV-based IRT; (4) Operational Planning, which provides guidance on optimizing UAV flight parameters and recommends suitable environmental conditions for effective thermographic surveys; (5) Challenges and Limitations of UAV-Based IRT, which highlights key technical and methodological limitations and identifies areas requiring further research; and the (6) Conclusion, which summarizes the major contributions of this review and outlines practical steps for performing UAV-based building energy assessment surveys.

2. Methodology of the Review

This review was conducted through a systematic survey of peer-reviewed literature on UAV-based infrared thermography (IRT) for building inspection and energy assessment. The search strategy, selection criteria, and classification process are described below to ensure transparency and reproducibility.

Relevant publications were identified through three major academic databases, Scopus, Web of Science, and Google Scholar, to ensure comprehensive coverage across engineering, building science, and remote sensing disciplines. The search was restricted to studies published between 2013 and 2025, a period that reflects the accelerated development and commercialization of UAV platforms and compact thermal imaging sensors. Search queries were constructed using combinations of the following keywords: “UAV thermography,” “infrared thermography building,” “drone thermal inspection,” “building envelope diagnostics,” “dynamic infrared thermography,” “UAV building inspection,” and “building energy assessment.” Boolean operators (AND, OR) were applied systematically to refine queries and isolate studies specifically addressing UAV-based thermographic inspection in the built environment.

To ensure the relevance and methodological quality of the reviewed body of literature, explicit inclusion and exclusion criteria were applied during the screening process. Studies were included if they: (i) investigated the application of UAV-mounted infrared thermography within the building sector; (ii) addressed at least one of the following themes, building envelope diagnostics, thermal anomaly detection, energy performance assessment, U-value estimation, or defect characterization; and (iii) provided sufficient methodological or experimental detail to enable meaningful comparison and synthesis. Peer-reviewed journal articles and conference papers constituted the primary source types. Internationally recognized technical standards, including relevant ASTM and ISO guidelines, were additionally incorporated to provide reference frameworks for thermographic inspection procedures, environmental operating conditions, and equipment performance specifications.

The screening process followed a structured, multi-stage workflow, as illustrated in Figure 1. The initial database search returned 423 records in total across Scopus, Web of Science, and Google Scholar. Following the removal of 111 duplicates, 312 unique records remained and were subjected to title and abstract screening against the inclusion and exclusion criteria, resulting in the exclusion of 198 records that were clearly outside the defined scope. Of the 114 remaining records sought for retrieval, 10 could not be accessed in full text and were excluded. Full-text review was subsequently performed on the remaining 104 candidate publications.

Following the selection process, the retained studies were systematically analyzed and classified according to application type (qualitative thermal anomaly detection versus quantitative thermal characterization); inspection methodology and data processing workflow; environmental and operational requirements; camera configuration and calibration approach; and UAV platform and flight parameters. This classification framework enabled structured cross-study comparison, identification of recurring methodological patterns, and systematic mapping of knowledge gaps within the current literature. The findings are presented in two dedicated sections, qualitative and quantitative IRT applications, supported by comparative figures and summary tables to facilitate interpretation.

3. Background

This section explains the function of each major component to help better understand how infrared cameras work. It also aims to provide practical guidance for selecting the right infrared camera for dynamic thermographic applications based on performance standards and operational needs. In addition, criteria for choosing a suitable UAV platform will be discussed, as the effectiveness of a dynamic IRT survey also depends on selecting a drone that can operate efficiently in the intended environment. Infrared cameras are available in various models, but they typically share a common structure comprising several key components: optics, detector, electronics, and, in some models, a cooling system. These elements are integrated into a housing unit that also includes a power supply and user interface or display.

The detector is the core component responsible for converting incoming infrared radiation emitted by an object into an electrical signal [13]. Detectors are generally categorized into two types based on their operating principles: photon and thermal detectors. Photonic detectors operate by detecting changes in the distribution of electronic energy within a semiconductor material, which occur due to variations in the concentration of free charge carriers when infrared radiation is absorbed [14]. However, the output signal from these detectors is weak and easily masked by thermal noise generated through the random generation and recombination of charge carrier [15]. To mitigate this issue, photonic detectors require a cryogenic cooling system, often a mechanical cryocooler, that typically lowers the operating temperature to around −196 °C [13]. While these systems enhance detector sensitivity and performance, they also make photonic infrared cameras bulkier, more expensive, and more power-intensive [16]. Additionally, the presence of moving parts and helium seals in cryocoolers necessitates periodic maintenance and replacement [14]. Despite these drawbacks, photonic detectors offer high sensitivity (±0.01 °C) [17], excellent signal-to-noise ratio, and rapid response times [15]. In contrast, thermal detectors detect infrared radiation by absorbing it, which causes a temperature change in the detector material. This temperature variation induces a measurable change in a physical property of the material, such as resistance, resulting in an electrical output signal [18]. Although thermal detectors generally have lower sensitivity (±0.1 K) and slower response times compared to photonic detectors [17], they offer significant advantages in terms of weight, cost, and power consumption, as they do not require cryogenic cooling [19]. In UAV-based thermographic surveys, thermal infrared cameras are predominantly used. The weight and high power consumption of photonic cameras make them unsuitable for most UAV applications, despite their higher sensitivity and performance compared to thermal camera.

Another essential component of infrared cameras is the optical system, which uses lenses and mirrors to focus and direct incoming infrared radiation onto the detector [20]. The choice of lens material depends on the specific wavelength range in which the camera operates. Germanium and zinc selenide are among the most commonly used materials for infrared lenses due to their high transmittance in the 8–14 μm wavelength range [21]. This spectral range is particularly significant because it aligns with the thermal radiation emitted by most objects at ambient temperatures and is also the typical operational range for thermal infrared cameras. The final key component of an infrared camera is the readout circuit. After the detector captures the infrared radiation, the resulting analog signal is transmitted to the camera’s electronic circuitry, where it is converted into a digital value for further processing before turn it into a thermal image as shown in Figure 2 [22]. Infrared cameras also include an integrated signal processing unit that handles various image processing tasks. Since detectors may exhibit non-uniform responses, especially in thermal systems, which are susceptible to thermal drift, the camera employs a calibration method known as Non-Uniformity Correction (NUC) [10]. This process typically involves periodic calibration against a uniform thermal reference, often achieved by briefly closing an internal shutter or activating a flag, allowing the system to adjust pixel-specific offset and gain values [23]. In addition to NUC, infrared cameras also perform other corrections such as bad pixel replacement, noise reduction, and other filtering techniques to enhance image quality and stability.

Several standards provide minimum performance requirements for infrared cameras used in building envelope thermographic surveys, helping practitioners select appropriate equipment for reliable diagnostics. One critical parameter is the Minimum Resolvable Temperature Difference (MRTD), which quantifies an operator’s ability to discern small temperature differences using an infrared imaging system [24]. According to ASTM E1213 [25] MRTD is measured by projecting a test chart with four periodic bars (aspect ratio 1:7) onto a blackbody source. The infrared camera is focused on this chart, and the temperature difference is gradually decreased until the target pattern becomes invisible. It is then increased until the pattern is just visible again. This threshold is recorded as the MRTD for that specific spatial frequency. The procedure is repeated across multiple spatial frequencies and typically conducted at least three times for consistency. For example, CNMCS 02 27 13 [26], the Canadian National Master Construction Specification for thermographic assessment of building envelopes, specifies that infrared cameras should be capable of detecting a Minimum MDTD of 0.1 °C at 30 °C, with an operational range of −20 °C to 100 °C or better. ASTM C1060 [27] provides guidance for qualitative thermal imaging of insulation deficiencies, offering a formula that incorporates thermal resistance (R-value), surface film coefficients, and the difference in ambient indoor and outdoor temperatures. For common residential applications (e.g., R-10 to R-15 insulation), the thermal imaging system should have a MRTD of at least 0.2 °C. ASTM C1153-10 [24] further stipulates that thermal cameras must operate within a spectral range of 2 to 14 µm and exhibit an MRTD no greater than 0.3 °C at 20 °C.

Another critical performance metric is the Instantaneous Field of View (IFOV), which defines the smallest angular resolution the system can achieve, representing its ability to distinguish thermal radiation from fine structural details [24]. In practical terms, IFOV determines the smallest object that can be resolved by a single detector pixel, and therefore directly influences the spatial resolution of thermal images. A smaller IFOV corresponds to higher spatial resolution and enables the camera to detect smaller defects or thermal anomalies on the building envelope. Because IFOV is an angular measure, the physical area represented by each pixel increases as the distance between the camera and the target increases. Consequently, when UAVs operate at larger distances from the building façade, each pixel represents a larger surface area, which reduces the ability to detect small-scale defects such as narrow air leakage paths or thin thermal bridges. CNMCS 02 27 13 [26] recommends a minimum IFOV of 1.5 milliradians for exterior inspections. For qualitative thermographic assessments, ASTM C1060 [27] recommends that the IFOV meet the condition IFOV ≤ (500 × s)/d, where s is the width of the framing member and d is the distance between the infrared camera and the surface. For instance, if a structural framing member is 0.0381 m wide and the viewing distance is 5 m, the IFOV must be less than 3.8 milliradians to resolve the framing detail adequately. According to ASTM C1153-10 [24], detecting surface anomalies from a drone-mounted system is feasible only if the anomaly width is at least 0.002 × IFOV × d. The maximum allowable IFOV for such conditions is given by: IFOV = 150/d, where d is the distance from the sensor to the target in meters.

Based on the aforementioned parameters, the selection of a suitable infrared camera, both in terms of specifications and performance class, should align with the objectives of the thermographic survey and available budget. Equally important is the choice of an appropriate UAV platform to support data acquisition. A UAV is defined as an aircraft that operates without a pilot onboard and is remotely controlled [28]. UAVs can be broadly categorized into three types based on their aerial platform: fixed-wing, multirotor, and hybrid UAVs [29]. Fixed-wing UAVs resemble conventional airplanes with rigid wings that generate lift during forward motion. They are energy-efficient in flight as they do not require continuous propulsion to remain airborne and their payload capacity is often limited to 2–15 kg [28]. However, they typically require a horizontal runway or open space for takeoff and landing, which limits their operational flexibility in confined environments [30]. In contrast, multirotor UAVs, such as quadcopters, hexacopters, and octocopters, achieve lift and propulsion through multiple vertically oriented rotors. These platforms are highly maneuverable, capable of vertical takeoff and landing [31], and can hover and move freely in all directions [29]. While multirotor UAVs offer better control and spatial flexibility, they exhibit higher energy consumption, particularly during hovering, and generally have shorter flight times usually around 20 to 40 min than fixed wing systems with 60 to 120 min [28]. Their payload capacity typically ranges between 0.5 and 3 kg, which is sufficient for most lightweight infrared cameras used in building inspections. Lastly, Hybrid UAVs combine the vertical takeoff and hovering capabilities of multirotor platforms with the efficient forward flight of fixed-wing designs [32]. This hybrid configuration provides enhanced maneuverability during takeoff and landing, along with extended range and endurance during cruise flight. Despite these advantages, hybrid UAVs are still considered technologically complex and are largely in the experimental phase, with limited commercial availability [28]. Given that most dynamic IRT applications in the building sector are conducted in dense urban environments, where space for runway operations is constrained, multirotor UAVs are the preferred choice by both researchers and professionals. Their ability to operate in tight spaces, ease of deployment, and capacity to carry infrared camera payloads make them highly suitable for urban thermal inspection missions. Stable hovering capability is essential for minimizing motion blur, maintaining consistent observation angles, and reducing variations in imaging distance during thermal data acquisition. UAVs with higher positional stability and precise flight control can therefore improve the reliability of surface temperature measurements and the overall quality of thermographic analysis. Consequently, despite their shorter flight endurance, multirotor UAVs remain the most suitable platform for UAV-based infrared thermography in urban building inspections, where stability, maneuverability, and controlled hovering are critical for accurate data collection.

The following sections will review current research on the application of dynamic infrared IRT, with a focus on the specific objectives pursued in each study. Subsequent chapters will explore the key phases involved in conducting dynamic IRT surveys, including environmental conditions, flight parameters, and equipment preparation. Finally, the challenges and limitations associated with dynamic IRT will be examined in detail.

4. Applications of UAV-Based IRT in the Building Sector

Infrared thermography applications in the building sector can be broadly categorized into two main groups based on their objective: qualitative and quantitative [33]. Qualitative IRT is primarily used to visualize temperature distribution patterns in order to detect potential deficiencies in the building envelope. This method identifies irregularities in thermal images, such as color gradients and intensity differences, to reveal thermal anomalies [34]. Common applications include the detection of moisture damage [35], air leakage [36,37], missing insulation [38] and thermal bridges. For instance, during external IRT inspections conducted in winter, areas with missing insulation, cracks, or air leaks typically appear as hot spots [39]. This occurs because a defect-free envelope exhibits uniform thermal behavior, while disruptions such as voids or material inconsistencies cause localized increases in heat flows and appear as hot spot in thermal image [40]. Air leakages, in particular, show as warmer areas resulting from indoor air escaping through the building envelope. These are intensified by increase in pressure differentials and air flow [41], and appear as linear thermal streaks in thermal image [37]. It is important to note that air leakage patterns are directional and plume-like, unlike insulation defects or thermal bridges which are broader or structural. Visibility of these patterns strongly depends on the pressure difference across the building envelope; during winter inspections, buildings are often pressurized or depressurized using a blower-door test so that warm indoor air is forced outward and appears clearly in thermal images. Without this artificial pressurization, leakage patterns can be weak or masked by environmental factors such as wind and solar radiation. Thermal bridge is a part of the building envelope where heat flows more easily due to differences in material, geometry, or construction, leading to localized heat loss [42]. These typically appear as regularly spaced hot spots caused by structural elements or hot spots in geometrically distinct areas like joints and corners [43]. Additionally, moisture accumulation within the envelope can be identified through non-uniform surface temperature patterns. Such areas often appear as cold spots, primarily due to the cooling effect of evaporation [39].

In contrast, quantitative IRT focuses on the measurement and numerical evaluation of thermal properties across the building envelope [44]. It is typically used to calculate parameters such as the U-value, which represents the thermal transmittance of a building element [45]. It should be emphasized that IR cameras do not measure the U-value directly; they only measure surface temperature. Quantitative estimates of U-value require combining surface temperature data with known material properties and boundary conditions. Both qualitative and quantitative IRT techniques have been applied widely in stationary and dynamic setups. The following section reviews recent studies on dynamic IRT in the building sector. The discussion highlights key aspects of these studies, including their specific objectives, methodological innovations, and current limitations. The analysis is organized according to the objective of each study, qualitative or quantitative.

4.1. Qualitative Studies

Recent studies have increasingly explored dynamic UAV-based infrared thermography for qualitative assessment of building envelope conditions. These applications primarily focus on detecting thermal anomalies associated with insulation defects, moisture intrusion, and air leakage in building envelopes

Early work demonstrated the effectiveness of UAV thermography for building inspections. Ortiz-Sanz et al. [46] investigated the detection of thermal anomalies in traditional semi-buried wine cellars located in northwest Spain. The study compared UAV-based inspections with inspections conducted using a pole-mounted thermal camera. The results showed that both methods were effective in identifying the loss of thermal insulation in roof tiles over time and the presence of moisture in the walls. The authors noted that UAV-based inspections were more efficient and timesaving for exterior surveys. In contrast, for indoor environments or areas inaccessible to drones, pole-mounted cameras offered a practical alternative for inspecting hard-to-reach locations.

Building on this, researchers explored how 3D thermal modeling could enhance spatial interpretation of thermographic data. Gil-Docampo et al. [47], who combined UAV-based and pole-mounted thermal imaging to generate a 3D thermal model of a wine cellar for detecting thermal anomalies. They utilized Structure from Motion and Multiview Stereo photogrammetry software, Agisoft Metashape™, to construct three-dimensional models. Since thermal cameras produce large pixel sizes and often lead to poor image alignment during 3D reconstruction, the researchers applied a preprocessing workflow in MATLAB to enhance image quality. This included converting color thermal images to greyscale to retain temperature data, adjusting histograms for better contrast, and using Gaussian filters to reduce noise and sharpen the images. As a result, the number of successfully aligned thermal images increased from 153 to 199. Despite the lower resolution compared to RGB images, the final 3D thermal model successfully identified areas of heat loss and moisture on the building’s roof and walls. Additional work has further expanded this concept. Additional work has further expanded this concept.

Roof inspection has emerged as one of the most active application areas for UAV-IRT, given the difficulty of traditional access methods. Moore et al. [48] utilized dynamic IRT for remote roof inspections to identify damage related to water, heat, and HVAC system issues. Using a dual-camera setup, they identified cracks in slate shingles caused by mineral inclusions, missing shingles, roof damage from long-term water accumula-tion, drainage issues indicated by small standing water pools, and rusting HVAC pipes due to steam condensation. The study concluded that this method offers a time-efficient approach for assessing rooftop conditions and conducting spot inspections of facilities. A persistent technical challenge in UAV-IRT roof inspections is the autogain control problem, where identical temperatures are scaled differently across images during orthomosaic generation. Zhang et al. [49] conducted a study at York University focusing on the thermal infrared inspection of roof insulation using UAVs. The study aimed to address the autogain control problem of thermal cameras and improve the accuracy of thermal anomaly detection. The autogain control issue arises during orthomosaic image generation, where points with the same apparent temperature are scaled differently in different photos. To mitigate this problem, the researchers developed a relative thermographic radiometric calibration algorithm. This algorithm uses histogram matching technic which use one image as a reference image and then compute histogram of each image and then use their discrete cumulative distribution function to calibrate image. For anomaly detection, the researchers proposed a superpixel Markov Random Field model, which treats aggregated pixels with similar grayscales and spatial patterns as basic units, rather than analyzing individual pixels. They state that this approach improved detection accuracy compared to traditional pixel-based methods.

A key limitation of most conventional thermal surveys is their reliance on single, steady-state snapshots, which miss transient anomalies that only appear under specific diurnal or seasonal conditions. Rakha et al. [50] addressed this by introducing a time-lapse, drone-based thermography workflow that captures sequential IR imagery and constructs four-dimensional models (3D geometry over time). These thermal datasets are directly integrated into a BEM to better reflect the dynamic behavior of building envelopes. In their field study, they capture thermographic data every two hours from 09:00 to 17:00. After temperature normalization the temperature range of each photo, the thermal images were processed in Agisoft Metashape to generate point clouds and orthomosaics for each timestep. These were then imported into Rhino/Grasshopper for defect mapping and subsequently linked to EnergyPlus using Honeybee. Parametric simulations, which involved gradually lowering the R-values of identified defective areas, revealed that these anomalies accounted for an estimated increase of 6.45 MWh/year in heat loss and a 2 kWh/m²/year rise in the building’s Energy Use Intensity. Notably, certain defects, such as roof moisture and slab-edge thermal bridges, were detectable only during midday or late afternoon.

As datasets from UAV-IRT campaigns have grown, researchers have turned to machine learning to automate defect classification. Dabetwar et al. [51] conducted a sensitivity analysis to evaluate the performance of different neural network models in classifying heat loss locations in buildings, focusing on the impact of dataset size and number of training epochs on model accuracy. They developed a dataset comprising three classes of heat loss: (1) leakage through doors and the adjacent wall, (2) leakage due to insulation damage in walls, and (3) leakage through window seals. Using dynamic IRT, they collected 98, 91, and 73 samples for each class, respectively. Four classification models were tested: a general five-layer convolutional neural network (CNN), VGG16, transfer learning with VGG16, and transfer learning with InceptionV3. The general CNN outperformed the others, achieving the highest classification accuracy even with a relatively small dataset. InceptionV3 with transfer learning showed the second-best performance, while VGG16 was the least accurate. Their results indicated that smaller datasets require more epochs to achieve high accuracy, while larger datasets reduce the needed epochs but increase training time. The optimal configuration was found to be the general CNN trained with 128 images over 5 epochs, which achieved perfect classification accuracy on the validation dataset. Mirzabeigi & Razkenari [52] conducted a study using dynamic IRT to automate the identification of thermal anomalies. Computer vision algorithms were then implemented to analyze the data and identify thermal anomalies in the building envelope. The algorithm successfully identified thermal bridges, material degradation, and air leakage on their test data. Alongside classification-based approaches, computer vision methods have been applied to directly automate anomaly identification in thermal imagery.

Accurate integration of thermal findings into spatial building models has become an important goal, motivating research into acquisition parameter optimization. Dabetwar et al. [53] a systematic sensitivity analysis was performed to establish flight-planning benchmarks for reconstructing centimetre-accurate 3-D point clouds from infrared-only UAV imagery and, in turn, to enable quantitative heat-loss mapping of buildings. The authors varied three key acquisition variables, surface-to-air temperature differential (∆T), image side-overlap, and camera oblique angle, through a two-part campaign that combined controlled laboratory heating of a defect-laden mock-up with field flights around a three-storey campus building. Laboratory trials showed that the workflow could still resolve insulation voids and moisture traps with ≥95% dimensional fidelity at a ∆T of just 1.3 °C. The second tier of the study moved outdoors where drone was flown while three side-overlap levels (90%, 80%, 70%) and four single oblique angles (15°, 20°, 25°, 30°) were tested. A 90% pattern kept the mean cloud-to-cloud error near 0.02 m, whereas reducing overlap to 80% and 70% inflated the error to 1.62 m and 2.34 m, respectively. Among the single-angle oblique flights, tilting the thermal camera 25° from nadir delivered the tightest cloud-to-cloud match. Taken together, the experiments establish practical lower bounds for infrared surveys: a ∆T of roughly 1.5 °C, a minimum 90% side overlap, and a nadir-plus-25° camera configuration are sufficient to generate BIM-ready thermal meshes in a single short flight.

With reliable acquisition parameters established, several studies developed frameworks for registering thermal data directly into BIM environments. Zhang et al. [54] aimed to develop a framework for generating thermal-textured BIM using UAV-based thermal and RGB images. The research addressed significant challenges in the automated registration of UAV thermal images with BIM, such as the low texture in thermal images and thermal inconsistencies. Key objectives included creating a cost-effective calibration target for thermal cameras, developing a method for fusing thermal and RGB images to improve image registration, and enhancing thermal images to correct inconsistencies and improve the quality of the thermal-textured BIM. They developed a 3D-printed calibration target designed to align and calibrate the thermal and RGB cameras. Homography transformation was employed to register and fuse the captured images, aligning them with the BIM. A histogram-based approach was used to correct thermal inconsistencies between adjacent thermal images to enhance the contrast and detail in the thermal orthophotos. In a case study, the framework demonstrated its ability to accurately register and map thermal images onto BIM, resulting in a mean error of approximately 5.713 pixels. This improved accuracy and detail in the thermal-textured BIM allowed for the clear identification of thermal damages. Despite these advancements, some challenges remain. The histogram-based correction method assumes consistent temperatures in overlapping areas of adjacent images, which may not be true over extended periods. Moreover, the method is currently suitable only for flat building façades, necessitating further development for curved surfaces and more generalized applications. Chen et al. [55] proposed a computational workflow for automatically registering and fusing thermal anomalies detected in aerial infrared images to a 3D building model. The study aimed to improve the localization and assessment of thermal anomalies. The primary objectives included developing a method to align IR images with corresponding visual RGB images and map them into façade reference images, and subsequently register these images to the 3D building coordinate system through a predefined transformation process. The methodology involved using UAVs equipped with dual RGB and IR cameras to capture multi-spectral images for close-up inspections of building façades. The captured RGB images were used to detect visible defects, while IR images identified thermal anomalies. The workflow included preprocessing steps such as undistorting the RGB images, aligning them with IR images using key point matches, and registering the IR images to the 3D building model using coordinate transformation techniques. Findings from a pilot case study demonstrated that the proposed method effectively registered and fused UAV-captured images and thermal anomaly information into a 3D building model. The study highlighted the improved precision in locating and measuring thermal anomalies, which is crucial for diagnosing and addressing building energy performance issues. The challenges identified in the study included the complexity of accurately aligning and registering multi-spectral images, the need for precise camera calibration parameters, and the potential distortions in close-up images captured by UAVs. Additionally, the study noted that environmental factors, such as direct radiation during IR image capture, could affect the accuracy of thermal anomaly detection.

Other researchers have combined UAV thermography with complementary spatial sensing technologies to further enrich 3D diagnostics. Huang et al. [56] developed a methodology combining 3D modeling from point clouds and IRT to identify facade defects in buildings. Their objective was to extend the life cycle of buildings by accurately identifying defects using non-destructive testing methods. They employed UAVs and Terrestrial Laser Scanning (TLS) to collect spatial data and generate a 3D model of the building. The methodology involved using image segmentation to cluster pixels with similar temperatures, making it easier to detect and locate defects based on thermal variations. The spatial data from UAV and TLS were integrated using ground control points to align their coordinate systems, and the segmented thermal images were mapped onto a 2D image of the 3D model. The study demonstrated that this approach could effectively identify and locate facade defects, providing both thermal and spatial information in a comprehensive 3D model. Daffara et al. [57] utilized the FLIR Duo R camera to capture dual visible-thermal image datasets. The system was calibrated using a passive target to align the visible and thermal sensors accurately. Reconstruction software was then employed to process the captured images and generate 3D thermal models of the buildings. The findings revealed that the system could accurately reconstruct 3D thermal models, allowing for the identification of thermal anomalies and potential areas of energy loss in buildings. The study highlighted the cost-effectiveness of the system, which relies on affordable, commercially available devices and user-friendly software, making it accessible for widespread use in building inspections.

Beyond envelope diagnostics, UAV-IRT has also been applied to structural damage assessment in post-disaster contexts. Zhang et al. [58] developed an automated system for detecting structural damage in buildings post-earthquake by integrating UAV-based oblique photography with infrared thermal imaging. The objective was to provide accurate and timely information crucial for emergency rescue and loss assessment. The methodology consisted of four main components: 3D live-action modeling and structural analysis using ultramicro oblique UAV images to create detailed 3D models of buildings, damage information extraction from these models, crack detection in walls using infrared thermal imaging to identify cracks based on temperature distribution, and the integration of these detection systems into a cohesive framework. Findings indicated that the integrated system effectively identified structural damage and wall cracks with an accuracy of 78%. The authors state that combination of 3D modeling and thermal imaging offered a more comprehensive assessment of building conditions compared to traditional methods, which often missed critical damage details.

This section provided an overview of qualitative UAV-IRT studies focusing on detecting thermal anomalies in various building types. While the reviewed studies collectively demonstrate the technical feasibility and operational efficiency of UAV-based thermography for outdoor inspections, particularly for identifying insulation defects, moisture intrusion, and HVAC-related problems, a critical appraisal of the underlying evidence base reveals important concerns about its overall robustness, internal consistency, and readiness for broader adoption beyond controlled research settings.

A fundamental limitation running through the majority of reviewed studies is their reliance on single-building or single-campaign case studies, with little to no replication across different climatic zones, construction typologies, building ages, or envelope materials. Several studies were validated on a limited number of facilities without systematic variation in building type, age, or material composition. This pattern of narrow validation is pervasive across the reviewed literature. Reported performance metrics should therefore be treated as indicative rather than definitive, as some studies achieved high accuracy figures under controlled laboratory conditions, and it remains unclear whether such levels of accuracy are reproducible in uncontrolled field environments where surface temperatures, emissivity, and atmospheric conditions fluctuate unpredictably. Similarly, damage detection accuracy figures reported in post-disaster inspection studies were not consistently benchmarked against alternative non-destructive testing methods or validated across multiple structures, making it difficult to assess their true operational reliability. Registration and alignment errors reported in BIM integration studies are likewise difficult to contextualize without standardized reference data or independent replication across different building geometries and environmental conditions.

The application of machine learning to automate thermal anomaly detection represents one of the most promising directions in the field, yet some studies in this area reveal several important constraints that temper optimism about near-term deployment. The dataset sizes employed in some studies are strikingly small relative to the complexity of the classification task, with certain models trained on fewer than 100 samples per defect class and covering only a narrow range of heat loss categories. While high classification accuracies were reported in optimal configurations, these results must be interpreted with considerable caution given the limited sample sizes and the narrow range of building types, materials, and environmental conditions represented. Perfect or near-perfect accuracy on small, homogeneous datasets is a well-recognized indicator of overfitting, wherein a model learns the specific characteristics of its training data rather than the underlying patterns that would generalize to unseen buildings or different inspection scenarios. Furthermore, some studies found that model performance was highly sensitive to dataset size and the number of training epochs, which underscores how fragile these configurations may be when training data does not adequately capture real-world variability. Computer vision approaches for thermal anomaly identification similarly demonstrated effectiveness within narrow experimental contexts. Until training datasets are substantially expanded to encompass diverse building stock, varied climatic conditions, multiple defect severities, and data collected across different seasons and times of day, the generalizability of these models to real-world inspection practice remains an open and important question.

The variability of real-world boundary conditions presents a persistent and partially intractable challenge. Wind speed, solar irradiance, ambient temperature, relative humidity, and surface emissivity all influence the apparent temperature recorded by thermal cameras, and these variables interact in complex, nonlinear ways that are difficult to model or correct systematically. Some studies have shown that certain envelope defects, including roof moisture anomalies and slab-edge thermal bridges, are only detectable during specific windows of the day, when solar loading and surface temperature gradients create sufficient contrast for identification. This finding has significant implications for the generalizability of UAV-IRT campaigns conducted under narrow temporal windows, as surveys performed at suboptimal times may systematically underreport the true extent of envelope deficiencies. The autogain control problem encountered in orthomosaic generation represents another unresolved methodological obstacle, where identical apparent temperatures are scaled differently across images. While partial solutions have been proposed, these depend on the availability of reliable reference images and may not perform robustly under conditions of rapid temperature change or heterogeneous scene content. Preprocessing workflows developed to improve thermal image alignment in 3D reconstruction, including greyscale conversion, histogram adjustment, and Gaussian filtering, similarly represent workarounds for fundamental limitations in thermal camera resolution and image quality, rather than solutions that eliminate the underlying constraints. These limitations become particularly significant when UAV-IRT data is intended for integration into BIM environments. Such frameworks rely on precise camera calibration, stable environmental conditions during data capture, and extensive manual post-processing including key point matching, homography transformation, and coordinate system registration. The multi-step nature of these workflows introduces multiple potential sources of error that compound across processing stages, and thermal consistency correction methods that assume stable temperatures in overlapping image regions may not hold over extended flight campaigns or under variable solar loading.

Overall assessment. Taken together, the qualitative UAV-IRT literature reviewed in this section represents a technically diverse and rapidly evolving body of work that has convincingly established the feasibility of drone-based thermography for building envelope diagnostics across a range of applications and building types. However, the evidence base as a whole remains in an early stage of maturity. The predominance of small-scale, single-site case studies, the absence of standardized datasets and evaluation benchmarks, the sensitivity of machine learning models to dataset size and composition, the unresolved challenges of radiometric consistency and multi-sensor data registration, and the influence of regulatory and institutional barriers collectively indicate that current knowledge, while promising, is not yet sufficiently robust or validated to support confident generalization to diverse real-world inspection practice. Addressing these gaps will require not only methodological innovation within individual research groups, but coordinated efforts to develop shared datasets, common evaluation frameworks, and cross-jurisdictional dialogue between researchers, regulators, and professional bodies.

Table 1. Summary of qualitative UAV-IRT studies, detailing objectives, IR camera and UAV models, building types, and country of study.

Study	Environmental Conditions During Data Acquisition	IR Camera	UAV	Building Type	Country
Ortiz-Sanz et al. [46]	Morning and noon surveys, wind speed < 1 km/h, summer conditions, air temperature ≈ 35 °C	Vue Pro R, FLIR B335	Quasar	Wine Cellar	Spain
Gil-Docampo et al. [47]	Ambient air temperature ≈ 14 °C	Vue Pro-R 640, FLIR B335	Quasar	winery	Spain
Dabetwar et al. [51]	After sunset in winter, external air temperature < −5 °C	Zenmuse XT2	DJI Matrice 200 V2	Educational building	USA
Mirzabeigi & Razkenari [52]	Not reported	-	DJI Mavic 2	Educational building	USA
Moore et al. [48]	Summer conditions, ambient temperature ≈ 31 °C	-	DJI Phantom 4	Educational building	USA
Zhang et al. [49]	Not reported	FLIR’s Quark	Aeryon Labs SkyRanger	-	Canada
Rakha et al. [50]	Winter surveys conducted between 9:00 AM and 5:00 PM	Zenmuse XT2	DJI Matrice 200	Courtyard building	USA
Dabetwar et al. [53]	Not reported	Zenmuse XT2	DJI Matrice 200 V2	Educational building	USA
Zhang et al. [54]	Sunny weather conditions	-	Mavic 2 Enterprise Dual	Pre-1920 two-story masonry building	New Zealand
Chen et al. [55]	Not reported	FLIR Tau 2 640, Zenmuse XT2, M2ED Dual Camera	Intel Falcon 8+, DJI Matrice 200, DJI Mavic Pro	Educational building	USA
Huang et al. [56]	Not reported	NEC Thermo Gear-G120	-	-	Taiwan
Daffara et al. [57]	Cloudy conditions, ambient temperature between 9–13 °C	FLIR C2, FLIR Duo R	Custom-made quadrotor	Building under restoration	Italy
Zhang et al. [58]	Winter conditions, ambient temperature ≈ −12 °C	Zenmuse XT2	QingTing-5S	building structures at the earthquake emergency rescue base	China

synthesizes the reviewed studies by summarizing their objectives, data acquisition approaches, processing techniques, and key findings, thereby highlighting both current capabilities and the remaining research gaps that future work must address.

4.2. Quantitative Studies

In recent years, dynamic IRT has gained prominence as a quantitative tool for building envelope assessment due to its ability to provide rapid, non-contact, and large-scale thermal measurements. The growing demand for accurate in situ U-value estimation, urban-scale energy screening, and energy model calibration has further driven research in this field. Numerous studies have demonstrated the potential of dynamic IRT for quantitative thermal analysis under diverse operational and environmental conditions.

A foundational concern in quantitative UAV-IRT is how flight parameters affect measurement accuracy. Zhang et al. [59] focused on using dynamic IRT to determine wall thermal transmittance. The study examined how different UAV flight distances, 15 m, 30 m, and 45 m, and speeds, 0.5 m/s and 1.5 m/s, affected measurement accuracy. Findings showed that as the distance between the camera and the wall increased, the recorded surface temperatures decreased when compared to readings from a heat flux meter. Temperature errors grew from 3.3% to 11.7%, and heat flux errors rose from 23.7% to 40%, primarily due to infrared radiation being partially absorbed by atmospheric elements like water vapor and CO₂, reducing atmospheric transmission. Additionally, faster flight speeds caused the drone to tilt more, slightly compromising image quality. However, the impact on measurement accuracy was limited, with total error rising only about 2%. The authors concluded that dynamic IRT remains a practical approach for evaluating thermal transmittance in buildings. Environmental timing plays an equally critical role in measurement reliability. Rodriguez et al. [60] investigated the use of dynamic IRT to estimate the thermal transmittance of building envelopes, comparing the results with those obtained through the thermometric method. Acknowledging the influence of environmental conditions on measurement accuracy, they proposed conducting tests across different seasons and times of day. To evaluate this, they performed every hour measurements for 24-h during both summer and winter. Their findings showed that greater indoor-outdoor temperature differences significantly improved measurement reliability. Winter nights, characterized by higher temperature differentials, produced the most stable and accurate U-values, while summer nights, with minimal temperature differences, led to inconsistent and distorted results. Dynamic IRT demonstrated a strong correlation with thermometric measurements, with correlation coefficients of 0.978 in winter and 0.902 in summer. Despite this, the RMSE indicated a 25–29.1% discrepancy between the two methods, with most errors occurring during the day. These inaccuracies were primarily attributed to insufficient thermal contrast. Based on their findings, the authors recommended avoiding measurements when the temperature difference between indoor and outdoor conditions is low.

Beyond individual building assessments, dynamic IRT has been scaled to urban-level analysis. Zheng et al. [61] introduced a dynamic IRT approach to create 3D thermal models of multiple buildings for the purpose of identifying those with high energy consumption. Thermal images captured during UAV flights were processed using Structure-from-Motion techniques and converted into a color-coded 3D model. To assess the accuracy of their method, the thermal data were compared against actual heating energy consumption records. The results demonstrated a strong alignment, confirming that this technique can effectively pinpoint buildings with poor energy performance. The study highlights the potential of dynamic IRT for large-scale urban energy efficiency evaluations.

Dynamic IRT has also proven effective as a tool for energy model calibration potential of dynamic IRT for large-scale urban energy efficiency evaluations. Bayomi et al. [62] employed dynamic IRT as a method to calibrate energy models by assessing thermal transmittance. They compared three simulations based on design specifications, U-value calculated from dynamic IRT, and actual measured energy use of the building. This approach reduces the normalized mean bias error from 21% to less than 1% compared to the design specification simulation. Similarly, the coefficient of variation decreased from 25% to 9%. Additionally, they conducted a sensitivity analysis on four key input parameters for dynamic IRT: emissivity, reflected temperature, indoor temperature, and convective heat coefficient. By varying each parameter by 15%, they assessed the impact on thermal transmittance values. Their results showed that emissivity and reflected temperature had minimal influence on the final U-value calculation, whereas indoor temperature and convective heat coefficient played a more significant role. Overall, their study demonstrated that dynamic IRT effectively reduces simulation errors and enhances the accuracy of energy model calibration. Similarly, Ficapal and Mutis [63] employed dynamic IRT to create a framework to identify and diagnose thermal bridges in building envelopes. In their case study, they utilized a FLIR Vue Pro-R camera to inspect a curtain wall and compared the results obtained through dynamic IRT with measurements from a thermometer and a simulation tool (THERM). Their findings demonstrated that the temperature values recorded using IRT exhibited a minimal discrepancy of less than 3% when compared to thermometer probe data and simulation outputs for exterior temperatures. Additionally, state that this variation fell within the 5% uncertainty range specified by the IR-sensing device manufacturer [63].

Camera stability and measurement drift represent persistent technical challenges. Zheng et al. [64] employed dynamic IRT to develop a 3D point cloud model with temperature data for evaluating the thermal behavior of building envelopes. Using Pix4Dmapper software, they constructed the 3D thermal model and compared the IR temperature readings to those from thermocouples. The study identified a maximum temperature deviation of 5 °C between the two methods, with larger discrepancies occurring when the camera was used immediately after startup. As the camera stabilized with environment, typically after 10 to 20 min, the measurement error decreased. They reported that 81.25% of the IR camera readings fell within a 3 °C range of the thermocouple data. Based on these findings, the authors recommended a 20-min stabilization period for the camera prior to initiating thermal surveys to improve accuracy. Mahmoodzadeh et al. [10] examined how effectively infrared cameras measure wall surface temperatures during aerial building inspections. In their initial test using the Zenmuse XT2, they observed temperature errors reaching up to 32 °C within the first 10 min of flight. A second test using a different device, the FLIR A65, showed improved stability, with a 7 °C error after 3 min in the air. These inaccuracies were attributed to airflow caused by the UAV’s propellers, which disrupted the thermal readings. To address this, they developed a shield to protect the camera. With the shield in place, temperature errors were significantly reduced to 4.5 °C after just 3 min and dropped to below 1 °C after 10 min. The researchers recommended a 30-min warm-up period for the camera before flight and advised more frequent application of NUC when conducting dynamic IRT to enhance measurement reliability. Yang et al. [65] applied a similar approach and used shield to reduce wind-induced errors during dynamic IRT. Their comparison of laboratory and dynamic test results revealed average temperature deviations of approximately +8 °C and −5 °C under dynamic conditions. After implementing the shield, the fluctuation between maximum and minimum recorded temperatures dropped to below 2.5 °C. While the shield effectively enhanced camera performance, the researchers pointed out a key limitation: it restricted the camera’s field of view to 90 degrees, making it less compatible with automated UAV operations due to possible interference. Yang et al. [65] applied a similar approach and used shield to reduce wind-induced errors during dynamic IRT. Their comparison of laboratory and dynamic test results revealed average temperature deviations of approximately +8 °C and −5 °C under dynamic conditions. After implementing the shield, the fluctuation between maximum and minimum recorded temperatures dropped to below 2.5 °C. While the shield effectively enhanced camera performance, the researchers pointed out a key limitation: it restricted the camera’s field of view to 90 degrees, making it less compatible with automated UAV operations due to possible interference.

The integration of machine learning has opened new directions for automating quantitative assessments. Sadhukhan et al. [66] combined dynamic IRT with machine learning to automate the assessment of heat loss in building envelopes. They employed the Mask R-CNN model to detect and classify building components, such as walls and windows, using bounding boxes and semantic segmentation. For U-value estimation, the model produced values ranging from 2.64 to 0.31 BTU/hr·ft² for walls across different days, compared to the ASHRAE reference value of 0.085 BTU/hr·ft². For window analysis, the estimated U-values ranged from 0.14 to 0.5 BTU/hr·ft² for single-glazed windows, relative to the ASHRAE benchmark of 0.35 BTU/hr·ft², while double-glazed windows showed values between 0.23 and 1.02 BTU/hr·ft², compared to the standard of 0.95 BTU/hr·ft². During thermal data collection, a significant discrepancy was noted between infrared readings and actual surface temperatures measured by thermocouples; for instance, the camera recorded 25 °C while the true temperature was only 7.9 °C. However, the study did not explore the reasons behind this deviation. Overall, they sate that the measurements followed the ASHRAE standard, and this method is reliable.

The studies summarized in

Table 2. Summary of quantitative UAV-IRT studies, detailing objectives, IR camera and UAV models, building types, and country of study.

Study	Environmental Conditions During Data Acquisition	IR Camera	UAV	Building Type	Country
Zhang et al. [59]	Winter experiments conducted over four days; outdoor temperature between −18 °C and −12 °C; wind speed ≈ 0.2 m/s; indoor temperature ≈ 22 °C	FLIR Zenmuse XT	DJI Inspire 1	Residential building	China
Rodríguez et al. [60]	Experiments conducted in both winter and summer; wind speed ≈ 3.47 m/s in winter and 2.27 m/s in summer; tests performed during both daytime and nighttime	-	DJI Mavic 2 Enterprise	Educational building	Spain
Zheng et al. [61]	Winter nighttime measurements; wind speed ≈ 2.1 m/s; air temperature between −1 °C and 1 °C	FLIR Zenmuse XT	DJI Matrice 210 RTK V2	Educational building	China
Bayomi et al. [62]	Early morning measurements (6:00–8:00 AM) in winter; air temperature ≈ 8 °C; indoor–outdoor temperature difference ≈ 10 °C	FLIR Zenmuse XT	DJI Inspire 1	Academic building	USA
Ficapal & Mutis [63]	Indoor temperature 27.2 °C with 65% RH; outdoor temperature 26.0 °C with 72% RH; calm wind conditions (<1 m/s)	FLIR Vue Pro-R 640	DJI Matrice 100	Educational building	USA
Zheng et al. [64]	Not reported	FLIR Zenmuse XT	DJI Matrice 210	2 Residential building, 3 Commercial building	China
Sadhukhan et al. [66]	Morning and evening measurements conducted during both summer and winter; air temperature between −11 °C and 15 °C	Mirage 650	-	Educational building	USA
Stokowiec & Sobura [67]	Winter morning measurements; air temperature ≈ −5 °C	Testo 890, FLIR Zenmuse XT2	DJI Matrice 210 RTK v2	Residential building	Poland
Gil-Docampo et al. [47]	Ambient air temperature ≈ 14 °C	Vue Pro-R 640, FLIR B335	Quasar	winery	Spain
Yang et al. [65]	Nighttime measurements; wind speed ≈ 0.9 m/s; air temperature between 25.1 °C and 26.1 °C	MAG62, DOC	DJI M300RTK	Educational building	China
Mahmoodzadeh et al. [10]	Indoor temperature ≈ 22 °C; wind speed ≈ 0.60 m/s; nighttime measurements; indoor and outdoor temperatures ≈ 22.12 °C and 7.22 °C	FLIR Zenmuse XT2, FLIR A65	DJI Matrice 200	Wooden structure	Canada

demonstrate that dynamic IRT is a promising quantitative tool for building envelope assessment, energy model calibration, and urban-scale energy analysis. The collective findings confirm its strong potential for non-invasive thermal characterization; however, they also reveal persistent challenges related to atmospheric effects, camera thermal drift, airflow disturbance, emissivity uncertainty, and environmental dependency. Future research should therefore focus on improving radiometric calibration under dynamic conditions, developing robust correction models for airflow and atmospheric attenuation, standardizing measurement protocols, and further integrating data-driven techniques to enhance reliability and automation. Addressing these challenges is essential for the large-scale and routine deployment of dynamic IRT in building energy diagnostics and retrofit decision-making.

Also,

Table 3. Comparative overview of UAV-based IRT applications in the building sector.

Category	Experimental Setup	Validation/Accuracy	Key Limitations
Thermal anomaly detection	UAV-based thermal imagery	Validation through visual inspection, blower-door tests, or ground thermography	Sensitive to ΔT, wind, and solar radiation; emissivity variability
3D thermal modeling	UAV thermal + RGB imaging; Structure-from-Motion photogrammetry	Cloud-to-cloud distance, reference geometric model, or point cloud alignment accuracy	Low thermal resolution; difficulty aligning thermal and RGB images
Automated defect detection	Thermal image datasets analyzed using CNN or computer vision algorithms	Classification accuracy, precision/recall, confusion matrix evaluation	Limited labeled datasets; model generalization issues
Energy performance assessment	Close-range UAV surveys combined with energy modeling or thermal analysis	Comparison with heat flux meters, surface temperature sensors, or reference models	Emissivity uncertainty; atmospheric attenuation; wind effects
Thermal–BIM integration	UAV thermal and RGB imagery integrated with BIM models	Image-to-model registration error and spatial localization accuracy	Complex registration workflows; camera calibration requirements
Structural damage detection	UAV thermography combined with oblique imaging or multi-sensor data	Detection success rate validated against visual inspection or structural assessment	Thermal noise and environmental effects may obscure defects

provides a generalized comparison of the main UAV-based infrared thermography applications in the building sector, highlighting their typical experimental setups, validation approaches, and key limitations. As shown in the table, qualitative applications such as thermal anomaly detection and structural damage assessment commonly rely on visual inspection, ground thermography, or other complementary diagnostic methods for validation. In contrast, quantitative-oriented applications require more rigorous validation through comparison with reference measurements such as heat flux meters or temperature sensors.

The table also illustrates that despite the diversity of applications; several limitations consistently appear across the literature. Measurement accuracy is often influenced by environmental factors such as temperature differential, wind speed, and solar radiation, as well as uncertainties related to emissivity and atmospheric attenuation. In addition, technical challenges associated with thermal image resolution, data fusion between thermal and RGB imagery, and the need for precise camera calibration remain significant barriers for advanced applications such as 3D modeling and BIM integration. Overall, the comparison highlights that while UAV-based IRT provides a powerful tool for building diagnostics, improvements in data processing, calibration procedures, and standardized validation methods are still required to enhance the reliability and broader adoption of these techniques.

5. Operational Planning for Dynamic IRT

Due to the high sensitivity of thermal imaging to both environmental and operational factors, this chapter introduces a structured framework for planning UAV-based IRT in three critical phases: pre-inspection planning, equipment preparation, and flight parameter selection (Figure 3). The first phase, Pre-Inspection Planning, focuses on the need to manage external environmental conditions that significantly affect thermal imaging results. Key factors, including temperature differential, wind speed, solar radiation, and surface moisture, must be evaluated, controlled, and monitored prior to the survey. The second phase, Equipment Preparation, focuses on configuring the thermal camera with precise input parameters. This includes calibrating emissivity, estimating reflected apparent temperature, correcting for atmospheric transmission losses, and allowing sufficient camera warm-up time. The final phase, Flight Parameter Selection, examines UAV-specific operational settings, such as flight altitude, camera viewing angle, UAV speed, and image overlap, that influence image resolution, thermal anomaly detection, and modeling accuracy. By integrating environmental controls, precise camera configuration, and customized flight strategies, this chapter provides a comprehensive guide to improving the accuracy, consistency, and repeatability of UAV-based dynamic IRT surveys.

5.1. Pre Inspection Planning

Conducting a successful IRT survey requires thoughtful pre-inspection planning, as thermal imaging is highly sensitive to several key variables. Environmental conditions, surface emissivity, and reflected temperature must all be accounted for to ensure the accuracy of diagnostic results. Neglecting these parameters can result in inaccurate estimations of heat loss, missed or misinterpreted defects, and unreliable data interpretation. This section outlines the major environmental factors influencing the quality of IRT assessments and summarizes best practices based on established standards.

One of the most critical parameters is the temperature differential (ΔT) between the interior and exterior of the building. A sufficient ΔT is essential for creating the thermal contrast necessary to detect anomalies in insulation or air leakage pathways. According to ASTM C1060 [27], a minimum temperature difference of 10 °C (either surface-to-surface or ambient air) sustained for at least four hours prior to inspection is recommended to reveal insulation voids and thermal bridging. ISO 6781-1 [68] further specifies that a ΔT of at least 15 °C is required under steady-state conditions for detecting moisture accumulation, while a 10 °C difference is acceptable for air leakage surveys. ASTM E1186 [69] also supports a minimum 5 °C differential during depressurizing test for qualitative air leakage assessments. CNMCS 02 27 13 [26] specifies that the minimum ΔT should be at least ΔT ≥ 10 °C for air leakage detection, ΔT ≥ 15 °C for identifying thermal bridging and insulation deficiencies, and ΔT ≥ 20 °C for effective moisture detection. For roof moisture detection, ASTM C1153 [24] prescribes higher thresholds: a ΔT of at least 10 °C for general surveys, 18 °C for ballasted roofs without insulation, and 23 °C for insulated roofs. Roof surveys should typically be conducted one hour after sunset and before sunrise to exploit the delayed cooling of moisture-retaining materials. A sunny day preceding the survey is ideal, as wet insulation absorbs more solar energy during the day and cools more slowly at night, enhancing thermal contrast [24]. In contrast to qualitative applications, quantitative infrared thermography lacks formally recognized standards or guidelines; however, researchers often adopt similar protocols to ensure consistency and accuracy. For example, Tejedor et al. [44], recommend maintaining a ΔT of at least 10–15 °C to enable measurable heat transfer through building components. Fokaides and Kalogirou [70] suggest the same ΔT threshold, along with ensuring quasi-steady-state thermal conditions for a period of 3–4 h before measurement. Mahmoodzadeh et al. [71] also advocate for a minimum 15 °C temperature difference, particularly for precise calculation of U-value.

Wind is another factor that must be carefully considered. ASTM C1060 [27] advises that wind speeds should not exceed 6.7 m/s, as air movement across surfaces can disrupt heat retention and obscure surface temperature patterns, particularly those associated with air leakage, both on the interior and exterior of the building envelope. CNMCS 02 27 13 [26] specifies a maximum allowable wind speed of 2.8 m/s during thermographic inspections. For roof moisture surveys, ASTM C1153 [24] permits a threshold of 25 km/h, acknowledging that convective heat transfer reduces the temperature differential between wet and dry areas and diminish the thermal contrast essential for accurate moisture detection. Although standard guidelines provide general recommendations, many researchers emphasize the need for stricter wind conditions to improve the accuracy and reliability of IRT surveys. Tejedor et al. [44] recommend conducting surveys under very low wind speeds, ideally below 1 m/s, to minimize convective heat dispersion. Zhang et al. [59] investigated the impact of wind speed on the accuracy of U-value calculations in a dynamic survey and found that as external wind speed increased from 0.2 m/s to 1.5 m/s, the measurement error, compared to a heat flux meter, rose significantly from 2.2% to 25.2%. They concluded that wind speeds during surveys should remain below 0.4 m/s. Tardy [72] also supports maintaining wind speeds below 1 m/s to reduce thermal errors. Kylili et al. [12] warn that wind speeds exceeding 5 m/s can introduce substantial errors and should be avoided. Similarly, Bienvenido-Huertas et al. [73] recommend keeping wind speeds within the 0.1–1 m/s range during IRT inspections. Lucchi [33] further reported that wind speeds above 1 m/s caused deviations of up to 80% in IRT results when compared to heat flux meter data, attributing this discrepancy to increased external convective heat transfer that amplifies thermal dispersion.

Solar radiation is another critical interference factor. Direct sunlight on exterior surfaces can distort thermal signatures and cause false anomalies. According to CNMCS 02 27 13 [26] inspections should be scheduled after sunset to minimize the influence of residual solar heat gain, which can mask or distort thermal anomalies. The minimum waiting period after sunset varies depending on the thermal mass and construction type of the building envelope. For high-mass solid masonry exterior walls, a delay of at least eight hours is recommended. Masonry-clad stud walls require a minimum of six hours, while low-mass cladding on stud walls should be inspected no sooner than four hours after sunset. For glass and metal curtain walls, which respond more quickly to ambient temperature changes, a minimum waiting period of two hours is considered sufficient. ASTM C1060 [27] strongly recommends that surveys be conducted after sunset or before sunrise to avoid solar loading effects. Additionally, post-solar exposure effects, such as temperature reversal where thermal bridging components appear warmer, can persist for hours. For light-frame buildings, a minimum of three hours without direct sun is recommended, and for masonry veneer, at least eight hours. Moreover, surfaces must be free of snow, ice, water, or debris to ensure consistent emissivity and thermal response. This also had been recommended by ASTM C1153-10 [24], stating that no significant rainfall should occur within 24 h prior to the survey. Recent precipitation can mask thermal anomalies by equalizing surface temperatures between wet and dry areas, thus reducing the accuracy of moisture detection. These recommendations have also been supported by several other researchers [44,70,71].

Overall, the reviewed studies indicate that environmental thresholds reported in standards should be interpreted as baseline conditions rather than strict universal requirements. In practice, the optimal parameters for thermographic inspections depend on the specific objective of the survey and the characteristics of the building envelope. For example, qualitative inspections that aim to detect anomalies such as air leakage, insulation defects, or thermal bridges primarily rely on sufficient thermal contrast and can therefore tolerate moderate variations in environmental conditions. In contrast, quantitative applications, including U-value estimation and heat-loss assessment, require more controlled environments because even small disturbances caused by wind or unstable temperature conditions can introduce significant measurement errors. The literature also suggests that building construction plays an important role in determining suitable inspection timing. Buildings with high thermal-mass envelopes, such as masonry or concrete structures, store solar heat for extended periods and therefore require longer cooling times before reliable measurements can be obtained. Lightweight façades, however, reach thermal equilibrium more quickly and can be inspected sooner after solar exposure. These observations highlight that environmental conditions should be evaluated together with inspection objectives and building characteristics when planning IRT surveys.

Table 4. Summary of recommended environmental and operational conditions.

IRT Application	Objective	Recommended ΔT	Wind Speed	Solar Exposure/Timing
Qualitative inspection	Detection of thermal anomalies (air leakage, insulation defects, thermal bridges)	≥10 °C	<2–3 m/s preferred	After sunset or early morning to avoid solar loading
Quantitative analysis	Estimation of U-value, heat loss, or energy modeling inputs	≥15 °C (preferably stable for several hours)	<1 m/s recommended	Night-time inspections under quasi-steady-state conditions
Moisture detection (roof/wall)	Identification of moisture accumulation within envelope	≥15–20 °C	<2 m/s	Typically evening or night following a sunny day
High thermal-mass envelopes (masonry, concrete)	General thermographic inspection	≥10–15 °C	<1–2 m/s	≥6–8 h after sunset
Lightweight envelopes (cladding systems, curtain walls)	General thermographic inspection	≥10 °C	<2 m/s	≥2–4 h after sunset

summarizes these recommended environmental parameters and Figure 4 also summarize a step-by-step flowchart for pre inspection planning. Adhering to these best practices not only improves measurement accuracy but also ensures repeatability of IRT surveys in diverse building diagnostics contexts.

5.2. Equipment Preparation

In addition to the environmental factors discussed in the previous chapter, several other parameters must be considered during infrared thermography surveys, as neglecting them can significantly compromise measurement quality. These include emissivity, camera warm-up time, reflected temperature, distance to the target, and atmospheric transmission rate. This chapter focuses on these factors and explains how to properly configure them within the camera’s software to improve the accuracy and reliability of thermographic results.

As previously discussed, most commercial infrared cameras are equipped with thermal detectors, particularly microbolometers [74]. Unlike photonic detectors, thermal cameras lack an internal cooling system to regulate their internal temperature [18]. As a result, the thermal radiation that camera detector receives is not only from the target object but also from the camera’s internal components, such as the housing and lens [75]. When the camera is powered on, its internal electronics begin to generate heat. It takes time for the system to reach thermal equilibrium, where the heat generated internally is balanced by heat loss to the surroundings [76]. If the camera is used before reaching this equilibrium, it may produce measurement errors due to the influence of internally generated heat. Therefore, the camera needs to go on a warm-up period before conducting a survey. The warm-up period required for infrared cameras can vary significantly depending on the camera model and environmental conditions. Zhao et al. [76] recommend a minimum of 30 min to allow the camera to reach thermal equilibrium with its surroundings. Similarly, Acorsi et al. [77] used a 15-min warm-up but emphasized that this duration is insufficient when there is a large temperature difference between the camera’s interior and the ambient environment. For aerial surveys, they advised including extra flight lines at the start of the mission to give the camera time to stabilize under actual flight conditions, such as changing air temperatures and wind. Kelly et al. [78] conducted a study to assess camera stabilization time and observed a dramatic change in temperature readings within the first 9 min, with discrepancies as high as 16 °C. They concluded that a minimum of 15 min is necessary, though a full hour is preferable for maximum accuracy. For dynamic UAV inspections, they also recommend adding additional flight lines at the beginning of the mission to allow for thermal stabilization in the air. Smigaj et al. [79] reported a similar trend, observing a temperature fluctuation of 3.5 °C within the first 30 min of operation. They also recommend a 30-min warm-up to avoid misleading data. In a controlled lab study, Wan et al. [80] tested two different infrared cameras by exposing them to a blackbody at constant temperature for over an hour. They found that one camera required 30 min, while the other needed 60 min to stabilize, with fluctuations of up to 1.5 °C before stabilization. Based on these results, they suggest a minimum one-hour warm-up for aerial inspections to ensure accurate readings. Mahmoodzadeh et al. [10] combined laboratory and field tests, finding that under dynamic UAV conditions, the camera stabilized after 20 min, at which point the measurement errors were significantly reduced. Prior to stabilization, errors of up to 14 °C were observed. Yang et al. [65], through lab tests simulating dynamic flight conditions, determined that a warm-up period of at least one hour is required for accurate measurements. They found that with proper warm-up, the measurement accuracy improved to within ±1 °C, whereas skipping the warm-up resulted in readings approximately 2.6 °C lower than the actual surface temperature. A summary of the aforementioned studies is presented in

Table 5. Summary of minimum recommended warm up times for infrared cameras as reported in the literature.

Inspection Type	Recommended Warm-Up Time
Laboratory setups	15 min
Field inspections	~30 min
UAV-based surveys	30–60 min

.

In summary, thermal infrared cameras require a warm-up period to reach thermal equilibrium before reliable measurements can be obtained. Across multiple studies, warm-up durations varied depending on camera model, environmental conditions, and whether the survey was conducted on the ground or from a UAV. Laboratory and field studies consistently show that insufficient warm-up can result in substantial temperature errors, ranging from several degrees up to 16 °C in extreme cases. Most authors recommend a minimum warm-up period of 15–30 min for laboratory or stationary inspections, with longer durations of up to 60 min preferable for dynamic or aerial surveys, where airflow, ambient temperature changes, and motion further affect camera stabilization. Overall, ensuring adequate warm-up is critical for minimizing measurement errors and improving the accuracy and reliability of both stationary and dynamic infrared thermography surveys.

Another critical factor that significantly affects the accuracy of infrared thermography is the emissivity of the object. Emissivity refers to a material’s ability to emit thermal radiation compared to an ideal blackbody, and it is expressed as a value between 0 and 1 [81]. The emissivity of a surface depends on several factors, including its material properties, surface texture, temperature, and the wavelength of observation [82]. Accurate emissivity calibration is essential, as even small errors in assumed emissivity can directly lead to substantial temperature measurement errors. There are two primary methods for determining the emissivity of building surfaces, both of which are recommended by ASTM E1933 [83] and ISO 6781-1 [68]. The first is the contact method, where a calibrated infrared camera is focused on the surface while the reflected apparent temperature is entered into the camera settings. A contact thermometer (or mirrored sensor) is then used to measure the actual surface temperature. Without moving the camera, the emissivity setting is manually adjusted until the infrared reading matches the contact measurement. This process is repeated at least three times, and the average value is the object emissivity. The second is the reference material method, which involves placing a material with a known emissivity, such as black electrical tape or high-emissivity paint, on or next to the surface being measured. After inputting the known emissivity and reflected temperature into the camera, the temperature of the reference material is recorded. The camera is then aimed at the untreated surface, and its emissivity value is adjusted until its displayed temperature matches that of the reference. This process is also repeated multiple times to obtain a reliable average emissivity value. While both methods have proven effective in stationary IRT applications, particularly the reference method due to its simplicity and speed, they present challenges in dynamic IRT, especially when inspecting entire buildings or multiple structures. Dynamic surveys often include many different materials within a single image, making manual emissivity calibration difficult and time-consuming. In practice, some researchers have resorted to using generalized emissivity values during dynamic inspections. For example, Ortiz-Sanz et al. [46] applied generalized emissivity values across surfaces in their UAV-based survey but cautioned against this practice, noting that infrared cameras are highly sensitive to emissivity variations and such approximations can lead to significant measurement errors. Gil-Docampo et al. [47] echoed this concern, emphasizing that generic emissivity assumptions should be avoided for accurate thermographic analysis. This is primarily because an object’s emissivity is influenced not only by its material composition but also by factors such as its age, surface roughness, shape, and temperature [84]. One practical strategy for managing emissivity heterogeneity in UAV-based inspections is the segmentation of façade materials using regions of interest. In this approach, different façade components are delineated as separate regions, allowing each region to be assigned its appropriate emissivity value rather than applying a single generalized emissivity across the entire image. Overall, there is a clear gap in the literature regarding reliable methods for determining emissivity during dynamic IRT surveys. The standard approaches used in stationary conditions are often impractical in aerial or large-scale applications, and needs for further research to develop efficient and accurate emissivity estimation techniques suited to dynamic thermography. Future research could explore automated and data-driven strategies to address emissivity variability. These approaches may include machine learning-based façade material classification that estimate emissivity through statistical or regression-based methods. For example, future workflows may combine RGB imagery and thermal data to automatically classify façade materials and assign appropriate emissivity values to each detected region. Similarly, data-driven calibration models could be developed to estimate emissivity corrections directly from large thermal datasets collected during UAV surveys. Such approaches have the potential to significantly reduce operator dependency and improve the consistency of emissivity estimation across complex building façades. However, these techniques require further investigation and validation before being implemented in large-scale building inspections.

Another crucial parameter that must be adjusted before conducting an infrared thermography survey is the reflected temperature. The radiation detected by an infrared camera is not solely emitted by the object itself, it also includes reflected infrared radiation from surrounding surfaces [12]. Accurately determining this reflected temperature is essential, as errors in its estimation can significantly impact thermal readings. As noted by Fokaides and Kalogirou [70], a 1 °C error in reflected temperature can result in up to 10% error in surface temperature calculations and as much as 100% error in U-value estimations. Nardi et al. [85] highlighted that lower the difference between ambient air temperature and reflected temperatures improve the accuracy of U-value calculations and therefore recommended conducting surveys on cloudy days, when environmental reflections are minimized. To measure reflected temperature, ASTM E1862 [86] suggests using crumpled aluminum foil as a reflective target. The foil is placed in front of the surface being measured, and the average temperature of the foil, recorded by the camera with the emissivity set to 1, is used as the reflected temperature value. This method has become widely used due to its ease of application and the quick setup time it requires. Similar to emissivity calibration, the procedure for measuring reflected temperature was originally developed for stationary setups. In dynamic surveys, especially when inspecting large buildings from varying angles and elevations, the reflected temperature can vary significantly. As a result, relying on a single reflected temperature value or one measured from a single location can lead to reduced measurement accuracy. A potential solution is to place crumpled aluminum foil at multiple locations and heights on the building façade, allowing for more accurate correction during post-processing. However, this approach can be labor-intensive and may still not fully capture the variation across all surfaces. Overall, much like emissivity, the determination of reflected temperature in dynamic IRT surveys remains an unresolved challenge in current literature and warrants further research to develop reliable, practical methods suited for aerial inspections.

The final parameter that must be considered in IRT surveys is the atmospheric transmission rate. As thermal radiation travels from the object to the camera, it can be absorbed or scattered by atmospheric components such as carbon dioxide, water vapor, and other particles [12]. The extent of this absorption depends on the wavelength of the infrared radiation. For instance, radiation in the 5–7 μm range is almost entirely absorbed by the atmosphere [87], whereas mid-wave (2–5 μm) and long-wave (8–14 μm) radiation bands experience much less absorption [39]. Most photonic infrared cameras, the most commonly used type, operate in the long-wave band [88]. In all bands, atmospheric transmission decreases with increased humidity, CO₂ concentration, and distance between the target and the camera [89]. Although the formulas used to calculate atmospheric transmission vary between camera manufacturers, they typically depend on variables such as humidity, ambient and atmospheric temperature, distance to the target, and the emissivity of the surface [90]. Accounting for atmospheric transmission is especially important in dynamic IRT surveys, where flight distances can reach up to 50 m. Minkina and Klecha [90] investigated this effect and found that, under conditions of 50% relative humidity and 15 °C atmospheric temperature, increasing the distance between the infrared camera and the target from 0 to 60 m reduced the atmospheric transmission rate from 1.0 to 0.95. This means that approximately 5% of the radiation emitted by the object is lost, which can significantly affect measurement accuracy if not properly corrected. To estimate the atmospheric transmission rate, one widely used tool is MODTRAN (MODerate resolution atmospheric TRANsmission), a computational model for simulating radiative transfer through the atmosphere [91]. By inputting parameters such as humidity, water vapor content, CO₂ levels, wavelength range, altitude, and distance, MODTRAN estimates the transmission rate across various atmospheric conditions. Another approach to adjusting the atmospheric transmission rate is to contact the camera manufacturer to obtain the specific formula used in their system, as these calculations can vary between different manufacturers. For example, in a study conducted by Mahmoodzadeh et al., [39] the transmission rate formula used by FLIR cameras was provided.

Overall, accurately estimating emissivity and reflected temperature across various surfaces, properly adjusting the atmospheric transmission rate, and allowing the camera to warm up for at least 30 min are essential steps to minimize errors and enhance the reliability of dynamic thermal inspections. Step by step procedure for equipment preparation has been shown in Figure 5. The next chapter will examine the impact of flight parameters and explore how their selection can influence the accuracy of thermal readings.

5.3. Flight Parameter

When conducting dynamic IRT surveys, selecting appropriate flight parameters is crucial to ensure high-quality data acquisition. Despite the growing adoption of UAV-based IRT, there is currently not a universally accepted standard or guideline that prescribes specific flight parameters tailored to various building inspection applications. Nevertheless, understanding and optimizing key flight parameters, such as camera angle, flight altitude and distance, UAV speed, image overlap, and flight mode (manual or automated), is essential, as these factors significantly influence the quality and reliability of thermal imaging results.

Each flight parameter involves trade-offs that must be balanced based on the specific objectives of the survey. For instance, increasing the UAV’s distance from the target area expands the camera’s field of view, allowing for broader coverage. However, this comes at the cost of reduced image resolution and potential degradation of thermal data quality due to atmospheric interference, such as the absorption of infrared radiation by airborne particles and moisture [60]. Similarly, higher flight altitudes can facilitate faster data collection over large areas but may compromise the detection of subtle thermal anomalies due to decreased spatial resolution. The camera’s viewing angle also plays a critical role; oblique angles can reveal features on vertical surfaces but may introduce geometric distortions, whereas nadir (downward facing) angles provide uniform coverage of horizontal surfaces like roofs. In the subsequent sections, current flight parameters reported in the literature will be examined in relation to their intended objectives (Qualitative or quantitative) and recommended best practices

In qualitative applications, Moore et al. [48] utilized dynamic IRT to detect and locate water damage on building rooftops. Thermal images were captured from various distances: 1, 2, 3, 4, 5, 10, 15, 20, 30, 40, and 50 m above the rooftop. The study found that objects with an area as small as 0.12 m² could still be identified from a height of 50 m. However, the authors noted that image resolution significantly decreased beyond 20 m, resulting in the target object appearing substantially smaller. At 50 m, the object’s visible area was reduced by approximately 90% compared to its original size, which could hinder accurate identification [48]. Rakha et al. [50] conducted a study using dynamic IRT for defect characterization and modeling of whole building. Thermal images were captured along an automated flight path at a height and distance of 30 m from the building, with the camera angled at 45 degrees. Images were taken at 0.5-m intervals along the flight path, maintaining a 95% overlap. The flight strategy included strip paths flown parallel to the building envelope to generate high-resolution thermal orthomosaics, as well as orbit or polygonal paths around the structure to create 3D photogrammetric models [50]. That was based on the recommendations of Eschmann and Wundsam [92], who suggested that vertical strip flight paths can lead to unfavorable lens movement and horizontal strip paths are considered more effective for thermal data collection [92]. In another study conducted by Rakha and Gorodetsky [93] recommended that strip patterns with at least 70% overlap are suitable for energy auditing and visualization purposes. For photogrammetry and 3D modeling, they suggested using an elliptical flight path with up to 95% overlap and noted that capturing data from multiple flight altitudes, specifically 18, 22, and 27 m, or approximately twice the height of the building, can significantly enhance the quality and completeness of the resulting 3D models [93]. Dabetwar et al. [53] conducted a study to evaluate the impact of different camera angles and image overlap percentages on the accuracy of 3D point cloud models generated from UAV-based infrared thermography. The study tested side overlaps of 90%, 80%, and 70%, combined with camera angles of 15°, 20°, 25°, and 30°, with all images captured from a consistent flight altitude of 75 m. The authors noted that increasing the camera angle (i.e., more oblique views) results in a greater distance between the UAV and the building façade, which leads to lower-resolution thermal images and reduced accuracy. Among the tested configurations, a 90% image overlap combined with a 25° camera angle yielded the most accurate 3D reconstruction, with an average error of just 0.01 m from the reference model. In contrast, overlaps of 80% and 70% resulted in larger errors of 1.62 m and 2.34 m, respectively. The study concluded that overlap has a greater influence on reconstruction accuracy than camera angle, with 25° at 90% overlap offering the optimal balance for thermal 3D modeling [53]. Zhang et al. [54] conducted a case study survey using a manual flight at a distance of 6 m from the building, following strip paths parallel to the building envelope [54]. Chen et al. [55] employed IRT to generate a 3D thermal model of a building, using an automated strip flight path at a distance of 10 m from the structure and a height of 15 m [55]. Hou et al. [94] conducted a study to examine how different flight parameters influence the quality of thermal images for generating 3D models. The tests were carried out at two flight altitudes, 35 m and 60 m, and two camera angles, 30° and 45°. The results indicated that flying at 35 m produced higher-quality thermal images. The authors attributed this to the fact that at higher altitudes, the thermal sensor is farther from the target surface, making it more difficult to accurately capture thermal values. They further recommended using a 30° camera angle when the focus is on rooftops, and a 45° angle when the objective is to capture building façades. This is because the 45° angle enables the sensor to collect more data points from vertical surfaces compared to rooftops and wiseversa [94]. Mirzabeigi and Razkenari [52] utilized dynamic IRT to detect thermal bridges on the building envelope. They employed a strip flight pattern with a distance of 6.5 m from the building during image capture [52]. Mayer et al. [95] conducted a comprehensive study to determine optimal UAV flight parameters by comparing thermal image quality from dynamic flights against a stationary reference. They evaluated three camera angles (0°, 45°, and 90°), two flight modes (manual and automated), three flight speeds (1, 3, and 5 m/s), flight altitudes ranging from 4 to 60 m, and distances of 4, 8, and 15 m from the building. For qualitative assessment, the results indicated that flight speeds up to 5 m/s did not noticeably degrade thermal image quality. The study confirmed that nadir (90°) angles, while beneficial for rooftop imaging, are unsuitable for façade analysis due to temperature distortions and reduced contrast caused by reflective surfaces. In contrast, a 45° camera angle provided clearer thermal imagery for both roofs and façades. Although higher altitudes such as 42 m enable broader coverage, they result in lower resolution and reduced visibility of thermal anomalies compared to lower altitudes like 22 m. As a result, the authors recommended using a 45° viewing angle and a 22-m flight distance for effective qualitative thermal inspections of buildings [95].

In quantitative applications, Zhang et al. [59] used dynamic IRT to estimate the thermal transmittance (U-value) of an exterior wall and tested three distances from the building, 15, 30, and 45 m, to evaluate the effect of distance on measurement accuracy. They also examined the impact of UAV speed, using two flight speeds: 0.5 m/s and 1.5 m/s, while maintaining an 80% image overlap and strip flight path. The results showed that as the drone’s distance from the wall increased, the accuracy of thermal transmittance measurements decreased, with errors rising from 6.2% to 21.4% compared to reference heat flux meter data. Additionally, higher flight speeds led to more pronounced thermal image distortions, though temperature reading fluctuations at the remained relatively minor [59]. Yang et al. [65] proposed a methodology to enhance surface temperature measurement accuracy in dynamic IRT. In their field experiment, the UAV was flown at 6.5 m from the building and at a height of 3 m [65]. Rodriguez et al. [60] employed dynamic IRT to calculate the thermal transmittance of an exterior wall using a manual flight path. To minimize measurement errors, they avoided camera angles greater than 50°, noting that steeper angles could distort results. Additionally, to reduce atmospheric absorption of the radiation emitted by the wall surface, they selected a minimal flight distance of 3 m [60]. Bayomi et al. [62] utilized dynamic IRT to calibrate energy models by evaluating thermal transmittance. They implemented both horizontal and vertical strip flight paths with a 90% image overlap [62]. Zheng et al. [61] applied dynamic IRT to generate 3D models of multiple buildings and identify high energy-consuming structures. The UAV surveys were conducted at distances ranging from 8 to 15 m from the buildings, following a strip flight path with 90% image overlap [61]. Mayer et al. [95] similar to their qualitative assessment, proposed optimal flight parameters for quantitative thermal analysis by comparing UAV-based measurements to stationary reference data. Their findings revealed that increasing the flight altitude significantly reduced temperature measurement accuracy, with deviations reaching up to 8 °C compared to handheld infrared readings. They also found that rooftop assessments were more reliable than façade inspections at altitudes between 22 and 42 m, as the impact of camera distance on temperature accuracy was less severe for horizontal surfaces. Based on these observations, the authors recommended performing close-range surveys at distances up to 15 m with a 0° (nadir) camera angle to enhance accuracy in quantitative thermographic evaluations [95]. Mahmoodzadeh et al. [10] applied dynamic IRT to measure the surface temperature of an exterior wall and proposed a methodology to enhance camera performance under dynamic conditions. In their experiment, the UAV was flown at a height of 3 m, and measurements were compared at two different distances from the wall: 2 m and 8 m. The goal was to assess whether forced convection, caused by drone-induced airflow, would influence thermal readings. Despite notable differences in wind speed at the wall surface (1.52 m/s at 2 m and 0.11 m/s at 8 m), the temperature trends remained consistent across both distances. This indicated that variations in convective heat transfer had little to no effect on measurement accuracy. The study suggests that UAVs can operate at close proximity to building envelopes for detecting small-scale defects or air leakage without compromising thermal data quality [10]. Zheng et al. [64] conducted a study using dynamic IRT to generate 3D point cloud models enriched with temperature data for evaluating the thermal performance of building envelopes. The UAV survey was carried out at a distance of 30 m from the buildings, with a camera angle of 45°, a flight speed of 2 m/s, and 90% image overlap [64]. Ortiz-Sanz et al. [46] employed dynamic IRT to measure surface temperatures and detect thermal anomalies in a wine cellar. They observed that when the camera angle was less than 45°, most of the radiation captured by the sensor originated directly from the target surface, resulting in more accurate measurements. However, measurement errors became more pronounced with errors up to 6 °C at tilt angles between 45° and 50° and increased significantly beyond that range due to reflections from the sky and sunlight. The authors emphasized that these reflection effects are highly variable and depend on several factors such as time of day, location, atmospheric conditions, and surface characteristics [46]. Furthermore, selecting appropriate flight parameters plays a vital role in enhancing the accuracy and overall quality of dynamic IRT surveys.

Table 6. Summary of UAV Flight Parameters for Dynamic IRT Applications.

Study	Application	Flight Altitude (M)	Distance (M)	Flight Path	Flight Mode	Camera Angle (°)	Speed (M/S)	Image Overlap (%)
Moore et al. [48]	Qualitative	20	-	-	-	-	-	-
Rakha et al. [50]	Qualitative	-	30	Strip and orbit path	Automated	45	-	95
Rakha & Gorodetsky [93]	Qualitative	12	-	Elliptical path	-	-	-	95
Dabetwar et al. [53]	Qualitative	75	-	Strip path	Automated	25	-	90
Zhang et al. [54]	Qualitative	-	6	Strip path	Manual	-	-	-
Chen et al. [55]	Qualitative	15	10	Strip path	Automated	-	-	-
Hou et al. [94]	Qualitative	35	-	-	-	30	-	-
Mirzabeigi & Razkenari [52]	Qualitative	-	6.5	Strip path	-	-	-	-
Mayer et al. [95]	Qualitative	-	22	-	Automated	45	-	-
Zhang et al. [59]	Quantitative	-	15	-	-	-	0.5	80
Yang et al. [65]	Quantitative	3	6.5	-	-	-	-	-
Bayomi et al. [62]	Quantitative	-	-	Horizontal and vertical strips path	Automated	-	-	90
Zheng et al. [61]	Quantitative	-	8 and 15	Strip path	Automated	-	-	90
Mahmoodzade et al. [10]	Quantitative	3	2 and 8	Static wall-facing	Manual	-	-	-
Zheng et al. [64]	Quantitative	-	30	-	Automated	45	2	90
Mayer et al. [95]	Quantitative	-	15	-	Manual	0	-	-

summarizes the key parameters reported in the literature, with only the recommended values included for studies that tested multiple configurations.

Overall, the reviewed studies indicate that the selection of flight parameters plays a critical role in determining the accuracy and quality of UAV-based dynamic infrared thermography surveys. Although the literature reports a wide range of operational configurations, several consistent trends emerge when the results are synthesized. For qualitative inspections aimed at identifying thermal anomalies such as insulation defects, moisture intrusion, or thermal bridges, moderate flight altitudes typically between 15 and 35 m provide an effective balance between spatial resolution and inspection coverage. In contrast, quantitative applications, such as surface temperature measurement or thermal transmittance (U-value) estimation, generally require closer survey distances below 15 m to minimize atmospheric attenuation and improve measurement accuracy. Camera viewing angles between 30° and 45° are commonly recommended for façade inspections because they enable effective imaging of vertical surfaces while limiting reflection and geometric distortion, whereas nadir views (90°) are more suitable for rooftop surveys. Additionally, high image overlap (typically ≥90%) has been shown to significantly improve the quality of thermal orthomosaics and 3D thermal models, while moderate UAV speeds around 2–3 m/s help maintain stable image acquisition and reduce motion-related distortions. These synthesized observations are summarized in

Table 7. Summary of recommended UAV flight parameters for dynamic IRT surveys in building inspection applications.

Parameter	Qualitative Inspection (Anomaly Detection)	Quantitative Analysis (U-Value/Temperature Measurement)	Key Considerations
Flight altitude distance from building	15–35 m typical; up to ~50 m for large-area screening	Preferably <15 m; often 3–10 m for accurate measurements	Lower distances improve spatial resolution and reduce atmospheric attenuation
Camera viewing angle	30–45° recommended for façade inspections; 90° (nadir) suitable for rooftops	Preferably ≤45°; near-nadir or shallow oblique angles preferred	Larger angles increase reflection and measurement error
Image overlap	≥70% for basic surveys; up to 90–95% for 3D reconstruction	≥80–90% recommended	Higher overlap improves orthomosaic and thermal model accuracy
Flight speed	Up to ~5 m/s acceptable for qualitative inspections	Preferably ≤2 m/s for stable temperature measurements	Lower speed reduces motion blur and improves thermal stability
Flight pattern	Strip or orbit flight paths around building	Strip paths parallel to façade	Consistent flight paths improve image alignment
Camera distance variability	Moderate variation acceptable	Maintain consistent distance from target surface	Consistency improves measurement comparability

, which presents indicative parameter ranges derived from the literature for both qualitative and quantitative UAV-based IRT applications. While a universal standard for UAV-based thermographic inspections has not yet been established, the parameter ranges presented in Table 7 provide practical guidance for selecting appropriate flight configurations based on the inspection objective and desired measurement accuracy.

6. Challenges and Limitations of Quantitative UAV-Based IRT

Like many emerging technologies, dynamic infrared thermography presents certain limitations. As previously noted in this paper, several studies have reported measurement inaccuracies when using dynamic IRT. These errors range from as little as 2 °C to as high as 35 °C. This chapter explores the underlying causes of these discrepancies and presents potential solutions to mitigate them, aiming to enhance the accuracy and reliability of dynamic IRT for building diagnostics.

As previously discussed, most commercial infrared cameras are equipped with microbolometer detectors [74]. Unlike photonic detectors, these systems lack an internal cooling mechanism to stabilize the sensor temperature and reduce noise [18]. As a result, they require a thermal equilibrium to achieve optimal performance. This equilibrium occurs when the heat generated by the camera’s internal electronics is balanced by the heat lost to the external environment (as detailed in Chapter 3.4) [76]. In dynamic setups, this balance is disrupted. The airflow generated by the UAV’s propellers can cool the camera housing, lens, and other components, preventing the system from reaching or maintaining thermal equilibrium. This fluctuation in the camera’s internal temperature, or in its immediate surroundings, introduces a phenomenon known as thermal drift, which is cause of these errors [96]. The summary of the maximum errors reported by researchers in the building sector is presented in

Table 8. Maximum Reported Errors in UAV-Based Infrared Thermography Studies.

Study	Maximum Reported Errors (°C)
Mayer et al. [95]	8
Sadhukhan et al. [66]	17.1
Stokowiec and Sobura [67]	1
Yang et al. [65]	8
Gil-Docampo et al. [47]	8
Mahmoodzadeh et al. [10]	32
Zheng et al. [64]	5

.

To address thermal drift, several solutions have been proposed, but the first and most commonly used method is performing NUC function. Most commercial infrared cameras are equipped with shutter-based NUC systems [97]. In this method, a non-transparent shutter is periodically placed between the lens and the detector to update the camera’s offset parameters after a specific time interval or temperature change [96]. The camera adjusts the gain and offset values based on the known temperature of the shutter, which is assumed to match the temperature of the camera interior components [23]. As reported by several researchers, activating NUC during flight can help stabilize the camera more quickly and reduce measurement errors [10,98]. However, NUC alone is not sufficient to eliminate the effects of thermal drift. One of the major limitations is that the method assumes the shutter temperature equals to the temperature of the entire camera interior. In dynamic UAV setups, the outer parts of the camera, like the lens and housing, are exposed to wind and may cool down faster than the inner components, leading to temperature differences that NUC do not consider [78]. Moreover, in dynamic setups, the heat gain from the inside and the heat loss to the outside are still not balanced. So even after each NUC activation and correction, because of this imbalance, thermal drift will continue to occur until it eventually reaches equilibrium. Another drawback is that each NUC activation briefly interrupts the thermal imaging process, causing frozen images for 1 to 2 s [97]. Excessive NUC activation can therefore affect data quality, especially in time-series thermal surveys. Despite these limitations, setting the camera to automatically perform NUC every minute can be beneficial in dynamic conditions, helping the system stabilize faster. Nevertheless, NUC alone cannot fully eliminate thermal drift, and additional correction methods are needed to minimize its impact on measurement accuracy.

Yuan and Hua [98] conducted a laboratory experiment and observed that under wind speeds of 2 m/s and 7 m/s, infrared cameras recorded errors of up to 15 °C. To minimize these errors, they recommended allowing the camera to warm up for 30 min prior to flight and suggested using an extra battery to keep the camera stabilized in the air or employing a portable fan to simulate and counteract wind effects before takeoff. Ribeiro-Gomes et al. [99] adopted a neural network-based calibration method using a laboratory setup with a blackbody source whose temperature varied between 5 °C and 65 °C. Their model successfully reduced the average measurement error from 3.55 °C to 1.37 °C. Similarly, Aragon et al. [100] implemented a multilinear regression model for camera calibration, using camera temperature readings and ambient air temperature as input features. They gather data across four ambient temperatures (4 °C, 22 °C, 33 °C, and 37 °C). The result showed that the model reduce bias and RMSE by 6.084 °C and 5.404 °C, respectively. However, during UAV field tests, they still recorded errors of up to 4 °C. Mahmoodzadeh et al. [10], proposed a shielding method to reduce airflow-induced errors, which lowered the maximum observed error from 7 °C to 4 °C. Yang et al. [65] employed a similar shielding approach, reporting a reduction in temperature deviation from 8 °C to 2.5 °C.

Some researchers have employed reference-based methods to calibrate thermal images using targets with known temperatures. For example, Al Maashri et al. [101], in the healthcare domain, proposed an algorithm that uses a person’s forehead temperature as a reference point to calibrate infrared camera readings. Their method reduces the measurement errors to less than 1 °C. Similarly, Pestana et al. [102] utilized the known temperature of melting snow as a calibration reference. Torres-Rua [103] placed a blackbody reference on-site to adjust the camera’s output accordingly. To enhance accuracy, Gómez-Candón et al. [104] recommended flying over the reference target at three stages of the survey, beginning, middle, and end, to ensure consistent calibration throughout the inspection.

Overall, thermal drift errors in dynamic setups have not yet been fully resolved. Although current calibration methods significantly reduce measurement errors, some inaccuracies still persist. This highlights a gap in the existing literature; there remains a need for a comprehensive approach that can fully calibrate infrared cameras under dynamic conditions. Most authors recommend a minimum warm-up period of 15–30 min for laboratory or stationary inspections, with longer durations of up to 60 min preferable for dynamic or aerial surveys.

7. Conclusions

This paper provides a comprehensive review of dynamic infrared thermography (IRT) applications in the building sector, evaluating its effectiveness for thermal inspection and diagnostics while synthesizing existing processes, associated technologies, prevailing research trends, key methodologies, and principal challenges. The review highlights how dynamic IRT, particularly when integrated with unmanned aerial vehicles (UAVs) has emerged as a powerful and promising approach for building envelope assessment.

The environmental conditions indicate that optimal inspection parameters vary depending on the specific IRT application and building envelope characteristics. A sufficient temperature differential (ΔT) is essential for generating thermal contrast; most qualitative inspections require at least ΔT ≥ 10 °C, while quantitative analyses such as U-value estimation typically require ΔT ≥ 15 °C under stable conditions. Wind speed also significantly influences measurement reliability because increased airflow enhances convective heat transfer and may distort surface temperature patterns. As a result, wind speeds should generally remain below 2–3 m/s for qualitative inspections and preferably below 1 m/s for quantitative measurements. Solar exposure must also be minimized to avoid false thermal anomalies; therefore, thermographic surveys are commonly conducted after sunset or during night-time conditions. Building envelope characteristics further influence inspection timing. High thermal-mass structures such as masonry or concrete may require 6–8 h after sunset to dissipate stored solar heat, whereas lightweight façade systems typically reach thermal equilibrium within 2–4 h after sunset. These findings demonstrate that environmental thresholds should be interpreted as application-specific guidelines rather than fixed limits, and inspection planning should consider both the diagnostic objective and the thermal response of the building envelope. Clear trends also emerge with respect to flight configuration: close-range surveys (typically <15 m) are better suited for quantitative assessments, while moderate altitudes (15–35 m) support effective qualitative inspections by balancing atmospheric effects and spatial resolution. Camera viewing angles between 30° and 45° are commonly recommended for façade inspections, whereas nadir views (90°) are preferred for roof surveys. High image overlap (≥90%) consistently improves thermal mapping and three-dimensional reconstruction, and flight speeds of 2–3 m/s are frequently adopted, with slower speeds favored for quantitative measurements.

The analysis of camera operation also indicates that stabilization of thermal detector infrared cameras is a critical factor affecting measurement accuracy. Based on the reviewed studies, a minimum warm-up period of approximately 30 min is generally required for reliable measurements, while 30–60 min is recommended for UAV-based dynamic inspections to ensure thermal equilibrium under changing environmental conditions.

Despite these advances, significant limitations remain. Procedures for determining reflected apparent temperature and emissivity were originally developed for stationary IRT applications and are not directly transferable to dynamic surveys. In UAV-based inspections, where measurement distance, viewing angle, and elevation vary continuously, reflected temperature and emissivity can fluctuate substantially, introducing considerable uncertainty when single-value assumptions are applied. Thermal drift in dynamic environments also remains insufficiently addressed; although existing calibration approaches can reduce errors, reported deviations of up to 32 °C indicate that current methods are inadequate for fully compensating dynamic measurement effects. These findings highlight a critical gap in the literature: the absence of a comprehensive calibration framework capable of ensuring accurate and reliable temperature retrieval under real flight conditions.

Future research should prioritize several key challenges to enable the reliable and standardized application of UAV-based dynamic infrared thermography building diagnostics. These challenges can broadly be categorized into measurement-related issues and methodological developments.

From a measurement perspective, improving the correction of thermal drift during UAV-based inspections is essential to ensure stable and reliable temperature measurements. Since thermal cameras may experience radiometric drift during operation, robust drift correction procedures are required to maintain measurement consistency and accuracy.

From a methodological perspective, the development of standardized procedures specifically tailored to dynamic IRT applications remains a critical priority. Particular attention should be given to improving emissivity estimation techniques, reflected temperature determination, and environmental correction methods during aerial inspections. Establishing standardized data acquisition protocols, flight planning guidelines, and validation procedures will be essential to ensure measurement repeatability and comparability across different studies and inspection conditions.

Overall, addressing these technological and methodological challenges will be crucial for advancing the field toward reliable, standardized, and widely adopted UAV-based dynamic IRT systems for quantitative building envelope diagnostics.