Thermal assessment system and method

Abstract

There is provided a thermal assessment system for generating a heat map of an object, the thermal assessment system comprising: a thermal image acquisition device configured to, in use, acquire thermal image data of the object; a visible light image acquisition device configured to, in use, acquire visible light image data of the object; and a processing module programmed to, in use: process the acquired visible light image data to generate a geometric model of the object; process the acquired thermal image data to generate a radiometric model of the object; combine the visual and radiometric models into an object model by matching features of the geometric model with features of the radiometric model.

Description

[0001] THERMAL ASSESSMENT SYSTEM AND METHOD

[0002] The invention relates to a thermal assessment system and method for generating a heat map of an object, particularly a building.

[0003] The problem of building energy inefficiency is typically addressed through manual inspections and audits. These inspections often involve professionals physically examining a building to identify areas of potential energy loss, such as poorly insulated walls or draughty windows. In some cases, thermal imaging cameras might be used to visually identify areas of heat loss.

[0004] However, these methods have several shortcomings:

[0005] • Time and Cost: Manual inspections are labour-intensive and time-consuming, making them expensive to carry out, especially for larger buildings or multiple properties.

[0006] • Lack of Precision: While thermal imaging cameras can be used by a trained thermographer to qualitatively identify areas of relative heat loss, they are generally not used quantitatively in building inspection and therefore do not provide a high level of detail or accuracy. They can fail to provide a comprehensive view of a building's overall energy performance, and so return on investment calculations for addressing identified heat leaks must be done manually.

[0007] • Lack of Scalability: Manual inspections do not scale well. Inspecting multiple buildings or a large portfolio of properties can be a logistical challenge.

[0008] • Subjectivity: The results of manual inspections can be influenced by the inspector's individual judgement and expertise, leading to potential inconsistencies in the assessment.

[0009] According to a first aspect of the invention, there is provided a thermal assessment system for generating a heat map of an object, the thermal assessment system comprising: a thermal image acquisition device configured to, in use, acquire thermal image data of the object; a visible light image acquisition device configured to, in use, acquire visible light image data of the object; and a processing module programmed to, in use:

[0010] • process the acquired visible light image data to generate a geometric model of the object; • process the acquired thermal image data to generate a radiometric model of the object;

[0011] • combine the visual and radiometric models into an object model by matching features of the geometric model with features of the radiometric model.

[0012] It will be appreciated that the thermal image acquisition device may be distinct from the visible light image acquisition device, or that the thermal and visible light image acquisition devices may form part of the same overall image acquisition apparatus.

[0013] In embodiments of the invention, the thermal image and visible light image acquisition devices may be operable to acquire the thermal image data and the visible light data at different times respectively.

[0014] In further embodiments of the invention, the acquired visible light image data may be processed separately from the acquired thermal image data.

[0015] In still further embodiments of the invention, the thermal image acquisition device may be configured to, in use, acquire the thermal image data of the object from oblique and / or vertical aspects, and / or wherein the visible light image acquisition device may be configured to, in use, acquire the visible light image data of the object from oblique and / or vertical aspects.

[0016] The thermal image data may be long wave infrared image data. The visible light image data may be or may include colour image data, such as RGB image data.

[0017] The geometric model may be, but is not limited to, a photogrammetric model.

[0018] In embodiments of the invention, the processing module may be programmed to, in use, perform geometric segmentation processing to identify at least one feature and / or at least one measurement of the object from the visible light image data.

[0019] In further embodiments of the invention, the processing module may be programmed to, in use, perform visual segmentation processing to identify at least one feature and / or at least one material of the object from the visible light image data. In such embodiments, the processing module may be programmed to, in use, perform the visual segmentation processing using a machine learning algorithm or model. The machine learning algorithm or model may be or may include, but is not limited to, a convolutional neural network. In still further embodiments of the invention, the processing module may be programmed to, in use, assign a surface emissivity value to the or each identified material of the object.

[0020] The processing module may be programmed to, in use, account (e.g. compensate or correct) for variation in pose (e.g. position and / or orientation) of the thermal and / or visible light image acquisition devices when combining the visual and radiometric models by matching features of the geometric model with features of the radiometric model. The processing module may be programmed to, in use, account (e.g. compensate or correct) for temporal variation when combining the visual and radiometric models by matching features of the geometric model with features of the radiometric model.

[0021] The processing module may be programmed to, in use, to use the radiometric model to identify, predict or estimate a thermal transmissivity value of one or more components of the object. The processing module may be programmed to, in use, to use the radiometric model to identify or classify at least one thermal anomaly of the object.

[0022] The thermal assessment system is for generating a heat map of a variety of objects, such as a building.

[0023] The processing module may include a processor and memory including computer program code. The memory and computer program code may be configured to, with the processor, enable the processing module at least to:

[0024] • process the acquired visible light image data to generate the geometric model of the object;

[0025] • process the acquired thermal image data to generate the radiometric model of the object;

[0026] • combine the visual and radiometric models into the object model by matching features of the geometric model with features of the radiometric model.

[0027] The processing module may be, may include or may form part of one or more of an electronic device, a portable electronic device, a portable telecommunications device, a mobile phone, a personal digital assistant, a tablet, a phablet, a laptop computer, a server, a cloud computing network, a smartphone, a smartwatch, smart eyewear, and a module for one or more of the same. It will be appreciated that references to a memory or a processor may encompass a plurality of memories or processors. According to a second aspect of the invention, there is provided a method of generating a heat map of an object, the method comprising the steps of: acquiring thermal image data of the object; acquiring visible light image data of the object; processing the acquired visible light image data to generate a geometric model of the object; processing the acquired thermal image data to generate a radiometric model of the object; combining the visual and radiometric models into an object model by matching features of the geometric model with features of the radiometric model.

[0028] The features and advantages of the first aspect of the invention and its embodiments apply mutatis mutandis to the features and advantages of the second aspect of the invention and its embodiments.

[0029] Non-limiting further features of the method of the invention are as follows.

[0030] The thermal image data and the visible light data may be acquired at different times respectively. The acquired visible light image data may be processed separately from the acquired thermal image data. The thermal image data of the object may be acquired from oblique and / or vertical aspects. The visible light image data of the object may be acquired from oblique and / or vertical aspects.

[0031] The thermal image data is long wave infrared image data. The visible light image data may be or may include colour image data, such as RGB image data.

[0032] The geometric model may be, but is not limited to, a photogrammetric model.

[0033] The method may further include the step of performing geometric segmentation processing to identify at least one feature and / or at least one measurement of the object from the visible light image data.

[0034] The method may further include the step of performing visual segmentation processing to identify at least one feature and / or at least one material of the object from the visible light image data. The visual segmentation processing may be performed using a machine learning algorithm or model. The machine learning algorithm or model may be or may include, but is not limited to, a convolutional neural network.

[0035] In the method of the invention, the processing module may be programmed to, in use, assign a surface emissivity value to the or each identified material of the object. The method may include the step of accounting (e.g. compensating or correcting) for variation in pose of the thermal and / or visible light image acquisition devices when combining the visual and radiometric models by matching features of the geometric model with features of the radiometric model.

[0036] The method may include the step of accounting (e.g. compensating or correcting) for temporal variation when combining the visual and radiometric models by matching features of the geometric model with features of the radiometric model.

[0037] The method may include the step of using the radiometric model to identify, predict or estimate a thermal transmissivity value of one or more components of the object. The method may include the step of using the radiometric model to identify or classify at least one thermal anomaly of the object.

[0038] The method may be for generating a heat map of a variety of objects, such as a building.

[0039] According to a third aspect of the invention, there is provided a computer-implemented method of generating a heat map of an object, the method comprising the steps of: processing visible light image data acquired from the object to generate a geometric model of the object; processing thermal image data acquired from the object to generate a radiometric model of the object; combining the visual and radiometric models into an object model by matching features of the geometric model with features of the radiometric model.

[0040] According to a fourth aspect of the invention, there is provided a computer program comprising computer code configured to perform the method of the third aspect of the invention.

[0041] The features and advantages of the first and second aspects of the invention and their embodiments apply mutatis mutandis to the features and advantages of the third and fourth aspects of the invention and their embodiments.

[0042] It will be appreciated that the use of the terms "first" and "second", and the like, in this patent specification is merely intended to help distinguish between similar features, and is not intended to indicate the relative importance of one feature over another feature, unless otherwise specified. Within the scope of this application it is expressly intended that the various aspects, embodiments, examples and alternatives set out in the preceding paragraphs, and the claims and / or the following description and drawings, and in particular the individual features thereof, may be taken independently or in any combination. That is, all embodiments and all features of any embodiment can be combined in any way and / or combination, unless such features are incompatible. The applicant reserves the right to change any originally filed claim or file any new claim accordingly, including the right to amend any originally filed claim to depend from and / or incorporate any feature of any other claim although not originally claimed in that manner.

[0043] Preferred embodiments of the invention will now be described, by way of non-limiting examples, with reference to the accompanying drawing in which:

[0044] Figure 1 illustrates a crosshatch flight pattern for acquiring images of an object from oblique and vertical aspects.

[0045] Current methods for assessing building energy efficiency are often time-consuming, expensive, and lack precision. They typically involve manual inspections and can miss subtleties that a more sophisticated approach could catch. Furthermore, these methods often fail to provide a comprehensive view of a building's overall energy performance, making it difficult to prioritise and plan retrofitting efforts effectively.

[0046] This problem is worth solving because improving energy efficiency in buildings is a key strategy for reducing greenhouse gas emissions and combating climate change, with 17% of global greenhouse gas emissions being generated from heating and cooling buildings. It also has the potential to reduce energy costs for building owners and occupants, and contribute to a more sustainable and energy-efficient economy.

[0047] The primary problem addressed by this invention is the significant energy inefficiency in existing buildings, which contributes to a substantial portion of global greenhouse gas emissions. The technical challenge lies in scalably and accurately identifying where and how large portfolios of buildings are losing heat, which is crucial for portfolio property owners to plan effective retrofitting strategies.

[0048] The invention's approach to this problem, which involves using artificial intelligence (Al), thermal imaging and building physics to create detailed heat leakage maps of the built environment at scale, has not been done before due to the complexity of the task and the advanced technology required. The integration of Al, thermal imaging and building physics simulation technology in this way has been applied by the inventors to the problem of building energy efficiency in a novel and innovative way.

[0049] The invention is a system that combines artificial intelligence, thermal imaging, and building physics to create a detailed map of heat leakage in buildings at scale. It provides a solution to the problem of lack of source data for planning, pricing and validation of retrofit works across large portfolios of property assets. This system involves several key features:

[0050] 1. Acquisition of Raw Image Data: The system acquires wide-scale radiometric LWIR (8-14 microns) images, as well as RGB images, in both oblique and vertical aspects. These two types of raw data are processed separately.

[0051] 2. RGB Data Processing: The RGB data is used to generate 3D models of buildings using photogrammetry. The 3D model then undergoes geometric segmentation to identify roofs and walls, and geometric measurements are taken of roof lines, wall shapes / sizes, and glazed areas. Visual segmentation is also performed to identify windows, doors, vegetation, external fixtures, and surface materials. The latter is used to determine LWIR emissivity of the surfaces and feeds into "Thermal Data Processing".

[0052] 3. Thermal Data Processing: The thermal data is processed using a radiation physics model to create a 3D radiometric model with accurate measurement of surface temperature for each surface of the building, compensating for distance, humidity, viewing angle, surface emissivity and thermal radiation from surrounding objects as well as the sky and the ground. The result is a surface temperature 3D point cloud. The system then uses the 3D radiometric model as input for 2 parallel purposes: a. a predictive building physics model that combines exterior surface temperature, atmospheric conditions, identified materials of building construction, and building age to estimate each building component's (wall, window, roof etc) U-value (thermal transmissivity). b. an anomaly detection classifier that identifies issues such as thermal bridges, poorly installed or degraded insulation, damp and sources of draughts.

[0053] 4. Integration of Data: The processed thermal and RGB data are then integrated into a per-property report.

[0054] What sets this invention apart from existing solutions is its combination of advanced Al, thermal imaging and building physics simulation technology to provide a detailed, accurate, and comprehensive view of a building's energy performance. Unlike manual inspections, which can be time-consuming and lack precision, this system can quickly and accurately identify and quantify heat loss in a building. This allows for more effective planning of retrofitting strategies and helps to ensure that these strategies are successful in improving the building's energy efficiency.

[0055] Features of the invention include:

[0056] • Acquisition of raw image data: The system requires both LWIR (8-14 microns) thermal images and RGB images of the building.

[0057] • Separate processing of thermal and RGB data: The two types of raw data are processed separately, each undergoing its own series of steps.

[0058] • Radiation physics: This process is used to create a 3D radiometric model from the thermal data, which provides an accurate measurement of surface temperature for each surface of the building's 3D model.

[0059] • Building physics: This process is used to get a more accurate estimate of the building component's U-value (the thermal leak value).

[0060] • Anomaly detection: This process identifies issues such as damp, poorly installed insulation, and draft sources in the thermal data.

[0061] • Geometric and visual segmentation: These processes are used to identify and measure various features in the RGB data, such as roofs, walls, windows, doors, vegetation and external fixtures.

[0062] • Integration of data: The processed thermal and RGB data are integrated to create a comprehensive view of the building's energy performance.

[0063] • Thermal image alignment

[0064] • Flight planning

[0065] The invention is applicable, but not limited, to:

[0066] • Building pricing assessment: Using the RGB data, the system can perform a building pricing assessment, which can provide additional useful information.

[0067] • Large-scale image acquisition: The ability to acquire wide-scale radiometric LWIR images can improve the system's effectiveness by providing a more comprehensive view of the building.

[0068] • Retrofit recommendations.

[0069] • Air source heat pump siting.

[0070] An exemplary implementation of the invention is described as follows:

[0071] 1. Acquisition of Raw Image Data a. The system begins by acquiring two types of raw image data: LWIR (8-14 microns) thermal images and RGB images using a camera or a different type of image acquisition device. For example, these images may be taken from both oblique and vertical aspects in a crosshatch flight pattern as shown in Figure 1, using an off-the-shelf drone, over a survey area to provide a comprehensive view of each building in the survey area. Individual survey areas are tessellated together to produce a large contiguous set of imagery across whole cities. b. The images are uploaded to a database, e.g. a cloud storage database, and indexed using location data, e.g. EXIF geolocation data, captured by the camera. c. The thermal and RGB images are processed separately, each undergoing its own series of steps.

[0072] 2. RGB Data Processing

[0073] RGB data is processed in 2 ways: to generate 3D geometric models and to identify surface materials of buildings. a. Generate 3D geometric models: i. For a given property, the geolocation of the property (from a reliable 3rd party source, e.g. Ordnance Survey) is used to determine the ranges of geolocations and camera angles of images that would include that property in the shot. ii. Images that meet those criteria are selected from the index. iii. The selected images are then pushed into a photogrammetry pipeline, which returns a 3D point cloud, mesh and orthomosaic of the target property. iv. An algorithm, e.g. a plane search algorithm, is applied to the point cloud, identifying planar intersections as edges and corners of the building. v. Walls, roofs, ground level are tagged as such based on a heuristic combination of plane normal and location relative to the other planes. vi. Corners, edges and faces are stored as annotations. vii. Having identified primary surfaces, the textures for those surfaces are segmented to identify 3D locations for windows, doors and other notable features. viii. The output of this process includes 3D mesh, RGB texture, surface, edge and corner annotations for each property surveyed. ix. The output is populated in a graph data structure to maintain relationships between neighbouring features of the property. b. Segmentation to identify surface materials of buildings that aids accurate determination of surface emissivity or LWIR radiation: i. A convolutional neural network (CNN) is trained to identify building materials in 2D RGB images. ii. To create the training data, RGB drone images containing high quality building images are identified, and the 3D meshes created in the above step 2a are used to select a region of each drone image for annotation. High quality images of buildings are identified by computing the proximity and angle of the geolocated 3D model to the dronecoordinates recorded in the image meta-data. The images may be sent to human annotators, who identify the building materials in each image. iii. The convolutional neural network is trained using the images and annotations obtained from step 2a to identify surface materials in 2D images. Each pixel in the image may be assigned a label denoting a material recognised by the CNN, or may be assigned an "Ignore" label. iv. The trained CNN can then be used to label the pixels in any image with the building materials present at that pixel. v. Identified materials are mapped using industry-standard emissivity lookup tables to give an emissivity value for each pixel in the 2D image. vi. The emissivity field for each image relevant to a given property and project onto the 3D model is provided by a weighted average blending of emissivity values for the same surface based on distance to surface and angle to surface normal. vii. The output of this process includes a "texture" of each 3D model for the emissivity, and labels of each surface identified in step 2a with corresponding material.

[0074] 3. Thermal Data Processing a. Alignment of thermal images to the 3D geometric model: a. The system employs a multi-model landmark detection and matching algorithm specifically designed to facilitate cross-spectral feature matching between visible (RGB) and thermal (LWIR) imagery. Unlike existing commercial solutions that require synchronized capture from the same platform, this novel approach enables the matching of imagery captured from different drones at different times. This is particularly crucial for thermal inspections that must be conducted at night for optimal measurement accuracy and for a second inspection to assess the effectiveness of the retrofit actions taken after the first assessment. Furthermore, there may be spatial variation in the images taken due to variation in camera pose, which may result in translation and / or rotation between the visible and thermal image data sets. Concretely, the algorithm identifies and matches distinctive features across both spectral domains, and then computes a precise 6-dimensional transformation matrix that accounts for different camera poses and temporal variations. This transformation enables accurate registration of thermal data onto the photogrammetric 3D model, resulting in an object model with spatially accurate thermal measurements across the entire structure. b. Radiation physics modelling for accurate measurement of surface temperature for each surface of the building's 3D model

[0075] For each thermal image having the given property in sight, the surface temperature in the image is determined as follows:

[0076] For each pixel in the aligned thermal image:

[0077] 1. Calculate the point in the building 3D model that corresponds to the pixel.

[0078] 2. Calculate a distance between the camera viewpoint and the point in the building 3D model based on Euclidean geometry.

[0079] 3. Read the base emissivity of the surface to which the point belongs in the 3D model (if that pixel intersects with the property).

[0080] 4. Calculate the angle between the surface normal and the camera viewpoint.

[0081] 5. Calculate the ratio between the amount of incident radiation coming from the sky to the total incident radiation.

[0082] 6. Calculate the ratio between the amount of incident radiation coming from the ground to the total incident radiation using a raytracing simulation, tracing emissions from sky and ground onto the building.

[0083] 7. Combine all of the above including atmospheric temperature and humidity recorded at the time of image acquisition using an equation for radiated heat flux to correct for emissivity and reflected temperature and using Beer's law and ideal gas law to correct for atmospheric humidity, atmospheric temperature and distance. 8. Correct for angle based on how emissivity varies with angle to the surface normal for different materials.

[0084] 9. The temperature field for each image relevant to a given property and project onto the 3D model is provided using a weighted average blending of temperature values for the same surface based on distance to surface and angle to surface normal.

[0085] 10. The output of this process includes surface temperature "texture" on a 3D mesh. c. Identify thermal anomalies:

[0086] 1. Generate 2D aspect render images by segmenting the surface temperature "texture" for each surface identified in step 2a.

[0087] 2. Apply computer vision techniques to detect temperature variations within the 2D aspect images, labeling these variations as anomalies.

[0088] 3. Refine the detected anomalies by filtering them based on criteria such as size, average temperature, and proximity to one another.

[0089] 4. Classify the refined anomalies into categories - such as thermal bridging, damp, failed insulation, open vents, etc. - based on the surface type. This classification employs a heuristic approach that considers spatial properties, anomaly size, and average temperature.

[0090] 5. Project the classified anomalies back into 3D space to create annotations for reporting and visualization purposes.

[0091] 6. The output of this process includes thermal annotations located in 3D space relative to the building 3D model.

[0092] 7. Segment the surface temperature "texture" based on surfaces annotated in step 2a. For each surface image, use a trained fully CNN to segment thermal anomalies in the following categories: thermal bridging, damp, failed insulation, draught, open vents (e.g. uncapped chimney).

[0093] 8. Project the anomaly segmentation back into 3D space as an annotation for reporting and visualisation.

[0094] 9. The output of this process includes annotations located in 3D space relative to the property 3D model.

[0095] 4. Building physics modelling: predict the U-values of building components:

[0096] 1. Segment 3D temperature "texture" as in step 3b 2. Simulate the heat flow across building components using thermodynamic simulation based on an estimated internal temperature and material conductivity. The simulation accounts for: a. Material, as labelled in step 2b. b. Thermal anomalies, as labelled in step 2b. c. Age of building (as determined from 3rd party sources such as Ordnance Survey). d. Archetype of property (house, flat, terrace etc). e. Geographic location - used to retrieve weather data.

[0097] 3. The simulation runs for a period of time, e.g. 72 hours, prior to thermal image capture, concluding at the point of capture.

[0098] 4. An optimization algorithm, e.g. a genetic optimization algorithm, determines the optimal building conductivity and internal temperature values that minimize the difference between predicted (based on thermodynamic simulation) and observed surface temperatures from the thermal image.

[0099] 5. Using the optimized parameters, the U-value is calculated from simulated heat flux values and compared to measured ground truth data. A correction factor is then derived to develop a physical-statistical model for predicting U-values.

[0100] 6. U-values determined for a window and wall for the same room (same indoor and outdoor temperatures) may yield exterior surface temperatures that are inconsistent with what was measured from the thermal image data and so may need a correction factor.

[0101] 5. Integration of Data

[0102] The geometric and radiometric models when combined results in an object model that provides a comprehensive, 3D view of the building's energy performance, including a detailed map of heat leakage and potential areas for improvement, with the following features: a. Dimensions, materials and U-values of each building surface are annotated onto a 3D model; b. Expected daily indoor-outdoor temperature difference can be determined based on historical outdoor temperature for the area and typical heating patterns for residents. c. The number of floors of a property can be determined by applying a heuristic combining building height, identified windows and footprint, determined from the 3D model in step 2a. d. The whole property space heating demand intensity can be determined by multiplying U-values of each building surface with dimensions of those surfaces and the indoor-outdoor temperature delta, and dividing by floor area to determine the overall space heating demand intensity in kWh / yr / m2.

[0103] This invention provides a significant improvement over traditional methods of assessing building energy efficiency, which often involve manual inspections and can be time-consuming, expensive, and lack precision. By leveraging advanced Al and thermal imaging technology, this system can quickly and accurately identify areas of heat loss in a building, allowing for more effective planning of retrofitting strategies.

[0104] It will be appreciated that the above numerical values are merely intended to help illustrate the working of the invention and may vary depending on the requirements of the thermal assessment system and the object.

[0105] The listing or discussion of an apparently prior-published document or apparently prior- published information in this specification should not necessarily be taken as an acknowledgement that the document or information is part of the state of the art or is common general knowledge.

[0106] Preferences and options for a given aspect, feature or parameter of the invention should, unless the context indicates otherwise, be regarded as having been disclosed in combination with any and all preferences and options for all other aspects, features and parameters of the invention.

Claims

CLAIMS1. A thermal assessment system for generating a heat map of an object, the thermal assessment system comprising: a thermal image acquisition device configured to, in use, acquire thermal image data of the object; a visible light image acquisition device configured to, in use, acquire visible light image data of the object; and a processing module programmed to, in use:• process the acquired visible light image data to generate a geometric model of the object;• process the acquired thermal image data to generate a radiometric model of the object;• combine the visual and radiometric models into an object model by matching features of the geometric model with features of the radiometric model.

2. A thermal assessment system according to Claim 1 wherein the thermal image and visible light image acquisition devices are operable to acquire the thermal image data and the visible light data at different times respectively.

3. A thermal assessment system according to any one of the preceding claims wherein the acquired visible light image data is processed separately from the acquired thermal image data.

4. A thermal assessment system according to any one of the preceding claims wherein the thermal image acquisition device is configured to, in use, acquire the thermal image data of the object from oblique and / or vertical aspects, and / or wherein the visible light image acquisition device is configured to, in use, acquire the visible light image data of the object from oblique and / or vertical aspects.

5. A thermal assessment system according to any one of the preceding claims wherein the thermal image data is long wave infrared image data.

6. A thermal assessment system according to any one of the preceding claims wherein the visible light image data is or includes colour image data.

7. A thermal assessment system according to Claim 6 wherein the colour image data is RGB image data.

8. A thermal assessment system according to any one of the preceding claims wherein the geometric model is a photogrammetric model.

9. A thermal assessment system according to any one of the preceding claims wherein the processing module is programmed to, in use, perform geometric segmentation processing to identify at least one feature and / or at least one measurement of the object from the visible light image data.

10. A thermal assessment system according to any one of the preceding claims wherein the processing module is programmed to, in use, perform visual segmentation processing to identify at least one feature and / or at least one material of the object from the visible light image data.

11. A thermal assessment system according to Claim 10 wherein the processing module is programmed to, in use, perform the visual segmentation processing using a machine learning algorithm or model.

12. A thermal assessment system according to Claim 11 wherein the machine learning algorithm or model is or includes a convolutional neural network.

13. A thermal assessment system according to any one of Claims 10 to 12 wherein the processing module is programmed to, in use, assign a surface emissivity value to the or each identified material of the object.

14. A thermal assessment system according to any one of the preceding claims wherein the processing module is programmed to, in use, account for variation in pose of the thermal and / or visible light image acquisition devices when combining the visual and radiometric models by matching features of the geometric model with features of the radiometric model.

15. A thermal assessment system according to any one of the preceding claims wherein the processing module is programmed to, in use, account for temporal variation when combining the visual and radiometric models by matching features of the geometric model with features of the radiometric model.

16. A thermal assessment system according to any one of the preceding claims wherein the processing module is programmed to, in use, to use the radiometric model to identify, predict or estimate a thermal transmissivity value of one or more components of the object.

17. A thermal assessment system according to any one of the preceding claims wherein the processing module is programmed to, in use, to use the radiometric model to identify or classify at least one thermal anomaly of the object.

18. A thermal assessment system according to any one of the preceding claims wherein the thermal assessment system is for generating a heat map of a building.

19. A thermal assessment system according to any one of the preceding claims wherein the processing module includes a processor and memory including computer program code, the memory and computer program code configured to, with the processor, enable the processing module at least to:• process the acquired visible light image data to generate the geometric model of the object;• process the acquired thermal image data to generate the radiometric model of the object;• combine the visual and radiometric models into the object model by matching features of the geometric model with features of the radiometric model.

20. A method of generating a heat map of an object, the method comprising the steps of: acquiring thermal image data of the object; acquiring visible light image data of the object; processing the acquired visible light image data to generate a geometric model of the object; processing the acquired thermal image data to generate a radiometric model of the object; combining the visual and radiometric models into an object model by matching features of the geometric model with features of the radiometric model.

21. A method according to Claim 20 wherein the thermal image data and the visible light data are acquired at different times respectively.

22. A method according to Claim 20 or Claim 21 wherein the acquired visible light image data is processed separately from the acquired thermal image data.

23. A method according to any one of Claims 20 to 22 wherein the thermal image data of the object is acquired from oblique and / or vertical aspects, and / or wherein the visible light image data of the object is acquired from oblique and / or vertical aspects.

24. A method according to any one of Claims 20 to 23 wherein the thermal image data is long wave infrared image data.

25. A method according to any one of Claims 20 to 24 wherein the visible light image data is or includes colour image data.

26. A method according to Claim 25 wherein the colour image data is RGB image data.

27. A method according to any one of Claims 20 to 26 wherein the geometric model is a photogrammetric model.

28. A method according to any one of Claims 20 to 27 further including the step of performing geometric segmentation processing to identify at least one feature and / or at least one measurement of the object from the visible light image data.

29. A method according to any one of Claims 20 to 28 further including the step of performing visual segmentation processing to identify at least one feature and / or at least one material of the object from the visible light image data.

30. A method according to Claim 29 wherein the visual segmentation processing is performed using a machine learning algorithm or model.

31. A method according to Claim 30 wherein the machine learning algorithm or model is or includes a convolutional neural network.

32. A method according to any one of Claims 29 to 31 wherein the processing module is programmed to, in use, assign a surface emissivity value to the or each identified material of the object.

33. A method according to any one of Claims 20 to 32 including the step of accounting for variation in pose of the thermal and / or visible light image acquisition devices when combining the visual and radiometric models by matching features of the geometric model with features of the radiometric model.

34. A method according to any one of Claims 20 to 33 including the step of accounting for temporal variation when combining the visual and radiometric models by matching features of the geometric model with features of the radiometric model.

35. A method according to any one of Claims 20 to 34 including the step of using the radiometric model to identify, predict or estimate a thermal transmissivity value of one or more components of the object.

36. A method according to any one of Claims 20 to 35 including the step of using the radiometric model to identify or classify at least one thermal anomaly of the object.

37. A method according to any one of Claims 20 to 36 wherein the method is for generating a heat map of a building.

38. A computer-implemented method of generating a heat map of an object, the method comprising the steps of: processing visible light image data acquired from the object to generate a geometric model of the object; processing thermal image data acquired from the object to generate a radiometric model of the object; combining the visual and radiometric models into an object model by matching features of the geometric model with features of the radiometric model.

39. A computer program comprising computer code configured to perform the method of Claim 38.