An infrared scene real-time ray tracing rendering method based on a GPU hardware ray tracing pipeline

Abstract

The application discloses an infrared scene real-time light tracing rendering method based on a GPU hardware light tracing pipeline and belongs to the field of infrared simulation of virtual scenes. The application is based on the path tracing principle and takes Unity as a platform to realize real-time rendering of infrared scenes. Based on the infrared physics principle, firstly, real-time rendering of the infrared scene is realized based on the path tracing; on the basis, the GPU hardware light tracing pipeline technology is applied to improve the speed of the infrared path tracing, and the importance sampling and the space-time filtering denoising algorithm are used to guarantee the image quality of the infrared scene, so that the real-time rendering of the infrared scene is realized, the realism and details of the infrared rendering result are greatly improved, and the similarity is higher compared with the actually shot infrared scene image.

Description

Technical Field

[0001] This invention relates to the field of infrared simulation of virtual scenes, and in particular to a real-time ray tracing rendering method for infrared scenes based on GPU hardware ray tracing pipeline. Background Technology

[0002] With the continuous development of infrared detection imaging technology, infrared detection imaging is gradually progressing towards higher resolution and greater detection range. Currently, advanced infrared detection instruments can achieve detection ranges of up to 100,000 meters, comparable to airborne radar. Simultaneously, the resolution of infrared imaging has gradually improved from 240P to 720P, and even 1080P. The imaging results of current infrared cameras have undergone a qualitative leap, and high-definition infrared images are now very common and routine.

[0003] Infrared imaging sensors are increasingly widely used in various applications, such as disaster relief, troubleshooting of factory systems, racing car performance analysis, building thermal analysis, leak analysis, and detecting people with fever in shopping malls and subways—all scenarios highly correlated with temperature. In the research and development of various infrared imaging, infrared guidance, and infrared stealth technologies, limitations such as location, funding, and unforeseen circumstances often prevent the full-weather, all-time infrared radiation characteristic measurement and analysis of a specific target scenario. Numerical simulation methods are not only cost-effective but also generate infrared radiation characteristic data for various environments and weather conditions, easily producing data for extreme situations. This allows for wide-area testing and analysis of infrared radiation characteristic suppression and infrared detection guidance technologies. However, scene infrared radiation characteristic analysis is a highly complex engineering project. The simulation inputs are complex; even just considering different detection solid angles can generate an extremely large dataset. Furthermore, to achieve infrared radiation characteristic detection results that closely approximate reality, various complex physical coupling processes need to be modeled and calculated, making real-time rendering of infrared scenes impossible.

[0004] Therefore, for the field of infrared simulation, it is of great significance to generate realistic infrared scene target images in real time. Summary of the Invention

[0005] To address the problems of existing technologies, this invention proposes a real-time ray tracing rendering method for infrared scenes based on GPU hardware ray tracing pipeline. Based on the principles of infrared physics, it first achieves non-real-time rendering of infrared scenes with realistic feel based on path tracing; on this basis, it utilizes GPU hardware ray tracing pipeline, importance sampling, and spatiotemporal filtering to accelerate the calculation speed, thereby realizing real-time rendering of infrared scenes.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A real-time ray tracing rendering method for infrared scenes based on GPU hardware ray tracing pipeline includes the following steps:

[0008] 1) Measure the infrared data of the target scene materials, including the infrared emissivity, infrared reflectivity and infrared absorptivity of each material in the target scene; and acquire the temperature field information, solar radiation, atmospheric background radiation and atmospheric transmission effect of the target scene.

[0009] 2) Combine the infrared data of the target scene material and the temperature field information of the target scene with the three-dimensional geometric model to build a virtual scene model;

[0010] 3) Based on the infrared data of the target scene material, as well as the temperature field information, solar radiation, atmospheric background radiation, and atmospheric transmission effect of the target scene, the Monte Carlo path tracing algorithm is used to calculate the infrared radiation data of the virtual scene. When calculating the infrared radiation data, the initial infrared radiation is jointly affected by solar radiation, atmospheric background radiation, and infrared radiation emitted by scene objects. The infrared radiation calculation process is accelerated by GPU hardware ray tracing pipeline and importance sampling. The infrared radiation emitted by the scene objects is jointly determined by temperature field information, infrared emissivity, infrared reflectivity, and infrared absorptivity.

[0011] 4) Perform spatiotemporal filtering and noise reduction on the infrared radiation data calculated by the Monte Carlo path tracing algorithm, and plot the final infrared scene simulation results.

[0012] Furthermore, in step 2), building the virtual scene model includes the following steps:

[0013] 2.1) Generate typical target material classification textures

[0014] Obtain the material of the items in the target scene, initialize the number of material categories to 2 to 4 times the actual number of categories, merge similar materials according to the clustering algorithm, and perform manual review to obtain the categories under the actual number of categories, and generate typical target material category textures;

[0015] 2.2) Generate scene infrared skybox texture

[0016] Acquire infrared irradiance data of the sky and generate scene infrared skybox texture;

[0017] 2.3) Calculation of typical target temperature field

[0018] Based on the 3D geometric model of the target scene object, weather data, and temperature material data, the temperature distribution is calculated using the GPU and computational shaders to generate a temperature field map.

[0019] Furthermore, the typical target temperature field calculation process includes:

[0020] a. Read the vertex data, UV data, and triangle facet data of the 3D geometric model of the target scene object;

[0021] b. Use vertex data and its UV data, and triangle patch data to perform soft rasterization to obtain the position, normal, and UV data of each pixel in the temperature field map;

[0022] c. Organize the data for each pixel, remove blank pixels, and calculate the direction of direct sunlight per minute for 24 hours based on latitude, longitude, time, and date; each pixel contains material information;

[0023] d. Input pixel set data, weather data (temperature, wind speed, thermal irradiance data of the sky hemisphere, etc.), temperature material data, 3D geometric model, and the direction of direct sunlight per minute for 24 hours into the GPU;

[0024] e. The GPU emits light rays in parallel towards the direction of direct sunlight at the current moment, based on pixel set data and 3D geometric model, and records whether each pixel is occluded;

[0025] f. The GPU uses a compute shader to perform parallel one-dimensional finite difference temperature distribution calculation and iteration based on pixel set data, weather data, temperature material data, and occlusion conditions;

[0026] g. After the calculation is completed, the GPU memory data is transferred to the main memory, and a temperature field map is generated according to the specified format.

[0027] Furthermore, in step 3), the ray tracing core in the graphics card is invoked to accelerate the operation of the Monte Carlo path tracing algorithm, and the sampling method in the Monte Carlo path tracing algorithm is replaced with the importance sampling method.

[0028] Furthermore, the aforementioned method of using the ray tracing core in the graphics card to accelerate the Monte Carlo path tracing algorithm specifically involves:

[0029] The emitted rays are detected by the computation shader;

[0030] The ray tracing core obtains and decodes the collision boxes generated based on the 3D geometric model, and performs intersection detection between the emitted ray and the collision box. If the emitted ray intersects with the collision box, it traverses all intersecting sub-collision boxes. For each intersecting sub-collision box, it performs an intersection test between the ray and the triangular facet of the sub-collision box, traverses all triangular facets, and returns the intersection point between the ray and the triangular facet.

[0031] The intersection region is colored by a computational color tracker.

[0032] Furthermore, in step 4), the spatiotemporal filtering noise reduction includes temporal filtering and spatial filtering. The result of the previous frame is reused during temporal filtering, and the result of the surrounding pixels is reused during spatial filtering.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] (1) Because the present invention introduces a path tracing algorithm, it overcomes the problem that the existing technology cannot simulate infrared radiation multiple reflections well, thus achieving more realistic simulation results.

[0035] (2) Because the present invention accelerates the path tracing algorithm through the GPU ray tracing pipeline, improves the sampling efficiency of Monte Carlo path tracing by using importance sampling, and then uses spatiotemporal filtering technology to denoise the infrared radiation image with a low sampling rate to ensure image quality, thereby realizing real-time rendering of infrared radiation data of virtual scenes. Attached Figure Description

[0036] Figure 1 The following are examples of noise caused by different sampling rates in embodiments of the present invention;

[0037] Figure 2 This is a schematic diagram illustrating the path tracing calculation of multiple ray ejections in an embodiment of the present invention;

[0038] Figure 3 This is a flowchart illustrating the soft ray tracing process according to an embodiment of the present invention;

[0039] Figure 4 This is a flowchart illustrating a hardware ray tracing pipeline according to an embodiment of the present invention;

[0040] Figure 5 This is a comparison chart of sampling results using different probability density functions (pdf) as shown in an embodiment of the present invention;

[0041] Figure 6 This is a comparison chart showing the impact of different sampling algorithms on the results in an embodiment of the present invention;

[0042] Figure 7 The impact of BRDF importance sampling and hemispherical importance sampling on image noise is illustrated in this embodiment of the invention.

[0043] Figure 8 The image shows a comparison between the infrared rendering of the test oil tank at noon and a real photograph, as illustrated in an embodiment of the present invention. The right image is the real photograph, and the left image is the simulation rendering.

[0044] Figure 9The image shown is a comparison between the infrared rendering of the test boss model at noon and the actual photograph, as illustrated in this embodiment of the invention. The right image is the actual photograph, and the left image is the simulation rendering. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings.

[0046] This invention proposes a real-time ray tracing rendering method for infrared scenes based on GPU hardware ray tracing pipeline, which mainly includes the following steps:

[0047] 1) Measurement of infrared data

[0048] This invention requires prior measurement of infrared data for various materials in the target scene. The infrared data includes the infrared emissivity, infrared reflectivity, and infrared absorptivity of each material, which can be obtained by measuring and estimating meteorological data, solar irradiance data, and surface temperature data of the target scene.

[0049] 2) Establishment of virtual scenes

[0050] The infrared simulation scenario of this invention needs to be pre-built. In this embodiment, a drone is used to conduct on-site oblique photogrammetry of the target scene, and a three-dimensional geometric model of its terrain is established based on the series of photos taken. The infrared emissivity, infrared reflectivity, infrared absorptivity, and temperature field information of various materials in the target scene are combined with the three-dimensional geometric model to build a virtual scene model.

[0051] In one specific embodiment of the present invention, the process of building a virtual scene model includes:

[0052] 2.1) Generate typical target material classification textures

[0053] Obtain the material properties of objects in the target scene. Initialize the number of material categories to be 2-4 times the actual number of categories. Merge similar materials according to a clustering algorithm, taking into account the visible light textures in oblique photogrammetry modeling. Figure 1 Generally, it contains lighting information. Lighting has a great influence on the actual color reflected by the object, which can easily cause clustering algorithms to misjudge. Therefore, manual review is carried out in the end to obtain the categories under the actual number of categories and generate typical target material classification textures.

[0054] 2.2) Generate scene infrared skybox texture

[0055] Infrared irradiance data of the sky is acquired, and an infrared skybox texture of the scene is generated. The method for acquiring the infrared irradiance data of the sky can be a direct measurement method or it can be acquired through third-party software. For example, in this embodiment, MODTRAN software is used to calculate the infrared sky irradiance data, and the infrared radiation of the sky dome in the infrared band can also be calculated by this software.

[0056] 2.3) Calculation of typical target temperature field

[0057] a. Read the vertex data, UV data, and triangle facet data of the 3D geometric model of the target scene object;

[0058] b. Use vertex data and its UV data, and triangle patch data to perform soft rasterization to obtain the position, normal, and UV data of each pixel in the temperature field map;

[0059] c. Organize the data for each pixel, remove blank information pixels, and calculate the direction of direct sunlight every minute for 24 hours based on latitude, longitude, time and date;

[0060] d. Input pixel set data, weather data, temperature and material data, 3D geometric model, and the direction of direct sunlight per minute for 24 hours into the GPU;

[0061] e. The GPU emits light rays in parallel towards the direction of direct sunlight at the current moment, based on pixel set data and 3D geometric model, and records whether each pixel is occluded;

[0062] f. The GPU uses a compute shader to perform parallel one-dimensional finite difference temperature distribution calculation and iteration based on pixel set data, weather data, temperature material data, and occlusion conditions;

[0063] g. After the calculation is completed, the GPU memory data is transferred to the main memory, and a temperature field map is generated according to the specified format.

[0064] 3) Solar radiation and atmospheric background radiation

[0065] This data can be obtained through direct measurement or simulation using third-party software. In this embodiment, the relatively mature third-party software Modtran is used to calculate solar radiation and atmospheric background radiation. The Modtran program is invoked by starting a new process. The main process and the Modtran process communicate via files. The parameters required for the Modtran program to perform calculations can be loaded by reading files, and the calculation results are stored in another file. The main process reads the files by polling, obtains the data, and then continues to perform subsequent calculations.

[0066] Since the methods of obtaining solar radiation and atmospheric background radiation through third-party software are well known in the field, they will not be elaborated here.

[0067] After obtaining solar radiation, the solar mid-wave radiation is treated as parallel light, and the initial thermal radiation of shadows and scene objects is calculated for subsequent calculations of scene infrared radiation. The influence of thermal radiation is considered in all calculations of scene infrared radiation. The initial infrared thermal radiation is jointly affected by solar radiation, atmospheric background radiation, and the thermal radiation of various objects in the scene. Atmospheric background radiation is considered as ambient light and added to the initial thermal radiation.

[0068] 4) Atmospheric transport effect

[0069] Based on the calculated solar radiation and atmospheric background radiation data, the radiance of each object surface in the scene can be obtained. However, the image actually seen in the simulation is the radiation that enters the camera from the object surface. Therefore, it is necessary to obtain atmospheric transmission effects, including parameters such as atmospheric transmittance and path radiation.

[0070] In this embodiment, based on the date and time input by the user, the measured scene environment parameters at that time are input into the Modtran software to obtain parameters such as atmospheric transmittance and path radiation under different bands and different meteorological conditions. Then, in the actual calculation, by performing difference calculation on the pre-calculated values, the data required by the present invention can be obtained quickly and relatively accurately.

[0071] 5) Calculate the scene's infrared radiation

[0072] Based on the various data obtained from the above steps, infrared radiation information for a specific time and scene is rendered using the Monte Carlo path tracing algorithm. In this process, the Monte Carlo path tracing algorithm is accelerated by means of GPU hardware ray tracing pipeline, importance sampling, spatiotemporal filtering and other methods.

[0073] In this embodiment, the calculation of scene infrared radiation includes the following steps:

[0074] 5.1) Monte Carlo Path Tracing

[0075] Path tracing is a ray transport algorithm derived from the path integral form of the light transport equation. It uses an inverse Monte Carlo method to create realistic radiating images of 3D scenes. When combined with physically based materials, lighting, and cameras, it can create incredibly lifelike static images. For example... Figure 1 As shown, two infrared radiation images were drawn using path tracing at different sampling rates. The images clearly demonstrate that the path tracing algorithm produces highly noisy final images with a low number of samples per pixel. Only by using a larger number of samples can a relatively high-quality image be obtained. Because the path tracing algorithm is relatively inefficient and requires a large number of samples to achieve the desired effect, it is primarily used for non-real-time rendering.

[0076] Figure 2 This demonstrates the path of light rays continuously emitted from the camera and eventually sampled at the light source. The path tracing algorithm emits countless light rays from the scene's camera position towards the actual shooting direction, calculates the radiance of each pixel in the path tracing algorithm's graph through continuous sampling of light rays, and saves the final radiance to the corresponding camera screen pixel position.

[0077] Due to limitations in GPU hardware, the function stack for each thread on a GPU is generally very small, which cannot support functions within a thread continuously recursively entering a deep recursion depth. A common practice is to manually define the maximum recursion depth of a function. However, since algorithms such as ray generation / emission and recursive shading in Monte Carlo path tracing are recursive, the recursion depth limits the number of ray bounces, thus requiring a manual limit on the depth.

[0078] 5.2) Hardware Ray Tracing Pipeline

[0079] Graphics card hardware manufacturers typically implement the most computationally intensive part of ray tracing—the calculation of the intersection between rays and the scene—using dedicated hardware ray tracing cores (RT Cores). Before hardware acceleration for ray tracing, software ray tracing was used. Simply put, this method uses the graphics card's general-purpose computing units to parallelize the software simulation of the intersection process between rays and the scene. The calculation process is as follows: Figure 3 As shown. In this calculation process, the intersection test of each ray and the scene requires thousands of hardware instructions.

[0080] This invention employs GPU hardware ray tracing pipeline technology, using ray tracing hardware acceleration. It improves speed by utilizing the RT Cores (ray tracing cores) within the graphics card. The computational process is as follows: Figure 4 As shown:

[0081] First, the emitted rays are detected by the computation shader;

[0082] Secondly, the ray tracing core acquires and decodes the collision boxes generated based on the 3D geometric model, and performs intersection detection between the emitted ray and the collision box. If the emitted ray intersects with the collision box, it traverses all intersecting sub-collision boxes. For each intersecting sub-collision box, it performs an intersection test between the ray and the triangular facet of the sub-collision box, traverses all triangular facets, and returns the intersection point between the ray and the triangular facet.

[0083] Finally, the intersection region is colored by a colorimeter.

[0084] Next, we will compare the GTX 1080Ti and RTX 2080Ti to see the difference in the number of rays emitted in the same amount of time between using and not using hardware ray tracing cores.

[0085] Table 1 compares graphics cards with and without hardware ray tracing.

[0086]

[0087] As shown in Table 1, for graphics cards of the same level, the RTX graphics card using ray tracing cores (68 RT Cores) emits rays almost 10 times faster than the GTX graphics card with general computing cores (11.3 TFLOPS).

[0088] 5.3) Importance Sampling

[0089] Importance sampling is a sampling method that differs from uniform sampling. It replaces the uniform probability density function with a probability density function that conforms to a functional distribution, so that more samples are sampled in the region that contributes more to the integral, thereby achieving higher sampling efficiency and convergence speed. Figure 5 The sampling results are shown under different probability density functions (pdfs). The sampling effect of a uniform probability density function is generally poor, while the sampling effect of a probability density function that conforms to the function distribution is good, and the sampling effect of a probability density function that does not conform to the function distribution is poor.

[0090] Compare the noise levels of images drawn by the Monte Carlo path tracing algorithm under the same time consumption and number of samples. For example, Figure 6 This illustrates the difference in effectiveness between importance sampling (right) and uniform sampling (left) for cosine weights. Figure 7 This illustrates the difference in effectiveness between BRDF importance sampling (right) and uniform sampling (left). Observation Figure 6 and Figure 7 It can be observed that the importance sampling method significantly reduces noise in the final drawn image under the same time consumption and the same number of samples, presenting a clearer image with less noise. This effectively improves the effective utilization rate of samples in the Monte Carlo path tracing algorithm and reduces variance.

[0091] 5.4) Spatiotemporal filtering and noise reduction

[0092] Spatial filtering refers to spatial noise reduction of an image. This embodiment uses a joint bilateral filter kernel J, with the input being an unfiltered (noisy) image. The denoised image is obtained by using a joint bilateral filter kernel. Represented as:

[0093]

[0094] in, For the unfiltered pixels in the k-th frame, Let represent the spatially filtered pixels of the k-th frame, where Filter[.] denotes the spatial filter.

[0095] The joint bilateral filter kernel used in this embodiment is defined as follows:

[0096]

[0097] Where J(.) represents the joint bilateral filter kernel, and i,j represent the positions of two pixels. σ represents the radiance value corresponding to the unfiltered pixel at position i, j. p σ c σ n σ d D is an adjustable weighting parameter. normal (ij) represents the angle between the normals of two pixels, D plane (i,j) represents the difference in depth between the faces of two pixels, and ||.| represents the absolute value.

[0098] Temporal filtering reduces noise by reusing previously calculated samples; in other words, it increases the number of actual samples. However, the camera and objects in the scene are constantly moving, so a way to find the position of a pixel in the previous frame is needed. In this embodiment, motion vectors are used to find the pixel position in the current frame corresponding to the pixel position in the previous frame. The motion vectors are mapped from the previous frame position to the current frame position by recording the MVP matrix of the previous frame and its inverse. This is a mature technology in the field, and the specific implementation process will not be described in detail here.

[0099] Assume the temporal filtering process for the pixels in this frame (the i-th frame) is as shown in the following formula:

[0100]

[0101] Where a represents the sampling ratio, C (k) This represents the final pixel of the i-th frame.

[0102] 6) Plot the simulation results

[0103] Based on the filtered infrared radiation data of the scene, rasterization is used to render the scene, resulting in a realistic 3D infrared radiation image. When combined with physically based materials, lighting, and camera, incredibly lifelike static images can be created.

[0104] In this embodiment, taking the oil tank and boss models in the scene as examples, the local simulation effect of the scene is as follows: Figure 8 , Figure 9 As shown. Among them, Figure 8The image shows a comparison between the infrared rendering of the test oil tank at noon and a real photograph, as illustrated in an embodiment of the present invention. The right image is the real photograph, and the left image is the simulation rendering. Figure 9 The image shown is a comparison between the infrared rendering of the test boss model at noon and the actual photograph, as illustrated in this embodiment of the invention. The right image is the actual photograph, and the left image is the simulation rendering.

[0105] The similarity between the simulation results of this invention and the actual photographs can be verified using SSIM and MSE. The following section compares the real-time and non-real-time rendering parts of the simulation results and analyzes the time consumption of real-time rendering. The similarity comparison is shown in Table 2, and the time consumption analysis for non-real-time rendering and real-time rendering is shown in Table 3.

[0106] Table 2 Similarity Results

[0107]

[0108] Table 3. Time Consumption Analysis Results

[0109]

[0110] The above data demonstrates the superiority of this invention.

[0111] The above examples are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments and many variations are possible. All variations that can be directly derived or conceived by those skilled in the art from the disclosure of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for real-time ray tracing rendering of infrared scenes based on GPU hardware ray tracing pipeline, characterized in that, Includes the following steps: 1) Measure the infrared data of the target scene materials, including the infrared emissivity, infrared reflectivity and infrared absorptivity of each material in the target scene; and acquire the temperature field information, solar radiation, atmospheric background radiation and atmospheric transmission effect of the target scene. 2) Combine the infrared data of the target scene material and the temperature field information of the target scene with the three-dimensional geometric model to build a virtual scene model; 3) Based on the infrared data of the target scene material, as well as the temperature field information, solar radiation, atmospheric background radiation, and atmospheric transmission effect of the target scene, the Monte Carlo path tracing algorithm is used to calculate the infrared radiation data of the virtual scene. When calculating the infrared radiation data, the initial infrared radiation is jointly affected by solar radiation, atmospheric background radiation, and infrared radiation emitted by scene objects. The infrared radiation calculation process is accelerated by GPU hardware ray tracing pipeline and importance sampling. 4) Perform spatiotemporal filtering and noise reduction on the infrared radiation data calculated by the Monte Carlo path tracing algorithm, and plot the final infrared scene simulation results.

2. The method for real-time ray tracing rendering of infrared scenes based on GPU hardware ray tracing pipeline as described in claim 1, characterized in that, Step 2) involves building the virtual scene model, which includes the following steps: 2.1) Generate typical target material classification textures Obtain the material of the items in the target scene, initialize the number of material categories to 2 to 4 times the actual number of categories, merge similar materials according to the clustering algorithm, and perform manual review to obtain the categories under the actual number of categories, and generate typical target material category textures; 2.2) Generate scene infrared skybox texture Acquire infrared irradiance data of the sky and generate scene infrared skybox texture; 2.3) Calculation of typical target temperature field Based on the 3D geometric model of the target scene object, weather data, and temperature material data, the temperature distribution is calculated using the GPU and computational shaders to generate a temperature field map.

3. The method for real-time ray tracing rendering of infrared scenes based on GPU hardware ray tracing pipeline as described in claim 2, characterized in that, The typical target temperature field calculation process includes: a. Read the vertex data, UV data, and triangle facet data of the 3D geometric model of the target scene object; b. Use vertex data and its UV data, and triangle patch data to perform soft rasterization to obtain the position, normal, and UV data of each pixel in the temperature field map; c. Organize the data for each pixel, remove blank information pixels, and calculate the direction of direct sunlight every minute for 24 hours based on latitude, longitude, time and date; d. Input pixel set data, weather data, temperature and material data, 3D geometric model, and the direction of direct sunlight per minute for 24 hours into the GPU; e. The GPU emits light rays in parallel towards the direction of direct sunlight at the current moment, based on pixel set data and 3D geometric model, and records whether each pixel is occluded; f. The GPU uses a compute shader to perform parallel one-dimensional finite difference temperature distribution calculation and iteration based on pixel set data, weather data, temperature material data, and occlusion conditions; g. After the calculation is completed, the GPU memory data is transferred to the main memory, and a temperature field map is generated according to the specified format.

4. The method for real-time ray tracing rendering of infrared scenes based on GPU hardware ray tracing pipeline as described in claim 1, characterized in that, In step 3), the ray tracing core in the graphics card is called to accelerate the operation of the Monte Carlo path tracing algorithm, and the sampling method in the Monte Carlo path tracing algorithm is replaced with the importance sampling method.

5. The method for real-time ray tracing rendering of infrared scenes based on GPU hardware ray tracing pipeline as described in claim 4, characterized in that, The aforementioned method of using the ray tracing core in the graphics card to accelerate the Monte Carlo path tracing algorithm specifically involves: The emitted rays are detected by the computation shader; The ray tracing core obtains and decodes the collision boxes generated based on the 3D geometric model, and performs intersection detection between the emitted ray and the collision box. If the emitted ray intersects with the collision box, it traverses all intersecting sub-collision boxes. For each intersecting sub-collision box, it performs an intersection test between the ray and the triangular facet of the sub-collision box, traverses all triangular facets, and returns the intersection point between the ray and the triangular facet. The intersection region is colored by a computational color tracker.

6. The method for real-time ray tracing rendering of infrared scenes based on GPU hardware ray tracing pipeline as described in claim 1, characterized in that, In step 4), the spatiotemporal filtering noise reduction includes temporal filtering and spatial filtering. The result of the previous frame is reused during temporal filtering, and the result of the surrounding pixels is reused during spatial filtering.

Images