A real-time rendering optimization method based on ray tracing

Abstract

The application discloses a real-time rendering optimization method based on ray tracing. The application combines the optimization of spatial data structure, ray beam tracing technology, adaptive sampling, denoising technology, light map precalculation, shader optimization, ray-scene intersection acceleration, light source clipping, importance sampling, hardware acceleration, software-hardware cooperation, machine learning prediction, hybrid rendering technology, asynchronous calculation and dynamic resolution adjustment and other technologies; thereby improving the calculation efficiency of ray tracing and reducing the demand for hardware performance, so that the real-time or near real-time rendering speed is realized without sacrificing the image quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rendering technology, and particularly relates to a real-time rendering optimization method based on ray tracing. BACKGROUND

[0002] With the improvement of computing power and the development of graphics processing technology, ray tracing has become an important means of rendering high-quality images. However, due to the complexity of the ray tracing algorithm, its application in the field of real-time rendering has been limited. Existing real-time rendering techniques usually rely on rasterization methods and cannot fully utilize the highly realistic visual effects brought by ray tracing.

[0003] Therefore, the prior art has defects and needs to be improved. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a real-time rendering optimization method based on ray tracing, to improve the computing efficiency of ray tracing and reduce its demand for hardware performance, so as to realize real-time or near real-time rendering speed without sacrificing image quality.

[0005] The technical solution of the present application is as follows: a real-time rendering optimization method based on ray tracing is provided, comprising the following steps.

[0006] S1: Spatial data structure optimization; construct an efficient spatial data structure such as a bounding volume hierarchy (BVH) or a kd-tree to quickly determine the intersection of a ray and an object and reduce unnecessary calculations.

[0007] S2: Ray bundle tracing; group rays into bundles and use the SIMD parallelism of GPU for batch processing to improve computing efficiency.

[0008] S3: Adaptive sampling; dynamically adjust the sampling rate according to the importance of the image area, and apply a higher sampling rate to important areas such as edges and high-contrast areas.

[0009] S4: Denoising; apply a denoising algorithm such as Temporal Anti-Aliasing (TAA) to reduce noise in real-time rendering.

[0010] S5: Light map pre-computation; pre-compute light maps when loading the scene and store lighting information for use in real-time rendering.

[0011] S6: Shader optimization; optimize the algorithm of the shader, reduce branching and division operations, and improve the running efficiency of the shader.

[0012] S7: Ray-scene intersection acceleration; use conservative rasterization and other techniques to quickly determine potential intersection geometries and reduce the amount of ray tracing calculations.

[0013] S8: Light source pruning; analyze the visibility of light sources in the scene, ignore occluded light sources, and reduce unnecessary ray casting.

[0014] S9: Importance sampling; perform importance analysis on light sources, prioritizing light sources that contribute significantly to rendering.

[0015] S10: Hardware acceleration; use dedicated ray tracing hardware, such as NVIDIA RTX GPU's RT Core, to accelerate ray tracing calculations. RT Core is a hardware unit designed specifically for accelerating ray tracing calculations. It speeds up the calculation of tree-based ray tracing structures, such as the bounding volume hierarchy (BVH), by quickly determining the intersection of rays with 3D geometry in the scene. Compared to traditional methods that rely on general-purpose processing units for software simulation, RT Core can significantly improve the performance of ray tracing, enabling real-time or near-real-time rendering speeds. In this step, the dedicated ray tracing hardware is any one of the NVIDIA RTX series of graphics cards, AMD RDNA series of graphics cards, Intel A770 graphics card, or Imagination Technologies' IMG CXT GPU.

[0016] S11: Software-hardware collaboration; reasonably distribute tasks between CPU and GPU, use CPU for complex calculations and data management, and focus GPU on ray tracing calculations.

[0017] S12: Machine learning prediction; use machine learning algorithms to predict the interaction of rays with the scene, optimizing rendering parameters.

[0018] S13: Hybrid rendering technology; combine ray tracing and rasterization rendering techniques to handle different visual effects.

[0019] S14: Asynchronous computation; perform ray tracing calculations in background threads to reduce the impact on the main thread.

[0020] S15: Dynamic resolution and quality adjustment; dynamically adjust rendering resolution and quality based on scene dynamics and camera movement.

[0021] Further, in step S1, the method for optimizing the data structure includes at least one of the following:

[0022] Pre-traversal optimization: Before traversing the tree, some information can be pre-computed, such as pre-computing some possible intersection points, to reduce the amount of calculation during actual traversal.

[0023] Multi-level BVH: Use multi-level BVH, where each level has different-sized bounding volumes, which can better handle objects of different sizes and shapes.

[0024] Adaptive Construction: Depending on the characteristics of the scene and the distribution of objects, adaptively choose to construct BVH or kd-tree, or mix both in different regions.

[0025] Parallel Construction: Utilize multi-core processors to construct spatial data structures in parallel to reduce preprocessing time.

[0026] Dynamic Update: For dynamic scenes, use spatial data structures that can be quickly updated, or periodically rebuild to adapt to scene changes.

[0027] Further, in step S2, the batch processing scheme includes at least one of the following:

[0028] SIMD Parallelism: Graphics processing units have a large number of SIMD units, each of which can process multiple data under one instruction; in ray beam tracing, all rays in the beam can be processed simultaneously in the same SIMD unit.

[0029] Intersection Calculation: Calculate the intersection of all rays in the beam simultaneously, which can share many intermediate results such as traversing the bounding volume hierarchy tree.

[0030] Further, in step S3, the adaptive sampling implementation method includes at least one of the following:

[0031] Layered Sampling: In the rendering process, use a multi-layer sampling strategy; first use a lower base sampling rate for preliminary rendering, then according to the results of preliminary rendering, sample important areas at a higher level.

[0032] Iterative Convergence: Adaptive sampling can be part of an iterative process, after each iteration, re-evaluate the importance of the region according to the current rendering result, and adjust the sampling rate accordingly.

[0033] Further, in step S4, Temporal Anti-Aliasing (TAA) algorithm is used to realize denoising, including at least one of the following:

[0034] Camera Shaking: By introducing a small random offset to the camera position or projection matrix in each frame, Temporal Anti-Aliasing (TAA) can capture more sub-pixel information; this shaking is particularly effective in static scenes, significantly improving the sampling rate.

[0035] Reprojection: For dynamic scenes, Temporal Anti-Aliasing (TAA) uses reprojection technology, i.e. according to the motion of objects and cameras, the pixels of the current frame are inversely mapped to the position of the previous frame to obtain historical sampling information.

[0036] Motion Vector: To accurately correspond the current pixel with the previous frame's pixel, Temporal Anti-Aliasing (TAA) uses motion vector to compensate the movement of objects and camera, ensuring the correct superposition of temporal information.

[0037] Further, in step S5, the light map pre-computation includes at least one of the following:

[0038] Multi-threaded computation: Utilize multi-core CPU to parallelize the light map computation, shortening the pre-computation time.

[0039] On-demand loading: Dynamically load and unload light maps based on the current view angle and object visibility, reducing memory occupation.

[0040] Further, in step S6, the optimization of the shader's algorithm includes at least one of the following:

[0041] Reducing branching: (1) Avoid using if-else statements, as they can cause GPU execution inefficiency, especially in dynamic branching cases. (2) Avoid branching by pre-computing or redesigning the algorithm, or use compile-time constants as branching conditions. (3) Use mathematical functions and interpolation functions instead of conditional branching; for example, use step functions and mix functions to replace if-else statements, which can convert branching logic into mathematical operations and reduce branching operations.

[0042] Reducing division operations: Replace real-time computation with pre-computation and lookup tables; for example, for complex calculations such as trigonometric functions, pre-compute and store them in a lookup table, and directly access the results through indexing, avoiding the overhead of real-time computation.

[0043] Further, in step S12, the specific steps are: S121: Application of deep learning models; use deep learning models such as convolutional neural networks or recurrent neural networks to analyze object and material properties in the scene, predict lighting effects and reflection / refraction behavior; S122: Preparation of training data; to train these models, a large amount of labeled data is needed, including light interaction effects in different scenes and corresponding rendering parameter settings; these data can be generated by existing ray tracing algorithms.

[0044] Further, in step S13, the hybrid rendering technique includes at least one of the following:

[0045] Direct lighting processing: Use rasterization rendering to process direct lighting, use shaders to calculate the lighting intensity and color of object surfaces; this method is fast and suitable for real-time rendering requirements.

[0046] Indirect lighting and global illumination: indirect lighting such as reflection and refraction, and global illumination effects are calculated using ray tracing techniques; this can improve the realism of scene lighting while maintaining computational efficiency.

[0047] With the above scheme, the present application provides a real-time rendering optimization method based on ray tracing, which has the following technical effects:

[0048] (1) Improve picture quality: through the optimization of ray tracing algorithm and the application of hardware acceleration, more realistic lighting effects can be realized, and higher quality pictures can be presented in real-time rendering.

[0049] (2) Improve rendering speed: through the optimization of rendering pipeline and computing resources, and the strategy of adaptive sampling, the rendering speed is improved. Under the premise of not losing picture quality, high-quality rendering results can be quickly generated.

[0050] (3) Enhance the practicality of real-time rendering application: due to the improvement of rendering effect and rendering speed, the present scheme can be widely applied in virtual reality, games and film production fields, and the practicality and user experience of real-time rendering application are improved. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0052] The present application will be described in detail below in combination with the drawings and specific embodiments.

[0053] Please refer to Figure 1 , the present application provides a real-time rendering optimization method based on ray tracing, which includes the following steps.

[0054] S1: Spatial data structure optimization; construct an efficient spatial data structure such as BVH (Bounding Volume Hierarchy) or kd-tree to quickly determine the intersection of rays and objects, and reduce unnecessary calculations.

[0055] S2: Ray bundle tracing; group rays into bundles and use the SIMD parallelism of GPU for batch processing to improve computational efficiency.

[0056] S3: Adaptive sampling; dynamically adjust the sampling rate according to the importance of image regions, and apply higher sampling rate to important regions such as edges and high-contrast regions.

[0057] S4: Denoising; apply denoising algorithms such as TAA (Temporal Anti-Aliasing) to reduce noise in real-time rendering.

[0058] S5: Lightmap Precomputation; Precompute lightmaps during scene loading and store lighting information for real-time rendering.

[0059] S6: Shader Optimization; Optimize shader algorithms, reduce branching and division operations, and improve shader efficiency.

[0060] S7: Ray-Scene Intersection Acceleration; Use techniques such as conservative rasterization to quickly determine potential intersection geometries, reducing the computational load of ray tracing.

[0061] S8: Light Source Clipping; Analyze the visibility of light sources in the scene, ignore occluded light sources, and reduce unnecessary ray casting.

[0062] S9: Importance Sampling; Perform importance analysis on light sources, prioritizing light sources that contribute significantly to rendering.

[0063] S10: Hardware Acceleration; Utilize dedicated ray tracing hardware, such as NVIDIA RTX GPU's RT Core, to accelerate ray tracing calculations. The RT Core is a hardware unit designed specifically for accelerating ray tracing calculations. It speeds up the computation of tree-based ray tracing structures, such as the bounding volume hierarchy (BVH), by quickly determining the intersection of rays with 3D geometry in the scene. Compared to traditional methods that rely on general-purpose processing units for software simulation, the RT Core can significantly improve the performance of ray tracing, enabling real-time or near-real-time rendering speeds. In this step, the dedicated ray tracing hardware is any of the NVIDIA RTX series of graphics cards, AMD RDNA series of graphics cards, Intel A770 graphics card, or Imagination Technologies' IMG CXT GPU.

[0064] S11: Software-Hardware Collaboration; Reasonably distribute tasks between CPU and GPU, using CPU for complex calculations and data management, and GPU for ray tracing calculations.

[0065] S12: Machine Learning Prediction; Use machine learning algorithms to predict the interaction of rays with the scene, optimizing rendering parameters.

[0066] S13: Hybrid Rendering Technique; Combine ray tracing and rasterization rendering techniques to handle different visual effects.

[0067] S14: Asynchronous Computing; Perform ray tracing calculations in background threads to reduce the impact on the main thread.

[0068] S15: Dynamic Resolution and Quality Adjustment; Dynamically adjust rendering resolution and quality based on scene dynamics and camera movement.

[0069] In this embodiment, in step S1, the method for optimizing the data structure includes at least one of the following:

[0070] Pre-traversal optimization: Before traversing the tree, some information can be pre-computed, such as pre-computing some possible intersection points, to reduce the amount of calculation during actual traversal.

[0071] Multi-level BVH: Using multi-level BVH, where each level has different size bounding volumes, can better handle objects of different sizes and shapes.

[0072] Adaptive construction: According to the characteristics of the scene and the distribution of objects, adaptively choose to construct BVH or kd-tree, or mix both in different areas.

[0073] Parallel construction: Use multi-core processors to construct spatial data structures in parallel to reduce preprocessing time.

[0074] Dynamic update: For dynamic scenes, use spatial data structures that can be quickly updated, or periodically rebuild to adapt to scene changes.

[0075] In this embodiment, in step S2, the batch processing scheme includes at least one of the following:

[0076] SIMD parallelism: Graphics processing units have a large number of SIMD units, each of which can process multiple data under one instruction; in ray beam tracing, all rays in the beam can be processed simultaneously in the same SIMD unit.

[0077] Intersection calculation: All rays in the beam are calculated simultaneously, and many intermediate results can be shared, such as traversing the bounding volume hierarchy tree.

[0078] In this embodiment, in step S3, the implementation method of adaptive sampling includes at least one of the following:

[0079] Layered sampling: In the rendering process, use a multi-layer sampling strategy; first use a lower base sampling rate for preliminary rendering, then according to the results of preliminary rendering, sample important areas at a higher level.

[0080] Iterative convergence: Adaptive sampling can be part of an iterative process, after each iteration, re-evaluate the importance of the region according to the current rendering result, and adjust the sampling rate accordingly.

[0081] In this embodiment, in step S4, Temporal Anti-Aliasing (TAA) algorithm is used to realize denoising, which includes at least one of the following:

[0082] Camera Shaking: Temporal Anti-Aliasing (TAA) can capture more sub-pixel information by introducing a small random offset to the camera position or projection matrix at each frame; this shaking is particularly effective in static scenes, significantly increasing the sampling rate.

[0083] Reprojection: For dynamic scenes, Temporal Anti-Aliasing (TAA) uses reprojection technology, i.e., according to the motion of objects and cameras, the pixels of the current frame are inversely mapped to the positions of the previous frame to obtain historical sampling information.

[0084] Motion Vector: To accurately correspond the current pixel with the pixel of the previous frame, Temporal Anti-Aliasing (TAA) uses motion vectors to compensate for the motion of objects and cameras, ensuring correct superposition of temporal information.

[0085] In this embodiment, in step S5, the light map pre-computation includes at least one of the following:

[0086] Multi-threaded Computing: Utilize multi-core CPU parallel computing light map, shorten the pre-computation time.

[0087] On-demand Loading: According to the current view angle and the visibility of objects, dynamically load and unload the light map, reduce the memory occupation.

[0088] In this embodiment, in step S6, the algorithm of the optimized shader includes at least one of the following:

[0089] Reducing Branches: (1) Avoid using if-else statements, as they will cause GPU execution efficiency to be low, especially in dynamic branching cases. (2) Avoid branching by pre-computing or redesigning the algorithm, or use compile-time constants as branching conditions. (3) Use mathematical functions and interpolation functions instead of conditional branching; for example, use step functions and mix functions instead of if-else statements, which can convert branching logic into mathematical operations and reduce branching operations.

[0090] Reducing Division Operations: Replace real-time calculation with pre-computation and lookup tables; for example, for complex calculations such as trigonometric functions, pre-compute and store in a lookup table, and directly obtain the result by indexing, avoiding the overhead of real-time calculation.

[0091] In this embodiment, in step S12, the specific steps are: S121: application of deep learning model; use a deep learning model such as a convolutional neural network or a recurrent neural network to analyze the object and material properties in the scene, predict the lighting effects and reflection / refraction behavior; S122: preparation of training data; in order to train these models, a large amount of labeled data is needed, including light interaction effects and corresponding rendering parameter settings in different scenes; these data can be generated by existing ray tracing algorithms.

[0092] In this embodiment, in step S13, the hybrid rendering technique includes at least one of the following:

[0093] Direct lighting processing: use rasterization rendering to process direct lighting, use shaders to calculate the lighting intensity and color of the object surface; this method is fast and suitable for real-time rendering requirements.

[0094] Indirect lighting and global lighting: use ray tracing technology to calculate indirect lighting such as reflection and refraction, and global lighting effects; this can improve the lighting realism of the scene while maintaining computational efficiency.

[0095] In summary, the present application provides a real-time rendering optimization method based on ray tracing, which has the following technical effects:

[0096] (1) Improve picture quality: through the optimization of the ray tracing algorithm and the application of hardware acceleration, more realistic lighting effects can be achieved, and higher quality pictures can be presented in real-time rendering.

[0097] (2) Improve rendering speed: through the optimization of the rendering pipeline and computing resources, and the adaptive sampling strategy, the rendering speed is improved. Under the premise of not losing picture quality, high-quality rendering results can be quickly generated.

[0098] (3) Enhance the practicality of real-time rendering applications: due to the improvement of rendering effect and rendering speed, the present application can be widely used in virtual reality, games and film production fields, and can improve the practicality and user experience of real-time rendering applications.

[0099] The above is only a preferred embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for optimizing real-time rendering based on ray tracing, characterized in that, The method comprises the following steps: S1: Spatial data structure optimization; build efficient spatial data structures to quickly determine the intersection of rays and objects; S2: Ray bundle tracing; group rays into bundles and use the SIMD parallelism of the GPU for batch processing; S3: Adaptive sampling; dynamically adjust the sampling rate according to the importance of the image area, and apply higher sampling rate to important areas; S4: Denoising; apply denoising algorithms to reduce noise in real-time rendering; S5: Light map pre-computation; pre-compute light maps when loading the scene, store lighting information for use in real-time rendering; S6: Shader optimization: optimize the algorithm of the shader, reduce branch and division operations, and improve the efficiency of the shader; S7: Ray-scene intersection acceleration; Quickly determine potential intersection geometry to reduce ray tracing calculations; S8: Light source clipping; analyze the visibility of light sources in the scene, ignore occluded light sources, and reduce unnecessary ray casting; S9: Importance sampling; Importance analysis of light sources, preferential processing of light sources with large contribution to rendering; S10: Hardware acceleration; use dedicated ray tracing hardware to accelerate ray tracing calculations; S11: Software-hardware collaboration; reasonably distribute tasks between CPU and GPU, use CPU for complex calculations and data management, and GPU for ray tracing calculations; S12: Machine learning prediction; Use machine learning algorithms to predict the interaction of rays and scenes, and optimize rendering parameters; S13: Hybrid rendering technology; combine ray tracing and rasterization rendering techniques to handle different visual effects; S14: Asynchronous computation; perform ray tracing calculations in background threads to reduce the impact on the main thread; S15: Dynamic resolution and quality adjustment; dynamically adjust rendering resolution and quality based on scene dynamics and camera movement.

2. The method of claim 1, wherein, In step S1, the method for optimizing the data structure includes at least one of the following: Pre-traversal optimization: pre-compute some possible intersection points before traversing the tree to reduce the amount of calculation during actual traversal; Multi-level BVH: use multi-level BVH, where each level has different size bounding volumes, which can better handle objects of different sizes and shapes; Adaptive construction: adaptively select to construct BVH or kd-tree according to the characteristics of the scene and the distribution of objects, or mix both in different areas; Parallel construction: use multi-core processors to construct spatial data structures in parallel to reduce preprocessing time; Dynamic update: for dynamic scenes, use spatial data structures that can be quickly updated, or periodically rebuild to adapt to scene changes.

3. The method of claim 1, wherein, In step S2, the batch processing scheme includes at least one of the following: SIMD parallelism: the graphics processing unit has a large number of SIMD units, each of which can process multiple data under one instruction; Intersection calculation: all rays in the ray bundle are calculated simultaneously, and many intermediate results can be shared.

4. The method of claim 1, wherein, In step S3, the implementation method of adaptive sampling includes at least one of the following: Hierarchical Sampling: During rendering, a multi-layer sampling strategy is adopted; first, a preliminary rendering is performed using a lower base sampling rate, then according to the results of the preliminary rendering, higher-level sampling is performed on important areas; Iterative Convergence: Adaptive sampling is part of the iterative process, after each iteration, the region importance is re-evaluated according to the current rendering result, and the sampling rate is adjusted accordingly.

5. The method of claim 1, wherein, In step S4, the Temporal Anti-Aliasing algorithm is used to achieve denoising, which includes at least one of the following: Camera Shaking: By introducing a small random offset to the camera position or projection matrix in each frame, Temporal Anti-Aliasing can capture more sub-pixel information; Reprojection: For dynamic scenes, Temporal Anti-Aliasing uses reprojection technology, that is, according to the motion of objects and cameras, the pixels of the current frame are inversely mapped to the positions of the previous frame to obtain historical sampling information; Motion Vector: In order to accurately correspond the current pixel with the pixel of the previous frame, Temporal Anti-Aliasing uses motion vector to compensate for the motion of objects and cameras, ensuring correct superposition of temporal information.

6. The method of claim 1, wherein, In step S5, light map pre-computation includes at least one of the following: Multi-threaded computation: use multi-core CPU to parallelize light map computation, shorten pre-computation time; On-demand loading: dynamically load and unload light maps according to the current view angle and object visibility, reduce memory occupancy.

7. The method of claim 1, wherein, In step S6, the algorithm for optimizing the shader includes at least one of the following: Reduce branching: Avoid using if-else statements; Avoid branching by pre-computing or redesigning the algorithm, or use compile-time constants as branching conditions; Use mathematical functions and interpolation functions instead of conditional branching; Reduce division operations: replace real-time calculation with pre-computation and lookup tables.

8. The method of claim 1, wherein, In step S12, the specific steps are: S121: Application of deep learning models; use convolutional neural network or recurrent neural network deep learning models to analyze object and material properties in the scene, predict lighting effects and reflection / refraction behavior; S122: Preparation of training data; in order to train these models, a large amount of labeled data is required, including light interaction effects in different scenes and corresponding rendering parameter settings.

9. The method of claim 1, wherein, In step S13, the hybrid rendering technique includes at least one of the following: Direct lighting processing: use rasterization rendering to process direct lighting, use shaders to calculate the lighting intensity and color of object surfaces; Indirect lighting and global lighting: use ray tracing technology to calculate indirect lighting and global lighting effects.

10. The method of claim 1, wherein, In step S10, the dedicated ray tracing hardware is any one of the NVIDIA RTX series graphics cards, AMD RDNA series graphics cards, Intel A770 graphics cards, and Imagination Technologies' IMG CXT GPU.

Images