A multi-dimensional feature-based adaptive hybrid rendering method, system and medium

Abstract

The present application relates to the technical field of scene rendering, in particular to a kind of adaptive hybrid rendering method, system and medium based on multidimensional feature, the method is generated by geometry channel G-Buffer containing scene information, comprehensively utilize depth, normal, material and motion and so on Multidimensional feature information;Using decision function, screen space pixel is adaptively divided into visual key area and non-visual key area;For non-visual key area, delay rendering technology is used, for visual key area, path tracking rendering technology is used;The rendering result is weighted and fused to obtain final rendering result.The present application can effectively reduce the invalid calculation overhead of non-visual key area, significantly improve the update efficiency of acceleration structure under dynamic scene, while guaranteeing the physical reality of large-scale dynamic three-dimensional scene, meet the performance requirements of real-time interaction.

Description

Technical Field

[0001] This invention belongs to the field of scene rendering technology, and in particular relates to an adaptive hybrid rendering method, system and medium based on multi-dimensional features. Background Technology

[0002] With the rapid development of cutting-edge applications such as virtual reality (VR), digital twins, and smart factories, real-time rendering technology for 3D scenes is facing increasingly stringent challenges in both performance and image quality. These applications not only require systems to have high frame rate real-time interactive capabilities, but also place extremely high demands on the physical realism of the scene.

[0003] Currently, mainstream rendering technologies fall into two main categories: one is rasterization-based rendering, which projects 3D geometric primitives onto a 2D screen, uses depth buffers for occlusion culling, and combines local lighting models for shading. Due to its high computational efficiency, it is widely used in various real-time interactive systems. However, rasterization, based on an approximate local lighting model, struggles to accurately simulate the multiple bounces of light across object surfaces, color overflow, and complex specular reflection chains, often resulting in images lacking depth and physical consistency. The other category is ray tracing-based rendering, which simulates the real-world propagation paths of light in a scene and uses Monte Carlo integration to solve rendering equations. This allows for precise rendering of global illumination, soft shadows, and complex material effects, making it the standard method for generating highly realistic images. However, its computational overhead is enormous, especially in dynamic scenes where the structure needs to be reconstructed and accelerated every frame, causing single-frame rendering time to far exceed the threshold for real-time interaction, resulting in stuttering.

[0004] To balance efficiency and quality, existing technologies employ hybrid rendering schemes, attempting to combine the advantages of both approaches. For example, ray tracing is used on specular surfaces while rasterization is used in other areas. However, existing hybrid schemes generally use fixed pipeline switching strategies. This coarse-grained allocation leads to unnecessary precise calculations for distant, flat areas, or non-focal objects, resulting in wasted computational resources. Furthermore, in visually sensitive areas such as geometric edges, grazing angles, specular materials, or moving objects, high-quality algorithms cannot be enabled due to overall performance budget constraints, resulting in insufficient realism. Therefore, achieving pixel-level adaptive rendering resource allocation to maximize overall rendering efficiency while maintaining realism in visually focal areas has become a pressing technical challenge in the field of real-time hybrid rendering. Summary of the Invention

[0005] To address the problems of high computational load for path tracing, insufficient physical realism in rasterization rendering, and difficulty in balancing efficiency and quality due to coarse resource allocation granularity in existing hybrid rendering schemes for large-scale dynamic 3D scene rendering, this invention provides an adaptive hybrid rendering method based on multi-dimensional features, comprising the following steps: Step S1: Extract multi-dimensional features from the dynamic 3D scene through the geometric channel. The multi-dimensional features include at least depth, normals, material and motion information. Step S2: Based on the multidimensional features, the visual importance of pixels is evaluated through a decision function to achieve scene segmentation and output non-visual key regions and visual key regions. Step S3: Perform deferred rendering on the non-visual key regions and path tracing rendering on the visual key regions. Step S4: The deferred rendering result and the path tracing rendering result are merged to output the final rendered image.

[0006] Based on the above scheme, step S2 includes: S21: Input the multidimensional features into the decision function, and the decision function outputs the visual importance weight of the pixel in the current view; S22: When the output value of the decision function is greater than the preset global threshold, the pixel is marked as the non-visual key region; otherwise, it is marked as the visual key region.

[0007] Preferably, the decision function is a weighted sum of a depth factor, a normal factor, a material factor, and a motion factor; the depth factor is determined based on the screen space depth gradient and is used to identify regions with rapidly changing depth; the normal factor is determined based on the angle between the normal and the line of sight and is used to identify regions with significant Fresnel effects; the material factor is determined based on surface roughness and is used to evaluate the computational complexity of lighting; and the motion factor is determined based on the screen space motion vector and is used to identify high-speed moving objects.

[0008] Based on the above scheme, step S3 performs deferred rendering on non-visually critical regions, specifically including: S301: Divide the screen space into multiple tiles, obtain the depth range of each tile based on the depth pyramid constructed by the depth buffer, and construct a list of local effective light sources for each tile by combining view frustum culling; S302, traverse the light sources in the list of locally effective light sources and perform cumulative illumination calculations on the pixels within the image block.

[0009] Further, step S3 performs path tracing rendering on the visual key regions, the path tracing rendering including incremental accelerated structure construction, specifically including: S311: Use the motion information to perform temporal reprojection on the pixels and calculate the residual information at the position corresponding to the previous frame; S312: Based on the residual information, determine the active regions of geometric change in the scene using a gradient-based adaptive threshold function; S313: Calculate the percentage of pixels that have changed covered by each object instance. If the change in the percentage of pixels exceeds the deformation threshold, perform local reconstruction on the underlying acceleration structure. If the change in the percentage of pixels does not exceed the deformation threshold, but displacement is detected, update the transformation matrix in the top-level acceleration structure of the object.

[0010] Preferably, the path tracing rendering further includes an energy compensation approximation, employing the Kulla-Conty approximation model, specifically including: S321: Pre-calculate the albedo in the single scattering direction and the average albedo in the hemisphere, and store it as a two-dimensional lookup table; S322: During real-time rendering, the two-dimensional lookup table is sampled based on the pixel roughness and viewing angle to obtain the energy compensation coefficient; S323: Construct a multiple scattering bidirectional reflection distribution function term and superimpose it onto the single scattering bidirectional reflection distribution function term.

[0011] Specifically, step S4 includes: Edge blending weights are calculated based on the depth and normal information of pixels, and the edge blending weights are used to perform weighted blending of deferred rendering results and path tracing rendering results. The mixed image is tone-mapped and post-processed to output the final rendered image.

[0012] On the other hand, the present invention also provides an adaptive real-time hybrid rendering system based on multi-dimensional features, the system comprising a data generation module, a scene partitioning module, a hybrid rendering module, and a result fusion module: The data generation module extracts multi-dimensional features from the dynamic 3D scene through a geometric channel. The multi-dimensional features include at least depth, normals, material, and motion information. The scene segmentation module is used to evaluate the visual importance of pixels based on the multidimensional features through a decision function, thereby achieving scene segmentation and outputting non-visual key regions and visual key regions. The hybrid rendering module is used to perform deferred rendering on the non-visual key areas and path tracing rendering on the visual key areas. The result fusion module is used to fuse the deferred rendering result and the path tracing rendering result to output the final rendered image.

[0013] Furthermore, the hybrid rendering module includes a deferred rendering unit and a path tracing unit; The deferred rendering unit is used to process the basic lighting in non-visually critical areas; The path tracing unit is used to process global illumination in key visual regions. This unit integrates an incremental acceleration structure construction submodule and an energy compensation submodule.

[0014] In another aspect, the present invention also provides a computer-readable storage medium having a computer program that, when executed by a processor, implements the steps of the adaptive hybrid rendering method based on multidimensional features as described above.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By adaptive resource allocation and incremental structural updates, the bottleneck of the incompatibility between high image quality and high frame rate is broken. It can effectively reduce the invalid computing overhead of non-visual key areas, significantly improve the efficiency of accelerating structural updates in dynamic scenes, and meet the performance requirements of real-time interaction while ensuring the physical realism of large-scale dynamic 3D scenes. 2. By identifying key visual regions through multi-dimensional feature decision functions, the traditional method avoids coarse division based solely on depth thresholds. High-fidelity path tracing is enabled in key visual regions, while efficient deferred rendering is used in non-key regions. This solves the problem of coarse-grained resource allocation in traditional hybrid rendering, balancing rendering efficiency and visual quality. 3. By adopting an incremental BVH dynamic update strategy, the accelerated structure is updated differently based on temporal reprojection and multi-dimensional feature residual detection, which significantly reduces the maintenance overhead of the accelerated structure in dynamic scenes, solves the real-time rendering performance bottleneck caused by full reconstruction, and realizes high frame rate rendering of large-scale dynamic scenes. 4. By using the Kulla-Conty energy compensation mechanism and pre-computed lookup tables, the multiple scattering energy of high-roughness materials can be recovered without increasing light bounce, which solves the problem of energy non-conservation of micro-surface models under low sampling rate and makes the rendering results close to the offline path tracing true value. 5. By using a depth normal-based edge blending mechanism, the results of different rendering regions are weighted and mixed to eliminate the visual discontinuity at the boundaries caused by partitioned rendering, thus ensuring the overall visual quality of the final image. Attached Figure Description

[0016] Figure 1 This is a flowchart of the hybrid rendering method of the present invention; Figure 2 This is a schematic diagram of the hybrid rendering method of the present invention; Figure 3 This is a schematic diagram of the light source removal in this invention; Figure 4 This is a schematic diagram of the hierarchical incremental update BVH method of the present invention; Figure 5 This is a schematic diagram of the hybrid rendering system of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] like Figure 1 and Figure 2 As shown, this invention provides an adaptive hybrid rendering method based on multi-dimensional features, comprising the following steps: Step S1: Extract multi-dimensional features from the dynamic 3D scene through the geometric channel. The multi-dimensional features include at least depth, normals, material and motion information. According to the present invention, step S1 specifically involves: preprocessing the dynamic 3D scene through the geometry rendering channel to generate a geometry buffer (G-Buffer) containing world space position, normal, material properties and screen linear depth information. According to the present invention, the multidimensional features include world spatial location, normal, material properties, and screen linear depth information contained within the geometric buffer.

[0019] It should be noted that, in order to reduce the bandwidth usage of video memory in subsequent calculations, this embodiment of the invention employs a depth compression coding method based on logarithmic distribution, which uses a logarithmic function to map linear depth to memory space: , Where C is the precision adjustment constant, and f is the distance to the far cutting plane. This is a linear depth. Through this mapping encoding, the depth precision is redistributed within a limited storage width, effectively alleviating the problem of insufficient precision of linear depth at distant locations, and maximizing precision within a limited storage width.

[0020] Step S2: Based on multi-dimensional features, the visual importance of pixels is evaluated through a decision function to achieve scene segmentation and output non-visual key regions and visual key regions. This step involves scene segmentation using multi-dimensional features, marking key and non-key visual regions, specifically including: S21: Input the multidimensional features into the decision function, and output the visual importance weight of the pixel in the current view through the decision function; Step S21 in this embodiment specifically includes: S211: Define the decision function The weighted sum of multidimensional feature factors: , in, Depth factor The weighting coefficients, is the weight coefficient of the normal factor , is the weight coefficient of the material factor , is the weight coefficient of the motion factor .

[0021] S212: Determine each weight based on multi-dimensional feature factors Depth factor: Use the screen space depth gradient to identify areas with rapid depth changes, and the lower the weight of areas with greater depth and flatter surfaces; Normal factor: Calculate the angle between the normal and the line of sight to identify complex Fresnel effect areas, and the higher the weight in the grazing angle area; Material factor: Use surface roughness to evaluate the lighting calculation overhead, and the higher the weight in areas with low roughness; Motion factor: Based on the screen space motion vector judgment, although high-speed moving objects have motion blur, they are prone to ghosting at low sampling rates, so the higher the weight coefficient.

[0022] S22: Compare the output value of the decision function with the preset global threshold T. If D(p)>T, mark the pixel p as a visually critical area; if D(p)<T, mark it as a non-visually critical area.

[0023] Based on the above steps, the present invention identifies visually critical areas and non-visually critical areas, and then performs partition rendering optimization, and executes different rendering pipelines for different areas, specifically see step S3.

[0024] Step S3: Perform deferred rendering on non-visually critical areas and path tracing rendering on visually critical areas; First, step S3 performs deferred rendering on non-visually critical areas, as Figure 3 shown, specifically including: S301: Divide the screen space into multiple tiles Title, obtain the depth range of each tile based on the depth pyramid constructed from the depth buffer, and combine frustum culling to construct a local valid light source list for each tile; According to this embodiment, S301 specifically includes: (1) Divide the screen space into regular tiles of 16×16 pixels, use the depth buffer generated in step S1 to construct a depth pyramid, and quickly query the minimum depth and maximum depth of each Title in the view space; (2) Combine the two-dimensional screen range and depth range of the tile to form a frustum slice, perform a culling test on all light sources in the scene, and eliminate invalid light sources that overlap in screen projection but do not intersect in depth, and construct a list of local effective light sources for each tile.

[0025] S302: Traverse the light sources in the list of locally effective light sources and perform cumulative illumination calculations on the pixels within the Tile.

[0026] By implementing deferred rendering in step S3, the overhead of branch prediction and unnecessary computation are significantly reduced. Especially in complex scenes, this effectively reduces the computational load and ensures real-time rendering of the region.

[0027] Furthermore, for visually critical regions, this invention employs path tracing rendering, including incrementally accelerated structure construction and energy-compensated approximation, wherein, as... Figure 4 As shown, the incremental acceleration structure construction specifically includes: S311: Temporally reproject the pixels of the current frame using the motion vectors in the G-Buffer, calculate their corresponding positions in the previous frame, and calculate the residual information between the current frame and the corresponding positions in the previous frame. This residual information is a three-dimensional feature residual, including depth residuals. Normal residual and world position residuals ; S312: Based on residual information, combined with a gradient-based adaptive threshold function Determine the active areas of geometric change in the scene; S313: Calculate the percentage of changed pixels covered by each object instance. If the change in pixel percentage exceeds the deformation threshold... If an object is determined to have undergone non-rigid deformation, a bottom-up local reconstruction is performed on the underlying acceleration structure of the object that has undergone non-rigid deformation. If the change in pixel ratio does not exceed the deformation threshold, but rigid displacement is detected, only the transformation matrix of the object that has undergone non-rigid deformation in the top-level acceleration structure is updated. If the object is stationary, the structure of the previous frame is reused.

[0028] The above steps employ an incremental BVH dynamic construction strategy based on temporal reprojection and multidimensional feature residual detection, which can solve the problem of accelerated structural failure caused by object deformation or displacement in dynamic scenes.

[0029] Furthermore, to address the energy loss issue caused by neglecting multiple scattering in the microsurface model at low sampling rates, an energy compensation approximation is performed using the Kulla-Conty approximation model, specifically including: S321: Offline pre-calculation of single-scattering directional albedo and hemispherical average albedo And store it as a two-dimensional lookup table. Let cosine be the angle between the line of sight and the normal. S322: During real-time rendering, a two-dimensional lookup table is sampled based on the pixel roughness and viewing angle to obtain the energy compensation coefficient; S323: Construct the Bidirectional Reflectance Distribution Function (BRDF) term. And superimposed on the single-scattering BRDF result: , in, The directional albedo is the direction of emission. The directional albedo is the incident direction. The average Fresnel reflectance.

[0030] Through the above energy compensation, the energy of multiple scatterings can be effectively compensated without increasing the number of light bounces, effectively restoring the energy conservation characteristics of high-roughness materials and enhancing the physical realism.

[0031] Step S4: The deferred rendering result and the path tracing rendering result are merged to output the final rendered image.

[0032] Specifically, step S4 includes: Edge blending weights are calculated based on pixel depth and normal information. These weights are then used to weight and blend the deferred rendering results and path tracing rendering results to eliminate visual discontinuities at region boundaries. The mixed image is tone-mapped and post-processed to output the final rendered image.

[0033] On the other hand, the present invention also provides an adaptive real-time hybrid rendering system based on multi-dimensional features, such as... Figure 5 As shown, the system includes a data generation module, a scene segmentation module, a hybrid rendering module, and a result fusion module.

[0034] The data generation module extracts multi-dimensional features from the dynamic 3D scene through the geometric channel, generating a geometric buffer containing multi-dimensional features, which include at least depth, normals, material, and motion information.

[0035] The scene segmentation module is used to perform visual importance evaluation of pixels based on multidimensional features in the geometric buffer through a decision function, thereby achieving scene segmentation, outputting non-visual key regions and visual key regions, and generating region masks.

[0036] The hybrid rendering module includes a deferred rendering unit and a path tracing unit. The deferred rendering unit is used to handle the basic lighting of non-visually critical areas, while the path tracing unit is used to handle the global lighting of visually critical areas. This unit integrates an incremental acceleration structure construction submodule and an energy compensation submodule.

[0037] The hybrid rendering module receives the region mask generated by the scene partitioning module, schedules GPU resources, and the deferred rendering unit starts the block lighting computation shader for the regions marked as deferred rendering (visually critical regions). The path tracing unit starts the ray tracing pipeline for the regions marked as path tracing (non-visually critical regions) and calls the computation shader to perform incremental BVH updates and Kulla-Conty energy compensation calculations.

[0038] The results fusion module is used to merge the deferred rendering results and the path tracing rendering results to output the final rendered image.

[0039] This invention selected representative scenarios such as Sponza, Bistro, and dynamic robotic arms for testing. The results show that, while ensuring image quality, the method of this invention achieves a speedup of approximately 2.6 to 2.8 times compared to full path tracing, with a stable frame rate that meets the requirements of real-time interaction. Therefore, the method of this invention can adaptively adjust the rendering strategy according to the geometric complexity, material distribution, and dynamic characteristics of the scene, adapting to industrial simulation and visualization application scenarios of different scales.

[0040] Specific applications: In a large-scale virtual simulation system embodiment, the technical solution of this invention serves as the core rendering engine, providing real-time visualization capabilities for complex geometric scenes. By deploying an adaptive adjustment strategy based on geometric load, rendering performance optimization is achieved under scenes of varying complexity. This hybrid rendering pipeline successfully overcomes the performance bottleneck of real-time rendering in large-scale scenes.

[0041] In an embodiment of an industrial digital twin system, the technical solution of the present invention serves as a dynamic scene rendering service, providing the system with real-time monitoring of high-load dynamic primitives. By configuring an incremental update strategy based on multi-dimensional feature residual detection to replace the traditional full reconstruction method, the system possesses high-efficiency operation capabilities under extreme loads, effectively avoiding severe squeezing of the frame budget and ensuring extremely high operating efficiency of the digital twin system in dynamic environments.

[0042] In an embodiment of a high-fidelity material rendering system, the technical solution of the present invention serves as a physically based lighting calculation module, providing the system with energy-conserving material rendering functionality. By introducing the Kulla-Conty energy compensation mechanism, the energy loss problem of the traditional single-scattering model on high-roughness, totally reflective metallic materials is solved. Under a white furnace test environment, the darkening phenomenon at the edges of the sphere is visually eliminated, enabling the system's rendering results to restore the high reflectivity of the metal and approximate the true value of offline path tracing.

[0043] It should be noted that the methods of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0044] Furthermore, the adaptive real-time hybrid rendering method according to the present invention can be recorded in a computer-readable recording medium. Specifically, according to the present invention, a computer-readable recording medium storing computer-executable instructions can be provided, which, when executed by a processor, cause the processor to execute the adaptive real-time hybrid rendering method as described above.

[0045] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or portion of code containing at least one executable instruction for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0046] In general, various exemplary embodiments of the present invention can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of the present invention are illustrated or described as block diagrams, flowcharts, or represented using certain other images, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or certain combinations thereof.

[0047] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0048] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. An adaptive hybrid rendering method based on multi-dimensional features, characterized in that, include: Step S1: Extract multi-dimensional features from the dynamic 3D scene through the geometric channel. The multi-dimensional features include at least depth, normals, material and motion information. Step S2: Based on the multidimensional features, the visual importance of pixels is evaluated through a decision function to achieve scene segmentation and output non-visual key regions and visual key regions. Step S3: Perform deferred rendering on the non-visual key regions and path tracing rendering on the visual key regions. Step S4: The deferred rendering result and the path tracing rendering result are merged to output the final rendered image.

2. The adaptive hybrid rendering method based on multi-dimensional features according to claim 1, characterized in that, Step S2 includes: S21: Input the multidimensional features into the decision function, and the decision function outputs the visual importance weight of the pixel in the current view; S22: When the output value of the decision function is greater than the preset global threshold, the pixel is marked as the non-visual key region; otherwise, it is marked as the visual key region.

3. The adaptive hybrid rendering method based on multi-dimensional features according to claim 2, characterized in that, The decision function is a weighted sum of depth factor, normal factor, material factor and motion factor; the depth factor is determined based on the screen space depth gradient and is used to identify regions with rapid depth changes; the normal factor is determined based on the angle between the normal and the line of sight and is used to identify regions with significant Fresnel effect. The material factor is determined based on surface roughness and is used to evaluate the computational complexity of lighting; the motion factor is determined based on screen space motion vectors and is used to identify high-speed moving objects.

4. The adaptive hybrid rendering method based on multi-dimensional features according to claim 1, characterized in that, Step S3 involves performing deferred rendering on non-visually critical regions, specifically including: S301: Divide the screen space into multiple tiles, obtain the depth range of each tile based on the depth pyramid constructed by the depth buffer, and construct a list of local effective light sources for each tile by combining view frustum culling; S302, traverse the light sources in the list of locally effective light sources and perform cumulative illumination calculations on the pixels within the image block.

5. The adaptive hybrid rendering method based on multi-dimensional features according to claim 1, characterized in that, Step S3 performs path tracing rendering on the visual key regions. The path tracing rendering includes incremental accelerated structure construction, specifically including: S311: Use the motion information to perform temporal reprojection on the pixels and calculate the residual information at the position corresponding to the previous frame; S312: Based on the residual information, determine the active regions of geometric change in the scene using a gradient-based adaptive threshold function; S313: Calculate the percentage of pixels that have changed covered by each object instance. If the change in the percentage of pixels exceeds the deformation threshold, perform local reconstruction on the underlying acceleration structure. If the change in the percentage of pixels does not exceed the deformation threshold, but displacement is detected, update the transformation matrix in the top-level acceleration structure of the object.

6. The adaptive hybrid rendering method based on multi-dimensional features according to claim 5, characterized in that, The path tracing rendering also includes an energy compensation approximation, employing the Kulla-Conty approximation model, specifically including: S321: Pre-calculate the albedo in the single scattering direction and the average albedo in the hemisphere, and store it as a two-dimensional lookup table; S322: During real-time rendering, the two-dimensional lookup table is sampled based on the pixel roughness and viewing angle to obtain the energy compensation coefficient; S323: Construct a multiple scattering bidirectional reflection distribution function term and superimpose it onto the single scattering bidirectional reflection distribution function term.

7. The adaptive hybrid rendering method based on multi-dimensional features according to claim 1, characterized in that, Step S4 specifically includes: Edge blending weights are calculated based on the depth and normal information of pixels, and the edge blending weights are used to perform weighted blending of deferred rendering results and path tracing rendering results. The mixed image is tone-mapped and post-processed to output the final rendered image.

8. An adaptive real-time hybrid rendering system based on multi-dimensional features, characterized in that, The system includes a data generation module, a scene segmentation module, a hybrid rendering module, and a result fusion module. The data generation module extracts multi-dimensional features from the dynamic 3D scene through a geometric channel. The multi-dimensional features include at least depth, normals, material, and motion information. The scene segmentation module is used to evaluate the visual importance of pixels based on the multidimensional features through a decision function, thereby achieving scene segmentation and outputting non-visual key regions and visual key regions. The hybrid rendering module is used to perform deferred rendering on the non-visual key areas and path tracing rendering on the visual key areas. The result fusion module is used to fuse the deferred rendering result and the path tracing rendering result to output the final rendered image.

9. The adaptive real-time hybrid rendering system based on multi-dimensional features according to claim 8, characterized in that, The hybrid rendering module includes a deferred rendering unit and a path tracing unit; The deferred rendering unit is used to process the basic lighting in non-visually critical areas; The path tracing unit is used to process global illumination in key visual regions. This unit integrates an incremental acceleration structure construction submodule and an energy compensation submodule.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium has a computer program that, when executed by a processor, implements the adaptive hybrid rendering method based on multidimensional features as described in any one of claims 1-7.

Images