Video quality detection method, device and equipment and readable storage medium

Abstract

The invention discloses a video quality detection method, device and equipment and a readable storage medium, and is applied to the technical field of image processing, and the method comprises the steps: obtaining a three-dimensional Gaussian sputtering model for rendering a to-be-detected volume video; parameters of the three-dimensional Gaussian sputtering model are converted into quality features, and feature values are obtained; determining an optimized viewpoint set according to the covariance matrix, and receiving a quality evaluation value of the user for the optimized viewpoint set; and determining the video quality of the volume video to be detected based on the characteristic value and the quality evaluation value. According to the method, abstract parameters of a 3DGS model are converted into quantizable and visual quality characteristic values, and quantitative control over geometric accuracy, color fidelity, motion smoothness and visual consistency of a 3DGS volume video is achieved; the subjective evaluation of the user for weak quality points and the above model objective detection are fused to finally ensure that the 3DGS volume video can meet the requirements of simulation degree, fluency and visual consistency, and the immersive experience of the user is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a video quality detection method, apparatus, device, and readable storage medium. Background Technology

[0002] 3D Gaussian Splatting (3DGS) technology, with its advantages of real-time high-fidelity rendering and detailed texture restoration, has become a core volumetric video generation solution for scenarios such as digital human social interaction, VR virtual concerts, and industrial simulation. However, the 3DGS volumetric videos generated by this technology have some problems. For example, sparse perspectives lead to geometric distortion, dynamic interactions cause inter-frame stuttering, holes / artifacts appear in high-uncertainty areas (edges / dynamic parts), and color deviations occur due to computational overload in multi-user scenes. Therefore, quality testing is required before presenting the video to users to ensure that the final 3DGS volumetric video presented to users meets requirements for realism, smoothness, and visual consistency.

[0003] Therefore, how to conduct comprehensive quality testing on 3DGS volumetric videos to ensure that the final 3DGS volumetric videos presented to users meet the requirements of realism, smoothness, and visual consistency, and enhance the user's immersive experience, is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a video quality detection method, apparatus, device and readable storage medium, which solves the problem that the incomplete 3DGS volumetric video quality detection in the prior art results in the 3DGS volumetric video presented to the user failing to meet the requirements of realism, smoothness and visual consistency, seriously affecting the user's immersive experience.

[0005] To address the aforementioned technical problems, this invention provides a video quality detection method, comprising:

[0006] Obtain a 3D Gaussian sputtering model of the video volume to be detected for rendering;

[0007] The parameters of the three-dimensional Gaussian sputtering model are transformed into quality features to obtain feature values; the parameters include covariance matrix, spherical harmonic coefficients, opacity, and Gaussian position variation between dynamic sequence frames; the quality features include geometric accuracy features, color fidelity features, motion smoothness features, and visual consistency features.

[0008] The optimized viewpoint set is determined based on the covariance matrix, and the user's quality evaluation value for the optimized viewpoint set is received.

[0009] The video quality of the volumetric video to be detected is determined based on the feature value and the quality assessment value.

[0010] Optionally, the parameters of the three-dimensional Gaussian sputtering model are converted into quality features to obtain feature values, including:

[0011] Based on the covariance matrix, spherical harmonic coefficients, opacity, and Gaussian position changes between dynamic sequence frames of each Gaussian sphere in the three-dimensional Gaussian sputtering model, statistical information of each parameter is calculated, and a parameter statistics graph is obtained; the Gaussian sphere is the constituent unit of the three-dimensional Gaussian sputtering model.

[0012] Based on the parameter statistics chart, the quality of the three-dimensional Gaussian sputtering model is evaluated to obtain geometric accuracy feature values, color fidelity feature values, motion smoothness feature values, and visual consistency feature values.

[0013] Optionally, based on the parameter statistics chart, the quality of the three-dimensional Gaussian sputtering model is evaluated to obtain geometric accuracy feature values, color fidelity feature values, motion smoothness feature values, and visual consistency feature values, including:

[0014] Based on the covariance matrix in the parameter statistics graph, the Gaussian covariance determinant of each Gaussian sphere is calculated, and the geometric accuracy characteristic value is obtained based on the Gaussian covariance determinant.

[0015] Based on the spherical harmonic coefficients in the parameter statistics graph, calculate the global or local variance of the spherical harmonic coefficients, and obtain the color fidelity feature value based on the variance of the spherical harmonic coefficients.

[0016] By comparing the position coordinates of the same Gaussian sphere in two consecutive frames of the parameter statistics graph, the inter-frame position change is calculated, and the motion smoothness feature value is obtained based on the inter-frame position change.

[0017] The visual consistency feature value is determined based on the uniformity of the opacity value distribution in the parameter statistics graph.

[0018] Optionally, an optimized viewpoint set is determined based on the covariance matrix, and a quality assessment value for the optimized viewpoint set is received from the user, including:

[0019] An uncertainty map is obtained based on the covariance matrix of each Gaussian sphere in the three-dimensional Gaussian sputtering model; the uncertainty map is marked with the uncertainty index values ​​of each region.

[0020] The optimized viewpoint set is determined based on the uncertainty map;

[0021] The three-dimensional Gaussian sputtering model corresponding to the optimized viewpoint set is presented to the user through a virtual reality device to obtain the user's quality evaluation value for the optimized viewpoint set.

[0022] Optionally, determining the optimized viewpoint set based on the uncertainty map includes:

[0023] The region in the uncertainty map whose uncertainty index value exceeds a preset threshold is designated as the target uncertainty region.

[0024] The virtual camera observation parameters that can cover the uncertain area of ​​the target, are unobstructed, and can fully capture geometric details and rendering effects are selected to form an initial optimized viewpoint set;

[0025] The initial optimized viewpoint set is filtered according to preset conditions to obtain the optimized viewpoint set.

[0026] Optionally, the three-dimensional Gaussian sputtering model corresponding to the optimized viewpoint set is presented to the user through a virtual reality device to obtain the user's quality evaluation value for the optimized viewpoint set, including:

[0027] Extract the rendering sequence corresponding to the optimized viewpoint set, and adapt it to the display resolution and field of view of the virtual reality device to ensure that the target uncertain area has no image distortion and loss of detail in the virtual reality scene;

[0028] A virtual reality immersive scene is constructed based on the adapted rendering sequence, allowing the user to focus on details of the target's uncertain area through perspective interaction;

[0029] Collect the quantitative evaluations and defect annotations of the users to form standardized, multi-dimensional quality evaluation values.

[0030] Optionally, after determining the video quality of the volumetric video to be detected based on the feature values ​​and the quality assessment values, the method further includes:

[0031] Based on the eigenvalues ​​and the quality assessment values, the parameters of the three-dimensional Gaussian sputtering model are dynamically corrected.

[0032] The present invention also provides a video quality detection device, comprising:

[0033] The model acquisition module is used to acquire a 3D Gaussian sputtering model for rendering the video of the volume to be detected;

[0034] The parameter conversion module is used to convert the parameters of the three-dimensional Gaussian sputtering model into quality features to obtain feature values; the parameters include the covariance matrix, spherical harmonic coefficients, opacity, and Gaussian position variation between dynamic sequence frames; the quality features include geometric accuracy features, color fidelity features, motion smoothness features, and visual consistency features;

[0035] The viewpoint set determination module is used to determine an optimized viewpoint set based on the covariance matrix and receive a quality evaluation value from the user for the optimized viewpoint set.

[0036] A quality assessment module is used to determine the video quality of the video volume to be detected based on the feature value and the quality assessment value.

[0037] The present invention also provides a video quality inspection device, comprising:

[0038] Memory, used to store computer programs;

[0039] A processor is used to implement the video quality detection method described above when executing the computer program.

[0040] The present invention also provides a computer-readable storage medium storing computer-executable instructions, which, when loaded and executed by a processor, implement the video quality detection method described above.

[0041] As can be seen, this invention obtains a 3D Gaussian sputtering model of the rendered volumetric video to be tested; transforms the parameters of the 3D Gaussian sputtering model into quality features to obtain feature values; the parameters include the covariance matrix, spherical harmonic coefficients, opacity, and Gaussian position variations between dynamic sequence frames; the quality features include geometric accuracy features, color fidelity features, motion smoothness features, and visual consistency features; an optimized viewpoint set is determined based on the covariance matrix, and the user's quality evaluation value for the optimized viewpoint set is received; the video quality of the volumetric video to be tested is determined based on the feature values ​​and the quality evaluation value. This invention transforms the abstract parameters of the 3DGS model into quantifiable and visualized quality feature values, solving the problem of the difficulty in evaluating the quality of the 3DGS model, and thus achieving quantitative control over the geometric accuracy, color fidelity, motion smoothness, and visual consistency of 3DGS volumetric videos; at the same time, by focusing on quality weaknesses through viewpoint optimization, and integrating the user's subjective evaluation of quality weaknesses and the model's objective detection, it ultimately ensures that the 3DGS volumetric video can meet the requirements of realism, smoothness, and visual consistency, greatly improving the user's immersive experience.

[0042] In addition, the present invention also provides a video quality detection device, equipment and readable storage medium, which also have the above-mentioned beneficial effects. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0044] Figure 1A flowchart of a video quality detection method provided in an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the structure of a video quality detection device provided in an embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of the structure of a video quality testing device provided in an embodiment of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Please refer to Figure 1 , Figure 1 A flowchart illustrating a video quality detection method provided in an embodiment of the present invention. The method may include:

[0049] S101: Obtain the 3D Gaussian sputtering model of the video volume to be detected for rendering.

[0050] The execution entity in this embodiment is the terminal. The volumetric video in this embodiment refers to a 3DGS volumetric video generated from a 3D Gaussian sputtering model. Volumetric video captures 3D dynamic content, supports six degrees of freedom (6DoF) interaction, and is suitable for virtual reality (VR), augmented reality (AR), and immersive application scenarios. 3DGS represents the scene using a Gaussian function, achieving efficient rendering. The core building block of the 3DGS model in this embodiment is a large number of 3D Gaussian spheres, each containing parameters such as position, covariance matrix (determining shape / size / orientation), spherical harmonic coefficients (determining color), and opacity.

[0051] S102: Convert the parameters of the three-dimensional Gaussian sputtering model into quality features to obtain eigenvalues.

[0052] The purpose of this step is to calculate feature values ​​that characterize quality using the parameters of the 3DGS model, quantifying potential artifacts (such as geometric distortion, color aberration, and motion discontinuities). This process requires no human observer intervention, ensuring rapid and automated evaluation. The parameters in this embodiment refer to Gaussian parameters, specifically including the covariance matrix, spherical harmonic coefficients, opacity, and Gaussian position variations between dynamic sequence frames. The quality features in this embodiment can include geometric accuracy features, color fidelity features, motion smoothness features, and visual consistency features. The video volume to be detected in this embodiment can be one or more, enabling batch video quality detection.

[0053] Furthermore, the above-mentioned transformation of the parameters of the 3D Gaussian sputtering model into quality features to obtain feature values ​​can specifically include: calculating the statistical information of each parameter based on the covariance matrix, spherical harmonic coefficients, opacity, and inter-frame Gaussian position changes of each Gaussian sphere in the 3D Gaussian sputtering model, and obtaining a parameter statistical graph; the Gaussian sphere is the constituent unit of the 3D Gaussian sputtering model; based on the parameter statistical graph, the quality of the 3D Gaussian sputtering model is evaluated to obtain geometric accuracy feature values, color fidelity feature values, motion smoothness feature values, and visual consistency feature values.

[0054] It should be noted that the parameter statistical chart in this embodiment refers to presenting the analyzed data (such as the distribution of covariance determinant, variance of spherical harmonic coefficients, distribution of opacity, and inter-frame positional changes) in the form of charts, such as histograms and line graphs, to intuitively show the overall distribution characteristics of the parameters (such as the proportion of areas with high uncertainty and the concentration range of color variance). In order to make the feature values ​​more accurate and more in line with the core needs of quality inspection, this embodiment performs statistics on each parameter to obtain a parameter statistical chart, and then calculates each feature value based on the data in the parameter statistical chart. This embodiment obtains the parameter statistical chart by calculating the statistical information of each parameter. The purpose is to: take the covariance matrix as an example to illustrate (1) classify all Gaussian covariance matrices according to spatial location (such as digital human face, hair, clothing, etc.); (2) count the distribution of the covariance matrix in each region (such as maximum value, minimum value, variance). Then calculate the covariance determinant of each region, and finally combine the global mean and local abnormal distribution to obtain the feature value. This can reflect the overall geometric accuracy and accurately capture local defects.

[0055] Furthermore, based on the parameter statistics chart, the quality of the 3D Gaussian sputtering model is evaluated to obtain geometric accuracy feature values, color fidelity feature values, motion smoothness feature values, and visual consistency feature values, which may specifically include:

[0056] Step 11: Calculate the Gaussian covariance determinant of each Gaussian sphere based on the covariance matrix in the parameter statistics graph, and obtain the geometric accuracy eigenvalue based on the Gaussian covariance determinant.

[0057] This step utilizes the covariance matrix to evaluate the degree of distortion in the 3D structure, corresponding to the geometric accuracy eigenvalue. The accuracy of the 3DGS model in spatial geometric reconstruction is measured using the Gaussian density gradient and covariance determinant; lower values ​​indicate less geometric distortion. Based on each Gaussian covariance determinant (det(Σ_i)) obtained from the previous statistics, the sum of all Gaussian determinants (∑det(Σ_i)) is calculated. Dividing the sum by the total number of Gaussians (N) yields the average determinant value (D), i.e., D = ∑det(Σ_i) / N. The covariance determinant reflects the spatial volume of the Gaussians; a larger average determinant value (D) indicates a more dispersed spatial distribution of Gaussians in the model, a more blurred geometric structure, and a higher degree of geometric distortion; conversely, a smaller value indicates better geometric accuracy.

[0058] Step 12: Calculate the global or local variance of the spherical harmonic coefficients based on the spherical harmonic coefficients in the parameter statistics graph, and obtain the color fidelity characteristic value based on the variance of the spherical harmonic coefficients.

[0059] This step calculates the color distribution by measuring the variance of the spherical harmonic coefficients. The spherical harmonic coefficients determine the color of a Gaussian sphere (especially its color performance under varying lighting conditions). A larger variance indicates more significant color differences between different Gaussian spheres in the model, reflecting uneven local texture reconstruction or color distortion. Therefore, this step uses pixel-level color as a benchmark to compare the model's rendered result with the real reference image; a smaller difference indicates higher color fidelity. Pixel color data rendered by the 3DGS model (based on the variance of the spherical harmonic coefficients calculated previously) is extracted, and the rendered color is compared pixel-by-pixel with the reference rendering (such as a real image captured by a high-precision camera or the rendering result of a traditional high-precision 3D reconstruction method). The color variance between the two (e.g., the mean square error of the RGB three channels) is calculated, and the variance value serves as a quantitative indicator of color fidelity. A smaller color variance indicates a smaller deviation between the model's rendered color and the real scene, resulting in more accurate color reproduction; a larger variance indicates significant color distortion (such as color cast or texture blurring).

[0060] Step 13: Compare the position coordinates of the same Gaussian sphere in two consecutive frames of the parameter statistics graph, calculate the position change between frames, and obtain the motion smoothness feature value based on the position change between frames.

[0061] This step calculates the change in position (e.g., Euclidean distance) by comparing the position coordinates of the same Gaussian sphere in two consecutive frames. Excessive or irregular changes indicate jitter or misalignment during dynamic reconstruction, leading to choppy and unstable video rendering. This step measures the continuity of Gaussian motion in a dynamic sequence (e.g., volumetric video) using inter-frame optical flow consistency; higher consistency indicates smoother motion. Based on the statistically analyzed inter-frame Gaussian position changes, a motion vector is generated for each Gaussian (i.e., the position offset direction and distance of the Gaussian in adjacent frames). The optical flow consistency corresponding to these motion vectors is calculated, determining whether the Gaussian motion direction is consistent and the motion amplitude is continuous in the same region of adjacent frames (e.g., avoiding sudden position jumps). Specifically, the optical flow consistency score can be used as a feature value for motion smoothness. A higher optical flow consistency score indicates more consistent Gaussian motion in the dynamic sequence, resulting in jitter-free, misaligned, and smoother motion rendering; a low score indicates a risk of stuttering or screen tearing.

[0062] Step 14: Determine the visual consistency feature value based on the uniformity of the opacity value distribution in the parameter statistics graph.

[0063] This step directly uses statistical analysis of the opacity distribution of all Gaussian spheres to characterize opacity uniformity. Opacity determines whether an area is visible; if the opacity value of a certain area fluctuates greatly (e.g., some Gaussian spheres are completely transparent while others are completely opaque), it may cause holes or overlapping artifacts during rendering, affecting visual quality. Therefore, this embodiment calculates the corresponding visual consistency feature value based on the opacity value.

[0064] S103: Determine the optimized viewpoint set based on the covariance matrix, and receive the user's quality assessment value for the optimized viewpoint set.

[0065] This step involves using the covariance matrix to identify weak areas in the 3DGS model, filtering out optimized viewpoints that can accurately observe these areas, and then collecting quality ratings from users for the rendered content corresponding to these viewpoints.

[0066] Furthermore, the process of determining the optimized viewpoint set based on the covariance matrix and receiving the user's quality assessment value for the optimized viewpoint set can specifically include: obtaining an uncertainty map based on the covariance matrix of each Gaussian sphere in the 3D Gaussian sputtering model; marking the uncertainty index values ​​of each region in the uncertainty map; determining the optimized viewpoint set based on the uncertainty map; and presenting the 3D Gaussian sputtering model corresponding to the optimized viewpoint set to the user through a virtual reality device to obtain the user's quality assessment value for the optimized viewpoint set.

[0067] It should be noted that the covariance matrix determines the spatial shape of a single 3D Gaussian sphere (such as whether it is a sphere, ellipsoid, and the direction of stretching). Its determinant value can quantify the spatial volume or morphological complexity of the Gaussian sphere. The larger the determinant value or the more drastic the fluctuation, the lower the 3D reconstruction accuracy and the higher the uncertainty in that area. Therefore, this embodiment further binds the covariance determinant with the spatial location of the 3D model to generate a 2D or 3D heatmap, i.e., an uncertainty map. The darker the color (such as red) in this map, the higher the uncertainty of the 3D reconstruction at that location, which can quickly locate the weak areas of the model. Then, the virtual camera observation parameters that can clearly cover these areas are selectively selected to form an optimized viewpoint set. Finally, the rendered content corresponding to this viewpoint set is presented to the user, and the user's subjective rating and defect feedback on core dimensions such as geometric accuracy and color fidelity are collected.

[0068] Furthermore, the above-mentioned determination of the optimized viewpoint set based on the uncertainty map may specifically include: taking the area in the uncertainty map where the uncertainty index value exceeds a preset threshold as the target uncertainty area; selecting virtual camera observation parameters that can cover the target uncertainty area, are unobstructed, and can fully capture geometric details and rendering effects to form an initial optimized viewpoint set; and filtering the initial optimized viewpoint set according to preset conditions to obtain the optimized viewpoint set.

[0069] Specifically, the process begins by selecting regions from the uncertainty map whose uncertainty index values ​​exceed a preset threshold as target uncertainty regions (i.e., core regions where the model quality is weak and prone to distortion). Then, virtual camera observation parameters (including position, angle, and field of view) that cover these regions without visual occlusion and can fully capture the geometric details and rendering effects of the regions are selectively chosen to form an initial optimized viewpoint set. Finally, further filtering is performed according to preset conditions (such as controllable quantity and no duplicate coverage) to obtain the final optimized viewpoint set. This approach accurately identifies weak regions, avoids indiscriminate sampling, and improves the targeting of quality detection; the occlusion-free and full-coverage parameter selection ensures complete capture of defect details, providing reliable material for subsequent evaluation; and the two rounds of filtering streamline the number of viewpoints, reducing redundant observations and lowering resource consumption for subsequent rendering and evaluation. This embodiment does not specifically limit the preset conditions or preset thresholds.

[0070] Furthermore, the above-mentioned optimization of the 3D Gaussian sputtering model corresponding to the viewpoint set will be presented to the user through a virtual reality device to obtain the user's quality assessment value for the optimized viewpoint set. Specifically, this may include: extracting the rendering sequence corresponding to the optimized viewpoint set, adapting it to the display resolution and field of view of the virtual reality device to ensure that the uncertain target area is free from image distortion and detail loss in the virtual reality scene; constructing a virtual reality immersive scene based on the adapted rendering sequence, supporting the user to focus on the details of the uncertain target area through viewpoint interaction; and collecting the user's quantitative evaluation and defect annotations to form a standardized multi-dimensional quality assessment value.

[0071] Specifically, the rendering sequence corresponding to the viewpoint set is extracted and optimized (for dynamic models, it is a short-duration continuous frame sequence focusing on uncertain areas of the target; for static models, it is a multi-view static image sequence). The format is then strictly adapted according to the display resolution and field of view of the virtual reality (VR) device. Through techniques such as adjusting the screen ratio and processing details for fidelity, it is ensured that uncertain areas of the target (i.e., areas with weak model quality and high risk) are presented in the VR scene without image distortion, and that key geometric details and rendering effects are fully preserved. Based on this, an immersive VR scene is constructed using the adapted rendering sequence, supporting users to control their heads, zoom in and out, and adjust their positions. The system employs natural interactive methods such as movement, allowing users to freely switch observation angles and focus on subtle features of uncertain areas of the target (such as geometric textures, color transitions, and motion states), simulating a deep observation experience in real-world usage scenarios. Subsequently, the VR interface guides users to complete multi-dimensional quantitative evaluations (such as geometric accuracy and color fidelity), while also supporting users to mark the location and type of defects they discover (such as artifacts and stuttering). Finally, the collected scoring data and defect information are structured and organized to form standardized, reusable multi-dimensional quality assessment values, providing a direct basis for subsequent 3DGS model quality judgment and parameter optimization.

[0072] In this way, the parameters of VR devices are specifically adapted to ensure the integrity of details and the stability of the image in uncertain areas of the target, avoiding evaluation deviations caused by device compatibility issues; the combination of immersive scenes and free interaction functions allows users to accurately capture subtle defects that are difficult to detect in a purely flat presentation, improving the accuracy of subjective evaluation; the combination of quantitative scoring and defect labeling forms a structured quality evaluation value, which not only facilitates subsequent fusion and analysis with objective feature values, but also provides clear directional guidance for model optimization.

[0073] S104: Determine the video quality of the volumetric video to be detected based on the feature value and quality assessment value.

[0074] Based on the user-assessed quality evaluation value obtained in step S103 and the objective feature values ​​of the model obtained in step S102, the video quality of the volumetric video to be detected is determined. Specifically, the two can be fused using a regression model to obtain the final quality detection result. Alternatively, a weighted normalization function can be used to achieve fusion. In this way, by combining objective feature values ​​(quantitative data such as geometry and color) with subjective quality evaluation values ​​(actual user feedback), the one-sidedness of single-dimensional evaluation is avoided, making the results more in line with actual usage scenarios. Objective data ensures the scientific nature of the evaluation, while subjective feedback compensates for the omission of visual details by pure numerical calculations. The two complement each other to improve the accuracy of quality judgment. It not only meets the quantitative standards at the technical level but also fits the actual user experience, and can flexibly adapt to the quality requirements of different scenarios (such as industrial simulation and virtual interaction). The dual-dimensional evaluation results can accurately locate the contradictions of numerical standards meeting the requirements but poor visual quality or poor subjective feedback but normal parameters, providing a clear direction for subsequent model optimization. For example, the platform can batch test the quality of volumetric videos, provide optimization suggestions for those that do not meet the requirements, and automatically optimize the corresponding Gaussian parameters; it can also identify motion artifacts, adjust rendering settings, and improve the user interaction experience; it can also filter artifact areas, subjectively verify and optimize rendering to improve the quality of the final output video; and it can identify problem areas to enhance learning immersion.

[0075] Furthermore, after determining the video quality of the volumetric video to be detected based on the feature values ​​and quality assessment values, the method may also include: dynamically correcting the parameters of the three-dimensional Gaussian sputtering model based on the feature values ​​and quality assessment values.

[0076] Specifically, geometric accuracy feature values ​​can be bound to the Gaussian covariance matrix along with user geometric evaluation feedback; color fidelity feature values ​​can be bound to the spherical harmonic coefficients along with user color evaluation feedback; motion smoothness feature values ​​can be bound to inter-frame interpolation parameters along with user fluency evaluation feedback; and visual consistency feature values ​​can be bound to parameters such as opacity distribution along with user consistency evaluation feedback. Based on the dual-dimensional evaluation results (feature values ​​+ quality evaluation values), the target regions corresponding to quality shortcomings and the set of Gaussian spheres within those regions are identified. The Gaussian parameters in geometrically distorted regions are optimized for shape; the parameters in color deviation regions are adjusted for color attributes; the parameters in motion-unsmooth regions are optimized for inter-frame transitions; and the parameters in visually inconsistent regions are adjusted for distribution. After parameter correction, the model feature values ​​and user subjective feedback are re-evaluated to confirm that the quality shortcomings have been improved.

[0077] The video quality detection method provided in this embodiment of the invention comprises the following steps: S101: Obtain a three-dimensional Gaussian sputtering model for rendering the volumetric video to be detected; S102: Convert the parameters of the three-dimensional Gaussian sputtering model into quality features to obtain feature values; S103: Determine an optimized viewpoint set based on the covariance matrix and receive the user's quality evaluation value for the optimized viewpoint set; S104: Determine the video quality of the volumetric video to be detected based on the feature values ​​and quality evaluation values. This method transforms the abstract parameters of the 3DGS model into quantifiable and visualized quality feature values, solving the problem of difficulty in evaluating the quality of the 3DGS model. This enables quantitative control over the geometric accuracy, color fidelity, motion smoothness, and visual consistency of the 3DGS volumetric video. Simultaneously, by focusing on quality weaknesses through viewpoint optimization and integrating the user's subjective evaluation of these weaknesses with the model's objective detection, the method ultimately ensures that the 3DGS volumetric video meets the requirements of realism, smoothness, and visual consistency, greatly enhancing the user's immersive experience.

[0078] The video quality detection device provided in the embodiments of the present invention will be described below. The video quality detection device described below can be referred to in correspondence with the video quality detection method described above.

[0079] Please refer to the details. Figure 2 , Figure 2 A schematic diagram of a video quality detection device provided in an embodiment of the present invention may include:

[0080] The model acquisition module 100 is used to acquire a three-dimensional Gaussian sputtering model for rendering the volume video to be detected;

[0081] The parameter conversion module 200 is used to convert the parameters of the three-dimensional Gaussian sputtering model into quality features to obtain feature values; the parameters include covariance matrix, spherical harmonic coefficients, opacity, and Gaussian position variation between dynamic sequence frames; the quality features include geometric accuracy features, color fidelity features, motion smoothness features, and visual consistency features;

[0082] The viewpoint set determination module 300 is used to determine an optimized viewpoint set based on the covariance matrix and receive a quality evaluation value from the user for the optimized viewpoint set.

[0083] The quality assessment module 400 is used to determine the video quality of the volume video to be detected based on the feature value and the quality assessment value.

[0084] Based on the above embodiments, the parameter conversion module 200 may include:

[0085] The parameter statistics graph acquisition unit is used to calculate the statistical information of each parameter based on the covariance matrix, spherical harmonic coefficient, opacity, and Gaussian position change between dynamic sequence frames of each Gaussian sphere in the three-dimensional Gaussian sputtering model, and obtain the parameter statistics graph; the Gaussian sphere is the constituent unit of the three-dimensional Gaussian sputtering model;

[0086] The feature value acquisition unit is used to perform quality assessment on the three-dimensional Gaussian sputtering model based on the parameter statistical graph, and obtain geometric accuracy feature values, color fidelity feature values, motion smoothness feature values, and visual consistency feature values.

[0087] Based on the above embodiments, the feature value acquisition unit may include:

[0088] The geometric accuracy feature value acquisition subunit is used to calculate the Gaussian covariance determinant of each Gaussian sphere based on the covariance matrix in the parameter statistics graph, and obtain the geometric accuracy feature value based on the Gaussian covariance determinant.

[0089] The color fidelity feature value acquisition subunit is used to calculate the global or local spherical harmonic coefficient variance based on the spherical harmonic coefficient in the parameter statistics graph, and obtain the color fidelity feature value based on the spherical harmonic coefficient variance.

[0090] The motion smoothness feature value acquisition subunit is used to compare the position coordinates of the same Gaussian sphere in two consecutive frames of the parameter statistics graph, calculate the inter-frame position change, and obtain the motion smoothness feature value based on the inter-frame position change.

[0091] The visual consistency feature value acquisition subunit is used to determine the visual consistency feature value based on the uniformity of the opacity value distribution in the parameter statistics graph.

[0092] Based on any of the above embodiments, the viewpoint set determination module 300 may include:

[0093] An uncertainty map acquisition unit is used to obtain an uncertainty map based on the covariance matrix of each Gaussian sphere in the three-dimensional Gaussian sputtering model; the uncertainty map is marked with uncertainty index values ​​for each region;

[0094] An optimized viewpoint set determination unit is used to determine the optimized viewpoint set based on the uncertainty map;

[0095] The quality assessment value determination unit is used to present the three-dimensional Gaussian sputtering model corresponding to the optimized viewpoint set to the user through a virtual reality device in order to obtain the user's quality assessment value for the optimized viewpoint set.

[0096] Based on the above embodiments, the optimized viewpoint set determination unit may include:

[0097] The target uncertainty area determination subunit is used to identify areas in the uncertainty map where the uncertainty index value exceeds a preset threshold as target uncertainty areas.

[0098] An initial optimized viewpoint set determination subunit is used to filter virtual camera observation parameters that can cover the uncertain area of ​​the target, are unobstructed, and can fully capture geometric details and rendering effects, thus forming an initial optimized viewpoint set;

[0099] The optimized viewpoint set determination subunit is used to filter the initial optimized viewpoint set according to preset conditions to obtain the optimized viewpoint set.

[0100] Based on the above embodiments, the quality assessment value determination unit may include:

[0101] The adaptation subunit is used to extract the rendering sequence corresponding to the optimized viewpoint set and adapt the format according to the display resolution and field of view of the virtual reality device so that the target uncertain area has no image distortion and loss of detail in the virtual reality scene.

[0102] The scene construction subunit is used to construct a virtual reality immersive scene based on the adapted rendering sequence, allowing the user to focus on the details of the target's uncertain area through viewpoint interaction;

[0103] The standardized subunit is used to collect the user's quantitative evaluation and defect labeling to form the standardized multi-dimensional quality evaluation value.

[0104] Based on the above embodiments, the video quality detection device may further include:

[0105] The parameter correction module is used to dynamically correct the parameters of the three-dimensional Gaussian sputtering model based on the feature values ​​and the quality assessment values ​​after determining the video quality of the video to be detected based on the feature values ​​and the quality assessment values.

[0106] It should be noted that the order of the modules and units in the aforementioned video quality detection device can be changed without affecting the logic.

[0107] The model acquisition module 100 provided in this embodiment of the invention is used to acquire a three-dimensional Gaussian sputtering model for rendering a video of the volume to be detected; the parameter conversion module 200 is used to convert the parameters of the three-dimensional Gaussian sputtering model into quality features to obtain feature values; the parameters include covariance matrix, spherical harmonic coefficients, opacity, and Gaussian position variation between dynamic sequence frames; the quality features include geometric accuracy features, color fidelity features, motion smoothness features, and visual consistency features; the viewpoint set determination module 300 is used to determine an optimized viewpoint set based on the covariance matrix and receive a quality evaluation value from the user for the optimized viewpoint set; the quality evaluation module 400 is used to determine the video quality of the video of the volume to be detected based on the feature values ​​and the quality evaluation value. This invention transforms the abstract parameters of 3DGS models into quantifiable and visualized quality feature values, solving the problem of difficulty in evaluating the quality of 3DGS models. This enables quantitative control over the geometric accuracy, color fidelity, motion smoothness, and visual consistency of 3DGS volumetric videos. Simultaneously, by focusing on quality weaknesses through viewpoint optimization and integrating subjective user assessments of these weaknesses with objective model detection, this invention ultimately ensures that 3DGS volumetric videos meet the requirements of realism, smoothness, and visual consistency, greatly enhancing the user's immersive experience.

[0108] The video quality detection device provided in the embodiments of the present invention will be described below. The video quality detection device described below and the video quality detection method described above can be referred to each other.

[0109] Please refer to Figure 3 , Figure 3 A schematic diagram of a video quality testing device provided in an embodiment of the present invention may include:

[0110] Memory 10 is used to store computer programs;

[0111] The processor 20 is used to execute computer programs to implement the video quality detection method described above.

[0112] The memory 10, processor 20, and communication interface 31 all communicate with each other through the communication bus 32.

[0113] In this embodiment of the invention, the memory 10 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment of the invention, the memory 10 may store programs for implementing the following functions:

[0114] Obtain a 3D Gaussian sputtering model of the video volume to be detected for rendering;

[0115] The parameters of the 3D Gaussian sputtering model are transformed into quality features to obtain eigenvalues. The parameters include the covariance matrix, spherical harmonic coefficients, opacity, and Gaussian position variation between dynamic sequence frames. The quality features include geometric accuracy features, color fidelity features, motion smoothness features, and visual consistency features.

[0116] The optimized viewpoint set is determined based on the covariance matrix, and the user's quality assessment value for the optimized viewpoint set is received; the video quality of the volume video to be detected is determined based on the eigenvalues ​​and the quality assessment value.

[0117] In one possible implementation, the memory 10 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; and the data storage area may store data created during use.

[0118] Furthermore, memory 10 may include read-only memory and random access memory, providing instructions and data to the processor. A portion of the memory may also include NVRAM. The memory stores operating systems and operating instructions, executable modules, or data structures, or subsets thereof, or extended sets thereof, wherein the operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and handling hardware-based tasks.

[0119] Processor 20 can be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic device. Processor 20 can be a microprocessor or any conventional processor. Processor 20 can call programs stored in memory 10.

[0120] Communication interface 31 can be an interface for the communication module, used to connect with other devices or systems.

[0121] Of course, it should be noted that, Figure 3 The structure shown does not constitute a limitation on the video quality detection device in the embodiments of the present invention. In practical applications, the video quality detection device may include more than Figure 3 More or fewer components as shown, or combinations of certain components.

[0122] The following describes the computer-readable storage medium provided in the embodiments of the present invention. The computer-readable storage medium described below can be referred to in correspondence with the video quality detection method described above.

[0123] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the video quality detection method described above.

[0124] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0126] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0127] Finally, it should be noted that in this document, relationships such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0128] The above provides a detailed description of a video quality detection method, apparatus, device, and computer-readable storage medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method of video quality detection, the method comprising: The method comprises the following steps: obtaining a three-dimensional Gaussian splatting model for rendering a video of a volume to be detected; converting parameters of the three-dimensional Gaussian splatting model into quality features to obtain feature values; the parameters include a covariance matrix, spherical harmonic coefficients, opacity, and dynamic sequence inter-frame Gaussian position changes; the quality features include geometric accuracy features, color fidelity features, motion smoothness features, and visual consistency features; determining an optimized viewpoint set according to the covariance matrix and receiving a quality evaluation value of the optimized viewpoint set from a user; determining the video quality of the volume video to be detected based on the feature values and the quality evaluation value.

2. The video quality detection method of claim 1, wherein, The method for converting the parameters of the three-dimensional Gaussian splatting model into quality features to obtain feature values comprises the following steps: According to the covariance matrix, spherical harmonic coefficients, opacity, and dynamic sequence inter-frame Gaussian position changes of each Gaussian sphere in the three-dimensional Gaussian splatting model, statistical information of each parameter is calculated to obtain a parameter statistical graph; the Gaussian sphere is a constituent unit of the three-dimensional Gaussian splatting model; According to the parameter statistical graph, quality evaluation of the three-dimensional Gaussian splatting model is performed to obtain geometric accuracy feature values, color fidelity feature values, motion smoothness feature values, and visual consistency feature values.

3. The video quality detection method of claim 2, wherein, According to the parameter statistical graph, quality evaluation of the three-dimensional Gaussian splatting model is performed to obtain geometric accuracy feature values, color fidelity feature values, motion smoothness feature values, and visual consistency feature values, which comprises the following steps: According to the covariance matrix in the parameter statistical graph, the Gaussian covariance determinant of each Gaussian sphere is calculated, and the geometric accuracy feature values are obtained according to the Gaussian covariance determinant; According to the spherical harmonic coefficients in the parameter statistical graph, the global or local spherical harmonic coefficient variance is calculated, and the color fidelity feature values are obtained according to the spherical harmonic coefficient variance; By comparing the position coordinates of the same Gaussian sphere in two consecutive frames in the parameter statistical graph, the inter-frame position change amount is calculated, and the motion smoothness feature values are obtained according to the inter-frame position change amount; According to the uniformity of the opacity numerical distribution in the parameter statistical graph, the visual consistency feature values are determined.

4. The video quality detection method of any of claims 1 to 3, wherein, According to the covariance matrix, an optimized viewpoint set is determined, and a quality evaluation value of the optimized viewpoint set is received from a user, which comprises the following steps: According to the covariance matrix of each Gaussian sphere in the three-dimensional Gaussian splatting model, an uncertainty map is obtained; the uncertainty map marks the uncertainty index values of each region; The optimized viewpoint set is determined according to the uncertainty map; The three-dimensional Gaussian splatting model corresponding to the optimized viewpoint set is presented to the user through a virtual reality device to obtain the quality evaluation value of the optimized viewpoint set from the user.

5. The video quality detection method of claim 4, wherein, According to the uncertainty map, the optimized viewpoint set is determined, which comprises the following steps: regarding the region with an uncertainty index value exceeding a preset threshold in the uncertainty map as a target uncertainty region; screening virtual camera observation parameters that can cover the target uncertainty region, have no occlusion, and can completely capture geometric details and rendering effects to form an initial optimized viewpoint set; screening the initial optimized viewpoint set according to a preset condition to obtain the optimized viewpoint set.

6. The video quality detection method of claim 4, wherein, The three-dimensional Gaussian sputtering model corresponding to the optimized viewpoint set is presented to the user through a virtual reality device to obtain a quality evaluation value of the user for the optimized viewpoint set, including: The rendering sequence corresponding to the optimized viewpoint set is extracted, and the display resolution and field of view angle of the virtual reality device are adapted to the format, so that the target uncertain area has no picture distortion and detail loss in the virtual reality scene; Based on the adapted rendering sequence, a virtual reality immersive scene is constructed to support the user to focus on the details of the target uncertain area through visual angle interaction; The user's quantitative evaluation and defect labeling are collected to form a standardized multi-dimensional quality evaluation value.

7. The video quality detection method of claim 1, wherein, After determining the video quality of the to-be-detected volumetric video based on the characteristic value and the quality evaluation value, further comprising: Based on the characteristic value and the quality evaluation value, the parameters of the three-dimensional Gaussian sputtering model are dynamically corrected.

8. A video quality detection apparatus, characterized by comprising: Including: A model acquisition module for acquiring a three-dimensional Gaussian sputtering model for rendering a to-be-detected volumetric video; A parameter conversion module for converting parameters of the three-dimensional Gaussian sputtering model into quality characteristics to obtain a characteristic value; the parameters include a covariance matrix, spherical harmonic coefficients, opacity, and dynamic sequence interframe Gaussian position changes; the quality characteristics include geometric accuracy characteristics, color fidelity characteristics, motion smoothness characteristics, and visual consistency characteristics; A viewpoint set determination module for determining an optimized viewpoint set according to the covariance matrix and receiving a quality evaluation value of the user for the optimized viewpoint set; A quality evaluation module for determining the video quality of the to-be-detected volumetric video based on the characteristic value and the quality evaluation value.

9. A video quality detection device, characterized by, Including: A memory for storing a computer program; A processor for executing the computer program to implement the video quality detection method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions, and when the computer executable instructions are loaded and executed by the processor, the video quality detection method of any one of claims 1 to 7 is implemented. The computer readable storage medium stores computer executable instructions, and when the computer executable instructions are loaded and executed by the processor, the video quality detection method of any one of claims 1 to 7 is implemented.

Images