Method and device for determining visual angle of three-dimensional model, and computing equipment

Abstract

The embodiment of the invention provides a three-dimensional model view angle determining method and device and computing equipment, and the three-dimensional model view angle determining method comprises the steps that a target three-dimensional model and a preset number of space parameters of initial view angles are obtained, and the target three-dimensional model comprises a plurality of vertexes; for the first vertex, determining the visibility probability of the first vertex under the preset number of initial visual angles based on the visual parameters of the first vertex and the space parameters of the preset number of initial visual angles; for the first initial view angle, based on the visibility probabilities of the plurality of vertexes under the first initial view angle, determining the visibility confidence of the target three-dimensional model under the first initial view angle; and determining a target view angle corresponding to the target three-dimensional model based on the visible confidence of the target three-dimensional model under the preset number of initial view angles. Accurate evaluation and sorting of the potential display view angle of the three-dimensional model are realized, so that the determination efficiency of the target display view angle of the three-dimensional model is improved, and the reconstruction and rendering efficiency of the three-dimensional scene is improved.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of computer technology, and in particular to a method for determining the perspective of a three-dimensional model, a device for determining the perspective of a three-dimensional model, and a computing device. Background Technology

[0002] With the rapid development of 3D visualization technology, 3D models are increasingly widely used on various multimedia terminals. Efficiently and accurately determining the perspective of a 3D model can provide a quick preview capability, which is crucial for improving visualization effects and work efficiency.

[0003] Currently, determining the perspective of a 3D model mainly relies on manual adjustment or preset rules based on geometric features. It usually requires multiple attempts by the human to adjust the position of the virtual camera in order to find a perspective that can clearly and accurately display the features of the 3D model. Some methods also use machine learning to assist in the selection and determination based on the geometric characteristics of the model.

[0004] However, the aforementioned methods, which rely on manual adjustment of the viewpoint, struggle to guarantee optimality and suffer from low processing efficiency, especially when dealing with large-scale 3D models or viewpoints, where they cannot efficiently determine the target viewpoint. Machine learning methods, on the other hand, depend on a large amount of original images of the 3D model or camera pose information as training data, rendering them unusable when data is missing or insufficient. Therefore, a more efficient and accurate method for determining the viewpoint of 3D models is urgently needed. Summary of the Invention

[0005] In view of this, embodiments of this specification provide a method for determining the perspective of a three-dimensional model. One or more embodiments of this specification also relate to a three-dimensional model perspective determination device, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0006] According to a first aspect of the embodiments of this specification, a method for determining the perspective of a three-dimensional model is provided, comprising:

[0007] Obtain the spatial parameters of the target 3D model and a preset number of initial viewpoints, wherein the target 3D model includes multiple vertices;

[0008] For the first vertex, based on the visual parameters of the first vertex and the spatial parameters of a preset number of initial viewpoints, determine the visibility probability of the first vertex under a preset number of initial viewpoints, where the first vertex is any vertex.

[0009] For a first initial viewpoint, based on the visibility probabilities of multiple vertices under the first initial viewpoint, the visibility confidence of the target 3D model under the first initial viewpoint is determined, where the first initial viewpoint is any initial viewpoint;

[0010] Based on the visibility confidence of the target 3D model under a preset number of initial viewpoints, the target viewpoint corresponding to the target 3D model is determined.

[0011] According to a second aspect of the embodiments of this specification, a three-dimensional model perspective determination device is provided, comprising:

[0012] The acquisition module is configured to acquire the spatial parameters of the target 3D model and a preset number of initial viewpoints, wherein the target 3D model includes multiple vertices;

[0013] The visibility probability determination module is configured to determine the visibility probability of a first vertex under a preset number of initial viewing angles based on the visual parameters of the first vertex and the spatial parameters of a preset number of initial viewing angles, wherein the first vertex is any vertex.

[0014] The visibility confidence determination module is configured to determine the visibility confidence of the target 3D model in the first initial viewpoint based on the visibility probabilities of multiple vertices in the first initial viewpoint, wherein the first initial viewpoint is any initial viewpoint;

[0015] The target viewpoint determination module is configured to determine the target viewpoint corresponding to the target 3D model based on the visibility confidence of the target 3D model under a preset number of initial viewpoints.

[0016] According to a third aspect of the embodiments of this specification, a computing device is provided, comprising:

[0017] Memory and processor;

[0018] The memory is used to store computer-executable instructions, and the processor is used to execute computer programs / instructions. When the computer programs / instructions are executed by the processor, they implement the steps of the above-described method for determining the perspective of a 3D model.

[0019] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the above-described three-dimensional model perspective determination method.

[0020] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described three-dimensional model perspective determination method.

[0021] One embodiment of this specification implements a method for determining the viewpoint of a 3D model, comprising: acquiring spatial parameters of a target 3D model and a preset number of initial viewpoints, wherein the target 3D model includes multiple vertices; for a first vertex, determining the visibility probability of the first vertex under the preset number of initial viewpoints based on the visual parameters of the first vertex and the spatial parameters of the preset number of initial viewpoints, wherein the first vertex is any vertex; for a first initial viewpoint, determining the visibility confidence of the target 3D model under the first initial viewpoint based on the visibility probabilities of multiple vertices under the first initial viewpoint, wherein the first initial viewpoint is any initial viewpoint; and determining the target viewpoint corresponding to the target 3D model based on the visibility confidence of the target 3D model under the preset number of initial viewpoints.

[0022] By calculating the visibility probability based on the visual parameters of each vertex in the 3D model and the spatial parameters of a preset number of initial viewpoints, and further integrating the visibility probabilities of multiple vertices to construct the visibility confidence of the 3D model under a specific initial viewpoint, the system achieves accurate evaluation and ranking of potential display viewpoints of the 3D model. By adaptively starting from the vertex distribution characteristics of the 3D model, without relying on the original image or camera pose information, the system can efficiently determine the target display viewpoint of the 3D model through probability field superposition, providing reliable technical support for rapid previewing and quality assessment of 3D content, and improving the accuracy and efficiency of viewpoint selection during 3D scene reconstruction and rendering. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a method for determining the perspective of a three-dimensional model, as provided in one embodiment of this specification.

[0024] Figure 2 This is a schematic diagram of the structure of a three-dimensional model perspective determination device provided in one embodiment of this specification;

[0025] Figure 3 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0026] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0027] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “the,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0028] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0029] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are obtained through open-source datasets or public datasets that comply with their license agreements, or are obtained with full authorization from the relevant parties. Moreover, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0030] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0031] Neural Radiance Field (NeRF) is a technique that uses neural networks to implicitly represent 3D scenes. By learning the continuous volume density and color distribution of the scene, it can accurately render high-quality 3D scene images from various perspectives.

[0032] Virtual Reality (VR) is a technology that uses computers to generate a three-dimensional virtual environment and immerse users in it. It achieves an immersive interactive experience by simulating the human sensory system, especially vision and hearing.

[0033] Augmented Reality (AR) is a technology that overlays virtual information onto the real world. It can be used to merge digital content with real-world scenes through mobile or wearable devices, providing an enhanced real-world experience. It is commonly used in applications such as navigation, games, and industrial maintenance guidance, allowing users to obtain additional virtual information while retaining their perception of the real world.

[0034] 3D Gaussian Splatting (3DGS) is a method for representing and rendering 3D scenes based on Gaussian distribution. By representing objects in 3D space as a set of multiple Gaussian functions, it enables efficient modeling and rendering of complex 3D scenes, and is especially suitable for dynamic environments or applications that require real-time performance.

[0035] Neural Radiance Field (NeRF) is a technique that uses neural networks to implicitly represent 3D scenes. By learning the continuous volume density and color distribution of the scene, it can accurately render high-quality 3D scene images from various perspectives.

[0036] Principal Axis Standard Deviation is a statistic that describes the dispersion of a three-dimensional Gaussian point along the principal axis. It is used to quantify the extent of the Gaussian point in space and is a key indicator for determining the size parameters of the Gaussian point, directly affecting the calculation results of the visibility probability.

[0037] Color values ​​(RGB values) represent the three-channel values ​​of a color, usually composed of red, green, and blue components. The RGB color space is used to accurately describe colors and is the basic way to represent colors in computer graphics. It is widely used in image processing and display technology and determines the accurate rendering of colors.

[0038] Transparency (Alpha value) is a parameter that represents the degree of transparency of an image or object. It is usually represented by an Alpha value, ranging from 0 (completely transparent) to 1 (completely opaque). In computer graphics, it is used to control the transparency and blending effects of an image. It is a key parameter for achieving semi-transparent effects and affects the calculation of visibility probability.

[0039] The Phone model is a common 3D model that can be used to simulate the appearance and structure of devices such as mobile phones. It is often used for the analysis and testing of back occlusion phenomena and is a typical test case for evaluating 3D model visibility algorithms. It can effectively verify the algorithm's ability to handle surface orientation and occlusion relationships.

[0040] Patch variance is a statistic used in image processing to measure the dispersion of color distribution in a local area. It assesses the richness of texture by calculating the color variance of a specific region (patch) in an image and is a key indicator in image texture analysis, affecting the calculation of Gaussian point texture parameters.

[0041] The Sobel operator is an operator used for image edge detection. It highlights edges by calculating gradients. It can use two 3x3 convolution kernels to calculate the gradients in the horizontal and vertical directions respectively. It is a commonly used edge detection tool in image processing and can effectively extract the contour information of an image.

[0042] The Gray-Level Co-occurrence Matrix (GLCM) is a statistical method for texture analysis that describes the co-occurrence relationship of pixel gray values ​​in an image. By calculating the co-occurrence frequency of different gray values ​​at specific distances and directions, it is used to extract texture features of an image and is a fundamental method for texture analysis.

[0043] Heuristic algorithms are optimization algorithms designed based on empirical rules. They are used to efficiently search the solution space of a problem. By utilizing the characteristics of the problem, they reduce the search range and improve search efficiency. They are commonly found in combinatorial optimization problems and can quickly find solutions close to the optimal solution.

[0044] The probabilistic gradient method is an optimization method that combines probabilistic models and gradient calculations to find optimal solutions in noisy environments. It guides the search process by estimating the gradient direction of the objective function and combining it with probability distributions. It is suitable for optimization problems in high-noise environments and makes the search more robust.

[0045] Monte Carlo sampling is a statistical method based on random sampling used to approximate complex integrals or estimate probability distributions. It approximates the true value by using the statistical properties of a large number of random samples and is widely used in scientific computing and financial modeling. It can effectively handle high-dimensional complex problems.

[0046] Particle swarm optimization is an optimization algorithm inspired by swarm behavior. It finds the optimal solution by simulating a flock of birds foraging. Each particle represents a potential solution, and the position is updated through swarm cooperation and individual experience. It is suitable for continuous optimization problems and has the characteristics of high computational efficiency and simple implementation.

[0047] Greedy search is a search strategy that selects the current optimal solution at each step without considering the global optimum. It gradually builds the global solution by using local optima. It is computationally efficient but may get stuck in local optima. It is often used in path planning and combinatorial optimization problems and can quickly obtain feasible solutions.

[0048] This specification provides a method for determining the perspective of a three-dimensional model. It also relates to a device for determining the perspective of a three-dimensional model, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0049] See Figure 1 , Figure 1 A flowchart of a method for determining the perspective of a three-dimensional model according to an embodiment of this specification is shown, which specifically includes the following steps.

[0050] Step 102: Obtain the spatial parameters of the target 3D model and a preset number of initial viewpoints, wherein the target 3D model includes multiple vertices.

[0051] The three-dimensional model perspective determination method provided in one or more embodiments of this specification can be applied to various scenarios, including three-dimensional content production and display, three-dimensional content interaction, etc.; specifically, it can be applied to various platforms and applications that provide three-dimensional content, including but not limited to social media applications, e-commerce applications, educational applications, and game applications; correspondingly, it can be applied to various terminal platforms such as mobile devices, VR / AR headsets, and web browsers.

[0052] A 3D model is a digital model with three-dimensional spatial coordinates constructed using computer graphics techniques. It is used to represent the geometric structure and surface properties of real-world objects or virtual scenes. 3D models can be generated using techniques such as Neural Radiation Field (NeRF) or 3D Gaussian Splatting (3DGS), or through geometric modeling methods such as polygonal meshes or voxel modeling. They can also be constructed using methods such as structured light scanning. Specifically, a 3D model can contain multiple vertex data points. These vertex data carry information about the model's geometric shape and visual features, and can serve as input data for determining the viewpoint of the 3D model.

[0053] A target 3D model is a specific 3D model whose target perspective needs to be determined. In practical applications, a target 3D model can be a reconstructed scene model or a model that has undergone secondary editing, requiring the determination of its target perspective for display or rendering. For example, a target 3D model could be a 3D model used for product display or a virtual character model used in immersive social applications.

[0054] The preset number is a pre-defined initial number of viewpoints used for sampling and evaluation during viewpoint determination. Its function is to control the scale of the initial viewpoints, ensuring the scope and accuracy of the viewpoint search and preventing too many or too few viewpoints from affecting computational efficiency and result accuracy. The preset number is related to the spatial parameters of the initial viewpoints; by setting reasonable values, computational complexity and viewpoint coverage can be balanced.

[0055] Initial viewpoints refer to a pre-defined set of viewpoints used to evaluate the visibility of a 3D model. Each initial viewpoint contains spatial parameters describing the virtual camera's position and orientation in space. Specifically, initial viewpoints are the foundation of viewpoint evaluation, providing candidate observation points. Their spatial parameters may include camera position parameters and viewpoint orientation parameters. Target viewpoints can be determined by calculating the visibility confidence level under each initial viewpoint and then filtering and ranking them.

[0056] Spatial parameters refer to the geometric properties describing the initial viewpoint in three-dimensional space. They can include camera position parameters and viewpoint orientation parameters, used to determine the position and viewing direction of the virtual camera in three-dimensional space. Specifically, the function of spatial parameters is to provide a mathematical representation of the viewpoint so as to calculate the relative position and orientation between vertices and the camera, thereby affecting the calculation of visibility probability.

[0057] Vertices are the basic geometric elements that make up a 3D model. They represent a point in 3D space and can have visual parameters such as position, color, and normal. Specifically, vertices can be points on the 3D model, including external and internal points; they can also be 3D Gaussian points that make up the 3D model, such as spheres or ellipsoids. Specifically, vertices serve as the basic unit for calculating the visibility of a 3D model. By analyzing the visibility of each vertex under different viewpoints, the visibility confidence of the entire 3D model can be obtained.

[0058] In practical applications, methods for acquiring the target 3D model can include reading from a local storage system, downloading from a cloud server, or dynamically generating it. Specifically, after creating the 3D model on a 3D model processing platform or system, the 3D model file can be saved in the local storage system for subsequent processing; alternatively, the stored 3D model file can be downloaded through a cloud service interface; or the 3D model can be dynamically generated based on the currently input parameter data (e.g., the original image). The specific method for acquiring the target 3D model can be dynamically determined based on the actual application scenario or requirements, and the embodiments in this specification do not impose specific limitations on this.

[0059] In the process of acquiring the target 3D model, the spatial parameters of the initial viewpoint can also be obtained. Specifically, the spatial parameters of the initial viewpoint can be obtained through a preset viewpoint list, uniform sampling, or intelligent generation methods based on model characteristics.

[0060] Specifically, the initial viewpoint can be a "reasonable" perspective for observing the target 3D model in 3D space, and it must meet preset conditions. Specifically, the camera position of the initial viewpoint must not overlap with the target 3D model, and its orientation must ensure that at least the target 3D model exists within the camera's field of view. The setting and selection of the initial viewpoint can be determined based on different types of target 3D models. For example, if the target 3D model is an object to be displayed, the initial viewpoint can be set outside or around the target 3D model, allowing observation of the entire target 3D model; if the target 3D model is a scene model to be displayed, the initial viewpoint can be set inside the target 3D model, allowing observation of the scene in a specific direction of the target 3D model.

[0061] Optionally, the initial viewpoint can be determined by first identifying the observable boundary of the target 3D model, and then further determining a reasonable viewpoint as the initial viewpoint. The observable boundary can be determined based on the coordinate distribution of the 3D model, and the determined observable boundary can include both inside-out and outside-in directions. The reasonable viewpoint can be determined by uniformly sampling the bounding box of the target 3D model to determine the camera position, and then determining the camera orientation of the initial viewpoint by traversing discrete orientations (e.g., by facing the field center or using a spherical coordinate system). That is, before obtaining a preset number of initial viewpoints, the following steps are also included:

[0062] Based on the coordinate distribution of the target 3D model, the observable boundary is determined;

[0063] Uniform sampling is performed on the bounding box of the target 3D model to determine the camera position, and the camera orientation is determined by traversing the discrete orientation of each camera position.

[0064] The initial viewpoint is determined based on the camera position and camera orientation.

[0065] In this step, by acquiring the spatial parameters of the target 3D model and a preset number of initial viewpoints, a data foundation is provided for subsequent calculation of vertex visibility probability and construction of visibility confidence. The target 3D model includes multiple vertices, each with visual parameters, enabling subsequent visibility probability calculations to be based on the specific visual characteristics of the vertices. This improves the accuracy and efficiency of 3D model viewpoint determination and enhances the adaptive capability of viewpoint selection during 3D scene reconstruction and rendering.

[0066] Step 104: For the first vertex, based on the visual parameters of the first vertex and the spatial parameters of a preset number of initial viewpoints, determine the visibility probability of the first vertex under a preset number of initial viewpoints, where the first vertex is any vertex.

[0067] The first vertex is any one of the multiple vertices in the target 3D model and can serve as the basic unit for calculating visibility probability. Specifically, the first vertex can be any point on the 3D model, including external points, internal points, or any 3D Gaussian point that constitutes the 3D model. The visual parameters (such as position parameters, size parameters, color parameters, and normal parameters) included in the first vertex can be used to calculate the visibility probability of that vertex under a preset number of initial viewing angles.

[0068] The visibility probability of a first vertex under a preset number of initial viewing angles is the probability or likelihood that the first vertex can be observed under each of the preset number of initial viewing angles. It can be used to measure the visibility of the first vertex under any given initial viewing angle. Specifically, the visibility probability can be a value between 0 and 1, with a higher value indicating that the first vertex is more visible under the corresponding initial viewing angle. The visibility probability is calculated and determined based on the visual parameters of the first vertex and the spatial parameters of each initial viewing angle. For example, if the normal direction of the first vertex is consistent with the observation direction of a certain initial viewing angle, the visibility probability can be close to 1, indicating that the first vertex is almost completely visible under that initial viewing angle.

[0069] In practical applications, given the target 3D model and the spatial parameters of a preset number of initial viewpoints, the visibility probability of the first vertex under the preset number of initial viewpoints can be determined based on its visual parameters and the spatial parameters of the first vertex. That is, for the i-th vertex out of m vertices, n visibility probabilities can be determined under n initial viewpoints. Therefore, the entire target 3D model has a total of m×n visibility probabilities under the preset number of initial viewpoints.

[0070] Specifically, for the first vertex, the visibility probability of the first vertex under a preset number of initial viewpoints can be determined by calculating the viewpoint distance parameter, which is the distance between the first vertex and the camera position of the virtual camera in the preset number of initial viewpoints, based on the vertex position parameter of the first vertex and the camera position parameter included in the spatial parameters of the initial viewpoints, as well as the viewpoint orientation parameter, which is the angle between the direction vector from the first vertex to the camera position and the camera observation direction.

[0071] Optionally, if the first vertex is a three-dimensional Gaussian point, the distance and size parameters can be further determined based on the view distance parameter and the Gaussian point size parameter (e.g., based on the variance or covariance of the three-dimensional Gaussian point, extract the attribute distribution characteristics of the obtained three-dimensional Gaussian point itself, such as the principal axis standard deviation, etc.) to reflect the relative size of the first vertex under the initial view.

[0072] Meanwhile, the visibility parameters can be determined based on the directional deviation between the Gaussian point normal parameters and the viewpoint orientation parameters, i.e., the dot product of the normal and the viewing direction, thereby simulating the surface visibility of the first vertex.

[0073] In this step, the visibility probability of the first vertex under multiple initial viewpoints is calculated based on the visual parameters of the first vertex and the spatial parameters of a preset number of initial viewpoints. This achieves an accurate assessment of the visibility of each vertex in the 3D model under different viewpoints, providing a data foundation for subsequently constructing the visibility confidence of the 3D model under each initial viewpoint. By calculating based on the visual parameters of each vertex, the visual characteristics of the vertex itself are considered. By fusing the visual parameters of the vertex with the viewpoint parameters to determine the visibility probability, the visibility degree of each vertex under a specific initial viewpoint can be more accurately reflected. This does not rely on the original image of the target 3D model or camera pose information, but adaptively starts from the visual characteristics of the vertex, improving the accuracy and efficiency of 3D model viewpoint determination. This provides technical support for rapid previewing and quality assessment of 3D content, and enhances the adaptive capability and reliability of viewpoint selection during 3D scene reconstruction and rendering.

[0074] Step 106: For the first initial viewpoint, based on the visibility probabilities of multiple vertices under the first initial viewpoint, determine the visibility confidence of the target 3D model under the first initial viewpoint, where the first initial viewpoint is any initial viewpoint.

[0075] The first initial viewpoint is any one of a preset set of initial viewpoints, used as the object of viewpoint evaluation for the target 3D model. Specifically, the first initial viewpoint is the basic unit of viewpoint evaluation, providing a basis for evaluating the visibility of the target 3D model under a specific viewing direction. Its spatial parameters may include camera position parameters and viewpoint orientation parameters, used to determine the specific position and viewing direction of the virtual camera in 3D space.

[0076] The visibility confidence of a target 3D model under the first initial viewpoint is an evaluation value determined based on the visibility probabilities of multiple vertices under that initial viewpoint. It is used to measure the visibility of the entire target 3D model under that initial viewpoint. Specifically, the visibility confidence can be calculated by combining the visual weights of multiple vertices and the visual texture parameters of the target 3D model. The visibility confidence is a value between 0 and 1. The higher the value, the more visible the target 3D model is under that initial viewpoint, and it is the core basis for determining the target viewpoint.

[0077] In practical applications, once the visibility probabilities of multiple vertices under a preset number of initial viewpoints are determined, the visibility confidence of the target 3D model under any one of the preset number of initial viewpoints can be further determined.

[0078] Specifically, determining the visibility confidence of the target 3D model under the first initial viewpoint can be based on the visual visibility parameters of multiple vertices, and the visual weights of the visibility probabilities of each vertex under the first initial viewpoint can be determined. These visual weights can be determined by the attribute feature distribution of each vertex to reflect the importance and salience of each vertex in the 3D model.

[0079] Furthermore, the visual texture parameters of the target 3D model under the first initial viewpoint can be determined based on the visual feature parameters of multiple vertices or the rendered image of the target 3D model as a whole under the first initial viewpoint.

[0080] Subsequently, by weighted calculation, the visual weights, visual texture parameters, and the visibility probabilities of multiple vertices can be fused to obtain the visibility confidence of the target 3D model under the first initial viewpoint.

[0081] In this step, by considering the visibility probability of multiple vertices under the first initial viewpoint, the visibility confidence of the target 3D model under that viewpoint is determined, thus achieving an accurate assessment of the visibility of the 3D model under different viewpoints. The calculation of visibility confidence can adaptively calculate based on the visual characteristics of each vertex, fully considering the contribution of each vertex to the visibility of the target 3D model. This accurately reflects the display integrity and visual quality of the target 3D model under the first initial viewpoint, improving the accuracy and efficiency of 3D model viewpoint determination. It provides reliable technical support for rapid previewing and quality assessment of 3D content, and enhances the adaptive capability and reliability of viewpoint selection during 3D scene reconstruction and rendering.

[0082] Step 108: Based on the visibility confidence of the target 3D model under a preset number of initial viewpoints, determine the target viewpoint corresponding to the target 3D model.

[0083] The target viewpoint is determined based on the visibility confidence of the target 3D model under a preset number of initial viewpoints. It is the final viewpoint used for displaying, interacting with, or rendering the 3D model. The target viewpoint can provide the optimal display points or key viewpoint set for the target 3D model to support rapid previewing, quality assessment, or interactive exploration of 3D content, thereby optimizing the user's visual perception and interactive experience of the 3D model.

[0084] Specifically, the target viewpoint can be determined by sorting or selecting the visibility confidence of the target 3D model under a preset number of initial viewpoints. Specifically, it can include a single best viewpoint for observing or displaying the target 3D model; it can also include multiple recommended viewpoints for displaying the target 3D model from multiple angles during the interaction, display, and rendering of the target 3D model; or it can include multiple key viewpoints with relatively rich information about the target 3D model.

[0085] Optionally, the target viewpoint can be further determined by ranking the visibility confidence from high to low. That is, the initial viewpoint with the highest visibility confidence and ranked first can be determined as the best viewpoint; the initial views ranked second to the preset ranking are multiple recommended viewpoints, and their preset ranking can be determined according to the actual application scenario. For example, if a platform needs one main image and six secondary images from multiple perspectives to jointly display an item, then the initial views ranked second to seventh can be determined as key viewpoints; for the remaining initial viewpoints, a visibility confidence threshold can also be set. If the visibility confidence of an initial viewpoint is greater than the visibility confidence threshold, it means that it contains relatively rich model information and can be determined as a key viewpoint.

[0086] In practical applications, once the visibility confidence of the target 3D model under a preset number of initial viewpoints is determined, the target viewpoint of the target 3D model can be further determined based on the visibility confidence of the preset number of initial viewpoints.

[0087] Specifically, the method for determining the target viewpoint may include sorting the preset number of initial viewpoints based on the visibility confidence of the target 3D model under the preset number of initial viewpoints to obtain the initial viewpoint sorting result; and determining the target viewpoint corresponding to the target 3D model from the preset number of initial viewpoints based on the initial viewpoint sorting result.

[0088] Among them, ranking based on visibility confidence can be achieved using various methods.

[0089] One alternative approach is to sort the initial viewpoints based on the direction of change in visible confidence using an optimization algorithm. This optimization algorithm may include heuristic algorithms or probabilistic gradient methods, etc.

[0090] Another alternative approach is to use a threshold filtering method based on visibility confidence, where the initial viewpoint with visibility confidence higher than a preset threshold is determined as the recommended viewpoint, and the initial viewpoint with the highest visibility confidence is determined as the optimal viewpoint.

[0091] Another alternative approach is to use a clustering algorithm to divide the initial viewpoints into three categories: best viewpoint, recommended viewpoint, and key viewpoint, based on the distribution of visibility confidence. Then, the initial viewpoints in each category are dynamically sorted and selected to find the target viewpoint that is locally optimal or globally optimal.

[0092] In this step, the method of determining the target viewpoint based on the visibility confidence of the target 3D model under a preset number of initial viewpoints is used to accurately evaluate and rank the potential display viewpoints of the 3D model. This method can determine the target viewpoints of the target 3D model, including the best display viewpoint, recommended viewpoint, and key viewpoint, making the display of 3D content more in line with user needs and improving the user's interactive experience with 3D content.

[0093] In the embodiments of this specification, the visibility probability is calculated based on the visual parameters of each vertex in the 3D model and the spatial parameters of a preset number of initial viewpoints. Furthermore, the visibility probabilities of multiple vertices are integrated to construct the visibility confidence of the 3D model under a specific initial viewpoint, thereby achieving accurate evaluation and ranking of potential display viewpoints of the 3D model. By adaptively starting from the vertex distribution characteristics of the 3D model, without relying on original images or camera pose information, the target display viewpoint of the 3D model can be efficiently determined through probability field superposition. This provides reliable technical support for rapid previewing and quality assessment of 3D content, improving the accuracy and efficiency of viewpoint selection during 3D scene reconstruction and rendering.

[0094] In one optional embodiment of this specification, the spatial parameters of the initial viewpoint include camera position parameters, and the visual parameters of the vertices include vertex position parameters;

[0095] For the first vertex, based on the visual parameters of the first vertex and the spatial parameters of a preset number of initial viewpoints, the visibility probability of the first vertex under the preset number of initial viewpoints is determined, including:

[0096] For the first vertex, based on the vertex position parameters of the first vertex and the camera position parameters of a preset number of initial viewpoints, determine the viewpoint distance parameters and viewpoint orientation parameters;

[0097] Based on the view distance parameter and view orientation parameter, the visibility probability of the first vertex under a preset number of initial viewpoints is determined.

[0098] Camera position parameters are geometric parameters that describe the specific position of a virtual camera in three-dimensional space from the initial viewpoint. They are typically represented using three-dimensional coordinates, including x, y, and z coordinate values, and are used to determine the camera's accurate position in three-dimensional space. Camera position parameters provide a mathematical representation of the viewpoint, thus providing an observer's positional reference for visibility probability calculation and influencing the calculation of visibility probability.

[0099] Vertex position parameters are geometric parameters describing the specific positions of each vertex in a target 3D model within 3D space. They can also be represented using 3D coordinates, including x, y, and z coordinate values, to determine the accurate position of each vertex in 3D space. Vertex position parameters provide a mathematical representation of the model's geometry, serving as the fundamental spatial information for visibility probability calculations. By comparing these parameters with camera position parameters, the geometric relationship between each vertex and the viewpoint can be determined, allowing for the calculation of visibility probability through the derivation of view distance and view orientation parameters.

[0100] The view distance parameter is the Euclidean distance between the first vertex and the camera position of the virtual camera at the initial viewpoint. It can be used to quantify the spatial distance between the first vertex and the camera; a closer distance usually indicates that the first vertex is clearer or more prominent at that initial viewpoint. The initial distance parameter can be calculated from the vertex position parameter and the camera position parameter.

[0101] The viewpoint orientation parameter is a directional quantification determined by calculating the vertex position parameter of the first vertex and the spatial parameters of the initial viewpoint. It can be determined by the angle between the direction vector from the first vertex to the camera position and the camera's observation direction vector. It reflects the alignment between the first vertex and the camera's observation direction; a higher alignment means the first vertex is more likely to be visible from the front in the initial viewpoint, serving as a key factor in evaluating surface visibility and back occlusion. The viewpoint orientation parameter can be calculated by the dot product of the direction vector from the vertex position to the camera position and the camera's observation direction vector.

[0102] In practical applications, once the target 3D model and a preset number of initial viewpoint spatial parameters are obtained, the viewpoint distance parameters and viewpoint orientation parameters can be determined for the first vertex based on the position parameters of the first vertex and the preset number of initial viewpoint spatial parameters.

[0103] Specifically, the view distance parameter can be determined by calculating the Euclidean distance between the vertex position and the camera position based on the vertex position parameter of the first vertex and the camera position parameters of a preset number of initial viewpoints; the view orientation parameter can be determined by calculating the dot product of the direction vector from the vertex position to the camera position and the camera observation direction vector.

[0104] Once the viewing distance parameter and viewing orientation parameter are determined, the visibility probability of the first vertex under a preset number of initial viewing angles can be further determined based on the viewing distance parameter and viewing orientation parameter.

[0105] Specifically, the visibility probability can be determined based on the view distance parameter and the view orientation parameter through a mathematical model or function mapping. Optionally, attenuation functions, such as Gaussian attenuation functions, linear attenuation functions, etc., or custom functions based on actual observation experience can be used to adapt to different types of 3D models and view determination requirements. At the same time, the attenuation coefficient can be dynamically adjusted based on the geometric characteristics of each vertex (such as the size and shape of the vertex) to make the calculation results more consistent with the actual observation effect.

[0106] In this embodiment, by using the vertex position parameters of the first vertex and the camera position parameters of a preset number of initial viewpoints, view distance parameters and view orientation parameters are determined. Based on these view distance and view orientation parameters, the visibility probability is determined, thus achieving an accurate assessment of the visibility of each vertex in a 3D model under different initial viewpoints. The camera position parameters and vertex position parameters provide the basic spatial data for calculation, enabling visibility assessment based on accurate geometric relationships. Simultaneously, the view distance parameter quantifies the impact of observation distance on the visibility of each vertex, simulating the distance attenuation effect in actual visual perception. The view orientation parameter quantifies the deviation between the observation direction and the actual orientation of each vertex, effectively identifying vertices visible from the front and invisible from the back. By determining the visibility probability based on the view distance and view orientation parameters, the accuracy and efficiency of 3D model viewpoint determination are improved, enhancing the adaptive capability of viewpoint selection during 3D scene reconstruction and rendering, and providing a more precise viewpoint determination scheme for the production and application of 3D content.

[0107] In one optional embodiment of this specification, the vertex includes a three-dimensional Gaussian point, and the visual parameters of the vertex include Gaussian point position parameters, Gaussian point size parameters, Gaussian point color parameters, and Gaussian point normal parameters.

[0108] Based on the view distance parameter and view orientation parameter, the visibility probability of the first vertex under a preset number of initial viewpoints is determined, including:

[0109] Based on the view distance parameter and the Gaussian point size parameter, the distance and size parameters of the first three-dimensional Gaussian point are determined, wherein the first three-dimensional Gaussian point is any three-dimensional Gaussian point;

[0110] Based on the directional deviation between the Gaussian point normal parameters and the view orientation parameters, the visibility parameters of the first three-dimensional Gaussian point are determined.

[0111] Based on the Gaussian point color parameters, distance size parameters, and visibility parameters, the visibility probability of the first three-dimensional Gaussian point under a preset number of initial viewing angles is determined.

[0112] A 3D Gaussian point is a set of 3D spatial data composed of a large number of discrete points. It is the basic unit for constructing a 3D Gaussian model. The attribute parameters of each 3D Gaussian point can include Gaussian point position parameters, Gaussian point size parameters, Gaussian point color parameters, and Gaussian point normal parameters, etc., used to accurately represent the visual features and spatial distribution of a specific location in a 3D scene. Specifically, 3D Gaussian points can exist in the form of a Gaussian point cloud, serving as the foundation for constructing a 3D Gaussian model. Each 3D Gaussian point in the Gaussian point cloud can be a Gaussian sphere, that is, a sphere, ellipsoid, or other shape with certain attribute parameters such as volume, color, and opacity.

[0113] Gaussian point location parameters are geometric parameters that describe the specific location of a 3D Gaussian point in 3D space. They are typically represented in 3D coordinate form and are used to accurately determine the location of a 3D Gaussian point in 3D space. Gaussian point location parameters provide geometric positioning information for the Gaussian point and serve as the basis for calculating visibility probability. Together with other parameters, they constitute a complete description of the 3D Gaussian point.

[0114] Gaussian point size parameters describe the distribution range or shape size of a three-dimensional Gaussian point. They are typically determined based on the variance or covariance matrix of the Gaussian point, such as the standard deviation of its principal axes. This parameter can represent the extent to which the Gaussian point extends in space in various directions. Gaussian point size parameters quantify the spatial extent of a three-dimensional Gaussian point, used to assess the size of its projection onto the imaging plane when viewed from different distances, thus affecting its visibility.

[0115] Gaussian point color parameters are parameters that describe the color information of the visual appearance of a 3D Gaussian point. They typically include color values ​​(RGB values) and transparency (Alpha values), representing the surface color and transparency characteristics of the Gaussian point. Specifically, Gaussian point color parameters can provide direct visual features of a 3D Gaussian point; areas with saturated colors are more attractive to the observer. Therefore, color parameters can be used to assess the visual importance of a 3D Gaussian point when calculating its visibility probability, and can also be used for subsequent texture analysis.

[0116] The Gaussian point normal parameter is a vector parameter describing the orientation of the local surface containing a 3D Gaussian point. It reflects the orientation of the 3D Gaussian point surface and is usually represented in a normalized 3D vector form. Specifically, the Gaussian point normal parameter can define the local orientation of a 3D Gaussian point, thereby evaluating the alignment between the Gaussian point surface and the camera's viewing direction.

[0117] The first three-dimensional Gaussian point is any three-dimensional Gaussian point in the target three-dimensional model. As the basic unit for calculating visibility probability, it is a basic component of the three-dimensional model.

[0118] The distance-size parameter is an intermediate parameter calculated based on the viewpoint distance parameter and the Gaussian point size parameter. It comprehensively reflects the combined effect of the initial viewing distance and the size of the 3D Gaussian point itself on visibility. Specifically, the distance-size parameter can transform absolute spatial distance relationships into relative relationships with the size of the 3D Gaussian point. For example, a larger 3D Gaussian point can maintain high visibility even at a greater viewing distance, while a smaller 3D Gaussian point may be indistinguishable at the same viewing distance.

[0119] The visibility parameter is an intermediate parameter calculated based on the directional deviation between the Gaussian point's normal parameter and the viewing orientation parameter. It quantifies the observable visibility of a 3D Gaussian point due to surface orientation. Specifically, the visibility parameter can simulate back occlusion phenomena in surface visibility, such as back occlusion in the Phone model. When the normal direction of the 3D Gaussian point is nearly aligned with the viewing direction (i.e., front view), its visibility parameter value is high; when the direction is opposite (i.e., back view), its visibility parameter value is low or zero. For example, the visibility parameter can be obtained by calculating the dot product of the Gaussian point's normal vector and the vector pointing from that point to the camera position, and taking the maximum value; this value can be between 0 and 1.

[0120] In practical applications, the distance and size parameters can be determined by calculating the relative size of the three-dimensional Gaussian point under the camera's viewpoint, based on the viewpoint distance parameter and the Gaussian point size parameter. Specifically, the distance and size parameters can be determined by calculating the ratio of the viewpoint distance parameter to the Gaussian point size parameter.

[0121] Visibility parameters can also be determined by calculating the dot product of the Gaussian point normal and the viewing direction vector, based on the directional deviation between the Gaussian point normal parameter and the viewing direction parameter. Specifically, different normalization methods or different dot product calculation methods can be used to adapt to different types of 3D models and viewing requirements. For example, for objects with smooth surfaces, more stringent normalization can be used to make the visibility parameters more distinguishable.

[0122] Once the distance and visibility parameters are determined, the visibility probability can be further determined by combining the Gaussian point color parameters through weighted combination or function mapping. Specifically, different weighting methods (e.g., based on color saliency, model importance, etc.) or different function forms (e.g., Gaussian decay function, linear decay function, etc.) can be used to adapt to different types of 3D models and viewpoint requirements.

[0123] For example, a specific calculation method will be used for further explanation.

[0124] Based on the view distance parameter and the Gaussian point size parameter, the distance and size parameters of the first three-dimensional Gaussian point are determined, which can be further expressed as:

[0125]

[0126] Where D is the distance dimension parameter, σ i p is a Gaussian dimension parameter. i The Gaussian point position parameters are the three-dimensional coordinates, which can include (x, y, z) coordinates, and t is the camera position parameter of the initial viewpoint.

[0127] Based on the directional deviation between the Gaussian point normal parameters and the view orientation parameters, the visibility parameters of the first 3D Gaussian point are determined, which can be further expressed as:

[0128] H = max(0,n) i ·d i,t )

[0129] Where H is the visibility parameter, n i Let d be the normal parameter of the Gaussian point. i,t The viewing angle parameter can be calculated using the following formula:

[0130]

[0131] Based on the Gaussian point's color parameters, distance and size parameters, and visibility parameters, the visibility probability of the first 3D Gaussian point under a preset number of initial viewing angles can be determined as follows:

[0132]

[0133] Among them, P i (t,R) represents the visibility probability of the i-th 3D Gaussian point at the initial viewpoint V(t,R) with camera position parameter t and viewpoint orientation parameter R, α i This refers to the transparency parameter in the Gaussian point color parameters of a 3D Gaussian point.

[0134] In the embodiments of this specification, by determining various visual parameters of the three-dimensional Gaussian point in the vertex, distance and size parameters are determined based on the view distance parameter and the Gaussian point size parameter; visibility parameters are determined based on the Gaussian point normal parameter and view orientation parameter; and further, by combining the Gaussian point color parameter, distance and size parameter, and visibility parameters, the visibility probability of the three-dimensional Gaussian point under the initial view is determined. This achieves accurate evaluation of the visibility of each three-dimensional Gaussian point in the three-dimensional model under different initial viewpoints. The Gaussian point position parameter and size parameter provide the basic geometric information for calculation, enabling the visibility evaluation to be based on accurate geometric relationships; view distance and size parameters... The size parameter quantifies the relative size of a 3D Gaussian point at a specific initial viewpoint, simulating the size change effect in actual visual perception. The visibility parameter quantifies the alignment between the surface of a 3D Gaussian point and the viewing direction, thus effectively identifying 3D Gaussian points that are visible from the front and invisible from the back. By determining the visibility probability based on the Gaussian point color parameter, distance size parameter, and visibility parameter, the accuracy and efficiency of 3D model viewpoint determination are improved, making the rapid preview and quality assessment of 3D content more accurate. This provides reliable technical support for viewpoint selection during 3D scene reconstruction and rendering, enhancing the display effect of 3D content and user experience.

[0135] In one optional embodiment of this specification, the visual parameters of the plurality of vertices include visual visibility parameters and visual feature parameters;

[0136] For the first initial viewpoint, based on the visibility probabilities of multiple vertices under the first initial viewpoint, the visibility confidence of the target 3D model under the first initial viewpoint is determined, including:

[0137] For the first initial viewpoint, based on the visual visibility parameters of multiple vertices, determine the visual weights of the visibility probabilities of multiple vertices under the first initial viewpoint;

[0138] Based on the visual feature parameters of multiple vertices, the visual texture parameters of the target 3D model under the first initial viewpoint are determined;

[0139] Based on visual weights, visual texture parameters, and the visibility probabilities of multiple vertices in the first initial viewpoint, a weighted calculation is performed to determine the visibility confidence of the target 3D model in the first initial viewpoint.

[0140] Visual visibility parameters describe the visibility characteristics of each vertex in a 3D model. They measure the visibility of each vertex from a specific initial viewpoint and can include visibility-related attributes such as the vertex's surface orientation, position, and color. Visual visibility parameters provide a quantitative description of the visibility of each vertex from a specific initial viewpoint and are used to determine the visual weight of each vertex from that viewpoint, thus affecting the calculation of visibility confidence. Optionally, taking a 3D Gaussian point as an example, visual visibility parameters can include the Gaussian point's position and color parameters, which together determine the visibility of each 3D Gaussian point from a specific initial viewpoint, thus providing basic data for determining visual weights.

[0141] Visual feature parameters are parameters that describe the visual characteristics of each vertex in a 3D model. They can include attributes related to visual performance, such as color gradient, texture, surface properties, and vertex Gaussian distribution, and can be used to characterize the visual features of each vertex. Specifically, visual feature parameters can provide visual information about vertices to determine the visual texture parameters of the target 3D model under a specific initial viewpoint, and are used together with visual weights to calculate visibility confidence.

[0142] Visual weights are weight values ​​determined based on the visual visibility parameters of each vertex. They reflect the importance and salience of each vertex in the target 3D model, quantifying the contribution of each vertex to the visibility of the target 3D model at a specific initial viewpoint. They can be used in conjunction with visibility probability and visual texture parameters to calculate visibility confidence. For example, visual weights can be determined based on the color salience and distribution location of vertices. Vertices with higher color salience have greater visual weights, and vertices closer to the initial viewpoint have higher weights. Weighted calculations can more accurately reflect the visibility of key regions in the model.

[0143] Visual texture parameters describe the visual texture characteristics of a target 3D model under a specific initial viewpoint. They characterize the texture representation of the target 3D model under that initial viewpoint and can include information such as local texture sharpness and color distribution. Visual texture parameters can be used in conjunction with visual weights and visibility probability to calculate visibility confidence.

[0144] In practical applications, visual weights are determined based on the visual visibility parameters of multiple vertices. This can be achieved by weighting the calculations based on the color salience or distribution location of each vertex, or by dynamically adjusting the weights in conjunction with the local importance of the target 3D model.

[0145] Specifically, for three-dimensional Gaussian points, the color saliency parameters of multiple three-dimensional Gaussian points can be determined first based on the color parameters of the Gaussian points. Then, the distribution location parameters of multiple three-dimensional Gaussian points in the target three-dimensional model can be determined based on the location parameters of the Gaussian points. Finally, the color saliency parameters and distribution location parameters are combined, and the visual weights are determined by weighted summation or function mapping.

[0146] The visual texture parameters of the target 3D model under the first initial viewpoint can be determined by analyzing the local color distribution of each vertex in the target 3D model, or the texture features of the overall rendered image of the target 3D model under the first initial viewpoint.

[0147] The local color distribution can be determined based on the color features of multiple vertices in the neighborhood of each vertex; the texture features can be determined based on the rendered image under the first initial viewpoint, for example by extracting the edge features and texture features of the image or by using a pre-trained texture analysis model.

[0148] In the embodiments of this specification, visual weights are determined based on visual visibility parameters of multiple vertices, and visual texture parameters are determined based on visual feature parameters of multiple vertices. Visibility confidence is then determined through a weighted calculation of visual weights, visual texture parameters, and visibility probability, achieving an accurate assessment of the visibility of a 3D model from different viewpoints. The visual visibility parameters and visual feature parameters provide basic visual information, enabling visibility confidence calculation to be based on accurate visual characteristics. Visual weights quantify the contribution of each vertex to the visibility of the target 3D model, identifying key areas within the model. Visual texture parameters quantify the texture representation of the target 3D model from the initial viewpoint, thereby simulating texture changes in actual visual perception. By weighting the visual weights, visual texture parameters, and visibility probability, the accuracy and efficiency of 3D model viewpoint determination are improved, providing reliable technical support for viewpoint selection during 3D scene reconstruction and rendering, and enhancing the display effect and user experience of 3D content.

[0149] In one optional embodiment of this specification, the vertex includes a three-dimensional Gaussian point, and the visual visibility parameters include Gaussian point position parameters and Gaussian point color parameters;

[0150] Based on the visual visibility parameters of multiple vertices, visual weights are determined for the visibility probabilities of multiple vertices under a first initial viewpoint, including:

[0151] Based on the color parameters of Gaussian points, determine the color saliency parameters of multiple three-dimensional Gaussian points;

[0152] Based on the Gaussian point location parameters, determine the distribution location parameters of multiple 3D Gaussian points in the target 3D model;

[0153] Based on color saliency parameters and distribution location parameters, the visual weights of the visibility probabilities of multiple 3D Gaussian points under the first initial viewpoint are determined.

[0154] A 3D Gaussian point is a set of 3D spatial data composed of a large number of discrete points. It is the basic unit for constructing a 3D Gaussian model. The attribute parameters of each 3D Gaussian point can include Gaussian point position parameters, Gaussian point size parameters, Gaussian point color parameters, and Gaussian point normal parameters, etc., used to accurately represent the visual features and spatial distribution of a specific location in a 3D scene. Specifically, 3D Gaussian points can exist in the form of a Gaussian point cloud, serving as the foundation for constructing a 3D Gaussian model. Each 3D Gaussian point in the Gaussian point cloud can be a Gaussian sphere, that is, a sphere, ellipsoid, or other shape with certain attribute parameters such as volume, color, and opacity.

[0155] Gaussian point location parameters are geometric parameters that describe the specific location of a 3D Gaussian point in 3D space. They are typically represented in 3D coordinate form and are used to accurately determine the location of a 3D Gaussian point in 3D space. Gaussian point location parameters provide geometric positioning information for the Gaussian point and serve as the basis for calculating visibility probability. Together with other parameters, they constitute a complete description of the 3D Gaussian point.

[0156] Gaussian point color parameters are parameters that describe the color information of the visual appearance of a 3D Gaussian point. They typically include color values ​​(RGB values) and transparency (Alpha values), representing the surface color and transparency characteristics of the Gaussian point. Specifically, Gaussian point color parameters can provide direct visual features of a 3D Gaussian point; areas with saturated colors are more attractive to the observer. Therefore, color parameters can be used to assess the visual importance of a 3D Gaussian point when calculating its visibility probability, and can also be used for subsequent texture analysis.

[0157] Gaussian point color parameters are parameters that describe the color information of the visual appearance of a 3D Gaussian point. They typically include color values ​​(RGB values) and transparency (Alpha values), representing the surface color and transparency characteristics of the Gaussian point. Specifically, Gaussian point color parameters provide direct visual features of a 3D Gaussian point. Regions with salience in color are more attractive to the observer. Therefore, color parameters can be used to assess the visual importance of a 3D Gaussian point when calculating visibility probability, and can also be used in subsequent texture analysis. Their relevance to related technical features lies in their use in determining color salience parameters, which in turn affects the calculation of visual weights.

[0158] Color saliency parameter is a quantitative index calculated based on the color parameters of Gaussian points. It describes the saliency of the color of a three-dimensional Gaussian point and measures the contrast and prominence of its color compared to the colors of other surrounding three-dimensional Gaussian points. Specifically, the color saliency parameter converts the color information of a three-dimensional Gaussian point into a comparable saliency value. The more vibrant the color and the higher the contrast with the surrounding environment, the larger the color saliency parameter value of the three-dimensional Gaussian point, indicating its greater visual importance.

[0159] Distribution location parameters are quantitative indicators calculated based on Gaussian point location parameters. They describe the spatial distribution of 3D Gaussian points within a target 3D model, representing the positional relationship of each 3D Gaussian point relative to the model center or a specific reference point (such as the camera position). Distribution location parameters quantify the spatial importance of 3D Gaussian points within the target 3D model and are used to assess their spatial strategic value. For example, 3D Gaussian points located in the central region of the outer surface of the target 3D model, or closer to the camera position at the initial viewpoint, contribute more to the overall visual perception of the target 3D Gaussian model.

[0160] In practical applications, the color saliency parameters and distribution location parameters can be determined based on the Gaussian point color parameters and Gaussian point location parameters by calculating the color difference and spatial distance.

[0161] Specifically, for the color saliency parameter, color saliency can be calculated based on the color gradient changes of each three-dimensional Gaussian point, or it can be calculated based on the color histogram of each three-dimensional Gaussian point.

[0162] For the distribution location parameters, the distribution location can be calculated based on the distance between each 3D Gaussian point and the model center, or it can be calculated based on the distance between each 3D Gaussian point and the camera position of the initial viewpoint.

[0163] Once the color saliency parameter and distribution location parameter are determined, the visibility probability visual weights of multiple three-dimensional Gaussian points under the first initial viewpoint can be further determined based on the color saliency parameter and distribution location parameter.

[0164] Specifically, calculations can be performed using linear weighted combinations, Gaussian weighted combinations, or dynamic weighting based on model importance. Weighted calculations can also be performed by combining the interaction between color saliency and distribution location.

[0165] For example, a linear weighted formula can be used to sum the color saliency parameter and the distribution location parameter to obtain the visual weight of each Gaussian point; a Gaussian function can also be used to fuse color saliency and distribution location to obtain a smoother weight distribution; or the weight coefficients of each 3D Gaussian point can be dynamically adjusted according to the key regions of the target 3D model to give the 3D Gaussian points in the key regions higher weights, thereby more accurately reflecting the visibility of the key regions in the model.

[0166] In the embodiments of this specification, color saliency parameters are determined based on Gaussian point color parameters, distribution position parameters are determined based on Gaussian point position parameters, and visual weights are further determined based on color saliency parameters and distribution position parameters. This achieves an accurate assessment of the visibility of 3D Gaussian points under a specific initial viewpoint. Specifically, color saliency parameters quantify the prominence of each 3D Gaussian point in terms of color, while distribution position parameters quantify the spatial importance of each 3D Gaussian point in the target 3D model. The visual weights determined by combining these two parameters can more comprehensively reflect the contribution of each 3D Gaussian point to the visibility of the target 3D model, improving the accuracy and efficiency of 3D model viewpoint determination. This makes rapid previewing and quality assessment of 3D content more accurate, enhancing the display effect and user experience of 3D content.

[0167] In one optional embodiment of this specification, the vertex includes a three-dimensional Gaussian point, and the visual feature parameters of the plurality of vertices include the color distribution features of the three-dimensional Gaussian point;

[0168] Based on visual feature parameters of multiple vertices, the visual texture parameters of the target 3D model under the first initial viewpoint are determined, including:

[0169] For the first three-dimensional Gaussian point, the Gaussian point texture parameters of the first three-dimensional Gaussian point are determined based on the color distribution characteristics of the three-dimensional Gaussian points included in the preset neighborhood of the first three-dimensional Gaussian point, wherein the first three-dimensional Gaussian point is any three-dimensional Gaussian point.

[0170] Based on visual weights, visual texture parameters, and the visibility probabilities of multiple vertices in the first initial viewpoint, a weighted calculation is performed to determine the visibility confidence of the target 3D model in the first initial viewpoint, including:

[0171] Based on the visual weights of the visibility probabilities of multiple 3D Gaussian points under the first initial viewpoint, the Gaussian point texture parameters of multiple 3D Gaussian points and the visibility probabilities of multiple 3D Gaussian points under the first initial viewpoint are weighted to determine the visibility confidence of the target 3D model under the first initial viewpoint.

[0172] Color distribution features are statistical parameters that describe the distribution patterns of each 3D Gaussian point and its neighborhood within a color space. They are used to characterize the color variations and distribution patterns in local areas of a target 3D model. Color distribution features can capture and characterize the microscopic texture patterns of a 3D model's surface from a color perspective. They can be used to identify areas with high color contrast and rich texture, thus providing fundamental data for determining the visual texture parameters of a target 3D model from a specific viewpoint. For example, areas with dramatic color changes may correspond to rich texture details, while areas with uniform color may correspond to smooth surfaces.

[0173] Specifically, color distribution features are quantified by calculating statistical indicators such as color gradient, color variance (patch variance), and color distribution entropy of 3D Gaussian points within a preset neighborhood. Color gradient measures the drasticness of color change, color variance measures the dispersion of the color distribution, and color distribution entropy measures the complexity of the color distribution. These color distribution features collectively reflect the richness and changing trends of colors within the neighborhood of the 3D Gaussian points, enabling a more accurate reflection of the visual characteristics of the target 3D model during the calculation of visibility confidence.

[0174] The first three-dimensional Gaussian point is any three-dimensional Gaussian point in the target three-dimensional model. As the basic unit for calculating visibility probability, it is a basic component of the three-dimensional model.

[0175] The preset neighborhood of the first 3D Gaussian point is a pre-defined spatial range surrounding the first 3D Gaussian point. It is used to determine the set of 3D Gaussian points within the neighborhood related to the first 3D Gaussian point during the calculation of its texture parameters. Specifically, the preset neighborhood can be a spherical region, a cubic region, or other shaped spatial region with a fixed radius. Its size and shape can be adjusted according to the complexity of the target 3D model and the application scenario. The preset neighborhood provides spatial contextual information for local texture analysis, ensuring that texture feature extraction is based on a meaningful local region, rather than an isolated single 3D Gaussian point.

[0176] Gaussian point texture parameters are quantitative indicators describing the visual texture characteristics of local regions of each 3D Gaussian point, and can characterize the texture performance of each Gaussian point in the target 3D model. Specifically, Gaussian point texture parameters can synthesize the color distribution characteristics within a preset neighborhood of each 3D Gaussian point into a scalar index to characterize the contribution of that 3D Gaussian point to the overall texture features of the target 3D model. Gaussian point texture, together with visual weights and visibility probability, affects the calculation result of the visibility confidence of the target 3D model under a specific initial viewpoint, so that the visibility confidence can more accurately reflect the visual quality of the target 3D model under a specific initial viewpoint.

[0177] In practical applications, the Gaussian point texture parameters of each three-dimensional Gaussian point can be determined as the visual texture parameters of the target three-dimensional model under the first initial viewpoint.

[0178] Specifically, this can be achieved by defining a predetermined neighborhood for the first 3D Gaussian point and acquiring the color information of all 3D Gaussian points within that neighborhood. Statistical characteristics of this color information can then be calculated, including color gradient (to measure the intensity of color change), color variance (to measure the dispersion of the color distribution), and color distribution entropy (to measure the complexity of the color distribution). Optionally, a weighted average can be used to combine the color gradient, color variance, and color distribution entropy into Gaussian point texture parameters, where the weights can be dynamically adjusted based on the type of the target 3D model and the actual application scenario. Alternatively, machine learning-based methods can be used to determine the Gaussian point texture parameters through a pre-trained model to adapt to different types of 3D models and visual requirements.

[0179] Once the Gaussian point texture parameters of each 3D Gaussian point are determined, the visibility confidence of the target 3D model under the first initial viewpoint can be determined based on visual weights and visibility probability.

[0180] Specifically, the visibility confidence of the target 3D model under the first initial viewpoint can be obtained by weighting and combining visual weights, Gaussian point texture parameters, and visibility probability. This can be done using a linear weighted combination method, adjusted according to the type of the target 3D model and the actual application scenario; alternatively, a non-linear weighted combination method, such as Gaussian weighted combination, can be used to give higher weights to important regions.

[0181] For example, a specific calculation method will be used for further explanation.

[0182] Based on the visual weights of the visibility probabilities of multiple 3D Gaussian points in the first initial viewpoint, the Gaussian point texture parameters of the multiple 3D Gaussian points and their visibility probabilities in the first initial viewpoint are weighted to determine the visibility confidence of the target 3D model in the first initial viewpoint, which can be further expressed as:

[0183]

[0184] Among them, P Scene (t,R) represents the spatial confidence of the target 3D model under an initial viewpoint V(t,R) with camera position parameter t and viewpoint orientation parameter R. i P is the visual weight of the visibility probability of the i-th 3D Gaussian point under the first initial viewpoint. i (t,R) represents the visibility probability of the i-th 3D Gaussian point under the first initial viewpoint, where T is the probability of visibility. i Let be the Gaussian point texture parameters for the i-th 3D Gaussian point.

[0185] In the embodiments of this specification, the Gaussian point texture parameters of the first three-dimensional Gaussian point are determined based on the color distribution characteristics of the three-dimensional Gaussian points included in the preset neighborhood of the first three-dimensional Gaussian point. These parameters are then weighted and calculated using visual weights and visibility probabilities to determine the visibility confidence of the target three-dimensional model under a specific initial viewpoint. This achieves accurate evaluation of the visual characteristics of the three-dimensional model under different viewpoints, accurately identifying key texture regions in the target three-dimensional model. This makes the visibility confidence calculation more consistent with actual visual perception. By determining the Gaussian point texture parameters, it is possible to effectively distinguish between texture-rich and texture-simple regions in the target three-dimensional model, making the determination of the target viewpoint more precise. This improves the accuracy and efficiency of viewpoint selection during three-dimensional scene reconstruction and rendering, providing reliable technical support for rapid previewing and quality assessment of three-dimensional content, and enhancing the display effect and user experience of three-dimensional content.

[0186] In one optional embodiment of this specification, the vertex includes a three-dimensional Gaussian point, and the visual feature parameters of the multiple vertices include the image texture features of the rendered image, which is obtained by rendering based on the three-dimensional Gaussian point under a first initial viewpoint;

[0187] Based on visual feature parameters of multiple vertices, the visual texture parameters of the target 3D model under the first initial viewpoint are determined, including:

[0188] Based on the image texture features of the rendered image, determine the image texture parameters of the target 3D model under the first initial viewpoint;

[0189] Based on visual weights, visual texture parameters, and the visibility probabilities of multiple vertices in the first initial viewpoint, a weighted calculation is performed to determine the visibility confidence of the target 3D model in the first initial viewpoint, including:

[0190] Based on the visual weights of the visibility probabilities of multiple 3D Gaussian points under the first initial viewpoint, the Gaussian point texture parameters of multiple 3D Gaussian points are weighted and then dot productd with the image texture parameters to determine the visibility confidence of the target 3D model under the first initial viewpoint.

[0191] A rendered image is a two-dimensional image generated using computer graphics techniques based on three-dimensional Gaussian points rendered at a specific initial viewpoint. It is used to characterize the visual appearance of a target three-dimensional model at that initial viewpoint. Specifically, it can be an orthographic projection image or a perspective projection image of the target three-dimensional model at the first initial viewpoint. From a specific viewing angle, it converts the geometric structure and visual features of the target three-dimensional model into two-dimensional visual information, providing basic data for the extraction of visual texture parameters. Optionally, the rendered image can be determined through rasterization projection in three-dimensional Gaussian sputtering, or it can be determined using methods such as neural radiation field rendering or voxel rendering.

[0192] Image texture features are statistical or structured features extracted from rendered images to characterize the local or global texture properties of an image. These features can include color gradient, color variance, edge features, texture energy, etc., and are used to describe the visual texture variation patterns in different regions of a rendered image. Image texture features can quantify the richness and distribution patterns of texture in a rendered image, thus providing a basis for evaluating the surface texture representation of a 3D model from a specific viewpoint.

[0193] Image texture parameters are quantitative metrics calculated based on the image texture features of a rendered image. They characterize the richness, sharpness, and visual information content of the overall surface texture of a target 3D model under a specific initial viewpoint. These parameters can be a single numerical value or a feature vector. Image texture parameters can aggregate multiple, potentially high-dimensional, image texture features extracted from a rendered image into a scalar or low-dimensional vector, thus providing a global texture quality metric. This metric, along with visual weights and visibility probability, is used to determine the visibility confidence of the target 3D model under that viewpoint.

[0194] In practical applications, the image texture parameters of the rendered image can be determined as the visual texture parameters of the target 3D model under the first initial viewpoint.

[0195] Specifically, the determination of image texture parameters can be based on feature extraction using image processing techniques, such as using the Sobel operator to calculate the color gradient of the rendered image, using the gray-level co-occurrence matrix to calculate texture quality, or using a pre-trained deep learning model to extract high-level texture features. Alternatively, methods based on statistical analysis can be employed, such as calculating the variance or entropy of the color distribution in the rendered image. Image segmentation techniques can also be used to divide the rendered image into multiple regions, calculate the texture features of each region separately, and then perform a weighted average. The specific methods for determining image texture parameters can be flexibly combined and determined according to the actual target 3D model or application requirements; this specification does not impose specific limitations on these methods in the embodiments.

[0196] Once the image texture parameters of the rendered image are determined, the visibility confidence of the target 3D model under the first initial viewpoint can be determined based on visual weights and visibility probability.

[0197] Specifically, the Gaussian point texture parameters of multiple 3D Gaussian points can be weighted based on the visual weights of the visibility probabilities of multiple 3D Gaussian points under the first initial viewpoint, and the weighted result can be multiplied by the image texture parameters to obtain the visibility confidence of the target 3D model under the initial viewpoint.

[0198] One approach is to treat visual weights, Gaussian point texture parameters, and image texture parameters as vectors and calculate their similarity through dot product operations to obtain the visibility confidence. Alternatively, a weighted summation method can be used, where the visual weights are summed with the products of the Gaussian point texture parameters and the image texture parameters to reflect the contribution of different factors to the visibility confidence.

[0199] For example, a specific calculation method will be used for further explanation.

[0200] Based on the visual weights of the visibility probabilities of multiple 3D Gaussian points under the first initial viewpoint, the Gaussian point texture parameters of the multiple 3D Gaussian points are weighted and then dot-producted with the image texture parameters to determine the visibility confidence of the target 3D model under the first initial viewpoint, which can be further expressed as:

[0201]

[0202] Among them, P Scene (t,R) represents the spatial confidence of the target 3D model under an initial viewpoint V(t,R) with camera position parameter t and viewpoint orientation parameter R. i P is the visual weight of the visibility probability of the i-th 3D Gaussian point under the first initial viewpoint. i (t,R) represents the visibility probability of the i-th 3D Gaussian point under the first initial viewpoint, where T is the probability of visibility. image The image texture parameters for rendering the image.

[0203] In the embodiments of this specification, image texture parameters are determined based on the image texture features of the rendered image, and the visibility confidence is calculated by performing a dot product operation with visual weights and Gaussian point texture parameters. This achieves an accurate evaluation of the visual characteristics of a 3D model from a specific viewpoint. Image texture features effectively capture the richness and distribution patterns of textures in the rendered image, making the visibility confidence calculation more consistent with actual visual perception. Image texture parameters, as quantitative indicators, can convert the texture information of the rendered image into calculable values, providing a data foundation for visibility confidence calculation. The dot product operation effectively combines visual weights, Gaussian point texture parameters, and image texture parameters to accurately reflect the visual quality of the target 3D model from a specific viewpoint. This makes the target viewpoint determination process for the target 3D model more accurate, thereby making the display of 3D content more in line with user needs. It improves the accuracy and efficiency of viewpoint selection during 3D scene reconstruction and rendering, provides reliable technical support for rapid previewing and quality assessment of 3D content, and enhances the display effect and user experience of 3D content.

[0204] In one optional embodiment of this specification, determining the target viewpoint corresponding to the target 3D model based on the visibility confidence of the target 3D model under a preset number of initial viewpoints includes:

[0205] Based on the visibility confidence of the target 3D model under a preset number of initial viewpoints, the preset number of initial viewpoints are sorted to obtain the initial viewpoint sorting results.

[0206] Based on the initial viewpoint sorting results, the target viewpoint corresponding to the target 3D model is determined from a preset number of initial viewpoints.

[0207] The sorting operation, in the 3D model viewpoint determination method, involves arranging the initial views in an ordered manner based on the visibility confidence of the target 3D model under a predetermined number of initial views. The sorting operation organizes the evaluated initial views in an orderly fashion, presenting the visibility quality of each initial view in a clear order, thus providing a direct basis for the subsequent efficient and objective selection of the target viewpoint. For example, a descending order based on the numerical value of visibility confidence can be used, placing the view with the highest visibility confidence at the top; alternatively, optimization algorithms based on the direction of visibility confidence change can be used for sorting, such as identifying local extrema of visibility confidence using the probability gradient method, thereby determining a more accurate sorting result.

[0208] The initial viewpoint ranking result is an ordered list of views obtained by sorting a preset number of initial views. These views can be arranged according to their visibility confidence level. The initial viewpoint ranking result visually displays the evaluation quality hierarchy of all candidate views, providing decision-makers or automated systems with a basis for selecting views from "best" to "worst" (or vice versa), making the determination of target views more systematic and standardized. For example, if the initial viewpoint ranking result is a list of views arranged from highest to lowest visibility confidence, the first-ranked initial viewpoint has the highest visibility confidence and can be considered the best viewpoint among the target views.

[0209] In practical applications, once the visibility confidence of the target 3D model under a preset number of initial viewpoints is determined, the preset number of initial viewpoints can be sorted based on the visibility confidence of the target 3D model under the preset number of initial viewpoints to obtain the initial viewpoint sorting results.

[0210] Specifically, the initial viewpoints can be ordered based on the visibility confidence of the target 3D model under a predetermined number of initial viewpoints. For example, a sorting method based on the magnitude of the visibility confidence can be used to arrange the initial viewpoints from high to low visibility confidence. Alternatively, the initial viewpoints can be sorted based on the direction of change of visibility confidence using optimization algorithms, such as heuristic algorithms or probabilistic gradient methods, to identify local or global extreme points of visibility confidence and determine a more accurate sorting result. Furthermore, based on the distribution of visibility confidence, clustering algorithms can be used to divide the initial viewpoints into different categories, such as best viewpoints, recommended viewpoints, and key viewpoints, and then sorting can be performed within each category to form a multi-level initial viewpoint sorting result.

[0211] Once the initial viewpoint sorting results are determined, the target viewpoint corresponding to the target 3D model can be further determined from the preset number of initial viewpoints based on the initial viewpoint sorting results.

[0212] Specifically, the initial view with the highest visibility confidence and ranked first in the initial view ranking results can be determined as the best view; alternatively, the initial views ranked 2nd to the preset ranking in the initial view ranking results can be determined as multiple recommended views, where the preset ranking can be dynamically adjusted according to the actual application scenario. For example, if a platform needs one main image and six secondary images from multiple perspectives to jointly display an item, the initial views ranked 2nd to 7th can be determined as recommended views; alternatively, a visibility confidence threshold can be set, and the initial views with visibility confidence greater than the threshold in the initial view ranking results can be determined as target views. This method can ensure that the target views contain relatively rich model information and avoid selecting views with poor visibility.

[0213] In this embodiment, the initial perspectives are ranked based on their visibility confidence under a preset number of initial viewpoints to obtain a ranking result. The target perspective is then determined based on this ranking result, achieving accurate evaluation and ranking of potential display perspectives for the 3D model. The ranking systematically arranges the initial perspectives according to their visibility confidence, providing a clear basis for subsequent target perspective selection. The ranking result clearly demonstrates the visibility advantages and disadvantages of each initial perspective, making the determination of the target perspective more objective and reliable. By combining multiple ranking methods and target perspective determination methods, the system can adapt to the needs of different application scenarios, such as optimal perspective display and multi-view recommendation, making the display of 3D content more in line with user needs. This improves the accuracy and efficiency of perspective selection during 3D scene reconstruction and rendering, provides reliable technical support for rapid previewing and quality assessment of 3D content, and enhances the display effect and user experience of 3D content.

[0214] In one optional embodiment of this specification, based on the visibility confidence of the target 3D model under a preset number of initial viewpoints, the preset number of initial viewpoints are sorted to obtain the initial viewpoint sorting result, including:

[0215] By optimizing the algorithm, the initial viewpoints are sorted based on the direction of change of the visibility confidence of the target 3D model under a preset number of initial viewpoints, and the initial viewpoint sorting results are obtained. The optimization algorithm includes heuristic algorithms or probabilistic gradient methods.

[0216] Optimization algorithms are a set of mathematical methods used to find the optimal solution of an objective function in the view space. They optimize the objective value step by step through iterative calculation and parameter adjustment, efficiently locating local or global maximum points of visibility confidence and avoiding the computational overhead of blind searches. Specifically, optimization algorithms can include heuristic algorithms or probabilistic gradient methods. Heuristic algorithms can further include Monte Carlo sampling, particle swarm optimization, greedy search, and other algorithms.

[0217] In practical applications, optimization algorithms can analyze the variation pattern of visibility confidence in the view space and find local maxima in the visibility confidence field of a preset number of initial viewpoints, thereby intelligently guiding the search path and improving sorting efficiency.

[0218] The direction of change refers to the trend of visibility confidence changing with initial viewpoint parameters in three-dimensional view space. It describes the increase or decrease of visibility confidence at different viewpoint positions, providing guidance for the search direction of the optimization algorithm, enabling the algorithm to search efficiently along the direction of increasing visibility confidence. Specifically, the direction of change is related to the gradient calculation of visibility confidence. By calculating the partial derivative of visibility confidence with respect to viewpoint parameters, the direction of change can be determined, thus judging which direction to move in at the current initial viewpoint to increase the visibility confidence of the initial viewpoint.

[0219] Heuristic algorithms are optimization methods based on problem characteristics and empirical rules. They guide the search process by simulating natural phenomena or using domain knowledge to design heuristic rules. In the embodiments of this specification, heuristic algorithms can quickly guide the algorithm to find an initial viewpoint with high visibility confidence during the determination of the target viewpoint of a 3D model, by utilizing known characteristics of the 3D model and viewpoint distribution patterns, thus avoiding getting trapped in local optima.

[0220] The probabilistic gradient method is an optimization approach that combines probabilistic models and gradient calculations. It guides the search process by estimating the gradient direction of the objective function and incorporating probability distributions. In the embodiments described in this specification, the probabilistic gradient method can effectively handle noise and uncertainty in the viewpoint space during the determination of the target viewpoint in a 3D model. It determines the direction of change in visibility confidence more smoothly through probability weighting. Specifically, the probabilistic gradient method treats each initial viewpoint as a candidate solution to be evaluated. By modeling the probability distribution of viewpoint ranking and utilizing the gradient information of visibility confidence relative to viewpoint parameters, it iteratively adjusts the sampling strategy to generate a ranking result that corresponds to a high-probability, high-quality viewpoint sequence.

[0221] In practical applications, when using optimization algorithms to sort based on the direction of change, once the visibility confidence of the target 3D model in the initial viewpoint set is determined, the direction of change of the visibility confidence can be determined by calculating the gradient of the visibility confidence in the viewpoint space. This direction of change can indicate the viewpoint movement path for the increase in visibility confidence. Furthermore, optimization algorithms can be used to search along the determined direction of change, and by iteratively adjusting the viewpoint parameters, the local or global maximum value of the visibility confidence can be gradually found, thereby sorting the initial viewpoints.

[0222] Specifically, this sorting process can combine gradient information of changing directions and search strategies of optimization algorithms to dynamically adjust the search step size and direction, improve sorting efficiency, and ensure that the sorting results can accurately reflect the visibility of the target 3D model from different perspectives, providing a high-quality sorting basis for subsequent determination of the target perspective.

[0223] In the embodiments of this specification, by utilizing an optimization algorithm to sort the initial viewpoints based on the changing direction of visibility confidence, the accurate evaluation and sorting of potential display viewpoints for 3D models is achieved, improving the efficiency and accuracy of viewpoint determination. Specifically, the optimization algorithm, combined with the analysis of the changing direction, can identify local extrema of visibility confidence, avoiding the additional computational overhead caused by blind searching, making the sorting process more intelligent and efficient. Optimized sorting based on the mathematical properties of visibility confidence makes the determination of target viewpoints more consistent with actual visual perception, providing reliable technical support for rapid previewing and quality assessment of 3D content. This effectively improves the accuracy and efficiency of viewpoint selection during 3D scene reconstruction and rendering, making the display of 3D content more in line with user needs and enhancing the user experience.

[0224] Corresponding to the above method embodiments, this specification also provides embodiments of a three-dimensional model perspective determination device, see [link to documentation]. Figure 2 , Figure 2 A schematic diagram of a three-dimensional model perspective determination device according to one embodiment of this specification is shown. Figure 2 As shown, the device includes:

[0225] The acquisition module 202 is configured to acquire the spatial parameters of the target 3D model and a preset number of initial viewpoints, wherein the target 3D model includes multiple vertices;

[0226] The visibility probability determination module 204 is configured to determine the visibility probability of a first vertex under a preset number of initial viewing angles based on the visual parameters of the first vertex and the spatial parameters of a preset number of initial viewing angles, wherein the first vertex is any vertex.

[0227] The visibility confidence determination module 206 is configured to determine the visibility confidence of the target 3D model in the first initial viewpoint based on the visibility probabilities of multiple vertices in the first initial viewpoint, wherein the first initial viewpoint is any initial viewpoint;

[0228] The target viewpoint determination module 208 is configured to determine the target viewpoint corresponding to the target 3D model based on the visibility confidence of the target 3D model under a preset number of initial viewpoints.

[0229] The acquisition module is a data input component in the 3D model perspective determination device. It can be used to acquire the spatial parameters of the target 3D model and a preset number of initial perspectives, providing the basic data required for the 3D model perspective determination process. For example, the target 3D model can be acquired by reading from the local storage system, downloading from the cloud server, or dynamically generating the data. At the same time, the spatial parameters of the initial perspectives can be acquired to provide the necessary input for subsequent calculations.

[0230] The visibility probability determination module is a visibility calculation component in the 3D model viewpoint determination device. It can be used to determine the visibility probability of a first vertex under a preset number of initial viewpoints, based on the visual parameters of the first vertex and the spatial parameters of a preset number of initial viewpoints. By calculating the visibility probability of each vertex under different initial viewpoints, it provides basic data for the visibility assessment of the entire 3D model.

[0231] The visibility confidence determination module is an evaluation and calculation component in the 3D model viewpoint determination device. It can be used to determine the visibility confidence of the target 3D model from a first initial viewpoint, based on the visibility probabilities of multiple vertices in the first initial viewpoint. By integrating the visibility probabilities of multiple vertices and performing weighted calculations, the visibility confidence of the target 3D model from a specific viewpoint is determined, serving as the basis for target viewpoint determination.

[0232] The target viewpoint determination module is a decision output component in the 3D model viewpoint determination device. It can be used to determine the target viewpoint corresponding to the target 3D model based on the visibility confidence of the target 3D model under a preset number of initial viewpoints. Through a sorting and filtering mechanism, it determines the best or recommended viewpoint from the initial viewpoint set, providing the user with the target viewpoint.

[0233] The 3D model viewpoint determination device provided in the embodiments of this specification efficiently acquires the target 3D model and initial viewpoint parameters through an acquisition module, accurately calculates the visibility probability of vertices under different viewpoints through a visibility probability determination module, comprehensively evaluates the visibility of the model under a specific viewpoint through a visibility confidence determination module, and intelligently sorts and determines the optimal viewpoint through a target viewpoint determination module. This achieves accurate evaluation and sorting of potential display viewpoints for 3D models without relying on original images or camera pose information. It can adaptively determine the target viewpoint based on the vertex distribution characteristics of the 3D model and through probability field superposition, improving the accuracy and efficiency of viewpoint selection during 3D scene reconstruction and rendering. It provides reliable technical support for rapid previewing and quality assessment of 3D content, making the display of 3D content more in line with user needs and enhancing the user experience.

[0234] Optionally, the spatial parameters of the initial viewpoint include camera position parameters, and the visual parameters of the vertex include vertex position parameters; the visibility probability determination module 204 is further configured to: for the first vertex, determine the viewpoint distance parameter and viewpoint orientation parameter based on the vertex position parameter of the first vertex and the camera position parameters of a preset number of initial viewpoints; and determine the visibility probability of the first vertex under a preset number of initial viewpoints based on the viewpoint distance parameter and viewpoint orientation parameter.

[0235] Optionally, the vertex includes a three-dimensional Gaussian point, and the visual parameters of the vertex include Gaussian point position parameters, Gaussian point size parameters, Gaussian point color parameters, and Gaussian point normal parameters; the visibility probability determination module 204 is further configured to: determine the distance and size parameters of a first three-dimensional Gaussian point based on the view distance parameters and the Gaussian point size parameters, wherein the first three-dimensional Gaussian point is any three-dimensional Gaussian point; determine the visibility parameters of the first three-dimensional Gaussian point based on the directional deviation between the Gaussian point normal parameters and the view orientation parameters; and determine the visibility probability of the first three-dimensional Gaussian point under a preset number of initial viewpoints based on the Gaussian point color parameters, distance and size parameters, and visibility parameters.

[0236] Optionally, the visual parameters of the multiple vertices include visual visibility parameters and visual feature parameters; the visibility confidence determination module 206 is further configured to: for a first initial viewpoint, determine the visual weights of the visibility probabilities of the multiple vertices under the first initial viewpoint based on the visual visibility parameters of the multiple vertices; determine the visual texture parameters of the target 3D model under the first initial viewpoint based on the visual feature parameters of the multiple vertices; and perform a weighted calculation based on the visual weights, visual texture parameters, and the visibility probabilities of the multiple vertices under the first initial viewpoint to determine the visibility confidence of the target 3D model under the first initial viewpoint.

[0237] Optionally, the vertices include three-dimensional Gaussian points, and the visual visibility parameters include Gaussian point position parameters and Gaussian point color parameters;

[0238] The confidence determination module 206 is further configured to: determine the color saliency parameters of multiple three-dimensional Gaussian points based on the color parameters of the Gaussian points; determine the distribution position parameters of multiple three-dimensional Gaussian points in the target three-dimensional model based on the position parameters of the Gaussian points; and determine the visual weights of the visibility probabilities of multiple three-dimensional Gaussian points in the first initial viewpoint based on the color saliency parameters and the distribution position parameters.

[0239] Optionally, the vertices include three-dimensional Gaussian points, and the visual feature parameters of multiple vertices include the color distribution features of the three-dimensional Gaussian points; the visibility confidence determination module 206 is further configured to: for the first three-dimensional Gaussian point, determine the Gaussian point texture parameters of the first three-dimensional Gaussian point based on the color distribution features of the three-dimensional Gaussian points included in the preset neighborhood of the first three-dimensional Gaussian point, wherein the first three-dimensional Gaussian point is any three-dimensional Gaussian point; and based on the visual weights of the visibility probabilities of multiple three-dimensional Gaussian points under the first initial viewpoint, weight the Gaussian point texture parameters of multiple three-dimensional Gaussian points and the visibility probabilities of multiple three-dimensional Gaussian points under the first initial viewpoint to determine the visibility confidence of the target three-dimensional model under the first initial viewpoint.

[0240] Optionally, the vertices include three-dimensional Gaussian points, and the visual feature parameters of multiple vertices include the image texture features of the rendered image, which is obtained by rendering based on the three-dimensional Gaussian points under the first initial viewpoint; the visibility confidence determination module 206 is further configured to: determine the image texture parameters of the target three-dimensional model under the first initial viewpoint based on the image texture features of the rendered image; and weight the Gaussian point texture parameters of multiple three-dimensional Gaussian points based on the visual weights of the visibility probabilities of multiple three-dimensional Gaussian points under the first initial viewpoint, and perform a dot product operation with the image texture parameters to determine the visibility confidence of the target three-dimensional model under the first initial viewpoint.

[0241] Optionally, the target viewpoint determination module 208 is further configured to: sort the preset number of initial views based on the visibility confidence of the target 3D model under the preset number of initial views, and obtain the initial viewpoint sorting result; and determine the target viewpoint corresponding to the target 3D model from the preset number of initial views based on the initial viewpoint sorting result.

[0242] Optionally, the target viewpoint determination module 208 is further configured to: sort the preset number of initial viewpoints based on the direction of change of the visibility confidence of the target 3D model under the preset number of initial viewpoints through an optimization algorithm, and obtain the initial viewpoint sorting result, wherein the optimization algorithm includes a heuristic algorithm or a probability gradient method.

[0243] The above is a schematic scheme of a three-dimensional model perspective determination device according to this embodiment. It should be noted that the technical solution of this three-dimensional model perspective determination device and the technical solution of the three-dimensional model perspective determination method described above belong to the same concept. For details not described in detail in the technical solution of the three-dimensional model perspective determination device, please refer to the description of the technical solution of the three-dimensional model perspective determination method described above.

[0244] Figure 3 A structural block diagram of a computing device 300 according to one embodiment of this specification is shown. The components of the computing device 300 include, but are not limited to, a memory 310 and a processor 320. The processor 320 is connected to the memory 310 via a bus 330, and a database 350 is used to store data.

[0245] The computing device 300 also includes an access device 340, which enables the computing device 300 to communicate via one or more networks 360. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 340 may include one or more of any type of wired or wireless network interface (e.g., a network interface controller (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0246] In one embodiment of this specification, the aforementioned components of the computing device 300 and Figure 3 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 3 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0247] The computing device 300 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 300 can also be a mobile or stationary server.

[0248] The processor 320 is used to execute the following computer program / instruction, which, when executed by the processor, implements the steps of the above-described three-dimensional model perspective determination method.

[0249] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above-described method for determining the perspective of a three-dimensional model belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-described method for determining the perspective of a three-dimensional model.

[0250] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the above-described three-dimensional model perspective determination method.

[0251] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the above-described three-dimensional model perspective determination method belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-described three-dimensional model perspective determination method.

[0252] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described three-dimensional model perspective determination method.

[0253] The above is an illustrative example of a computer program according to this embodiment. It should be noted that the technical solution of this computer program and the technical solution of the above-described method for determining the perspective of a 3D model belong to the same concept. Details not described in detail in the computer program's technical solution can be found in the description of the technical solution of the above-described method for determining the perspective of a 3D model.

[0254] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0255] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0256] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0257] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0258] The preferred embodiments disclosed above are merely illustrative of this specification. Optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A method for determining the perspective of a three-dimensional model, characterized in that, include: Obtain spatial parameters of the target 3D model and a preset number of initial viewpoints, wherein the target 3D model includes multiple vertices; For a first vertex, based on the visual parameters of the first vertex and the spatial parameters of the preset number of initial viewing angles, the visibility probability of the first vertex under the preset number of initial viewing angles is determined, wherein the first vertex is any vertex; For a first initial viewpoint, based on the visibility probabilities of the plurality of vertices under the first initial viewpoint, the visibility confidence of the target 3D model under the first initial viewpoint is determined, wherein the first initial viewpoint is any initial viewpoint; Based on the visibility confidence of the target 3D model under the preset number of initial viewpoints, the target viewpoint corresponding to the target 3D model is determined.

2. The method according to claim 1, characterized in that, The spatial parameters of the initial viewpoint include camera position parameters, and the visual parameters of the vertex include vertex position parameters; The step of determining the visibility probability of a first vertex under the preset number of initial viewing angles, based on the visual parameters of the first vertex and the spatial parameters of the preset number of initial viewing angles, includes: For the first vertex, based on the vertex position parameters of the first vertex and the camera position parameters of the preset number of initial viewpoints, the viewpoint distance parameters and viewpoint orientation parameters are determined. Based on the view distance parameter and the view orientation parameter, the visibility probability of the first vertex under the preset number of initial viewpoints is determined.

3. The method according to claim 2, characterized in that, The vertex includes a three-dimensional Gaussian point, and the visual parameters of the vertex include Gaussian point position parameters, Gaussian point size parameters, Gaussian point color parameters, and Gaussian point normal parameters. Determining the visibility probability of the first vertex under the preset number of initial viewing angles based on the viewing distance parameter and the viewing orientation parameter includes: Based on the view distance parameter and the Gaussian point size parameter, the distance size parameter of the first three-dimensional Gaussian point is determined, wherein the first three-dimensional Gaussian point is any three-dimensional Gaussian point; Based on the directional deviation between the Gaussian point normal parameters and the view orientation parameters, the visibility parameters of the first three-dimensional Gaussian point are determined. Based on the Gaussian point color parameter, the distance size parameter, and the visibility parameter, the visibility probability of the first three-dimensional Gaussian point under the preset number of initial viewing angles is determined.

4. The method according to claim 1, characterized in that, The visual parameters of the plurality of vertices include visual visibility parameters and visual feature parameters; The step of determining the visibility confidence of the target 3D model from the first initial viewpoint, based on the visibility probabilities of the multiple vertices from the first initial viewpoint, includes: For a first initial viewpoint, based on the visual visibility parameters of the multiple vertices, determine the visual weights of the visibility probabilities of the multiple vertices under the first initial viewpoint; Based on the visual feature parameters of the multiple vertices, the visual texture parameters of the target 3D model under the first initial viewpoint are determined; Based on the visual weights, the visual texture parameters, and the visibility probabilities of the multiple vertices under the first initial viewpoint, a weighted calculation is performed to determine the visibility confidence of the target 3D model under the first initial viewpoint.

5. The method according to claim 4, characterized in that, The vertex includes a three-dimensional Gaussian point, and the visual visibility parameters include Gaussian point position parameters and Gaussian point color parameters; The visual weights for determining the visibility probabilities of the multiple vertices under the first initial viewpoint based on the visual visibility parameters of the multiple vertices include: Based on the color parameters of the Gaussian points, the color saliency parameters of multiple three-dimensional Gaussian points are determined. Based on the Gaussian point position parameters, determine the distribution position parameters of the plurality of three-dimensional Gaussian points in the target three-dimensional model; Based on the color saliency parameter and the distribution location parameter, the visual weights of the visibility probabilities of the plurality of three-dimensional Gaussian points under the first initial viewpoint are determined.

6. The method according to claim 4, characterized in that, The vertices include three-dimensional Gaussian points, and the visual feature parameters of the plurality of vertices include the color distribution features of the three-dimensional Gaussian points; The determination of the visual texture parameters of the target 3D model under the first initial viewpoint based on the visual feature parameters of the multiple vertices includes: For a first three-dimensional Gaussian point, based on the color distribution characteristics of the three-dimensional Gaussian points included in the preset neighborhood of the first three-dimensional Gaussian point, the Gaussian point texture parameters of the first three-dimensional Gaussian point are determined, wherein the first three-dimensional Gaussian point is any three-dimensional Gaussian point; The step of determining the visibility confidence of the target 3D model under the first initial viewpoint by performing a weighted calculation based on the visual weights, the visual texture parameters, and the visibility probabilities of the multiple vertices under the first initial viewpoint includes: Based on the visual weights of the visibility probabilities of the plurality of three-dimensional Gaussian points under the first initial viewpoint, the Gaussian point texture parameters of the plurality of three-dimensional Gaussian points and the visibility probabilities of the plurality of three-dimensional Gaussian points under the first initial viewpoint are weighted to determine the visibility confidence of the target three-dimensional model under the first initial viewpoint.

7. The method according to claim 4, characterized in that, The vertex includes a three-dimensional Gaussian point, and the visual feature parameters of the plurality of vertices include the image texture features of the rendered image, which is obtained by rendering based on the three-dimensional Gaussian point under the first initial viewpoint; The determination of the visual texture parameters of the target 3D model under the first initial viewpoint based on the visual feature parameters of the multiple vertices includes: Based on the image texture features of the rendered image, determine the image texture parameters of the target 3D model under the first initial viewpoint; The step of determining the visibility confidence of the target 3D model under the first initial viewpoint by performing a weighted calculation based on the visual weights, the visual texture parameters, and the visibility probabilities of the multiple vertices under the first initial viewpoint includes: Based on the visual weights of the visibility probabilities of the plurality of three-dimensional Gaussian points under the first initial viewpoint, the Gaussian point texture parameters of the plurality of three-dimensional Gaussian points are weighted and a dot product operation is performed with the image texture parameters to determine the visibility confidence of the target three-dimensional model under the first initial viewpoint.

8. The method according to any one of claims 1-7, characterized in that, The step of determining the target viewpoint corresponding to the target 3D model based on the visibility confidence of the target 3D model under the preset number of initial viewpoints includes: Based on the visibility confidence of the target 3D model under the preset number of initial viewpoints, the preset number of initial viewpoints are sorted to obtain the initial viewpoint sorting result; Based on the initial viewpoint sorting results, the target viewpoint corresponding to the target 3D model is determined from the preset number of initial viewpoints.

9. The method according to claim 8, characterized in that, The process of ranking the preset number of initial viewpoints based on the visibility confidence of the target 3D model under the preset number of initial viewpoints to obtain the initial viewpoint ranking result includes: By using an optimization algorithm, the preset number of initial viewpoints are sorted based on the direction of change of the visibility confidence of the target 3D model under the preset number of initial viewpoints to obtain the initial viewpoint sorting result. The optimization algorithm includes a heuristic algorithm or a probability gradient method.

10. A three-dimensional model perspective determination device, characterized in that, include: The acquisition module is configured to acquire spatial parameters of a target 3D model and a preset number of initial viewpoints, wherein the target 3D model includes multiple vertices; The visibility probability determination module is configured to determine the visibility probability of a first vertex under the preset number of initial viewing angles, based on the visual parameters of the first vertex and the spatial parameters of the preset number of initial viewing angles, wherein the first vertex is any vertex. The visibility confidence determination module is configured to determine the visibility confidence of the target 3D model in the first initial viewpoint based on the visibility probability of the plurality of vertices in the first initial viewpoint, wherein the first initial viewpoint is any initial viewpoint; The target viewpoint determination module is configured to determine the target viewpoint corresponding to the target 3D model based on the visibility confidence of the target 3D model under the preset number of initial viewpoints.

11. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the three-dimensional model perspective determination method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, It stores a computer program / instruction that, when executed by a processor, implements the steps of the three-dimensional model perspective determination method according to any one of claims 1 to 9.

13. A computer program product, characterized in that, Includes a computer program / instruction that, when executed by a processor, implements the steps of the three-dimensional model perspective determination method according to any one of claims 1 to 9.

Images