Method for constructing foreign matter model of nuclear power plant foreign matter digital museum

Abstract

The invention belongs to the technical field of three-dimensional model modeling, and particularly relates to a method for constructing a foreign body model of a nuclear power plant foreign body digital museum. Comprising the following steps: step 1, collecting close-range image data of a target foreign body, generating point cloud data by using feature matching and a neural radiation field, calculating a normal vector of each point, and segmenting the point cloud data through a region growing algorithm to obtain a point cloud model of the target foreign body; 2, generating a surface grid model of the foreign matter by adopting a Poisson reconstruction algorithm, and finishing and optimizing the grid model by analyzing the solid structure of the foreign matter to obtain a refined foreign matter fine model; and 3, generating high-fidelity texture mapping by adopting a method based on texture reconstruction and rendering in combination with foreign matter surface features and close-range image data, and rendering illumination and reflection effects for the model to realize visual optimization. The method has the beneficial effects that the high-fidelity three-dimensional model of the foreign matter can be efficiently generated, and the damaged area can be effectively repaired.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional model modeling, and particularly relates to a method for constructing a foreign object model of a digital museum of foreign objects in a nuclear power plant. Background Art

[0002] During the operation of this nuclear power plant, various foreign objects may appear, and these foreign objects may affect the safety and operation efficiency of equipment. In order to effectively manage and maintain these foreign objects, a detailed digital model needs to be established. Traditional three-dimensional modeling usually uses modeling software such as 3dsMax, Google Sketchup, Solidworks, and CATIA, and manually models based on the contour information of foreign objects estimated from CAD two-dimensional vector maps, image data, or manually taken photos.

[0003] The models produced by traditional three-dimensional model construction methods have beautiful appearances, but have many defects such as low precision, a large amount of manual participation required during the production process, and a long production cycle.

[0004] 1. Precision Limitations

[0005] In scenarios that require high-precision modeling, such as the engineering design field, the precision of 3dsMax may not be sufficient. It focuses more on the presentation of visual effects and is relatively weak in the precise control of dimensions.

[0006] 2. Data Compatibility Issues

[0007] There may be incompatibility or loss of some information in the data exchange with some professional engineering software. For example, when performing data interaction with a specific CAD software, problems such as loss of model details and inability to fully transfer material properties may occur.

[0008] 3. High Learning Cost

[0009] Solidworks is powerful, but for beginners, the learning curve is relatively steep. It contains numerous tools and functions, and it takes a certain amount of time and effort to become familiar with and master.

[0010] 4. High System Resource Occupation

[0011] CATIA is a software with extremely powerful functions, but it also requires a high hardware configuration. When running large and complex models, it may occupy a large amount of system memory and processor resources, resulting in slow computer operation or even freezing.

[0012] 5. Complex Operation

[0013] Due to the complexity of its functions, the operation interface and process of CATIA are relatively complex. For some simple modeling tasks, it may seem overly cumbersome. Summary of the Invention

[0014] The object of the present invention is to provide a method for constructing a foreign object model in a digital museum of foreign objects in nuclear power plants, which can solve the problems of high human resource requirements and low accuracy in the prior art based on traditional modeling techniques.

[0015] The technical solution of the present invention is as follows: A method for constructing a foreign object model in a digital museum of foreign objects in nuclear power plants includes the following steps:

[0016] Step 1: Collect the close-range image data of the target foreign object, generate point cloud data by feature matching and neural radiance fields, calculate the normal vector of each point, and segment the point cloud data by the region growing algorithm to obtain the point cloud model of the target foreign object;

[0017] Step 2: Use the Poisson reconstruction algorithm to generate the surface mesh model of the foreign object, and trim and optimize the mesh model through the analysis of the entity structure of the foreign object to obtain a refined foreign object fine model;

[0018] Step 3: Adopt a method based on texture reconstruction and rendering, combine the surface features of the foreign object and the close-range image data, generate a high-fidelity texture mapping, and render the lighting and reflection effects for the model to achieve visual optimization.

[0019] The process of segmenting the point cloud data is as follows:

[0020] Step 101: Use the feature matching algorithm to match the common feature points in the multi-view images to generate three-dimensional point cloud data;

[0021] Step 102: Improve the density and accuracy of the point cloud through the neural radiance field point cloud generation method, and calculate the normal vector for each point cloud to describe the geometric information of the foreign object surface;

[0022] Step 103: Use the region growing algorithm to set a threshold, segment the generated point cloud data according to the point cloud normal vector, separate the foreign object from the background information, and finally obtain the point cloud model of the target foreign object.

[0023] In step 101, the specific process of feature matching uses the Euclidean distance calculation, and the formula is as follows:

[0024]

[0025] where d i，k and d j，k are the descriptors of the feature points in two images respectively, and n is the feature dimension.

[0026] The formula for the neural radiance field point cloud generation process in step 102 is as follows:

[0027]

[0028] Among them, σ is the density function, T(t) is the transmittance, and c(r(t)) is the color value.

[0029] In the said step 2, the generation process of the foreign object fine model includes the following:

[0030] Step 201: Use the Poisson reconstruction algorithm to perform three-dimensional meshing on the generated point cloud data to generate an initial three-dimensional mesh model;

[0031] Step 202: Detect the damaged or missing areas of the initial mesh model, and use an automatic repair algorithm to repair the damaged or missing areas of the initial mesh model.

[0032] The Poisson equation used in the Poisson reconstruction algorithm in the said step 201 is as follows:

[0033]

[0034] Among them, Φ(x) is the implicit surface function, and ρ(x) is the discretized density of the point cloud.

[0035] In the said step 202, the process of automatic repair of the mesh model is as follows:

[0036] Step 2201: Identify the damaged or missing areas in the mesh model through normal vector discontinuity detection, boundary detection, and vertex sparsity detection;

[0037] Step 2202: Use a hole filling algorithm to generate new mesh faces to fill the damaged or missing areas;

[0038] Step 2203: Perform mesh Laplacian smoothing on the repaired areas to ensure geometric continuity.

[0039] In the said step 3, the texture reconstruction and rendering process is as follows:

[0040] Step 301: By capturing the geometric features and multi-view image data of the foreign object surface, extract and match feature points, generate a seamlessly integrated texture map, and then map these maps onto the surface of the three-dimensional model to ensure texture clarity and surface consistency;

[0041] Step 302: Perform UV unwrapping on the model, unfold the model surface into a two-dimensional plane, generate a high-resolution texture map corresponding to the surface details, and then combine the normal map and displacement map to enhance the surface details and generate multi-resolution textures to optimize the rendering performance;

[0042] Step 303: Use physically based rendering method, combined with normal, roughness, and metallicity maps, to simulate real lighting and reflection effects. Use ambient occlusion and ray tracing methods to further enhance the realism of lighting and shadows, so that the model presents a highly realistic visual effect.

[0043] The lighting effect is calculated by the Phong lighting model, and the formula is as follows:

[0044] I = I a + I d (L·N) + I s (R·V) n ;

[0045] Where, I a is the ambient light, I d is the diffuse light, I s is the specular reflection light, L is the light source direction, N is the surface normal vector, R is the reflection direction, V is the viewing direction, and n is the specular reflection exponent.

[0046] The beneficial effects of the present invention are as follows: The present invention aims to provide a method for generating a foreign object model in a digital museum of foreign objects in a nuclear power plant. By processing multi-view images, combining technologies such as neural radiance fields, region growing algorithms, Poisson reconstruction, and ray tracing rendering, it realizes fine three-dimensional modeling and visual optimization of foreign objects in the nuclear power plant. This method can efficiently generate high-fidelity three-dimensional models of foreign objects and effectively repair damaged areas. Detailed implementation manners

[0047] The following further elaborates the present invention in detail with reference to specific embodiments.

[0048] A method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant includes the following steps:

[0049] Step 1: Collect the close-range image data of the target foreign object, generate point cloud data by using feature matching and neural radiance field methods, calculate the normal vector of each point, and segment the foreign object point cloud through the region growing algorithm to obtain the point cloud model of the target foreign object;

[0050] A further improvement of the technical solution of the present invention is that in the above Step 1, the segmentation process of the foreign object point cloud is as follows:

[0051] Step 101: Use the feature matching algorithm to match the common feature points in the multi-view images to generate three-dimensional point cloud data;

[0052] Step 102: Improve the density and accuracy of the point cloud through the neural radiance field point cloud generation method, and calculate the normal vector for each point cloud to describe the geometric information of the foreign object surface;

[0053] Step 103: Set a threshold using the region growing algorithm, segment the generated point cloud data based on the point cloud normal vectors, separate the foreign object and background information, and finally obtain the point cloud model of the target foreign object.

[0054] A further improvement of the technical solution of the present invention is that the specific process of the feature matching algorithm is to calculate using the Euclidean distance, and the formula is as follows:

[0055]

[0056] where d i，k and d j，k are the descriptors of the feature points in two images respectively, and n is the feature dimension;

[0057] The formula for the generation process of the neural radiance field point cloud is as follows:

[0058]

[0059] where σ is the density function, T(t) is the transmittance, and c(t(t)) is the color value.

[0060] Step 2: Use the Poisson reconstruction algorithm to generate the surface mesh model of the foreign object, and through the analysis of the entity structure of the foreign object, trim and optimize the mesh model to obtain a refined foreign object fine model;

[0061] In the said Step 2, the generation process of the foreign object fine model includes the following:

[0062] Step 201: Use the Poisson reconstruction algorithm to perform three-dimensional meshing on the generated point cloud data to generate an initial three-dimensional mesh model;

[0063] Step 202: Detect the damaged or missing areas of the initial mesh model, and use an automatic repair algorithm to repair the damaged or missing areas of the initial mesh model.

[0064] A further improvement of the technical solution of the present invention is that the Poisson equation used in the Poisson reconstruction algorithm in Step 201 is as follows:

[0065]

[0066] where Φ(x) is the implicit surface function and ρ(x) is the discretized density of the point cloud.

[0067] A further improvement of the technical solution of the present invention is that in the said Step 202, the process of automatically repairing the mesh model is:

[0068] Step 2201: Identify the damaged or missing areas in the mesh model through normal vector discontinuity detection, boundary detection, and vertex sparsity detection;

[0069] Step 2202: Use the hole filling algorithm to generate a new mesh surface to fill the damaged or missing areas;

[0070] Step 2203: Perform mesh Laplacian smoothing on the repaired area to ensure geometric continuity.

[0071] Step 3: Adopt a method based on texture reconstruction and rendering, combine the surface features of foreign objects and the near-infrared image data, generate a high-fidelity texture mapping, and render the lighting and reflection effects for the model to achieve visual optimization.

[0072] In the said Step 3, the texture reconstruction and rendering process is as follows:

[0073] Step 301: By capturing the geometric features and multi-view image data of the surface of the foreign object, extract and match the feature points, generate a seamlessly integrated texture map, and then map these maps onto the surface of the 3D model to ensure texture clarity and surface consistency;

[0074] Step 302: Unwrap the UV of the model, unfold the model surface into a two-dimensional plane, generate a high-resolution texture map corresponding to the surface details, and then combine the normal map and displacement map to enhance the surface details, generate multi-resolution textures to optimize the rendering performance;

[0075] Step 303: Use physically based rendering (PBR) technology, combine normal, roughness, and metallicity maps, simulate real lighting effects and reflection effects, use ambient occlusion and ray tracing technology to further enhance the realism of lighting and shadows, and make the model present a highly realistic visual effect.

[0076] PBR technology is a rendering technology based on physical laws. When rendering, it uses algorithms and materials based on physical rules to interact. Different from ordinary materials that use a simple lighting model (diffuse reflection, specular highlight, ambient light), this patent uses physical parameters to describe the characteristics of the object surface based on PBR technology, such as core parameters like metallicity and roughness. By directly specifying the physical characteristics of the foreign object into the PBR material, high-quality rendering is achieved.

[0077] Ray tracing technology is a technology that renders by simulating the physical propagation path of light in a virtual scene. In this invention patent, the hardware ray tracing function is directly enabled through the ray tracing interface provided by the graphics API. When constructing the rendering scene, a ray tracing acceleration structure is simultaneously constructed, and ray tracing is used for lighting rendering during the rendering process to achieve a near-real lighting effect.

[0078] A further improvement of the technical solution of the present invention lies in: the lighting effect is calculated by the Phong lighting model, and the formula is as follows:

[0079] I = Ia +I d (L·N)+I s (R·V) n ;

[0080] Among them, I a is the ambient light, I d is the diffuse reflection light, I s is the specular reflection light, L is the light source direction, N is the surface normal vector, R is the reflection direction, V is the viewing direction, and n is the specular reflection index.

[0081] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention.

[0082] Embodiment 1:

[0083] The present invention provides a method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant. First, multi-view close-range image data collection and point cloud generation of the target foreign object need to be carried out, including the following steps:

[0084] Step 101: Image collection

[0085] In the actual scenario of foreign objects, there are usually factors such as different ambient light conditions, surface materials, sizes, and shapes of foreign objects, which will all affect the image data collection. Therefore, when collecting data, the multi-view image collection technology is used to maximize the coverage of all surface information of the target foreign object.

[0086] The specific collection process can be carried out by using a drone or a hand-held camera device to take pictures of the target foreign object from multiple different angles. It is recommended to take pictures during a period with moderate lighting conditions to ensure clear image data and rich surface details. If the foreign object is in a dark or uneven lighting environment, a fill light device can also be used to improve the collection effect. In addition, multi-view collection can also be performed by an unmanned aerial vehicle with an automated path planning to improve the collection efficiency and accuracy.

[0087] Step 102: Feature matching and point cloud generation

[0088] After the multi-view image data collection is completed, these images need to be processed for feature matching. The feature matching algorithm realizes the registration between images of different views by extracting common feature points in the images. This process first extracts the key points of the images, and then calculates the similarity through descriptors, calculating the feature points in each image and the matching points in other images. The specific Euclidean distance calculation formula is as follows:

[0089]

[0090] Among them, di，k and d j，k are the descriptors of the feature points in two images respectively, and n is the feature dimension.

[0091] After the feature point matching is completed, the 3D point cloud data can be generated by triangulation to form the basic 3D geometric structure of the foreign object.

[0092] Step 103: Optimize point cloud generation using neural radiance fields

[0093] After the initial point cloud generation is completed, the neural radiance field technology can be used to further optimize the point cloud data to improve its density and accuracy. The neural radiance field inputs a series of 2D images into a neural network, learns the 3D scene information corresponding to these images, and then generates a high-precision 3D point cloud. The basic formula of the neural radiance field is as follows:

[0094]

[0095] where σ is the density function, T(t) is the transmittance, and c(r(t)) is the color value. This formula describes the color accumulation value of light when passing through different positions, and then simulates a high-precision 3D point cloud.

[0096] Step 104: Segment the point cloud using the region growing algorithm

[0097] The generated point cloud data can be segmented by the region growing algorithm to further distinguish the foreign object from the background information. During the segmentation process, first, a certain threshold is set according to the normal vector of the point cloud, and then the region growing algorithm is used to detect the connected regions in the point cloud. The region growing algorithm starts from the initial seed points and merges the neighboring points into the region according to the consistency of the normal vector and the distance threshold, and finally forms connected regions. Through this process, the point cloud model of the foreign object can be segmented, and the background clutter data can be removed.

Claims

1. A method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant, characterized in that, It includes the following steps: Step 1: Collect the close-range image data of the target foreign object, generate point cloud data using feature matching and neural radiance fields, calculate the normal vector of each point, and segment the point cloud data through the region growing algorithm to obtain the point cloud model of the target foreign object; Step 2: Use the Poisson reconstruction algorithm to generate the surface mesh model of the foreign object, and through the analysis of the entity structure of the foreign object, trim and optimize the mesh model to obtain a refined fine model of the foreign object; Step 3: Adopt a method based on texture reconstruction and rendering, combine the surface features of the foreign object and the close-range image data, generate a high-fidelity texture mapping, and render the lighting and reflection effects for the model to achieve visual optimization.

2. The method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant according to claim 1, wherein In the said Step 1, the process of segmenting the point cloud data is as follows: Step 101: Use the feature matching algorithm to match the common feature points in the multi-view images to generate three-dimensional point cloud data; Step 102: Improve the density and accuracy of the point cloud through the neural radiance field point cloud generation method, and calculate the normal vector for each point cloud to describe the geometric information of the foreign object surface; Step 103: Use the region growing algorithm to set a threshold, segment the generated point cloud data according to the point cloud normal vector, separate the foreign object from the background information, and finally obtain the point cloud model of the target foreign object.

3. The method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant according to claim 2, wherein: The specific process of feature matching in Step 101 uses the Euclidean distance calculation, and the formula is as follows: where d i，k and d j，k are the descriptors of the feature points in two images respectively, and n is the feature dimension.

4. The method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant according to claim 2, characterized in that: The formula for the neural radiance field point cloud generation process in Step 102 is as follows: Where, σ is the density function, T(t) is the transmittance, and c(r(t)) is the color value.

5. The method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant according to claim 1, characterized in that, In the said Step 2, the generation process of the foreign object fine model includes the following: Step 201: Use the Poisson reconstruction algorithm to perform three-dimensional meshing on the generated point cloud data to generate an initial three-dimensional mesh model; Step 202: Detect the damaged or missing areas of the initial mesh model, and use the automatic repair algorithm to repair the damaged or missing areas of the initial mesh model.

6. The method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant according to claim 5, characterized in that: The Poisson equation used in the Poisson reconstruction algorithm in Step 201 is as follows: Where, Φ(x) is the implicit surface function, and ρ(x) is the discretized density of the point cloud.

7. The method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant according to claim 5, wherein: In the said Step 202, the process of automatically repairing the mesh model is as follows: Step 2201: Identify the damaged or missing areas in the mesh model through normal vector discontinuity detection, boundary detection, and vertex sparsity detection; Step 2202: Use the hole filling algorithm to generate new mesh faces to fill the damaged or missing areas; Step 2203: Perform mesh Laplacian smoothing on the repaired areas to ensure geometric continuity.

8. The method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant according to claim 1, wherein, In the said Step 3, the texture reconstruction and rendering process is as follows: Step 301: By capturing the geometric features of the foreign object surface and the multi-view image data, extract and match the feature points, generate a seamlessly integrated texture map, and then map these maps onto the surface of the three-dimensional model to ensure texture clarity and surface consistency; Step 302: Unwrap the UV of the model, unfold the model surface into a two-dimensional plane, generate a high-resolution texture map corresponding to the surface details, and then combine the normal map and displacement map to enhance the surface details, and generate multi-resolution textures to optimize the rendering performance; Step 303: Use physically based rendering methods, combined with normal, roughness, and metalness maps, to simulate realistic lighting and reflection effects. Use ambient occlusion and ray tracing methods to further enhance the realism of lighting and shadows, making the model present a highly realistic visual effect.

9. The method for constructing a foreign object model in a digital museum of foreign objects in a nuclear power plant according to claim 8, wherein The lighting effect is calculated using the Phong lighting model, and the formula is as follows: I = I a +I d (L·N) + I s (R·V) n ; Among them, I a is the ambient light, I d is the diffuse reflection light, I s is the specular reflection light, L is the light source direction, N is the surface normal vector, R is the reflection direction, V is the viewing direction, and n is the specular reflection exponent.

Images