Digital human duplicate holographic display system based on skeleton driving and gridding algorithm

Abstract

The invention provides a digital human duplicate holographic display system based on skeleton driving and a gridding algorithm, and belongs to the technical field of holographic display. The system is composed of a 3D modeling module, a data conversion module, a point cloud denoising module, a holographic generation module and an optical reconstruction module. Wherein the 3D modeling module is used for constructing a Bip skeleton system, binding a 3D character model and setting a key frame to generate an action sequence; the data conversion module exports the action sequence as a glb file and converts the glb file into pcd point cloud data; the point cloud denoising module adopts a bilateral filtering algorithm to denoise the pcd point cloud; the holographic generation module imports the denoised pcd file into a TRS-PCG system to generate a holographic image; the optical reconstruction module optically reconstructs the hologram to present a digital human avatar holographic image. According to the method, the defects of a depth camera are avoided through skeleton driving action sequence generation, multiple functions are integrated to improve the holographic generation efficiency, and under the resolution of 1920 * 1080, the digital human holographic display speed and quality are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the fields of holographic display and three-dimensional computational imaging, and relates to a digital human avatar holographic display system based on bone driving and meshing algorithms. Background Art

[0002] With the rapid development of virtual reality (VR), augmented reality (AR) and metaverse technologies, holographic display technology has become a research hotspot because it can provide an immersive three-dimensional visual experience. Since traditional holographic display technology mainly relies on depth cameras to collect human motion data, there are many limitations in the holographic images generated by point cloud modeling. The main manifestations are as follows: ① Limitations in data acquisition of depth cameras: Depth cameras are prone to data loss, noise interference and edge blurring problems in complex lighting or occlusion scenarios, resulting in insufficient accuracy of three-dimensional point cloud reconstruction and affecting the action authenticity and detail expressiveness of digital avatars; ② Complexity of three-dimensional model construction: Existing technologies usually need to obtain three-dimensional information of objects through multi-view stereo vision or structured light scanning, with a cumbersome process and high equipment costs. In addition, action capture based on depth cameras requires real-time processing of a large amount of data, with high computational complexity and difficulty in meeting the requirements of real-time holographic display; ③ Computational bottleneck of holographic generation algorithms: Traditional holographic generation algorithms (such as Wavefront-Recording Plane (WRP), PCG, etc.) rely on point-by-point calculation of light field diffraction, with a large amount of calculation and high memory occupancy. To solve the problems existing in the prior art, the present invention proposes a digital human avatar holographic display system based on bone driving and meshing algorithms, which constructs a virtual bone framework to replace depth cameras to collect actions, and combines the Taylor Rayleigh-Sommerfeld point cloud meshing algorithm (TRS-PCG) to optimize the holographic generation efficiency, providing a new solution for real-time and high-fidelity holography. Summary of the Invention

[0003] Aiming at the problems of the prior art, the purpose of the present invention is to propose a technology for generating an action sequence driven by bones. By constructing a virtual Biped Skeleton bone system (Bip bone) and binding a 3D model, and using keyframe animation, the occlusion error of depth cameras in complex scenes is avoided. In the holographic generation link, the TRS-PCG algorithm approximates the radial value of Fresnel diffraction through a Taylor expansion (retaining up to the quadratic term), and combines parallel computing of point clouds with multiple depth layers to increase the holographic generation efficiency by more than 47%.

[0004] In order to achieve the above invention purpose, the technical solution adopted by the present invention is as follows:

[0005] A digital human avatar holographic display system based on bone driving and meshing algorithms, applied to a digital human avatar display system, is characterized in that it includes:

[0006] S1. Model acquisition and processing: Construct a Bip skeleton in 3D modeling software and bind it to a 3D human model. Generate an action sequence containing joint kinematic parameters by manually setting key frames.

[0007] S2. Data conversion: Export the action sequence as a glb format file and convert it to high-density pcd point cloud data using a point cloud generation tool.

[0008] S3. Point cloud denoising: Use the bilateral filtering algorithm to denoise the pcd point cloud and retain the model edge features.

[0009] S4. Hologram generation: Import the converted pcd format file into the Taylor Rayleigh - Sommerfeld Point Cloud Gridding (TRS-PCG) system. Optimize the relevant algorithm parameters of TRS-PCG according to the point cloud data characteristics in the pcd format file to generate high-quality holographic images.

[0010] S5. Optical reconstruction: Perform optical reconstruction on the hologram to finally achieve the holographic display of the digital human clone.

[0011] Preferably, the method of step S1 is as follows: In 3D modeling software (3ds Max 2023), first create a Biped skeleton system (Bip skeleton) and set joint constraints, including the rotation ranges of the shoulder joint of ±90°, the knee joint of 0 - 180°, and the ankle joint of ±45°. Subsequently, use the Skin modifier to bind the skeleton system to the 3D human model, and manually adjust the skin weights through the Vertex Paint tool to ensure that the skin deformation error is <0.02 mm when the joint moves. Manually set 12 - 24 key frames in the Track View to generate an action sequence containing kinematic parameters such as joint angles and displacements, such as walking, running, and jumping.

[0012] Preferably, the digital conversion method of step S2 is as follows: In step S2, the action sequence designed in 3D modeling software is saved frame by frame as a glb format file through the "Export" function. This file contains information such as bone hierarchy, vertex positions, and UV maps. Write Python code based on the Open3D library to achieve the conversion from glb to pcd: Use read_triangle_mesh to read the glb file, uniformly sample 1.5 million high-density pcd points through sample_points_uniformly, with a spatial resolution of 0.5 mm, and finally save it as a pcd format through write_point_cloud.

[0013] Preferably, the specific steps of the point cloud denoising method in step S3 are as follows:

[0014] Step S31, determination of bilateral filtering parameters: Determine the parameters of the bilateral filtering algorithm according to the features of the model in the pcd point cloud data, where the parameters include the spatial domain standard deviation σ s and the range domain standard deviation σ r , the spatial domain standard deviation σ s is used to control the size of the neighborhood during filtering, and the range domain standard deviation σ r is used to control the sensitivity to the difference in gray values during filtering;

[0015] Step S32, bilateral filtering denoising step: Use the bilateral filtering algorithm with the determined parameters to perform denoising processing on the pcd point cloud, and retain the edge features of the model while removing noise. The specific filtering formula is: where p i is the original value of the i-th point in the point cloud data, p i ′ is the value of the i-th point after filtering, N i is the neighborhood of the i-th point, w ij is the weight coefficient, which is obtained by multiplying the spatial domain weight w s and the range domain weight w r , that is, w ij = w s (d ij )×w r (|p i -p j |), d ij is the spatial distance between the i-th point and the j-th point,

[0016] Preferably, the hologram generation method in step S4 is specifically as follows:

[0017] Step S41, hierarchical processing: Divide the point cloud into L layers of parallel sub-grids according to the depth value (L≥50), and extract the gray value data of each layer to form a two-dimensional matrix P l (x, y)(l = 1, 2,..., L);

[0018] Step S42, Fresnel diffraction calculation: Perform Fresnel diffraction calculation on each layer of the point cloud, and the calculation formula is: where k = 2π / λ is the wave number, λ is the laser wavelength, and z is the propagation distance;

[0019] Step S43, Taylor expansion acceleration: Perform Taylor expansion on the radial value and retain the quadratic term: Combine the two-dimensional fast Fourier transform (2D-FFT) to reduce the computational complexity from O(N 2 ) to O(NlogN);

[0020] Step S44, conjugate gradient method optimization: Convert the optical field propagation equation into a sparse matrix form \(Au = b\), and solve it iteratively: where \(r\) n \(= b - Au\) n is the residual, and \(p\) n is the conjugate direction;

[0021] Step S45, phase superposition: Calculate the hologram \(H\) layer by layer l \(= \angle(U\) l ), and the final hologram is where \(\varphi\) l is the phase compensation factor for the depth layer.

[0022] Preferably, in the S41 layering process step, the grid resolution of each layer is set to 1920×1080, and it can also be adjusted according to the spatial light modulator.

[0023] Preferably, in the S43 Taylor expansion acceleration step, by selecting \(x_0 = x'\), the expansion error \(\delta\leq0.03\) mm is satisfied, and:

[0024] Preferably, in the S44 conjugate gradient method optimization step, by constructing a preconditioner matrix \(M\approx A\) -1 , the number of iterations is reduced by more than 40%.

[0025] As a preference, the specific method of optical reconstruction is: A holographic display system based on a single red light is constructed. The system uses a 638 nm red laser as the light source. The laser passes through an attenuator, a beam expander, and an aperture in sequence for beam shaping and then is projected onto a spatial light modulator (SLM). The SLM is connected to a personal computer through a USB interface and loads a monochromatic hologram. The light field modulated by the SLM is finally irradiated by the laser beam to generate a red three-dimensional holographic reconstruction image.

[0026] The beneficial effects of the present invention are:

[0027] (1) Compared with the holographic display technology based on a depth camera, the present invention has achieved breakthroughs in terms of computational efficiency, reconstruction accuracy, and environmental adaptability, providing a more feasible solution for the commercial application of holographic technology.

[0028] (2) The present invention uses a virtual bone framework to replace the depth camera, eliminates the influence of light and occlusion on data acquisition, improves data integrity, reduces the dependence on hardware such as depth cameras, and reduces the system deployment cost by more than 30%. It is applicable to mobile devices or resource-constrained environments.

[0029] (3) The present invention adopts a skeleton-driven action sequence generation technology to precisely control the model actions through joint kinematic parameters, avoiding the occlusion error of the depth camera, with a reconstruction error as low as 0.03 mm. The skeleton-driven modeling technology is built based on the 3ds Max Biped skeleton system, binding the model through a skin modifier and setting key frames to ensure high-precision synchronization of joint movement and skin deformation.

[0030] (4) Based on the TRS-PCG algorithm with Taylor expansion, at a resolution of 1024×1024, the GPU running time is only 0.39 seconds (the traditional WRP method requires 8.13 seconds), and the speedup ratio reaches 20.8 times; at a resolution of 2048×2048, the GPU time of TRS-PCG is 2.23 seconds (the traditional PCG method is 5.74 seconds), and the speedup ratio is 2.57 times, achieving a multiple increase in the hologram generation speed. Description of the Drawings

[0031] Figure 1 It is a flowchart of the steps of the digital human avatar holographic display system based on skeleton driving and meshing algorithm of the present invention;

[0032] Figure 2 It is a diagram of the skeleton construction of the 3D character model in the embodiment of the present invention;

[0033] Figure 3 It is the skinned image of the 3D character model in the embodiment of the present invention;

[0034] Figure 4 It is the walking action image of the 3D character model in the embodiment of the present invention;

[0035] Figure 5 It is the running action image of the 3D character model in the embodiment of the present invention;

[0036] Figure 6 It is the jumping action image of the 3D character model in the embodiment of the present invention;

[0037] Figure 7 It is the TRS-PCG flowchart in the embodiment of the present invention;

[0038] Figure 8 It is the original hologram generation diagram of the 3D character model action in the embodiment of the present invention;

[0039] Figure 9 It is the hologram generation diagram of the 3D character model action in the embodiment of the present invention;

[0040] Figure 10 It is the imaging diagram of optical reconstruction at distances of 20 mm, 40 mm, and 60 mm in the embodiment of the present invention. Detailed Embodiment

[0041] To clarify the content of the present invention in detail, the following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that the embodiments are the preferred solutions of the present invention, aiming to illustrate the implementation conditions that can be used to implement the present invention, rather than limiting the experimental conditions.

[0042] The technical solutions in the embodiments of the present invention will be described below with reference to the accompanying drawings of the specification.

[0043] As Figure 1 shown, this embodiment is a digital human avatar holographic display system based on bone drive and meshing algorithm, which specifically includes the following steps:

[0044] First aspect: Method for collecting action data of digital human model

[0045] The present invention provides a method for collecting action data of a digital human model, which is applied to 3D modeling and model processing devices. The method includes: In 3D modeling software (3ds Max 2023), install the corresponding software plug-ins and dependency libraries, and then construct a bip bone system and bind it to the 3D human model. Generate an action sequence containing joint kinematic parameters by manually setting key frames. The electronic device uses the built-in script interface of the software to extract data such as joint angles and displacements in each frame of the action sequence, save it to a local file in the glb data format, and at the same time save the initial state information of the 3D model (including vertex coordinates, texture information, etc.) together to provide a complete data basis for subsequent data processing.

[0046] Second aspect: Method for converting action data format

[0047] The present invention provides a method for converting action data format, which is applied to electronic devices. The method includes:

[0048] Use the point cloud generation tool Open3D to process the glb file and convert it into high-density pcd point cloud data. During the conversion process, by setting appropriate sampling parameters, ensure that the generated pcd point cloud data can accurately reflect the action characteristics and geometric shape of the digital human model. At the same time, perform a preliminary quality check on the converted pcd point cloud data, such as checking the integrity and density uniformity of the point cloud. If problems are found, corresponding repairs and adjustments are made.

[0049] Third aspect: Method for denoising point cloud data

[0050] The present invention provides a method for denoising point cloud data, which is applied to electronic devices. The method includes:

[0051] The electronic device loads the converted pcd point cloud data and performs noise reduction processing on it using the bilateral filtering algorithm. When implementing the bilateral filtering, according to the characteristics and application requirements of the pcd point cloud data, the parameters of the bilateral filtering algorithm are reasonably set, such as the standard deviation in the spatial domain and the standard deviation in the value domain. By continuously adjusting the parameters and observing the filtering effect, while removing noise, the edge features of the model are retained to ensure that the detailed information of the point cloud data is not lost. To verify the denoising effect, the point cloud quality evaluation index is used to compare and analyze the point cloud data before and after denoising to ensure that the denoised point cloud data meets the requirements of subsequent hologram generation.

[0052] Fourth aspect: Holographic image generation method

[0053] The present invention provides a holographic image generation method, which is applied to an electronic device. The method includes:

[0054] The electronic device imports the denoised pcd format file into the TRS-PCG system. The system first conducts a detailed analysis of the point cloud data features in the pcd format file, including the density distribution, spatial range, curvature change, etc. of the point cloud. According to these features, the relevant algorithm parameters of the TRS-PCG system are optimized and set, such as the number of layers for hierarchical processing, the depth interval of each layer, the grid resolution, etc., as well as the relevant parameters of the Taylor expansion and the parameters of the conjugate gradient method (PCG). During the parameter optimization process, using the preview function provided by the system, the holographic image generation effect under different parameter settings is observed in real time, and the parameters are continuously adjusted according to the preview results until a satisfactory effect is achieved. After completing the parameter optimization, click the "Generate" button of the system, and the system will process the point cloud data in the pcd format according to the set parameters to generate a high-quality holographic image. During the generation process, the system will display a progress bar in real time to prompt the generation progress and record the key information during the generation process, such as the calculation time of each layer, the error value, etc., for subsequent analysis and optimization.

[0055] Fifth aspect: Holographic image optical reconstruction method

[0056] The present invention provides a holographic image optical reconstruction method, which is applied to an electronic device. The method includes:

[0057] The electronic device transmits the generated holographic image data to the hologram optical reconstruction platform. The electronic device loads the holographic image onto the spatial light modulator and modulates the phase and amplitude of the laser by controlling the spatial light modulator, so that the laser can reproduce the three-dimensional shape of the object after modulation. The camera captures and records the modulated laser, thereby realizing the optical reconstruction of the holographic image.

[0058] The specific implementation steps are as follows:

[0059] 3D Character Model Acquisition and Processing Method: Obtain 3D character models from the AIUNI website. When obtaining, according to specific requirements, such as the style, gender, body type of the model, etc., select a suitable model. Download the obtained model to a specified local folder. Note that the model download format should be Original High Model (.glb) for convenient subsequent operations. After the model is downloaded, start the 3ds Max 2023 software. Select the "Import" option in the "File" menu. In the pop-up file selection dialog box, select the 3D character model file to complete the model import operation. At the same time, ensure that the correct scale factor and unit are selected to ensure the accuracy of the model size.

[0060] In the 3ds Max 2023 software, enter the "Create" panel and select the "Biped" tool under the "Systems" category. Click at an appropriate position in the view window to create a bip bone system. According to the principles of human anatomy, precisely adjust the positions of the bone nodes to ensure they match the body structure of the model. At the same time, appropriate joint rotation ranges need to be set to simulate the movement mode of a real human body. By continuously adjusting the bone weights, using the weight adjustment tools in the software, such as the "Paint Weights" function, when the model performs actions, the skin can naturally follow the movement of the bones, avoiding situations such as skin penetration or unnatural deformation. The 3D character model bone construction is Figure 2 .

[0061] After completing the bone setup, perform the skinning operation. Select the model and the built bones, and execute the "Skin" modifier command. In the parameter settings of the "Skin" modifier, add the bones to the skinning list through the "Add" button. Use the "Paint Weights" tool to precisely distribute the skin weights according to the influence area of the bones on the skin, so that when the model walks, runs, etc., the skin deformation conforms to the human movement logic. After 3D character skinning and weight adjustment, as Figure 3 .

[0062] In the "Motion" panel of 3ds Max 2023, turn on the "Auto Key" mode. Taking the walking action as an example, at the 0th frame, adjust the model pose to the starting walking action; move the time slider to the 6th frame, and according to the pose of a real human walking, adjust the bending degree of the model's legs, the swinging amplitude of the arms, etc., to simulate the leg-lifting action of taking a step. Continue to move the time slider to the 12th frame and adjust the model pose to the state of the other foot stepping out, and so on, to complete the design of a complete walking cycle action. The walking, running, and jumping actions of the 3D character model are successively as Figure 4 、 Figure 5 、 Figure 6 .

[0063] Similarly, for the running motion, focus on reflecting the forward inclination angle of the body, the rapid alternating movement of the legs, and the swinging rhythm of the arms during running; for the jumping motion, precisely design the model postures at stages such as takeoff, mid-air, and landing, including the stretching and bending of the body and the movement changes of the legs.

[0064] After the motion design is completed, select the "Export" option in the "File" menu, select the glb format in the file type, and then export each key-frame motion frame by frame.

[0065] When converting the point cloud, ensure that the Python environment is installed on the computer. Open the command-line tool and use the command to install the required Open3D library: pip install open3d numpy

[0066] After the installation is complete, open a text editor, create a new Python file, such as glb_to_pcd.py. Write the following code in the file and perform the corresponding operations:

[0067] First, specify the path of the 3D model file. For example, define a string variable path_glb to store the path of the glb file. Use the o3d.io.read_triangle_mesh() function to read the glb file at the specified path and load it as a triangular mesh object. To ensure the correct lighting effect, call the mesh.compute_vertex_normals() function to calculate and update the normal vectors of each vertex in the mesh, which is very important for subsequent rendering and lighting calculations.

[0068] Then, use the o3d.visualization.draw_geometries() function to visualize the mesh object in the window to view the read and processed 3D model. Next, generate point cloud data by uniformly sampling from the mesh surface. Extract 1.5 million points uniformly from the mesh surface through the pcd = mesh.sample_points_uniformly(number_of_points = 1500000) method and save these points to the pcd variable. Use the o3d.visualization.draw_geometries() function to visualize the generated point cloud object in the window. Specify the path to save the point cloud file. For example, define a string variable save_path to store the path of the pcd point cloud file to be saved.

[0069] Next, to ensure that all necessary directories in the save path already exist, use os.makedirs(os.path.dirname(save_path), exist_ok=True) to create all necessary directories in the save path.

[0070] Finally, call the o3d.io.write_point_cloud(save_path, pcd) function to write the point cloud data to the pcd file at the specified path.

[0071] The complete code is as follows:

[0072] import open3d as o3d

[0073] import os

[0074] path_glb = ("Actual glb file path")

[0075] mesh = o3d.io.read_triangle_mesh(path_glb)

[0076] mesh.compute_vertex_normals()

[0077] o3d.visualization.draw_geometries([mesh])

[0078] pcd = mesh.sample_points_uniformly(number_of_points = 1500000)

[0079] o3d.visualization.draw_geometries([pcd])

[0080] save_path = "Path where you hope to save the pcd file"

[0081] os.makedirs(os.path.dirname(save_path), exist_ok=True)

[0082] o3d.io.write_point_cloud(save_path, pcd)

[0083] Point cloud preprocessing mainly involves calling the bilateral filtering function of Open3D and setting σ S = 0.2, σ R= 0.05, and the number of iterations is; 5 Histogram equalization is used to enhance the contrast, and the gray - level dynamic range is extended to 0 - 255 levels.

[0084] Hologram generation method. The hologram generation method of the present invention relies on the TRS - PCG system. Open the TRS - PCG system. In the file import interface of the system, select the pcd - format file obtained by previous conversion. The system will automatically load the point - cloud data and display the preview effect of the point cloud in the interface.

[0085] As Figure 7 shown in the flowchart of TRS - PCG, according to the characteristics of the point - cloud data in the pcd file, optimize the algorithm parameters of the TRS - PCG system. By adjusting the parameters of point - cloud meshing, the point cloud can be divided into grids more reasonably, improving the quality of hologram generation; optimize the relevant parameters of the Taylor expansion to balance the calculation speed and the accuracy of the hologram. When adjusting the parameters, the preview function provided by the system can be used to observe the hologram generation effect under different parameter settings in real - time. Figure 8 is the state of the original hologram image after preliminary processing by the TRS - PCG system. Continuously adjust the parameters according to the preview results until a satisfactory effect is achieved.

[0086] After completing the parameter optimization, click the "Generate" button of the TRS - PCG system. The system will process the point - cloud data in pcd format according to the set parameters to generate a hologram. During the generation process, the system will display a progress bar to indicate the generation progress. After generation is completed, the system will display the original hologram. The walking, running, and jumping of hologram generation are as Figure 9 .

[0087] Using optical reconstruction, write an SLM driver interface program to support a 60Hz refresh rate. Achieve synchronous triggering of the laser and the SLM through GPIO pins to realize the holographic display of the intelligent digital human clone, as Figure 10 are the imaging diagrams at distances of 20mm, 40mm, and 60mm in sequence. When the distance is 40mm, the imaging effect is the best.

[0088] As described above, it is only the optimal implementation mode in the present invention, but the protection scope of the present invention is not limited thereto. Any transformation or replacement that can be understood and conceived by those skilled in the art within the technical scope disclosed by the present invention should be covered within the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A digital human avatar holographic display system based on bone drive and meshing algorithm, which is applied to the digital human avatar display system, and is characterized in that, It includes the following steps: S1. Model acquisition and processing: Construct a Bip skeleton in 3D modeling software and bind it to a 3D human model, and generate an action sequence containing joint kinematic parameters by manually setting key frames; S2. Data conversion: Export the action sequence as a glb format file and convert it into high-density pcd point cloud data using a point cloud generation tool; S3. Point cloud denoising: Use a bilateral filtering algorithm to denoise the pcd point cloud and retain the model edge features; S4. Hologram generation: Import the converted pcd format file into the TRS-PCG system, optimize the relevant algorithm parameters of the TRS-PCG according to the point cloud data characteristics in the pcd format file, and generate a high-quality holographic image; S5. Optical reconstruction: Optically reconstruct the hologram to finally realize the holographic display image of the digital human clone.

2. The system according to claim 1, wherein In the above S1, in the 3D modeling software, first create a Bip skeleton and set joint constraints, including the rotation ranges of the shoulder joint of ±90°, the knee joint of 0-180°, and the ankle joint of ±45°; then use the skin modifier to bind the Bip skeleton system to the 3D human model, and manually adjust the skin weights through the Vertex Paint tool to ensure that the skin deformation error is <0.02 mm when the joint moves; manually set 12-24 key frames to generate an action sequence containing kinematic parameters such as joint angles and displacements, such as walking, running, and jumping.

3. The system according to claim 1, characterized in that, In the above S2, the action sequence designed in the 3D modeling software is saved frame by frame as a glb format file through the "export" function. This file contains information such as bone hierarchy, vertex positions, and UV maps; write Python code based on the Open3D library to achieve the conversion from glb to pcd: use read_triangle_mesh to read the glb file, uniformly sample 1.5 million high-density pcd points through sample_points_uniformly, with a spatial resolution of 0.5 mm, and finally save it as a pcd format through write_point_cloud.

4. The system according to claim 1, wherein The point cloud denoising method in the above S3 includes the following steps: Step S31, determination of bilateral filtering parameters: Determine the parameters of the bilateral filtering algorithm according to the features of the model in the pcd point cloud data, where the parameters include the spatial domain standard deviation σ s and the range domain standard deviation σ r , the spatial domain standard deviation σ s is used to control the size of the neighborhood during filtering, and the range domain standard deviation σ r is used to control the sensitivity to the difference in gray values during filtering; Step S32, bilateral filtering denoising step: Use the bilateral filtering algorithm with determined parameters to perform noise reduction on the pcd point cloud, while removing noise and retaining the edge features of the model. The specific filtering formula is: where p i is the original value of the i-th point in the point cloud data, p i ′ is the value of the i-th point after filtering, N i is the neighborhood of the i-th point, w ij is the weight coefficient, which is obtained by multiplying the spatial domain weight w s and the range domain weight w r , that is, w ij = w s (d ij ) × w r (|p i - p j |), d ij is the spatial distance between the i-th point and the j-th point.

5. The system according to claim 4, wherein The spatial standard deviation σ s has a value range of 1.5 - 2.5, and the value range of the value domain standard deviation σ r is 0.03 - 0.07; the neighborhood N i is a spherical neighborhood centered on the i-th point with a radius of r, where the radius r is dynamically adjusted according to the density of the point cloud data and the complexity of the model.

6. The system according to claim 1, wherein The hologram generation in the above S4 includes the following steps: Step S41, layering process: The point cloud is divided into L layers of parallel sub-grids according to the depth value, where L≥50, and the grayscale value data is extracted from each layer to form a two-dimensional matrix P l (x, y) (l = 1, 2,..., L); Step S42, Fresnel diffraction calculation: Perform Fresnel diffraction calculation on each layer of point cloud, and the calculation formula is: where k = 2π / λ is the wave number, λ is the laser wavelength, and z is the propagation distance; Step S43, Taylor expansion acceleration: Perform Taylor expansion on the radial value and retain the quadratic term: Combine with two-dimensional fast Fourier transform to reduce the computational complexity from O(N 2 ) to O(NlogN); Step S44, conjugate gradient method optimization: Convert the optical field propagation equation into a sparse matrix form Au = b, and solve it iteratively: u n+1 = u n + α n p n , where r n = b - Au n is the residual, and p n is the conjugate direction; Step S45, phase superposition: calculate the hologram H layer by layer l = angle(U l ), and the final hologram is: where φ l is the phase compensation factor for the depth layer.

7. The system according to claim 6, characterized in that, In the layering processing step of the above S41, the grid resolution of each layer is set to 1920×1080, and it can also be adjusted according to the spatial light modulator.

8. The system according to claim 6, wherein In the S43 Taylor expansion acceleration step, by selecting x0 = x′, the expansion error δ ≤ 0.03 mm, satisfying:

9. The system according to claim 6, characterized in that, In the S44 conjugate gradient method optimization step, by constructing a preconditioner matrix M≈A -1 , the number of iterations is reduced by more than 40%.

10. The system according to claim 1, wherein The specific method of the optical reconstruction in the above S5 is: Construct a holographic display system based on a single red light. The system uses a 638 nm red laser as the light source. The laser passes through an attenuator, a beam expander, and an aperture in sequence for beam shaping and then projects onto the spatial light modulator. The spatial light modulator is connected to a personal computer through a USB interface and loads a monochromatic hologram. The light field modulated by the spatial light modulator is finally irradiated by the laser beam to generate a red three-dimensional holographic reconstruction image.

Images