CN-121033276-B - Laser radar-based live-action three-dimensional achievement product display method and system

CN121033276BCN 121033276 BCN121033276 BCN 121033276BCN-121033276-B

Abstract

The invention belongs to the technical field of three-dimensional models and discloses a method and a system for displaying a three-dimensional achievement product based on a laser radar live-action, wherein the method comprises the steps of scanning a target product by adopting three-dimensional scanning equipment to obtain three-dimensional data of the live-action of the product, carrying out multi-step denoising, layered collaborative hole filling and optimization on the three-dimensional data to obtain preprocessed three-dimensional data, constructing a three-dimensional model of the target product by utilizing preprocessed three-dimensional data through three-dimensional modeling software, creating a virtual display environment in a virtual reality engine, importing the three-dimensional model into the environment, designing an interaction function for the virtual reality scene, and receiving user feedback data in real time to optimize the three-dimensional model of the target product.

Inventors

Tie qinxiu
CHAI LIXIANG
Duan Sizhe
DUAN JIANWEN
ZHANG YIJIE
DU JUNJIE
LIU YANG
Ran Fengqi
FAN JIA
YAN WEIKANG

Assignees

山西万鼎空间数字股份有限公司
北京泰岳领航科技有限公司

Dates

Publication Date: 20260505
Application Date: 20250814

Claims (8)

1. The laser radar live-action based three-dimensional achievement product display method is characterized by comprising the following steps of: Scanning a target product by adopting three-dimensional scanning equipment to obtain product live-action three-dimensional data; performing multi-step denoising, layering collaborative hole filling and optimization on the three-dimensional data to obtain preprocessed three-dimensional data; The method for carrying out layered collaborative hole filling on the three-dimensional data comprises the following steps: Calculating the average distance of a k neighborhood point set of each denoising three-dimensional data point, wherein k represents the neighborhood point number, if the average distance of an ith point is greater than twice the global average distance, marking the ith point as a low-density point, identifying a continuous low-density area as a hole through connected area analysis, and dividing the hole according to voxel resolution; respectively carrying out layered collaborative hole filling on the divided holes: For boundary points of small holes, carrying out dynamic voxel division on a hole area by adopting a preset rule, filling missing data in a mode of weighted average based on adjacent point coordinates and normal vectors, constructing a local implicit function by combining voxel side lengths and hole coordinates, calculating SDF values, traversing each voxel in the small holes, calculating SDF values of 8 corresponding vertexes, determining intersection points of an isosurface and voxel sides according to a predefined basic configuration, generating continuous triangle grids by a lookup table corresponding to the basic configuration, and carrying out Laplace smoothing on the triangle grids; For the middle hole, the boundary point characteristic of the middle hole in the coding is position embedding, curvature coding is introduced, and attention weight is obtained through calculation; For a big hole, constructing GAN, taking coordinates of a hole boundary, gaussian curvature and normal vector as input of a generator to obtain filling data points; In the junction areas of different holes, adopting a distance weighting fusion mode to fusion and optimize hole supplementing data; carrying out cooperative optimization on hole filling results by combining with the surface textures of the product; constructing a three-dimensional model of the target product by using the preprocessed three-dimensional data through three-dimensional modeling software; creating a virtual display environment in a virtual reality engine, and importing a three-dimensional model into the virtual display environment; and designing an interaction function for the virtual reality scene, and receiving user feedback data in real time to optimize the three-dimensional model of the target product.
2. The method for displaying a three-dimensional achievement product based on laser radar reality according to claim 1, wherein the method for identifying the continuous low-density region as a hole through the connected region analysis comprises: dividing the denoising three-dimensional data space into cube voxels with fixed side length v; Counting the number of denoising three-dimensional data points contained in each voxel; the method comprises the steps of marking voxels with density lower than a preset density threshold as low-density candidate voxels, calculating average distances of S neighborhood points of each voxel, and dynamically adjusting the low-density threshold through an amplification factor according to the average distances of the S neighborhood points; traversing all voxels, and taking unlabeled voxels as seed points; Expanding to all connected neighborhood by taking any seed point as a starting point, and counting low-density voxels belonging to the same connected region; Recording a voxel set and a voxel volume of each connected region; And extracting the surface grids of all low-density candidate voxels from the reserved connected areas as hole boundaries to obtain holes.
3. The laser radar realistic three-dimensional achievement product display method based on claim 1, wherein the gaussian curvature obtaining method comprises: Obtaining J neighborhood points of a boundary point, calculating the mass center of the neighborhood point, translating the boundary point according to the difference between the coordinates and the mass center, extracting principal components of the translated boundary point to obtain a normal vector and two tangential directions of a tangential plane, establishing a local coordinate system according to the normal vector and the two tangential directions, projecting the neighborhood point onto the tangential plane formed by the two tangential directions to obtain projection coordinates and a distance from the tangential plane; And obtaining a second derivative matrix of the square curved surface according to the secondary function coefficient, solving two eigenvalues of the second derivative matrix, and calculating the product of the two eigenvalues to obtain the Gaussian curvature of the boundary point.
4. The lidar-based realistic three-dimensional achievement product display method of claim 1, wherein the GAN training method comprises: Taking coordinates, gaussian curvature and normal vector of the hole boundary as input of a generator, obtaining filling data points as generator generated data, receiving the generated data of the generator by a discriminator, and judging the probability that the generated data belongs to real data; randomly extracting real data of a batch from a preset training set, taking the extracted real data and generated data as input of a discriminator, and outputting corresponding discrimination probability by the discriminator; and (3) designing the optimization loss of the GAN, including constraint set loss and countermeasures loss, updating the network parameters of the GAN, stopping until the GAN reaches the preset optimal performance condition, and taking the GAN corresponding to the preset optimal performance condition as the GAN which is finally output.
5. The laser radar realistic three-dimensional achievement product display method based on claim 1, wherein the method for collaborative optimization of hole filling results in combination with product surface textures comprises: calculating the average distance of Q neighborhood points for each point, and deleting outliers with the distance exceeding the sum of the global mean value and 2 times of standard deviation; in the hole supplementing region and the hole radius of 2 times of the periphery, taking the average space of three-dimensional data in the hole radius as a reference, and supplementing missing points through uniform resampling; E neighborhood points of the hole filling point are obtained, and the hole filling point is corrected based on the distance weighted average result of the coordinates of the E neighborhood points, wherein the weighted weight is obtained by calculating the distance between the neighborhood points and the hole filling point; Acquiring RGB color values of Y neighborhood points of each three-dimensional data point, respectively taking a median value according to a channel, and replacing the current point color; The method comprises the steps of obtaining the colors of H nearest neighbor points of hole filling points of a colored hole filling region, carrying out weighted average on the hole filling points of the colored hole filling region, generating the color with the highest occurrence frequency of the hole filling points of the non-colored hole filling points according to the color distribution of the neighbor points, and enabling the color of the hole filling points to meet preset color constraint.
6. The laser radar realistic three-dimensional effort product display method of claim 1, wherein the method of multi-step denoising three-dimensional data comprises: s1, acquiring N neighborhood points of each data point by calculating Euclidean distance, judging the data point as an outlier and deleting if the distance from the data point to any neighborhood point exceeds a preset neighborhood distance threshold value, traversing and filtering all outliers to obtain filtered three-dimensional data; S2, selecting M neighborhood points with the nearest Euclidean distance from the data points in the three-dimensional data after filtering, and sorting the data points according to the distance from small to large; S3, carrying out Gaussian filtering on the coarse denoising data, and carrying out weighted average on the neighborhood points through a Gaussian kernel function to obtain fine denoising data; S4, analyzing the principal component of the covariance matrix of each fine denoising data point through k neighborhood points, taking a feature vector corresponding to the minimum feature value as a normal vector, projecting the neighborhood points to a tangent plane of the current point, calculating the gravity centers of the neighborhood points in the tangent plane, calculating smooth displacement only in the normal direction by combining a smoothing factor, and updating the fine denoising data points based on smooth traversal to obtain denoising three-dimensional data.
7. The method for displaying a three-dimensional achievement product based on laser radar live-action according to claim 1, wherein in the method for denoising three-dimensional data in multiple steps, after each step of S1-S3, point cloud feature points of a current step are extracted, the number of feature points of the current step and feature point data before filtering are counted, a retention rate is calculated, and if the retention rate is lower than a preset retention threshold, error compensation is triggered: calculating and obtaining covariance matrixes of characteristic point displacement before and after filtering; adjusting the next stage of filtering parameters according to the retention rate and the covariance matrix and a preset dynamic parameter adjustment rule; The method for optimizing the three-dimensional data comprises the following steps: Constructing a weight function according to the Gaussian curvature of the three-dimensional data point, and applying different smooth intensities in the normal direction and the tangential direction; Calculating a feature vector of each point in the three-dimensional data, carrying out principal component analysis on the feature vector, reducing the dimension to a three-dimensional feature space, classifying the feature vector after the dimension reduction by combining FCM clustering optimized by Lagrangian multiplier with a preset cluster high threshold value and a preset cluster low threshold value, wherein the classification result comprises a plane point, a transition point and a feature point; obtaining the real three-dimensional product the method for data comprises the following steps: obtaining preliminary data of a product through scanning of a sensor, and selecting a corresponding parameter combination according to a pre-established knowledge base; Collecting a scanning parameter range of R groups of scanning equipment; Scanning the sensor to obtain product preliminary data and the scanning parameter range of the R group scanning equipment as the input of a parameter adjustment model to obtain an optimal parameter combination; And carrying out adjustment detection on the preliminary data of the product, and carrying out compensation detection on the region meeting the preset adjustment requirement and combining with a preset adjustment strategy, wherein the compensation detection comprises reflection precompensation and shielding precompensation.
8. A laser-radar-based real-scene three-dimensional achievements product display system for implementing the laser-radar-based real-scene three-dimensional achievements product display method according to any one of claims 1 to 7, comprising: The real-scene scanning module is used for scanning a target product by adopting three-dimensional scanning equipment to obtain real-scene three-dimensional data of the product; The preprocessing module is used for carrying out multi-step denoising, layered collaborative hole filling and optimization on the three-dimensional data to obtain preprocessed three-dimensional data; The model building module is used for building a three-dimensional model of the target product through three-dimensional modeling software by utilizing the preprocessed three-dimensional data; The virtual display module is used for creating a virtual display environment in the virtual reality engine and importing the three-dimensional model into the virtual display environment; And the man-machine interaction module is used for designing an interaction function for the virtual reality scene and receiving user feedback data in real time so as to optimize the three-dimensional model of the target product.

Description

Laser radar-based live-action three-dimensional achievement product display method and system Technical Field The invention relates to the technical field of three-dimensional models, in particular to a laser radar live-action-based three-dimensional achievement product display method and system. Background With the development of technology, the product display modes are increasingly abundant, but the traditional display modes have certain limitations, such as two-dimensional pictures, videos and the like, which cannot comprehensively and three-dimensionally display the product. The realistic three-dimensional result has higher fidelity and stereoscopic impression, but the application in the field of product display is not wide. The Chinese patent application with the publication number of CN118840492A discloses an intelligent VR three-dimensional digital product display system and method, which comprises the steps of obtaining a product picture and a point cloud data set of a target product, extracting a product corner point of the target product according to the product picture, extracting a product contour of the target product according to the product corner point, matching point cloud coordinate points corresponding to the product contour in the point cloud data set according to the product contour, carrying out point cloud registration on the point cloud coordinate points to obtain initial alignment point cloud coordinate points, carrying out color matching on the initial alignment point cloud coordinate points to obtain matching point cloud coordinate points, carrying out point cloud filtering on the matching point cloud coordinate points to obtain target point cloud coordinate points, calculating a curved surface center point according to the target point cloud coordinate points, generating a curved surface supporting domain according to the curved surface center point, fitting a three-dimensional curved surface of the target product according to the curved surface supporting domain, and carrying out three-dimensional modeling on the three-dimensional curved surface to obtain a three-dimensional model of the target product. The method comprises the steps of extracting a product contour according to product corner points, determining point cloud coordinate points corresponding to the product contour in a point cloud data set, carrying out point cloud registration on the point cloud coordinate points to obtain initial registration point cloud coordinate points, effectively reducing complexity of the point cloud coordinate points, removing noise of point cloud data, carrying out color matching and point cloud filtering on the initial registration point cloud coordinate points, enabling real features in the point cloud data to be more obvious, obtaining high-quality target point cloud coordinate points, and improving the accuracy of subsequent three-dimensional model construction. Although the method can meet most of the scenes, research and practical application of the method and the prior art find that the method and the prior art have at least the following partial defects: Depending on the image extraction contour, the three-dimensional scanning equipment is not combined, so that the data integrity is insufficient, and the point cloud registration depends on static parameters, so that complex geometric structures or reflective surfaces are difficult to deal with, and registration accumulated errors are generated. In view of the above, the present invention provides a method and a system for displaying a three-dimensional achievement product based on laser radar reality to solve the above problems. Disclosure of Invention In order to overcome the defects in the prior art and achieve the purposes, the invention provides the technical scheme that the laser radar-based realistic three-dimensional achievement product display method comprises the following steps: Scanning a target product by adopting three-dimensional scanning equipment to obtain product live-action three-dimensional data; performing multi-step denoising, layering collaborative hole filling and optimization on the three-dimensional data to obtain preprocessed three-dimensional data; constructing a three-dimensional model of the target product by using the preprocessed three-dimensional data through three-dimensional modeling software; creating a virtual display environment in a virtual reality engine, and importing a three-dimensional model into the virtual display environment; and designing an interaction function for the virtual reality scene, and receiving user feedback data in real time to optimize the three-dimensional model of the target product. Further, the method for carrying out layered collaborative hole filling on the three-dimensional data comprises the following steps: Calculating the average distance of k neighborhood point sets of each denoising three-dimensional data point, wherein k represents the neighborhood point num