A multi-scale object three-dimensional reconstruction method, device and equipment

Abstract

The application provides a multi-scale object three-dimensional reconstruction method, device and equipment, and relates to the technical field of three-dimensional reconstruction. The method comprises the following steps: acquiring original point cloud data of a target object; constructing a three-dimensional space voxel of the target object according to the original point cloud data and a preset scanning resolution; determining the neighborhood voxel of each voxel in the three-dimensional space voxel; determining a feature region and a non-feature region from the three-dimensional space voxel according to the neighborhood voxel; and performing three-dimensional reconstruction by using different resolutions according to the feature region and the non-feature region to obtain a three-dimensional model of the target object. The application guarantees high fidelity of the model of the three-dimensional reconstruction of the object and improves the reconstruction speed.

Description

Technical Field

[0001] This application relates to the field of three-dimensional reconstruction technology, and more specifically, to a method, apparatus, and device for multi-scale three-dimensional reconstruction of objects. Background Technology

[0002] Laser 3D scanning, as an active, non-contact 3D measurement technology, is widely used in industrial manufacturing and precision inspection, cultural heritage and building digitization, medical and dental modeling, and other fields. The accuracy and efficiency of 3D reconstruction results are highly dependent on the resolution of the spatial voxels. High resolution can accurately restore the geometric details of an object's surface, but it significantly increases data storage requirements and computation time; while low resolution can improve reconstruction speed and save memory resources, it is prone to causing surface feature distortion, affecting subsequent analysis and application results. Currently, intelligent resolution 3D reconstruction methods mainly divide feature regions based on the curvature of the object's surface. Within the voxel reconstruction framework, combined with an octree data structure, higher resolution voxels are assigned to regions with higher curvature, while lower resolution is maintained for regions with lower curvature, thus achieving adaptive configuration of spatial resolution. Curvature calculation is mostly based on point clouds or voxel normal vectors in the local neighborhood. The degree of surface curvature is characterized by fitting the surface or statistical normal changes, and this is used as a criterion for region classification.

[0003] However, existing curvature calculation techniques are highly sensitive to neighborhood scale. Within a small neighborhood, it is difficult to distinguish between gently curving surfaces such as spheres and cylinders and planes, leading to insufficient feature region identification. A large number of surface details that should be finely modeled are classified as smooth regions, resulting in the loss of detail in the reconstruction results. Furthermore, noise interference in the normal vector estimation amplifies the curvature response, causing some noisy flat areas to be incorrectly identified as feature regions, leading to unnecessary high-resolution subdivision, which not only wastes storage space but also prolongs the overall reconstruction time. Summary of the Invention

[0004] The purpose of this application is to address the shortcomings of the prior art by providing a multi-scale object 3D reconstruction method, apparatus, and device to ensure high fidelity of the object 3D reconstruction model while improving the reconstruction speed.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, one embodiment of this application provides a multi-scale three-dimensional reconstruction method for objects, the method comprising: Obtain the raw point cloud data of the target object; Based on the original point cloud data and the preset scanning resolution, a three-dimensional spatial voxel of the target object is constructed; Determine the neighborhood voxels of each voxel in the three-dimensional space; Based on the neighboring voxels, characteristic regions and non-characteristic regions are determined from the three-dimensional spatial voxels; Based on the feature regions and the non-feature regions, three-dimensional reconstruction is performed using different resolutions to obtain a three-dimensional model of the target object.

[0006] Optionally, constructing the three-dimensional spatial voxels of the target object based on the original point cloud data and a preset scanning resolution includes: According to the preset scanning resolution, the original point cloud data is divided to obtain a three-dimensional spatial voxel composed of multiple unit voxels, and the size of each unit voxel is the voxel size corresponding to the preset scanning resolution.

[0007] Optionally, determining the neighborhood voxels of each voxel in the three-dimensional space includes: Based on the distance of each voxel to the object surface and a preset distance threshold, invalid voxels and valid voxels in each voxel are determined, and invalid voxels are removed. Determine the neighboring voxels of each effective voxel.

[0008] Optionally, determining the invalid and valid voxels among the voxels based on the distance from each voxel to the object surface and a preset distance threshold includes: If the distance from the voxel to the object surface is less than the preset distance threshold, the voxel is determined to be a valid voxel; If the distance from the voxel to the object surface is greater than or equal to the preset distance threshold, the voxel is determined to be an invalid voxel.

[0009] Optionally, determining the feature region and non-feature region from the three-dimensional spatial voxels based on the neighboring voxels includes: Calculate the tensor voting feature of each voxel based on all its neighboring voxels; Based on the tensor voting features, obtain multiple feature vectors of the voxel, and the feature values ​​of each feature vector; The surface feature intensity of the voxel is obtained based on the feature values ​​of each feature vector; The feature region and the non-feature region are determined from the three-dimensional voxel based on the surface feature intensity of the voxel.

[0010] Optionally, before determining the feature region and the non-feature region from the three-dimensional voxel based on the surface feature intensity of the voxel, the method further includes: The surface feature intensity of the voxel is filtered and denoised based on the surface feature intensity of the neighboring voxels.

[0011] Optionally, calculating the tensor voting feature of each voxel based on all its neighboring voxels includes: Based on the normal vectors of each neighboring voxel, obtain the normal vector tensor encoding result of the neighboring voxel; The voting weight of a neighboring voxel to a voxel is determined based on the distance between each neighboring voxel and the voxel, and the angle between the normal vectors of the neighboring voxel and the voxel. The tensor voting feature of the voxel is calculated based on the tensor encoding results of the normal vectors of all the neighboring voxels and the voting weight of each neighboring voxel to the voxel.

[0012] Optionally, determining the feature region and the non-feature region from the three-dimensional voxel based on the surface feature intensity of the voxel includes: If the surface feature intensity of the voxel is less than a preset intensity threshold, then the voxel is determined to be the initial feature voxel in the feature region; If the surface feature intensity of the voxel is greater than or equal to the preset intensity threshold, then the voxel is determined to be a non-feature voxel within the feature region; The initial feature voxel is divided into sub-voxels to obtain multiple leaf voxels corresponding to the initial feature voxel; Determine whether each of the described leaf voxels is a characteristic voxel of a leaf; If all the leaf voxels corresponding to the initial feature voxel are leaf feature voxels, then the voxel is determined to be a feature voxel in the feature region.

[0013] Secondly, another embodiment of this application provides a three-dimensional reconstruction apparatus for an object, the apparatus comprising: The acquisition module is used to acquire the raw point cloud data of the target object; The construction module is used to construct the three-dimensional spatial voxels of the target object based on the original point cloud data and the preset scanning resolution. The first determining module is used to determine the neighboring voxels of each voxel in the three-dimensional space voxel; The second determining module is used to determine feature regions and non-feature regions from the three-dimensional spatial voxels based on the neighboring voxels; The reconstruction module is used to perform three-dimensional reconstruction at different resolutions based on the feature regions and the non-feature regions to obtain a three-dimensional model of the target object.

[0014] Thirdly, another embodiment of this application provides a computer device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the multi-scale object three-dimensional reconstruction method as described in any of the first aspects above.

[0015] Fourthly, another embodiment of this application provides a storage medium storing a computer program, which, when executed by a processor, performs the steps of the multi-scale object three-dimensional reconstruction method as described in any of the first aspects above.

[0016] The beneficial effects of this application are: This application provides a multi-scale object 3D reconstruction method, apparatus, and device. The method involves acquiring the original point cloud data of the target object; constructing 3D spatial voxels of the target object based on the original point cloud data and a preset scanning resolution; determining the neighboring voxels of each voxel in the 3D spatial voxels; identifying feature regions and non-feature regions from the 3D spatial voxels based on the neighboring voxels; and performing 3D reconstruction using different resolutions based on the feature regions and non-feature regions to obtain a 3D model of the target object. In this embodiment, the space is rapidly structured using voxels at a preset scanning resolution, reducing the complexity of initial data processing. Subsequently, by analyzing the geometric relationships of the voxel neighborhoods, non-feature regions and feature regions are distinguished. Only feature regions rich in detail are subjected to voxel subdivision and high-resolution reconstruction, while flat regions retain coarse voxel representations. This avoids the computational cost of high-precision reconstruction across the entire scene and ensures that key features such as edges and corners are not lost or blurred. Ultimately, this improves reconstruction speed while maintaining high model fidelity. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a multi-scale three-dimensional object reconstruction method provided in this application embodiment; Figure 2 This is a schematic diagram illustrating the process of determining neighborhood voxels in a multi-scale object 3D reconstruction method provided in this application embodiment; Figure 3 A flowchart illustrating the process of determining feature regions in a multi-scale three-dimensional reconstruction method for objects provided in this application embodiment; Figure 4 A flowchart illustrating the determination of tensor voting features in a multi-scale object 3D reconstruction method provided in this application embodiment; Figure 5 A flowchart illustrating the process of determining feature regions in another multi-scale object 3D reconstruction method provided in this application embodiment; Figure 6 A schematic diagram of the structure of a three-dimensional reconstruction device for an object provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0020] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0022] To clearly describe the multi-scale 3D object reconstruction method provided in this application, the method will be explained below with reference to several accompanying drawings. The multi-scale 3D object reconstruction method in this application is applied to laser 3D scanning. Laser scanning obtains corresponding original point cloud data. Based on the original point cloud data, the multi-scale 3D object reconstruction method of this application performs 3D reconstruction of the object at different resolutions, achieving multi-scale 3D object reconstruction and obtaining a 3D model of the target object. The target object in this application can be any object, such as industrial parts, ceramics, buildings, bridges, etc., and this application does not impose any limitations on this.

[0023] Figure 1 A flowchart illustrating a multi-scale three-dimensional object reconstruction method provided in this application embodiment is shown below. Figure 1 As shown, the method includes: Step 101: Obtain the original point cloud data of the target object.

[0024] The original point cloud data is a set of three-dimensional coordinate points output by the laser scanner. Each three-dimensional coordinate point includes the horizontal position (X), vertical position (Y), and vertical position (Z) of the point in space.

[0025] Optionally, a laser scanner can be used to scan the target object to obtain the raw point cloud data of the target object.

[0026] Step 102: Construct the three-dimensional spatial voxels of the target object based on the original point cloud data and the preset scanning resolution.

[0027] The preset scanning resolution is the voxel side length of each unit voxel in the three-dimensional space. A three-dimensional space voxel is a set of voxels of a target object composed of multiple unit voxels. A unit voxel can be a cubic grid, or any cube that can cover the target object, such as a cube, cuboid, or polygon. This application embodiment does not impose any restrictions on this.

[0028] Optionally, the spatial range of the three-dimensional space voxels of the target object is determined based on the original point cloud data. Multiple unit voxels of the same size are constructed by using a preset scanning resolution. The spatial range of the three-dimensional space voxels is filled by the unit voxels. The coordinates of each point in all the original point cloud data are traversed, and the coordinates of each point are determined to the corresponding unit voxel. The original point cloud data is divided into the corresponding unit voxels, thereby obtaining the three-dimensional space voxels of the target object.

[0029] Step 103: Determine the neighborhood voxels of each voxel in the three-dimensional space.

[0030] Among them, the neighboring voxel is the other voxels surrounding the voxel in the three-dimensional space.

[0031] Optionally, each voxel in the three-dimensional space is taken as the central voxel, and the neighboring voxels of the central voxel are determined by a preset distance.

[0032] Step 104: Based on the neighboring voxels, determine the feature regions and non-feature regions from the three-dimensional space voxels.

[0033] Feature areas are the parts of an object's surface that change dramatically, such as edges, sharp corners, the edges of holes, and areas with large curves. Non-feature areas are the flat, smooth parts of the object's surface, such as walls and tabletops.

[0034] Optionally, the degree of surface variation of each voxel is determined based on the neighboring voxels, and characteristic regions and non-characteristic regions are determined from the three-dimensional voxels based on the degree of surface variation of each voxel.

[0035] Step 105: Based on the feature regions and non-feature regions, perform 3D reconstruction using different resolutions to obtain a 3D model of the target object.

[0036] The reconstruction resolution of non-feature regions is the preset scanning resolution, and the reconstruction resolution of non-feature regions is greater than the preset scanning resolution.

[0037] Optionally, non-feature regions are reconstructed according to a preset scanning resolution, while feature regions are reconstructed according to a resolution greater than the preset scanning resolution, thereby obtaining a 3D model of the target object. This ensures that the reconstruction resolution of the feature regions is greater than that of the non-feature regions, thus achieving multi-scale object reconstruction.

[0038] In this embodiment, the original point cloud data of the target object is acquired; based on the original point cloud data and a preset scanning resolution, a three-dimensional spatial voxel of the target object is constructed; the neighboring voxels of each voxel in the three-dimensional spatial voxel are determined; based on the neighboring voxels, feature regions and non-feature regions are determined from the three-dimensional spatial voxels; based on the feature regions and non-feature regions, three-dimensional reconstruction is performed using different resolutions to obtain a three-dimensional model of the target object. In this embodiment, the space is quickly structured using voxels at a preset scanning resolution, reducing the complexity of initial data processing; subsequently, by analyzing the geometric relationships of the voxel neighborhood, non-feature regions and feature regions are distinguished, and only feature regions with rich details are subjected to voxel subdivision and high-resolution reconstruction, while flat regions retain coarse voxel representations, achieving multi-scale object reconstruction. This application avoids the computational cost of full-scene high-precision reconstruction, while ensuring that key features such as edges and corners are not lost or blurred, ultimately improving reconstruction speed while maintaining high model fidelity.

[0039] Based on the above embodiments, this application also provides a process for determining three-dimensional spatial voxels in a multi-scale object three-dimensional reconstruction method. In step 102 above, the three-dimensional spatial voxels of the target object are constructed based on the original point cloud data and a preset scanning resolution, including: Based on the preset scanning resolution, the original point cloud data is divided to obtain a three-dimensional spatial voxel composed of multiple unit voxels.

[0040] The size of each voxel is the same as the voxel size corresponding to the preset scan resolution.

[0041] Optionally, based on a preset scanning resolution and using the Truncated Signed Distance Function (TSDF) algorithm, the original point cloud data is divided into three-dimensional spatial voxels composed of multiple voxels, where the size of each voxel is the same as the voxel size corresponding to the preset scanning resolution. Specifically, the spatial volume of the three-dimensional spatial voxels is determined based on the original point cloud data. Then, based on the spatial volume of the three-dimensional spatial voxels and the volume of each voxel, the spatial volume of the three-dimensional spatial voxels is divided to determine the number of voxels. Finally, the original point cloud data is filled into the corresponding voxels to obtain three-dimensional spatial voxels composed of multiple voxels.

[0042] In this embodiment, the number of voxel grids is determined only by the preset resolution and spatial range, and does not grow infinitely with the original point cloud density, thus avoiding excessive storage and saving computing resources.

[0043] Based on the above embodiments, this application also provides a process for determining neighborhood voxels in a multi-scale object 3D reconstruction method. Figure 2 This is a schematic diagram illustrating the process of determining neighborhood voxels in a multi-scale object 3D reconstruction method provided in an embodiment of this application, as shown below. Figure 2 As shown, determining the neighborhood voxels of each voxel in the three-dimensional space in step 103 above includes: Step 201: Based on the distance of each voxel to the object surface and the preset distance threshold, determine the invalid voxels and valid voxels in each voxel, and remove the invalid voxels.

[0044] The distance from each voxel to the object surface is the vector distance from the voxel's center point to the actual object surface. When the distance from a voxel to the object surface is 0, the voxel is on the object surface; when the distance is greater than 0, the voxel is outside the object; and when the distance is less than 0, the voxel is inside the object. The preset distance threshold is determined based on the preset scan resolution, which is [presumably a value]. The preset distance threshold is Effective voxels are those closer to the surface of an object, while ineffective voxels are those farther from the surface.

[0045] Optionally, based on the distance of each voxel to the object surface and a preset distance threshold, voxels that are closer to the object surface are determined to be valid voxels, and voxels that are farther from the object surface are determined to be invalid voxels.

[0046] Step 202: Determine the neighboring voxels of each effective voxel.

[0047] Optionally, the neighboring voxels of each effective voxel are determined based on each effective voxel and the preset scanning resolution. Specifically, in the x, y, and z directions, the neighboring voxels are determined based on the effective voxels. Voxels with a maximum distance of no more than 2L are considered neighboring voxels of the effective voxel. ,in, The coordinates of the central voxel. The coordinates of the neighboring voxels.

[0048] In this embodiment, invalid voxels far from the surface are completely removed by using distance values ​​and preset thresholds, leaving only valid voxels close to the object surface. This eliminates noise interference and computational redundancy in open areas. By associating discrete voxels into structured blocks through neighborhood voxels, the purity of the input data for subsequent feature analysis is ensured. Furthermore, the algorithm efficiency is improved by pre-computing the neighborhood of the regular grid.

[0049] Based on the above embodiments, this application also provides a process for determining voxel categories in a multi-scale object 3D reconstruction method. In step 201 above, invalid and valid voxels are determined based on the distance of each voxel to the object surface and a preset distance threshold, including: If the distance from a voxel to the object's surface is less than a preset distance threshold, the voxel is determined to be a valid voxel.

[0050] Optionally, if the distance D from the voxel to the object surface is less than the preset distance threshold Dt, it means that the voxel is close to the object surface, and the voxel is determined to be a valid voxel.

[0051] If the distance from a voxel to the object's surface is greater than or equal to a preset distance threshold, the voxel is determined to be an invalid voxel.

[0052] Optionally, if the distance D from the voxel to the object surface is greater than the preset distance threshold Dt, it means that the voxel is far from the object surface, and the voxel is determined to be noise, that is, an invalid voxel.

[0053] In this embodiment, only effective voxels closely attached to the object's surface are retained, while redundant voxels in open areas far from the surface and deeply buried within are eliminated. This eliminates noise interference and significantly reduces the amount of data required for subsequent calculations. This ensures the purity of the data used for feature analysis while allowing computational resources to be fully concentrated on key areas that truly contain surface information.

[0054] Based on the above embodiments, this application also provides a flowchart for determining feature regions in a multi-scale object 3D reconstruction method. Figure 3 This is a flowchart illustrating the process of determining feature regions in a multi-scale object 3D reconstruction method provided in an embodiment of this application, as shown below. Figure 3 As shown, in step 104 above, feature regions and non-feature regions are determined from three-dimensional voxels based on neighboring voxels, including: Step 301: Calculate the tensor voting features of each voxel based on all its neighboring voxels.

[0055] Among them, the tensor voting feature is the geometric shape feature of the neighborhood voxels of each voxel.

[0056] Optionally, the geometric shape features of each voxel can be calculated based on the normal vectors in all neighboring voxels, which are the tensor voting features of each voxel.

[0057] Step 302: Based on the tensor voting features, obtain multiple feature vectors of the voxel and the feature values ​​of each feature vector.

[0058] Among them, multiple feature vectors are the feature vectors of each voxel in the horizontal, vertical and vertical directions, and the feature values ​​of each feature vector are the length or importance in the direction of each feature vector.

[0059] Optionally, tensor voting features for voxels The decomposition yields three eigenvectors. The three eigenvalues ​​corresponding to the three eigenvectors , , .in and That is to say... Three eigenvalues , , The magnitude of reflects the strength of the normal vectors of each voxel in the three eigenvector directions after eigenvalue decomposition of the neighboring voxels.

[0060] Step 303: Obtain the surface feature intensity of the voxel based on the eigenvalues ​​of each eigenvector.

[0061] Optionally, the surface feature intensity of the voxel is obtained based on the eigenvalues ​​of each feature vector and a preset feature intensity formula. The preset feature intensity formula is: .

[0062] Optionally, when Approximately 1, and When the value is approximately 0, the voxel lies on a plane, and the normal vector of the voxel in the neighborhood has only one main orientation. The intensity in the other two directions is mainly caused by noise, and the surface feature intensity of the voxel... Very large.

[0063] Optionally, when and They are of similar size, and When the value is approximately 0, the voxel is located in feature regions such as chamfers and cylindrical surfaces. The voxel normal vector in the neighborhood has two main orientations, and the intensity in the remaining directions is mainly caused by noise. The surface feature intensity of the voxel... Smaller.

[0064] Optionally, when , and When the sizes are similar, voxels are located in feature regions with rich surface details, such as corners. The normal vectors of neighboring voxels have three main orientations, and the surface feature intensity of the voxels... Smaller.

[0065] Step 304: Determine the characteristic regions and non-characteristic regions from the three-dimensional voxels based on the surface feature intensity of the voxels.

[0066] Optionally, if the surface feature intensity of a voxel is less than a preset feature intensity threshold, the voxel is determined to be a feature region; if the surface feature intensity of a voxel is greater than or equal to the preset feature intensity threshold, the voxel is determined to be a non-feature region. Feature regions and non-feature regions are determined from the voxels in three-dimensional space based on the voxels corresponding to the feature regions and the voxels corresponding to the non-feature regions. The preset feature intensity threshold can be 500.

[0067] In this embodiment, the tensor voting mechanism is used to fuse the normal vector information of all voxels in the neighborhood, which enhances the noise resistance and robustness of feature discrimination. By performing eigenvalue decomposition on the voting results, the shape of each voxel can be clearly distinguished. The surface feature intensity provides a reliable decision basis for adaptive resolution reconstruction, ensuring that key details such as edges and corners are accurately captured, while flat areas are not misjudged as features, thus balancing reconstruction accuracy and computational efficiency.

[0068] Based on the above embodiments, this application also provides a process for filtering surface feature intensity in a multi-scale object 3D reconstruction method. Before determining the feature region and non-feature region from the 3D space voxels based on the surface feature intensity of the voxels in step 304 above, the method further includes: The surface feature intensity of the voxels is filtered and denoised based on the surface feature intensity of the neighboring voxels.

[0069] Optionally, the surface feature intensity of neighboring voxels is determined, and the surface intensity features of each voxel are filtered based on each voxel and its corresponding neighboring voxels to obtain the filtered surface intensity features of each voxel. Here, each voxel and its neighboring voxels are considered valid voxels.

[0070] Specifically, based on the surface feature intensity of each voxel and its corresponding neighborhood voxel, and the preset filtering formula... The surface intensity characteristics of each voxel after filtering were obtained. .in, Let the surface feature intensity be that of neighboring voxel j. Let be the filtering weight for neighboring voxels j, representing the curvature difference between neighboring voxels and the central voxel. , The surface intensity characteristics of the central voxel i. Let i be the set of neighborhood voxels of the central voxel i.

[0071] In this embodiment, the feature intensity information of the neighboring voxels is used to correct the current voxel, which can effectively suppress the jitter of feature intensity values ​​caused by scanning noise or local anomalies, making the feature intensity on the same smooth surface more uniform and continuous. At the same time, the principle of neighborhood similarity is used to keep the feature mutations at the edges and corners from being blurred, thereby improving the signal-to-noise ratio of the feature intensity field.

[0072] Based on the above embodiments, this application also provides a process for determining tensor voting features in a multi-scale object 3D reconstruction method. Figure 4 The flowchart illustrating the determination of tensor voting features in a multi-scale object 3D reconstruction method provided in this application is shown below. Figure 4 As shown, in step 301 above, calculating the tensor voting feature of each voxel based on all its neighboring voxels includes: Step 401: Obtain the normal vector tensor encoding result of the neighborhood voxels based on the normal vectors of each neighborhood voxel.

[0073] In this context, the normal vector of each neighboring voxel is used to represent the orientation of the neighboring voxel. The normal vector of each neighboring voxel is used to determine the centroid of the original point cloud data within each voxel. The covariance matrix of the original point cloud data is constructed based on the centroid of the original point cloud data within each voxel. The eigenvalues ​​and eigenvectors of this matrix are determined through the covariance matrix of the original power data. The eigenvector corresponding to the smallest eigenvalue is used as the normal vector.

[0074] Optionally, the normal vector tensor encoding result of the neighboring voxels is obtained based on the normal vectors of each neighboring voxel and the transpose of the normal vectors of the neighboring voxels. Specifically, the normal vector tensor encoding result is... , Let be the normal vector of the j-th neighborhood voxel. It is the transpose of the normal vector of the j-th neighborhood voxel.

[0075] Step 402: Determine the voting weight of the neighboring voxel to the voxel based on the distance between each neighboring voxel and the angle between the normal vectors of the neighboring voxel and the voxel.

[0076] The distance between each neighboring voxel is the spatial distance from the center point of the neighboring voxel to the center point of the voxel. The angle between the normal vectors of the neighboring voxels and the central voxel is the angle between the normal vector of the neighboring voxel and the normal vector of the central voxel, reflecting the orientation difference between the neighboring voxels and the central voxel. The voting weight of a neighboring voxel on a voxel represents the importance of the neighboring voxel to the voxel; the larger the weight, the more important the neighboring voxel.

[0077] Optionally, the voting weight of a neighboring voxel is determined based on the distance between each neighboring voxel and the voxel itself, as well as the angle between the normal vectors of the neighboring voxels and the voxel. Specifically, ,in, Let be the distance from neighboring voxel j to the central voxel i. The preset scan resolution, Let be the normal vector of the j-th neighborhood voxel. Let be the normal vector of voxel i. Let be the cosine of the angle between the normal vector of the j-th neighboring voxel and the normal vector of voxel i. It can be set to When neighboring voxels With voxels Angle greater than When the attenuation is severe, its weight is lower; when the included angle is less than... At that time, the weight decay was not significant.

[0078] Step 403: Calculate the tensor voting features of the voxels based on the tensor encoding results of the normal vectors of all neighboring voxels and the voting weights of each neighboring voxel to the voxel.

[0079] Optionally, the tensor voting features of the voxels are calculated based on the tensor encoding results of the normal vectors of all neighboring voxels and the voting weights of each neighboring voxel to the voxel. .in, Let be the normal vector of the j-th neighborhood voxel. The transpose of the normal vector of the j-th neighboring voxel, and the voting weight of neighboring voxel j for voxel i. , Let i be the set of neighborhood voxels of voxel i.

[0080] In this embodiment, the normal vectors of the neighboring voxels are encoded into tensor matrices, eliminating the ambiguity of orientation. The weights are determined based on the spatial distance between the neighboring voxels and the central voxel and the angle between their normal vectors, thus suppressing interference from directions far away or perpendicular to the center. The tensor voting features of the central voxel are obtained through weighted fusion. These features integrate all effective information within the neighborhood, providing a robust and information-rich data foundation for the subsequent accurate differentiation of geometric attributes such as planes, edges, and corners.

[0081] Based on the above embodiments, this application also provides a process for determining non-feature regions in a multi-scale object 3D reconstruction method. In step 104 above, determining the feature regions and non-feature regions from the 3D spatial voxels based on the surface feature intensity of the voxels includes: If the surface feature intensity of a voxel is greater than or equal to a preset intensity threshold, then the voxel is determined to be a non-feature voxel within a non-feature region.

[0082] The preset intensity threshold can be 500, and the specific value can be determined according to actual needs.

[0083] Optionally, if the surface feature intensity of a voxel is greater than or equal to a preset intensity threshold, it indicates that the voxel is a planar region, i.e., a non-feature region, and the voxel is a non-feature voxel within the non-feature region.

[0084] Based on the above embodiments, this application also provides a flowchart for determining feature regions in another multi-scale object 3D reconstruction method. Figure 5 A flowchart illustrating the process of determining feature regions in another multi-scale object 3D reconstruction method provided in this application embodiment is shown below. Figure 5 As shown, in step 104 above, determining the feature region and the non-feature region from the three-dimensional voxel based on the surface feature intensity of the voxel includes: Step 501: If the surface feature intensity of the voxel is less than the preset intensity threshold, then the voxel is determined to be the initial feature voxel in the feature region.

[0085] The initial feature voxel is the voxel that initially passes the screening, meaning that the voxel may be a feature voxel.

[0086] Optionally, if the surface feature intensity of a voxel is less than a preset intensity threshold, the surface of the voxel is determined to be relatively complex, and the voxel is determined to be the initial feature voxel in the feature region.

[0087] Step 502: Divide the initial feature voxel into sub-voxels to obtain multiple leaf voxels corresponding to the initial feature voxel.

[0088] Optionally, based on the octree voxel data storage method, all feature voxels at the original scale are treated as parent nodes, and they are split into segments with side lengths of... The leaf voxels with the preset scanning resolution are split into 8 leaf voxels for each parent node, resulting in multiple leaf voxels corresponding to the initial feature voxels.

[0089] Step 503: Determine whether each leaf voxel is a characteristic voxel of a leaf.

[0090] Optionally, based on the original point cloud information inside each leaf voxel, the normal vector of the leaf voxel and the distance from the leaf voxel to the object surface are recalculated. With weight The effective leaf voxels are determined from the leaf voxels based on a preset distance threshold. When the side length of a leaf voxel is... The preset distance threshold for the corresponding leaf voxel can be .

[0091] Optionally, based on all neighboring effective leaf voxels of each effective leaf voxel, the tensor voting features of the effective leaf voxels are calculated. Based on the tensor voting features, multiple feature vectors of the effective leaf voxels are obtained, as well as the feature values ​​of each feature vector. Based on the feature values ​​of each feature vector, the surface feature intensity of the effective leaf voxels is obtained.

[0092] Optionally, leaf feature voxels are determined from the effective leaf voxels based on the surface feature intensity of the effective leaf voxels and the feature intensity threshold of the effective leaf voxels. If the surface feature intensity of an effective leaf voxel is less than the feature intensity threshold of the effective leaf voxel, then the effective leaf voxel is a leaf feature voxel. The feature intensity threshold of the effective leaf voxel is less than a preset intensity threshold corresponding to the voxel.

[0093] Step 504: If all leaf voxels corresponding to the initial feature voxel are leaf feature voxels, then the voxel is determined to be a feature voxel in the feature region.

[0094] Optionally, if all leaf voxels corresponding to the initial feature voxel are leaf feature voxels, then the voxel is determined to be a feature voxel in the feature region. If any of the leaf voxels corresponding to the initial feature voxel are not leaf feature voxels, then the initial feature voxel is determined to be a non-feature voxel.

[0095] Optionally, if the voxel is a feature voxel in the feature region and the 3D reconstruction of the object requires higher resolution, the leaf feature voxel can be further subdivided and judged to extract a higher level of effective spatial voxel. Through the moving cube algorithm, a 3D mesh data containing multiple resolutions can be constructed, wherein the feature region is generated by high-resolution effective spatial voxels.

[0096] In this embodiment, a preliminary screening is performed using a relatively low preset intensity threshold to ensure that no potential feature voxels are missed. Subsequently, the initial feature voxels obtained from the preliminary screening are divided into sub-voxels. The feature attributes of each leaf voxel are recalculated and determined at a higher resolution. Only when all leaf voxels are confirmed as features are the original voxels finally identified as feature voxels. This prevents feature omissions and, through high-resolution verification, eliminates interference from noise and transition regions, improving the accuracy and reliability of feature region identification.

[0097] Based on the same inventive concept, this application also provides a three-dimensional reconstruction device for an object corresponding to the multi-scale three-dimensional reconstruction method. Since the principle of the device in this application is similar to the multi-scale three-dimensional reconstruction method of the object described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0098] Figure 6 This is a schematic diagram of the structure of a three-dimensional reconstruction device for an object provided in an embodiment of this application, as shown below. Figure 6 As shown, the device includes: an acquisition module 601, a construction module 602, a first determination module 603, a second determination module 604, and a reconstruction module 605; wherein, the acquisition module 601 is used to acquire the original point cloud data of the target object; The construction module 602 is used to construct the three-dimensional spatial voxels of the target object based on the original point cloud data and the preset scanning resolution. The first determining module 603 is used to determine the neighborhood voxels of each voxel in the three-dimensional space; The second determining module 604 is used to determine feature regions and non-feature regions from three-dimensional space voxels based on neighboring voxels; The reconstruction module 605 is used to perform three-dimensional reconstruction based on feature regions and non-feature regions at different resolutions to obtain a three-dimensional model of the target object.

[0099] In one possible implementation, the construction module 602 is specifically used to: divide the original point cloud data according to a preset scanning resolution to obtain a three-dimensional spatial voxel composed of multiple unit voxels, wherein the size of each unit voxel is the voxel size corresponding to the preset scanning resolution.

[0100] In one possible implementation, the first determining module 603 is specifically used to: determine invalid voxels and valid voxels in each voxel according to the distance of each voxel to the object surface and a preset distance threshold, and remove invalid voxels. Determine the neighboring voxels of each effective voxel.

[0101] In one possible implementation, the first determining module 603 is specifically used to: determine the voxel as a valid voxel if the distance from the voxel to the object surface is less than a preset distance threshold; If the distance from a voxel to the object's surface is greater than or equal to a preset distance threshold, the voxel is determined to be an invalid voxel.

[0102] In one possible implementation, the second determining module 604 is specifically used to: calculate the tensor voting features of the voxel based on all neighboring voxels of each voxel. Based on the tensor voting feature, obtain multiple feature vectors of the voxel, as well as the feature values ​​of each feature vector; The surface feature intensity of the voxel is obtained based on the eigenvalues ​​of each eigenvector; Based on the surface feature intensity of voxels, characteristic and non-characteristic regions are determined from voxels in three-dimensional space.

[0103] In one possible implementation, the second determining module 604 is further configured to: filter and denoise the surface feature intensity of the voxel based on the surface feature intensity of the neighboring voxels.

[0104] In one possible implementation, the second determining module 604 is specifically used to: obtain the normal vector tensor encoding result of the neighborhood voxels based on the normal vectors of each neighborhood voxel; The voting weight of a neighboring voxel is determined based on the distance between each neighboring voxel and the angle between the normal vectors of the neighboring voxels and the voxel. The tensor voting features of the voxels are calculated based on the tensor encoding results of the normal vectors of all neighboring voxels and the voting weights of each neighboring voxel to the voxel.

[0105] In one possible implementation, the second determining module 604 is specifically used to: if the surface feature intensity of the voxel is less than a preset intensity threshold, then determine the voxel as the initial feature voxel in the feature region; If the surface feature intensity of a voxel is greater than or equal to a preset intensity threshold, then the voxel is determined to be a non-feature voxel within a non-feature region. The initial feature voxel is divided into sub-voxels to obtain multiple leaf voxels corresponding to the initial feature voxel; Determine whether each leaf voxel is a characteristic voxel of a leaf; If all leaf voxels corresponding to the initial feature voxel are leaf feature voxels, then the voxel is determined to be a feature voxel in the feature region.

[0106] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0107] This application also provides a computer device. Figure 7 This application provides a schematic diagram of the structure of a computer device, as shown in the embodiment of the present application. Figure 7 As shown, the system includes a processor 701 and a memory 702, and optionally, a bus 703. The memory 702 stores machine-readable instructions executable by the processor 701. When the computer device is running, the processor 701 and the memory 702 communicate via the bus 703. When the machine-readable instructions are executed by the processor 701, the steps of the multi-scale object 3D reconstruction method described above are performed.

[0108] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the multi-scale object 3D reconstruction method described above.

[0109] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0110] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0111] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multi-scale method for three-dimensional reconstruction of objects, characterized in that, The method includes: Obtain the raw point cloud data of the target object; Based on the original point cloud data and the preset scanning resolution, a three-dimensional spatial voxel of the target object is constructed; Determine the neighborhood voxels of each voxel in the three-dimensional space; Based on the neighboring voxels, characteristic regions and non-characteristic regions are determined from the three-dimensional spatial voxels; Based on the feature regions and the non-feature regions, three-dimensional reconstruction is performed using different resolutions to obtain a three-dimensional model of the target object.

2. The method according to claim 1, characterized in that, The step of constructing a three-dimensional spatial voxel of the target object based on the original point cloud data and a preset scanning resolution includes: According to the preset scanning resolution, the original point cloud data is divided to obtain a three-dimensional spatial voxel composed of multiple unit voxels, and the size of each unit voxel is the voxel size corresponding to the preset scanning resolution.

3. The method according to claim 1, characterized in that, Determining the neighborhood voxels of each voxel in the three-dimensional space includes: Based on the distance of each voxel to the object surface and a preset distance threshold, invalid voxels and valid voxels in each voxel are determined, and invalid voxels are removed. Determine the neighboring voxels of each effective voxel.

4. The method according to claim 3, characterized in that, The step of determining the invalid and valid voxels among the voxels based on the distance of each voxel to the object surface and a preset distance threshold includes: If the distance from the voxel to the object surface is less than the preset distance threshold, the voxel is determined to be a valid voxel; If the distance from the voxel to the object surface is greater than or equal to the preset distance threshold, the voxel is determined to be an invalid voxel.

5. The method according to claim 1, characterized in that, The step of determining feature regions and non-feature regions from the three-dimensional spatial voxels based on the neighboring voxels includes: Calculate the tensor voting feature of each voxel based on all its neighboring voxels; Based on the tensor voting features, obtain multiple feature vectors of the voxel, and the feature values ​​of each feature vector; The surface feature intensity of the voxel is obtained based on the feature values ​​of each feature vector; The feature region and the non-feature region are determined from the three-dimensional voxel based on the surface feature intensity of the voxel.

6. The method according to claim 5, characterized in that, Before determining the feature region and the non-feature region from the three-dimensional voxel based on the surface feature intensity of the voxel, the method further includes: The surface feature intensity of the voxel is filtered and denoised based on the surface feature intensity of the neighboring voxels.

7. The method according to claim 5, characterized in that, The step of calculating the tensor voting feature of each voxel based on all its neighboring voxels includes: Based on the normal vectors of each neighboring voxel, obtain the normal vector tensor encoding result of the neighboring voxel; The voting weight of a neighboring voxel to a voxel is determined based on the distance between each neighboring voxel and the voxel, and the angle between the normal vectors of the neighboring voxel and the voxel. The tensor voting feature of the voxel is calculated based on the tensor encoding results of the normal vectors of all the neighboring voxels and the voting weight of each neighboring voxel to the voxel.

8. The method according to claim 5, characterized in that, The step of determining the feature region and the non-feature region from the three-dimensional voxel based on the surface feature intensity of the voxel includes: If the surface feature intensity of the voxel is less than a preset intensity threshold, then the voxel is determined to be the initial feature voxel in the feature region; If the surface feature intensity of the voxel is greater than or equal to the preset intensity threshold, then the voxel is determined to be a non-feature voxel within the non-feature region; The initial feature voxel is divided into sub-voxels to obtain multiple leaf voxels corresponding to the initial feature voxel; Determine whether each of the described leaf voxels is a characteristic voxel of a leaf; If all the leaf voxels corresponding to the initial feature voxel are leaf feature voxels, then the voxel is determined to be a feature voxel in the feature region.

9. A three-dimensional reconstruction device for an object, characterized in that, The device includes: The acquisition module is used to acquire the raw point cloud data of the target object; The construction module is used to construct the three-dimensional spatial voxels of the target object based on the original point cloud data and the preset scanning resolution. The first determining module is used to determine the neighboring voxels of each voxel in the three-dimensional space voxel; The second determining module is used to determine feature regions and non-feature regions from the three-dimensional spatial voxels based on the neighboring voxels; The reconstruction module is used to perform three-dimensional reconstruction at different resolutions based on the feature regions and the non-feature regions to obtain a three-dimensional model of the target object.

10. A computer device, characterized in that, include: The processor and memory, the memory storing machine-readable instructions executable by the processor, which, when the computer device is running, are executed by the processor to perform the steps of the multi-scale object 3D reconstruction method as described in any one of claims 1 to 8.

Images