Three-dimensional human body measurement method and system based on HoloLens

Abstract

The present invention discloses a three-dimensional human body measurement method based on HoloLens, comprising the following steps: Step 1, collecting rough three-dimensional human body mesh data; Step 2, preprocessing the rough three-dimensional human body mesh data; Step 3, performing conversion, approximate voxel sampling, key point extraction and optimization, coarse registration, fine registration, and fusion to obtain a complete human body point cloud; Step 4, implementing measurement, completing interception, projection, extraction of convex hull vertices, and summation of Euclidean distances; Step 5, designing a three-dimensional human body measurement system to implement the various functions of Steps 1 to 4. The present invention also discloses a three-dimensional human body measurement device based on HoloLens, which is used to implement the above-mentioned three-dimensional human body measurement method based on HoloLens. The method and device of the present invention reduce the measurement cost and acquisition difficulty of the three-dimensional dimensions of the human body and significantly improve the measurement accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional human body measurement, and relates to a three-dimensional human body measurement method based on HoloLens. The present invention also relates to a three-dimensional human body measurement system based on HoloLens. Background Art

[0002] With the development and introduction of three-dimensional measurement technology, human body measurement has shifted from traditional contact measurement to non-contact measurement. Non-contact measurement can use computer algorithms to automatically measure the three-dimensional human body data obtained by scanning equipment. It has the characteristics of simple operation, strong scalability, high accuracy and efficiency.

[0003] Efficient and high-precision registration is a critical preprocessing step in 3D anthropometric technology to ensure the integrity of 3D human body data. Given the availability of complete 3D human body data, the accuracy of 3D segmentation technology directly impacts the precise measurement of human dimensions. A complete 3D model can be segmented into its sub-sections. The speed and accuracy of segmentation are crucial for analyzing human body features, making 3D segmentation a crucial step in 3D human body measurement.

[0004] While 3D body measurement technology has gained fundamental applications in various fields, it's still primarily used in academia and industry, far from becoming universally applicable in everyday life. Therefore, there's still significant room for improvement. Exploring the use of HoloLens scan data for 3D body measurement has become a promising research direction. Summary of the Invention

[0005] The purpose of the present invention is to provide a 3D body measurement method based on HoloLens, which solves the problem in the prior art of requiring the construction of an additional environment and difficulty in quickly obtaining 3D data, thereby reducing the cost and difficulty of 3D body measurement.

[0006] Another object of the present invention is to provide a three-dimensional body measurement system based on HoloLens.

[0007] The technical solution adopted by the present invention is a three-dimensional body measurement method based on HoloLens, which is implemented according to the following steps:

[0008] Step 1: Collect rough 3D human body mesh data;

[0009] Step 2: Preprocessing the rough 3D human body mesh data;

[0010] Step 3: Perform conversion, approximate voxel sampling, key point extraction and optimization, coarse registration, fine registration, and fusion to obtain a complete human body point cloud;

[0011] Step 4: Implement measurement, complete interception, projection, extraction of convex hull vertices, and summation of Euclidean distances;

[0012] Step 5: Use the three-dimensional human body measurement system to visualize the functions of steps 1 to 4.

[0013] Another technical solution adopted by the present invention is a three-dimensional body measurement device based on HoloLens, which is used to implement the above-mentioned three-dimensional body measurement method based on HoloLens.

[0014] The HoloLens-based 3D body measurement device integrates the algorithm functions of steps 1 to 4 to develop a 3D body measurement system for HoloLens scan data. The main control process of the 3D body measurement system is described as follows:

[0015] First, use CMake to recompile VTK locally and add Qt support for VTK. Then, copy the "QVTKWidgetPlugin.dll" generated by the VTK compilation to the designer path of the msvc*_64 version of Qt. After completing the above operations, add the QVTKWidget control in the msvc*_64 version of Qt Designer.

[0016] Create a new QtWidgetsApplication project in VS, use Qt Designer to open the UI file in the project to design the interface, select the corresponding control from the component list, manually drag it to the design window on the right, and use a reasonable layout to design the display position and area of ​​the button according to the system module and functional requirements logic;

[0017] The architecture design of the 3D human body measurement system was carried out based on VS2019 and Qt. The algorithm functions of steps 1 to 4 were integrated and divided into a 3D human body mesh data preprocessing module, a point cloud registration module, and a 3D human body segmentation and measurement module. Corresponding system program sub-interfaces were developed for different functional modules, and multiple UI buttons for 3D human body data preprocessing, post-processing, and measurement were preset for user convenience.

[0018] The beneficial effects of the present invention include two aspects:

[0019] 1) The data acquisition method of deploying the MRTK spatial perception system can avoid the problems of poor convenience and unsatisfactory visualization effect in the default HoloLens device spatial mapping mesh data acquisition method; the pre-processing process for denoising and smoothing rough 3D human body mesh data can solve the problems of low resolution, many noise holes, and uneven topology structure of the 3D human body mesh data obtained after the HoloLens deploys the MRTK spatial perception system; the modified RANSAC and Point-to-Plane ICP's point cloud registration method efficiently and accurately registers and fuses the point cloud data converted from 3D human mesh data, which can significantly improve the registration speed while ensuring higher registration accuracy; using the K-Means clustering method, the human point cloud is divided into 8 parts by customizing the initial cluster center, and the measurement of the corresponding parts is completed based on the local human point cloud after clustering segmentation, and the global measurement problem on the overall point cloud model is converted into a local measurement problem, thereby improving the measurement efficiency; based on the implemented 3D data preprocessing, post-processing and human body measurement technology, a 3D human body measurement system based on HoloLens is designed and developed. It preprocesses the 3D human mesh data obtained by HoloLens, registers the point cloud data converted from the 3D human mesh data, performs point cloud segmentation on the complete point cloud after registration and fusion, and automatically measures the local human point cloud after segmentation, meeting the measurement requirements of standard posture human bodies for HoloLens scan data.

[0020] 2) The method of the present invention not only reduces the measurement cost and difficulty of obtaining three-dimensional human body dimensions, but also proposes a preprocessing process, a point cloud registration method, a point cloud clustering and segmentation method, and an automatic measurement method after clustering and segmentation suitable for three-dimensional human body mesh data acquired by HoloLens devices. Their combined use can further improve the efficiency of three-dimensional human body measurement and significantly enhance the measurement accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the overall process of the method of the present invention;

[0022] Figure 2a It is a visualization image of the original human body 3D data scanned by HoloLens. Figure 2b It is to remove outlier noise and visualize the image. Figure 2c is a retopological visualization image, Figure 2d is to smooth the visual image, Figure 2e is the hole completion visualization image, Figure 2f This is a visualization of Poisson disk sampling;

[0023] Figure 3a It is for point cloud data visualization images, Figure 3bis an improved approximate voxel sampling visualization image, Figure 3c It is ISS3D key point extraction and optimization visualization image, Figure 3d Modified RANSAC coarse registration visualization image, Figure 3e Point-to-Plane ICP precise registration visualization image, Figure 3f It is a point cloud fusion visualization image;

[0024] Figure 4a They are the complete human body point cloud visualization image, Figure 4b It is K-Means clustering segmentation visualization image, Figure 4c It is a visualization image of the K-Means custom initial cluster center;

[0025] Figure 5a It is to obtain the local point cloud visualization image of the part to be tested, Figure 5b It is a point cloud visualization image that captures the height of the part to be measured. Figure 5c It is a visual image of the plane projection height point cloud to be measured, Figure 5d The convex hull algorithm is used to extract the convex hull vertices and visualize the image. DETAILED DESCRIPTION

[0026] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Reference Figure 1 The three-dimensional body measurement method of the present invention is implemented according to the following steps:

[0028] Step 1: Collect rough 3D human body mesh data.

[0029] Deploy the MRTK spatial awareness system on the HoloLens device (the MRTK spatial awareness system is also known as the SpatialAwareness System, which is an existing technology).

[0030] First, start and activate the spatial perception system in the MRTK tool on the Unity platform. Configure and compile the relevant properties of the Spatial Mesh Observer in the MRTK spatial perception system. After compiling and generating the project, use Visual Studio to deploy the observer to the HoloLens.

[0031] Then, by putting on the HoloLens device, the wearer checks the reconstruction status of the 3D human body model at the gaze position in real time, obtains the original 3D human body data, and obtains rough 3D human body mesh data.

[0032] Step 2: Preprocess the rough 3D human body mesh data. The specific process is as follows:

[0033] 2.1) Denoising by using the connectivity between triangles and limiting the number of triangles in a triangle cluster;

[0034] 2.2) While ensuring accuracy, a unified local smoothing operator is used to optimize the vertex positions and edge directions of the output rough 3D human body mesh data, re-mesh the surface into an isotropic triangular mesh, and re-divide the mesh topology to achieve a uniform and regular mesh topology.

[0035] 2.3) Using the Taubin algorithm, we generalize the classic discrete Fourier analysis to two-dimensional discrete surface signals, i.e., functions defined on the surface of a polyhedron of arbitrary topology. This simplifies the surface smoothing problem into a low-pass filtering problem, and performs smoothing filtering to obtain three-dimensional human body mesh data with smooth surfaces and no significant shrinkage.

[0036] 2.4) For each hole in the 3D human body mesh data obtained in step 2.3), first find all its boundary edges and then sort them. Then, from all the sorted boundary edges, find the two boundary edges with the smallest angles and (add) a third boundary edge to form a new triangle facet. Finally, set a termination condition and continue iterating to complete the completion of all holes.

[0037] Step 3: Perform conversion, approximate voxel sampling, key point extraction and optimization, coarse registration, fine registration, and fusion to obtain a complete human body point cloud.

[0038] Based on the point cloud registration method of Modified RANSAC and Point-to-Plane ICP, the point cloud data converted from the 3D human body mesh data is registered and fused. The specific process is as follows:

[0039] 3.1) Convert the pre-processed 3D human body mesh data acquired by the HoloLens device into a point cloud format using Poisson disk sampling, defining the source point cloud as P1 and the target point cloud as P2;

[0040] 3.2) Sampling the improved approximate voxels of the source point cloud P1, using the hash function to quickly approximate the center of gravity of the non-empty voxels to construct a new point set P a {p a ,a=1,2,...};Use Kd-tree algorithm to traverse the point cloud P a , find the point cloud P a The nearest neighboring points of each point in form a new point set P b {p b ,b=1,2,...};for the downsampled point cloud P b Extract ISS3D key points to form a new point set P c {pc ,c=1,2,...}; then use the direction vector threshold to optimize the key points to form a new point set P d {p d ,d=1,2,...};

[0041] Similarly, the target point cloud P2 is subjected to improved approximate voxel sampling and ISS3D key point extraction and optimization to form a new point set P q {p q ,q=1,2,...};

[0042] 3.3) According to the point set P of the source point cloud d n (n≥3) random sampling points p i ,i∈{1,...,n} and its point set P in the target point cloud q The corresponding point q in i ,i∈{1,...,n}, establish FPFH feature description, use Modified RANSAC algorithm to solve the approximate transformation matrix, and perform coarse registration;

[0043] 3.4) The point cloud after coarse registration is then fine-registered using the Point-to-Plane ICP algorithm to obtain the optimal transformation matrix M opt (The approximate transformation matrix is ​​the initial inaccurate transformation matrix obtained by coarse registration, and the optimal transformation matrix is ​​the final optimal transformation matrix solved on the basis of the approximate transformation matrix). At this time, the distance between the point in the source point cloud P1 and the corresponding point in the target point cloud P2 is the smallest. The point in the source point cloud is p i =(p ix ,p iy ,p iz ,1) T , the corresponding point in the target point cloud is q i =(q ix ,q iy ,q iz ,1) T , click p i The unit normal vector at is n i =(n ix ,n iy ,n iz ,0) T ;

[0044] Iteratively solve the optimal transformation matrix M opt , perform M on the source point cloud opt Transformation, so that under a specific error metric, the distance from the point in the source point cloud to the plane where the corresponding point in the target point cloud is located is minimized. The expression is:

[0045]

[0046] 3.5) After completing the precise registration, the registered point clouds are fused to obtain a complete human body point cloud.

[0047] Step 4: Implement measurement, complete interception, projection, and Euclidean distance summation.

[0048] Use the K-Means clustering method and customize the initial cluster center to divide the human body point cloud obtained in step 3.5) into eight parts. Then, based on the local human body point cloud segmented by the K-Means clustering method, complete the measurement of the corresponding parts. The specific process is as follows:

[0049] 4.1) Customize the initial cluster centers. Use the "Point list picking" function in CloudCompare software to select the centers of eight segmented parts: head, left arm, right arm, chest, waist, hip, left leg, and right leg. Set the coordinates of the centers of the eight segmented parts as the initial cluster centers for K-Means clustering. Repeat the iterations and use the average value of the cluster sample points generated in each iteration as the cluster center until the clustering results stabilize.

[0050] 4.2) Calculate the subject's height H based on the maximum and minimum z-coordinate values ​​in the human point cloud data; determine the height of the subject's body part based on the linear relationship between the height of each characteristic part and the height H; and determine the maximum perimeter of a horizontal slice within a certain range at the corresponding height position, and intercept the corresponding segmented point cloud;

[0051] Preferably, the point cloud of 0.01m above and below the height of the characteristic part is intercepted, so as to ensure that the data of the subsequent circumference measurement is close to the true value with the point cloud within a sufficient height;

[0052] 4.3) Performing planar projection on the intercepted segmentation point cloud to obtain a planar point set that approximates the human body circumference, and using the RANSAC algorithm to perform plane fitting, projecting the intercepted segmentation point cloud onto the fitting plane;

[0053] 4.4) Use the convex hull algorithm to find the smallest convex set that encloses the feature point cloud, that is, the convex polygon with the smallest area that completely contains a certain point set;

[0054] 4.5) Calculate the Euclidean distance between adjacent vertices of the convex polygon obtained by cutting the contour, and the sum is the measured circumference to generate the convex hull vertex point set (X i ,Y i ), the expression of the sum of the Euclidean distance L is:

[0055]

[0056] Among them, X i , Yi It is the coordinate expression of the vertex point set of the convex hull, i = 1, 2, 3, ..., n represents the point set.

[0057] Step 5: Use the 3D human body measurement system to visualize the functions of steps 1 to 4.

[0058] Integrate the algorithm functions of steps 1 to 4 to design and develop a 3D body measurement system for HoloLens scan data. The 3D body measurement system is a software program. The following three steps describe the main control process (or software architecture) of the 3D body measurement system. The details are as follows:

[0059] 5.1) First, use CMake to recompile VTK locally and add Qt support for VTK. Then, copy the "QVTKWidgetPlugin.dll" generated by VTK compilation (using the Release build as an example) to the designer path of the msvc*_64 version of Qt. After completing the above operations, add the QVTKWidget control in the msvc*_64 version of Qt Designer.

[0060] 5.2) Create a new QtWidgetsApplication project in VS. Use Qt Designer to open the UI file in the project to design the interface. Select the corresponding controls from the component list and manually drag them to the design window on the right. Use a reasonable layout based on the system module and functional requirements to design the display position and area of ​​the buttons.

[0061] 5.3) Design the architecture of the 3D human body measurement system based on VS2019 and Qt (or Qt Designer). Integrate the algorithm functions of steps 1 to 4 and divide them into different functional modules, namely, a 3D human body mesh data preprocessing module, a point cloud registration module, and a 3D human body segmentation and measurement module. Develop corresponding system program sub-interfaces for each functional module, and preset multiple UI buttons for 3D human body data preprocessing, post-processing, and measurement to facilitate user use.

[0062] The method of the present invention can pre-process 3D human body mesh data; simultaneously register and fuse to generate complete 3D human body point cloud data; and can also segment the complete human body point cloud and automatically measure the local human body point cloud after clustering segmentation. When used together, it can effectively complete the basic size measurement of the 3D human body in standard posture, and has strong scalability. The measurement results are automatically displayed in the text display box of the 3D human body measurement system interface.

[0063] The device of the present invention is embodied in step 5, which is equivalent to integrating all the functions of steps 1 to 4 into three interfaces, namely the 3D human body mesh preprocessing interface, the 3D human body point cloud registration interface, and the 3D human body segmentation and anthropometry interface, forming a complete system. The relevant technical processes of each step from step 1 to step 4 correspond to a function button in the three interfaces developed in step 5. The developed functions all support real-time visualization and can be converted into images. Figure 2a 、 Figure 2b 、 Figure 2c 、 Figure 2d 、 Figure 2e 、 Figure 2f They are computer visualization images of 3D human body mesh data preprocessing and format conversion developed by 3D human body mesh preprocessing interface, Figure 3a 、 Figure 3b 、 Figure 3c 、 Figure 3d 、 Figure 3e 、 Figure 3f They are the 3D human point cloud registration interface developed for the registration and fusion of 3D human point cloud data and computer visualization images. Figure 4a 、 Figure 4b 、 Figure 4c They are computer visualization images of 3D human body segmentation and 3D human body point cloud clustering segmentation developed by human body measurement interface, Figure 5a 、 Figure 5b 、 Figure 5c 、 Figure 5d They are computer visualization images of three-dimensional human body segmentation and automatic circumference measurement developed by human body measurement interface.

[0064] Example 1

[0065] The measurement object Human1 of Example 1;

[0066] First, according to the steps of the method of the present invention, the calculated chest, waist, hip and other size data are measured; then, they are compared with the measured true values. For specific comparison results, please refer to Table 1.

[0067] Table 1. Analysis of automatic measurement results of Example 1

[0068]

[0069] The specific implementation of this embodiment 1 is visualized through the following figures:

[0070] The visualization of the steps of the 3D human body mesh data preprocessing process includes six small images, among which: Figure 2a HoloLens scans the original human body 3D data, Figure 2b is to remove outlier noise, Figure 2c It is retopology, Figure 2dIt is smoothing, Figure 2e Hole filling, Figure 2f is Poisson disk sampling.

[0071] The visualization of the point cloud registration step based on Modified RANSAC and Point-to-Plane ICP includes six small images, among which: Figure 3a It is for point cloud data visualization, Figure 3b is an improved approximate voxel sampling, Figure 3c It is ISS3D key point extraction and optimization, Figure 3d Modified RANSAC rough registration, Figure 3e Point-to-Plane ICP precision registration, Figure 3f It is point cloud fusion.

[0072] The 3D human body data display image includes three small images, among which, Figure 4a They are the complete human body point cloud, Figure 4b is K-Means clustering segmentation, Figure 4c This is an image of the K-Means custom initial cluster centers.

[0073] The visualization images of the circumference measurement after point cloud segmentation and cluster segmentation include four small images, among which, Figure 5a Is to obtain the local point cloud of the part to be measured, Figure 5b It is to intercept the point cloud of the height of the part to be measured, Figure 5c It is the plane projection of the height point cloud to be measured, Figure 5d It is the convex hull algorithm that extracts the convex hull vertices.

[0074] Example 2

[0075] The measurement object Human2 of Example 2;

[0076] First, according to the steps of the above-mentioned method of the present invention, the calculated chest circumference, waist circumference, hip circumference and other size data are measured; then compared with the measured true values. For specific comparison results, please refer to Table 2.

[0077] Table 2. Analysis of automatic measurement results of Example 2

[0078]

[0079] The visual images during the steps of Example 2 are omitted here and will not be shown.

[0080] Example 3

[0081] The measurement object Human3 of Example 3;

[0082] First, according to the steps of the method of the present invention, the calculated chest, waist, hip and other dimensions are measured;

[0083] Then compare it with the true value of the measurement. For specific comparison results, please refer to Table 3.

[0084] Table 3. Analysis of automatic measurement results of Example 3

[0085]

[0086] The visual images during the steps of Example 3 are omitted here and will not be shown.

[0087] By comparing the chest circumference, waist circumference, hip circumference and other dimensional data measured by the above three embodiments with the measurement true values, it is obvious that the measurement error percentages of the three circumferences measured by the method of the present invention are all less than 1.4%, which fully demonstrates the feasibility and accuracy of the method of the present invention.

[0088] In summary, the main purpose of the present invention is to improve the efficiency and accuracy of three-dimensional human body measurement. First, the MRTK spatial perception system is deployed on the HoloLens device to complete the acquisition of three-dimensional human body mesh data. Then, the three-dimensional human body mesh data is preprocessed, and the three-dimensional human body mesh data is converted into point cloud data for registration. Then, the complete human body point cloud after registration and fusion is segmented, and the three-dimensional size of the human body is automatically measured based on the segmented local human body point cloud. Finally, the above functions are integrated, Qt development is completed based on VS, and a three-dimensional human body measurement system based on HoloLens is designed and developed. The method of the present invention is fully functional, reduces the measurement cost and acquisition difficulty of human body size, has high measurement efficiency, high measurement accuracy, and high intelligence. Compared with existing methods, the method of the present invention can effectively complete the basic size measurement of three-dimensional human body data in standard posture at a lower cost and faster efficiency. It has strong scalability and provides a new technical means for three-dimensional human body measurement.

Claims

1. A three-dimensional body measurement method based on HoloLens, characterized in that: Follow these steps to implement: Step 1: Collect rough 3D human body mesh data; Step 2: Preprocess the rough 3D human body mesh data. The specific process is: 2.1) Denoising by using the connectivity between triangles and limiting the number of triangles in a triangle cluster; 2.2) Using a unified local smoothing operator, we optimize the vertex positions and edge directions of the output rough 3D human body mesh data, re-mesh the surface into an isotropic triangular mesh, and re-divide the mesh topology to achieve a uniform and regular mesh topology. 2.3) Using the Taubin algorithm, we generalize the classic discrete Fourier analysis to two-dimensional discrete surface signals, i.e., functions defined on the surface of a polyhedron of arbitrary topology. This simplifies the surface smoothing problem into a low-pass filtering problem, and performs smoothing filtering to obtain three-dimensional human body mesh data. 2.4) For each hole in the 3D human body mesh data obtained in step 2.3), first identify all its boundary edges and then sort them. Then, from all the sorted boundary edges, find the two with the smallest angles and add a third boundary edge to form a new triangle. Finally, set a termination condition and continue iterating to complete the completion of all holes. Step 3: Perform conversion, approximate voxel sampling, key point extraction and optimization, coarse registration, fine registration, and fusion to obtain a complete human body point cloud. Based on the point cloud registration method of Modified RANSAC and Point-to-Plane ICP, the point cloud data converted from the 3D human body mesh data is registered and fused. The specific process is as follows: 3.1) The pre-processed 3D human body mesh data acquired by the HoloLens device is converted into a point cloud format through Poisson disk sampling. The source point cloud is defined as , the target point cloud is ; 3.2) Source point cloud Improved approximate voxel sampling, using hash functions to quickly approximate the center of gravity of non-empty voxels to construct a new point set ; Use Kd-tree algorithm to traverse the point cloud , find and point cloud The nearest neighboring points of each point in form a new point set ; For the downsampled point cloud P b Extract ISS3D key points to form a new point set ; Then use the direction vector threshold to optimize the key points to form a new point set ; Similarly, the target point cloud Improved approximate voxel sampling and ISS3D key point extraction and optimization to form a new point set ; 3.3) According to the point set of the source point cloud middle Random sampling points and its point set in the target point cloud The corresponding points in , establish FPFH feature description, use Modified RANSAC algorithm to solve the approximate transformation matrix, and perform coarse registration; 3.4) Use the Point-to-Plane ICP algorithm to perform fine registration on the point cloud after coarse registration to obtain the optimal transformation matrix , at this time, the source point cloud Inlier point to target point cloud The distance between the corresponding points in the plane is the smallest, and the point in the source point cloud is , the corresponding point in the target point cloud is ,point The unit normal vector at is ; Iteratively solve the optimal transformation matrix , the source point cloud Transformation, so that under a specific error metric, the distance from the point in the source point cloud to the plane where the corresponding point in the target point cloud is located is minimized. The expression is: （1） 3.5) After completing the precise registration, the registered point clouds are fused to obtain a complete human body point cloud; Step 4: Perform the measurement, complete the interception, projection, extraction of convex hull vertices, and Euclidean distance summation. Use the K-Means clustering method and customize the initial cluster center to divide the human body point cloud obtained in step 3.5) into eight parts. Then, based on the local human body point cloud segmented by the K-Means clustering method, complete the measurement of the corresponding parts. The specific process is as follows: 4.1) Customize the initial cluster centers. Use the "Point list picking" function in CloudCompare software to select the centers of eight segmented parts: head, left arm, right arm, chest, waist, hip, left leg, and right leg. Set the coordinates of the eight segmented part centers as the initial cluster centers for K-Means clustering. Repeat the iterations, using the average of the cluster sample points generated in each iteration as the cluster center until the clustering results stabilize. 4.2) Calculate the subject's height H based on the maximum and minimum z-coordinate values ​​in the human point cloud data. Determine the height of the target body part based on the linear relationship between the height of each characteristic body part and the height H. The three measurements are the maximum perimeter of a horizontal slice within a certain range at the corresponding height position, and the corresponding segmented point cloud is intercepted. 4.3) Perform a planar projection on the intercepted segmentation point cloud to obtain a planar point set that approximates the human body circumference. Use the RANSAC algorithm to perform plane fitting and project the intercepted segmentation point cloud onto the fitting plane. 4.4) Use the convex hull algorithm to find the smallest convex set that encloses the feature point cloud, that is, the convex polygon with the smallest area that completely contains a certain point set; 4.5) Calculate the Euclidean distance between adjacent vertices of the convex polygon obtained by cutting the contour, and the sum is the measured circumference to generate the convex hull vertex point set ( X i ,Y i ), the sum of the Euclidean distances The expression is: （2） in, X i , Y i is the coordinate expression of the convex hull vertex point set, i =1,2,3,…, n Represents a point set; Step 5: Use the 3D human body measurement system to visualize the functions of steps 1 to 4.

2. The HoloLens-based three-dimensional body measurement method according to claim 1, characterized in that: In step 1, the specific process is: deploy the MRTK spatial perception system on the HoloLens device, First, start and activate the spatial perception system in the MRTK tool on the Unity platform. Configure and compile the relevant properties of the spatial grid observer in the MRTK spatial perception system. After compiling and generating the project, use Visual Studio to deploy the observer to the HoloLens. Then, by putting on the HoloLens device, the wearer checks the reconstruction status of the 3D human body model at the gaze position in real time, obtains the original 3D human body data, and obtains rough 3D human body mesh data.

3. The HoloLens-based three-dimensional body measurement method according to claim 1, characterized in that: In step 5, the specific process is: Integrate the algorithm functions of steps 1 to 4 to design and develop a 3D body measurement system for HoloLens scan data. The following three steps describe the main control process of the 3D body measurement system, as follows: 5.1) First, use CMake to recompile VTK locally and add Qt support for VTK. Then, copy the "QVTKWidgetPlugin.dll" generated by the VTK compilation to the designer path of the msvc*_64 version of Qt. After completing the above operations, add the QVTKWidget control in the msvc*_64 version of Qt Designer. 5.2) Create a new QtWidgetsApplication project in VS. Use Qt Designer to open the UI file in the project to design the interface. Select the corresponding controls from the component list and manually drag them to the design window on the right. Design the display position and location of the buttons according to the logical layout of the system modules and functional requirements. 5.3) Design the architecture of the 3D human body measurement system based on VS2019 and Qt. Integrate the algorithm functions of steps 1–4 and divide them into a 3D human body mesh data preprocessing module, a point cloud registration module, and a 3D human body segmentation and measurement module. Develop corresponding system program sub-interfaces for different functional modules, and preset multiple UI buttons for 3D human body data preprocessing, post-processing, and measurement to facilitate user experience.

4. A HoloLens-based 3D body measurement device, used to implement the HoloLens-based 3D body measurement method according to claim 1, characterized in that: The HoloLens-based 3D body measurement device integrates the algorithm functions of steps 1 to 4 to develop a 3D body measurement system for HoloLens scan data.

5. The HoloLens-based three-dimensional body measurement device according to claim 4, characterized in that: The main control process of the three-dimensional body measurement system is described as follows: First, use CMake to recompile VTK locally and add Qt support for VTK. Then, copy the "QVTKWidgetPlugin.dll" generated by the VTK compilation to the designer path of the msvc*_64 version of Qt. After completing the above operations, add the QVTKWidget control in the msvc*_64 version of Qt Designer. Create a new QtWidgetsApplication project in VS, use Qt Designer to open the UI file in the project to design the interface, select the corresponding control from the component list, manually drag it to the design window on the right, and use a reasonable layout to design the display position and area of ​​the button according to the system module and functional requirements logic; The architecture design of the 3D human body measurement system was carried out based on VS2019 and Qt. The algorithm functions of steps 1 to 4 were integrated and divided into a 3D human body mesh data preprocessing module, a point cloud registration module, and a 3D human body segmentation and measurement module. Corresponding system program sub-interfaces were developed for different functional modules, and multiple UI buttons for 3D human body data preprocessing, post-processing, and measurement were preset for user convenience.

Images